Datasets:

Daniel-P-Gonzalez

/

OCD

like 0

# Introduction Although traditionally thought to be of minor significance in pharmacology and toxicology, the clinical importance of CYP2B6 was recently established with the identification of increasing numbers of CYP2B6 substrates, including clinically important drugs such as the anticancer agents, cyclophosphamide, ifosfamide and tamoxifen, the anti-retrovirals efavirenz and nevirapine, the anesthetics ketamine and propofol, and the central nervous system-active bupropion and methadone. It was estimated that CYP2B6 is involved in the metabolism of nearly 25% of drugs on the market. Marked intra- and inter-individual variation of CYP2B6 expression and activity have been described in the literature and were considered to be attributed mainly to the highly-inducible and polymorphic nature of this gene. Many drugs and environmental chemicals can alter the expression of CYP2B6, potentially leading to clinically significant drug-drug interactions. Given the growing importance of CYP2B6 in drug metabolism, the need for a better understanding of the molecular mechanisms governing *CYP2B6* gene expression is evident. Our understanding of mechanisms underlying transcriptional regulation of CYP2B6 expression has grown substantially during the past two decades, and it has been well-established that induction of CYP2B6 expression by xenobiotics is mediated primarily by the constitutive androstane receptor (CAR, NR1i3) and the pregnane X receptor (PXR, NR1i2) through interactions with phenobarbital responsive enhancer module (PBREM) and xenobiotic responsive module (XREM) located in the distal region of CYP2B6 promoter. Nevertheless, large interindividual variability in the expression of CYP2B6 cannot be entirely explained by this simplified CAR/PXR-based model. Activation of CAR and PXR is essential but not sufficient for the optimal regulation of *CYP2B6* gene transcription. Ectopic expression of CAR or PXR alone failed to fully restore the basal and inductive expression of CYP2B6 in non-hepatic cells or hepatoma cell lines that express extremely low levels of liver enriched transcription factors (LETFs). Several previous reports including our own data demonstrated that expression of *CYP2B6* gene can be influenced by interactions between CAR/PXR and LETFs such as the hepatic nuclear factor 4α (HNF4α) and CCAAT/enhancer-binding protein α (C/EBPα), suggesting hepatic factors other than CAR and PXR contribute to the large individual variations in CYP2B6 expression. The hepatocyte nuclear factor 3β (HNF3β), also known as forkhead box protein A2 (FOXA2), is a DNA-binding protein that is encoded by the *FOXA2* gene in human, and plays a pivotal role in the regulation of metabolism-related gene expression in the liver and pancreas. As a hepatic transcription factor, HNF3β influences the expression of numerous genes involved in energy metabolism, bile acid homeostasis, drug metabolism and transport by interacting with other LETFs such as HNF1α and HNF4α, and nuclear receptors such as the glucocorticoid receptor and PXR. Disruption of HNF3β binding sites located in the promoter of glucose-6-phosphatase or CYP3A4 repressed the transcriptional activity of each respective gene. Recently, Lamba et al., reported that hepatic expression of CYP3A4 was positively correlated with that of HNF3β. Given the pleotropic roles of HNF3β in hepatic gene regulation and the known transcriptional cross-talk between CYP2B6 and CYP3A4, we hypothesized that HNF3β may play a role in the transcription of *CYP2B6* gene and contribute to the observed large inter- and intra-individual variations in CYP2B6 expression. In the current study, we provide experimental evidence to demonstrate that expression of CYP2B6 is closely associated with the expression of HNF3β in human primary hepatocytes (HPH). Overexpression of HNF3β enhanced expression and promoter activity of CYP2B6, while knockdown of HNF3β significantly repressed CYP2B6 expression in HepG2 cells. Utilizing *in silico* analysis, chromatin immunoprecipitation (ChIP) and surface plasmon resonance (SPR) binding affinity assays, characterization of the CYP2B6 promoter revealed two functional enhancer modules that are responsible for HNF3β-mediated CYP2B6 transcription. # Material and Methods ## Reagents Phenobarbital (PB) and 6-(4-Chlorophenyl) imidazo\[2,1-b\]thiazole-5-carbaldehyde-O-(3,4-dichlorobenzyl) oxime (CITCO) were purchased from Sigma-Aldrich (St. Louis, MO). Oligonucleotide primers were synthesized by Integrated DNA Technologies, Inc. (Coralville, IA). The Dual-Luciferase Reporter Assay System was purchased through Promega (Madison, WI). Antibodies against CYP2B6 and HNF3β were from Santa Cruz (Dallas, TX). β-Actin antibody was from Sigma-Aldrich. Matrigel, insulin, and ITS⁺ (insulin/transferrin/selenium) were obtained from BD Biosciences (Bedford, MA). Other cell culture reagents were purchased from Life Technologies (Grand Island, NY) or Sigma-Aldrich. ## Plasmid Construction Luciferase reporter plasmids containing -300bp, -387bp, -600bp, -1kb, -1.4kb and -2kb fragments of the CYP2B6 promoter region were PCR-amplified by using forward primers: `5’- GGGGTACCTAGACATACATATACCCAC-3’`, `5’- GGGGTACCCATACAGGGATGCAAGCAG-3’`, `5’- GGGGTACCGGGATTACAGGTGTGAGC-3’`, `5’- GGGGTACCTCAGCATCTGCAGGCTTC-3’`, `5’- GGGGTACCACACACCTGGAGCTCAAG-3’`, `5’- GGGGTACCGGACAATGTAGCCCCAACCC-3’` and the same reverse primer: `5’- AGTCTACTCGAGCTGCACCCTGCTGCAGCCT-3’`. The PCR products were sub-cloned into the KpnI and XhoI sites of pGL3-basic vector, resulting in constructs termed 2B6-300bp, 2B6-387bp, 2B6-600bp, 2B6-1kb, 2B6-1.4kb and 2B6-2kb, and the correct orientation was verified by sequencing. The pCR3-hCAR, pMEX-C/EBPα, pcDNA3.1/HNF3β expression vectors and the CYP2B6 reporter constructs (2B6-1.6kb and 2B6-1.8kb) were obtained or generated as described previously. pRL-TK was used as an internal control. ## Site-directed Mutagenesis Site-directed mutagenesis was performed by PCR, using the 2B6-2kb construct as the template, based on the protocol of the QuickChange Multi Site-Directed Mutagenesis Kit from Stratagene (Santa Clara, CA). Mutated nucleotides in the HNF3β binding sites are underlined for the HNF3β-a-mut: `CACTAAGAGTGTACCGCCTGAGTTACTGTGTG`, the HNF3β-b-mut: `CCCCTTTACAT GTACCAGTCATATAAGCACATAC`, and the HNF3β-c-mut: `CAAAGCTAAGTACCAGAGTGCA AGCTCACC`. The constructs were sequenced to confirm the presence of the mutation(s). ## HPH Cultures Human hepatocytes were isolated as described previously from human liver specimens obtained from University of Maryland Medical Center with prior approval by the Institutional Review Board at the University of Maryland at Baltimore or obtained from Bioreclamation In Vitro Technologies (Baltimore, MD). Hepatocytes with viability over 90% were seeded at 0.75 × 10⁶ cells/well in 12-well biocoat plates in DMEM supplemented with 5% FBS, 100 U/ml penicillin, 100 μg/ml streptomycin, 4 μg/ml insulin, and 1 μM dexamethasone. After attachment at 37°C in a humidified atmosphere of 5% CO2, hepatocytes were cultured in complete William’s Medium E (WME) and overlaid with Matrigel (0.25mg/ml). Cell culture medium was replaced on a daily basis. ## Real-Time PCR Analysis Total RNA from human primary hepatocytes was isolated using the RNeasy Mini Kit (Qiagen, Valencia, CA) and reverse transcribed using High Capacity cDNA Archive kit (Applied Biosystems, Foster City, CA) following the manufacturers’ instructions. Real-time PCR assays were performed in 96-well optical plates on an ABI StepOnePlus Real-Time PCR system (Applied Biosystems) with SYBR Green PCR Master Mix (Qiagen). Primer sequences for the CYP2B6, CAR, HNF3β, C/EBPα, HNF4α, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) are: CYP2B6, `5’-AGACGCCTTCAATCCTGACC-3’` and `5’-CCTTCACCAAGACAAATCCGC-3’`; CAR, `5’- GAGCTGAGGAACTGTGTGGTA-3’` and `5’-CTTTTGCTGACTGTTCTCCTGAA-3’`; HNF3β, `5’- GGAGCAGCTACTATGCAGAGC-3’` and `5’-CGTGTTCATGCCGTTCATCC-3’`; C/EBPα, `5’-TCGGTGGACAAGAACAGCAA-3’` and `5’-TTTCAGGAGGCACCGGAATCT-3’`; HNF4α, `5’-CGAAGGTCAAGCTATGAGGACA-3’` and `5’- ATCTGCGATGCTGGCAATCT-3’`; and GAPDH,`5’-CCCATCACCATCTTCCAG GAG-3’` and `5’-GTTGTCATGGATGACCTTGGC-3’`. Fold induction values were calculated according to the equation 2^ΔΔCt, where ΔCt represents the differences in cycle threshold numbers between the target gene and GAPDH, and ΔΔCt represents the relative change in these differences between control and treatment groups. ## Transient Transfection in HepG2 Cells HepG2 cells obtained from American Type Culture Collection (Manassas, VA) were transfected with different CYP2B6 reporter constructs in the presence of HNF3β, with or without hCAR and C/EBPα expression vectors using X-tremeGENE 9 DNA Transfection Reagent (Roche Diagnostics Corporation, Indianapolis, IN). Twenty- four hours after transfection, cells were treated with solvent (0.1% DMSO), PB (1mM), or CITCO (1μM) for 24 h. Subsequently, cell lysates were assayed for firefly luciferase activities normalized against the activities of co- transfected Renilla luciferase using Dual-Luciferase Kit (Promega, WI). Data were represented as mean ± S.D. of three individual transfections. ## HNF3β overexpression and knockdown in HepG2 cells Twenty-four hours after seeding, HepG2 cells were infected with negative control adenovirus or increasing concentrations of adenovirus expressing HNF3β purchased from Vigene Biosciences (Rockville, MD) for 48 h. For HNF3β-RNAi lentivirus, DNA oligonucleotides encoding short hairpin HNF3β RNA (`GATCAAGAACATGTCGTCGTACGTGCTCGAGCACGTACGACGACATGTTCTTTTTTTG`) was inserted into the BamHI and EcoRI restriction sites of the pGreenPuro™ shRNA expression lentivector from System Biosciences (Mountain View, CA). Lentivirus was prepared as described previously. In HNF3β knockdown assay, HepG2 cells were infected with the lentiviral particles for 96 h. Total RNA and Proteins were prepared for real-time PCR and western blot analysis. ## Western Blot Analysis Cell homogenate proteins harvested from Ad-HNF3β overexpressing HepG2 cells were resolved on SDS—polyacrylamide gels (12%) and electrophoretically transferred onto blotting membranes. Subsequently, membranes were incubated with antibodies against HNF3β (diluted 1:200), CYP2B6 (diluted 1:200), or β-actin (diluted 1:50,000). Blots were washed and incubated with horseradish peroxidase secondary antibodies, and developed using enhanced chemiluminescence Western blotting detection reagent from GE Healthcare (Pittsburgh, PA). ## CYP2B6 Activity Assay CYP2B6 metabolic activity assay was performed in adenovirus HNF3β infected HepG2 cells on 24-well plate using P450-Glo ™ CYP2B6 Assay kit (Promega, WI) following the manufacturer’s instruction. In brief, 48 h after HNF3β adenovirus infection, culture medium was completely removed and cells were washed twice with Krebs- Henseleit buffer (BioreclamationIVT, MD), followed by incubation at 37°C for 2 h in 300 μL Krebs-Henseleit buffer containing 3 μM of Luciferin-2B6 (CYP2B6 specific substrate) and 3 mM of salicylamide. Equal volume of Luciferin Detection Reagent was added into each well, and mixed at room temperature for 20 min before luminescence detection (Promega, WI). Data were represented as mean ± S.D. of three individual infections. ## Chromatin Immunoprecipitation Assays Experiments were performed using a ChIP assay kit according to the manufacturer’s protocol (Millipore Corporation, Billerica, MA). In brief, 1 × 10⁶ cultured human primary hepatocytes were cross-linked with 1% formaldehyde for 10 min at 37°C, washed with ice-cold phosphate buffered saline containing a protease inhibitor cocktail. Nuclear extracts were sonicated. Immunoprecipitation was performed overnight at 4°C using 5μg of anti-HNF3β antibody (Santa Cruz Biotechnology) or normal rabbit IgG (Cell Signaling Technology) followed by precipitation using protein A coupled to agarose beads. After de-crosslinking and protease digestion, DNA fragments were recovered by QIAquick PCR purification kit (QIAGEN). CYP2B6 promoters containing HNF3β-a (distal), HNF3β-c (proximal), and the region between -1.4/-1.6k were amplified by PCR using primers: `5’-TGGACAATGTAGCCCCAACCC-3’` and `5’-GATTGGGTGCTCATTGCAGCC-3’`; `5’-CTCATACACATGCAAGGATAC-3’` and `5’-GAGCAAGTGAATGTGTGGG TG-3’`; and `5’-GTCAGGCGTAGGATGAGACAG-3’`and `5’-TCTTGAGCTCCAGGTGTGTGC-3’`. A proximal promoter region of *SULT1E1* gene was used as a negative control as reported previously. ## Plasmon Resonance Binding Assay Recombinant HNF3β protein purchased from Abcam (Cambridge, MA) was covalently linked to the surface of a BIAcore CM5 sensor chip (BR-1005-30, Lot#10222950) by direct immobilization as described previously. Oligonucleotides containing HNF3β-a (`AGAGTGTAAAGACTGAG`), HNF3β-b (`TACATGTAAAAATCATA`) and HNF3β-c (`GCTAAGTAAAAAAGTGC`) were used as analytes. The binding assay was performed by injecting 60 μl each of the oligonucleotides at 10 μM in 10 mM Hepes, pH 7.4, containing 150 mM NaCl, 3 mM EDTA, and 0.005% P-20 at the flow rate of 30 μl per minute at 25°C. The association and dissociation between analytes and HNF3β protein were recorded respectively by SPR with a Biacore 3000 (GE Healthcare, Piscataway, NJ) following the manufacturer’s instructions. Sensorgrams of the interaction generated by the instrument were analyzed by the software BIAeval 3.2. ## General data analysis All data represent at least three independent experiments and were expressed as mean ± S.D. Statistical significance was determined using one-way analysis of variance followed by post-hoc Dunnett’s test or Student’s t test where appropriate. Statistical significance was set at *p* \< 0.05 or *p* \< 0.01. Linear regression was analyzed using Pearson’s Correlation Coefficient (JMP 7.0; SAS, NC). # Results ## Correlation of CYP2B6 and HNF3β expression in HPH Correlation between the expression of a number of hepatic transcription factors and CYP2B6 was initially evaluated in a collection of HPH prepared from 35 human liver donors. Basal mRNA expression of these genes in hepatocytes was determined by real-time PCR assays. As expected, positive correlations were observed between the expression of CYP2B6 and that of CAR (R = 0.6844; *p* \< 0.01), C/EBPα (R = 0.5935; *p* \< 0.01), and HNF4α (R = 0.5479; *p* \< 0.01), respectively. Interestingly, the abundance of CYP2B6 in these donors was also significantly correlated with the expression of HNF3β (R = 0.5429; *p* \< 0.01). These results indicate that besides the known role of CAR, HNF4α and C/EBPα in CYP2B6 expression, HNF3β may represent another LETF that contributes to the transcription of *CYP2B6* gene in human liver. ## Ectopic expression of HNF3β alters CYP2B6 expression and activity in HepG2 cells To further investigate the effects of HNF3β on CYP2B6 expression, adenovirus- driven HNF3β was used to ectopically over-express HNF3β in HepG2 cells. As shown in, mRNA expression of HNF3β by virus infection was associated with enhanced expression of CYP2B6 in HepG2 cells in a concentration-dependent manner. A similar pattern of increased CYP2B6 protein content and enzyme activity was also observed. On the other hand, knockdown of HNF3β by lentiviral-shRNA was associated with decreased expression of CYP2B6 in HepG2 cells. Together, these results indicate that the intracellular level of HNF3β influences expression of CYP2B6 in HepG2 cells and support the positive correlation between hepatic expression of HNF3β and CYP2B6 observed in human liver donors. ## HNF3β activates the transcriptional activity of a CYP2B6 reporter gene It is well-known that expression of CYP2B6 is predominantly regulated at the transcriptional level. Thus, we next investigated whether HNF3β could alter the transcriptional activity of CYP2B6 via a putative luciferase construct containing the first 2 kb of the CYP2B6 5’-flanking region in a pGL3-Basic reporter vector. This region contains the PBREM regulatory module that responds to the nuclear receptor CAR. Consistent with the notion that CAR is constitutively activated in immortalized cell lines, transfection of CAR robustly increased the luciferase activity of CYP2B6-2kb, while additional treatment with PB or CITCO only moderately increased the CYP2B6 luciferase activity. Notably, transfection of HNF3β dramatically enhanced the activation of the CYP2B6 reporter independent of chemical treatment. Co-transfection of hCAR/HNF3β/C/EBPα/CYP2B6-2kb resulted in an additive increase of the luciferase activity in HepG2 cells. Likewise, co-transfection of hCAR and/or HNF3β/C/EBPα as aforementioned also increased expression of endogenous CYP2B6 mRNA in HepG2 cells. Together, these findings suggest that the first 2 kb of the CYP2B6 promoter may contain specific sequences that coordinate HNF3β-mediated CYP2B6 transactivation. ## Identification of HNF3β-Response Elements in CYP2B6 promoter To elucidate the molecular basis underlying HNF3β-mediated CYP2B6 transcription, we carried out an *in silico* analysis of the first 2 kb upstream sequence of the CYP2B6 promoter using the MatInspector release professional program. Two clusters of potential HNF3β binding sites were localized at -1887/-1871bp and -434/-350bp regions. As shown in, three predicted HNF3β responsive motifs in the two clusters were designated as HNF3β-a, HNF3β-b, and HNF3β-c. Serial deletion reporter assays revealed that maximal transactivation of CYP2B6 promoter by HNF3β was achieved with the CYP2B6-2kb construct that contained all predicted enhancers. Deletion of HNF3β-a or HNF3β-c significantly repressed the role of HNF3β in CYP2B6 promoter activation, while elimination of HNF3β-b only exhibited negligible consequence. In site-directed mutagenesis experiments, transactivation of CYP2B6-luciferase activity through HNF3β was remarkably attenuated by the mutation of HNF3β-a or HNF3β-c, while only moderately affected by HNF3β-b mutation. Interestingly, although our *in silico* analysis failed to predict a consensus HNF3β binding site between -1.6 kb and -1.4 kb, our luciferase reporter assay showed that deletion of this 200 bp sequence significantly reduced HNF3β-mediated activation of CYP2B6 promoter, indicating the existence of a functional yet unidentified HNF3β binding site in this region. Together, these data suggest that transcriptional activation of CYP2B6 by HNF3β is mediated through multiple enhancer modules including the newly identified HNF3β binding sites, particularly the HNF3β-a and HNF3β-c motifs. ## Interaction between HNF3β and enhancers identified in the CYP2B6 promoter Potential physiological recruitment of HNF3β to the CYP2B6 promoter was assessed using ChIP assays in cultured HPH from two liver donors (HL#98 and \#107). As demonstrated in, endogenous HNF3β protein was efficiently recruited to the distal and proximal promoter regions of CYP2B6 containing HNF3β-a and HNF3β-c, respectively. Lack of binding to the promoter region of SULT1E1 was used as a negative control as reported previously. Notably, our results showed that HNF3β protein was also enriched in the -1.6/-1.4 kb region, further supporting the presence of a functional HNF3β binding site in this region. The binding kinetics between HNF3β and CYP2B6 enhancers were further validated using a SPR binding affinity assay. As shown in, DNA sequences (analytes) containing HNF3β-a or HNF3β-c bind to HNF3β efficiently, while HNF3β-b exhibits undetectable association with HNF3β. Detailed kinetics calculation revealed that HNF3β-a had a higher association rate constant (2.42 × 10⁵ 1/Ms) in comparison to that of HNF3β-c (4.82 × 10³ 1/Ms) though both shared similar dissociation rate constants (2.04 ×10⁻⁶ and 3.54 × 10⁻⁶ 1/s, respectively). In consequence, HNF3β-a exhibited a higher binding affinity (KD = 8.45 ×10⁻¹² M) than HNF3β-c (KD = 7.34 ×10⁻¹⁰ M). Together, these results indicate that HNF3β can efficiently interact with the CYP2B6 promoter with higher binding affinity to the HNF3β-a than the HNF3β-c motifs, while there is no detectable binding to the HNF3β-b site. # Discussion CYP2B6, an inducible cytochrome P450 isoform predominantly present in the liver, exhibits large intra- and inter-individual variations in human populations. Expression of CYP2B6 is controlled by many transcription factors including drug and hormone responsive nuclear receptors such as CAR, PXR, the vitamin D and glucocorticoid receptors, and constitutively activated LETFs including HNF4α and C/EBPα. Our current study reveals that HNF3β represents a novel LETF that can up-regulate *CYP2B6* gene expression by recognizing and interacting with multiple enhancer modules including two consensus HNF3β binding sites located at -1887/-1871bp and -347/-331bp upstream of the CYP2B6 transcriptional start site. Site-directed mutation of these responsive elements resulted in significant reduction of the CYP2B6 promoter activity. Moreover, HNF3β protein was efficiently recruited to the CYP2B6 promoter through direct interaction with these enhancer modules. To date, human CAR and PXR have been recognized as the major transcription factors mediating drug-induced CYP2B6 expression. However, optimal expression of CYP2B6 was only achieved in HPH that maintain physiologically relevant expression of most LETFs. Hepatocyte nuclear factors are a heterogeneous group of evolutionarily conserved transcription factors that are pivotal for the development and maintenance of liver specific features. Positive correlations between expression of CYP2B6 and HNF4α or C/EBPα (also named HNF2) were demonstrated previously in human liver samples. Knockdown of HNF4α expression decreased the mRNA level of CYP2B6, CAR, and PXR in HPH, while ectopic co- expression of HNF4α, CAR, and C/EBPα markedly increased *CYP2B6* gene transcription in HepG2 cells. Using HPH from 35 individuals, we found that expression of CYP2B6 was positively associated with the expression of HNF3β, in addition to its known correlation with CAR, HNF4α and C/EBPα. These initial observations led to a more focused investigation on the role of HNF3β in hepatic CYP2B6 transcription. Regarded as one of the master regulators of hepatocyte differentiation and maturation, HNF3β regulates the expression of numerous hepatic genes by interacting with respective cis-acting binding elements in the promoters of these genes, including the gluconeogenic phosphoenolpyruvate carboxykinase, insulin-like growth factor-binding protein 1, and tyrosine aminotransferase. Unlike ligand-activated transcription factors, presence of the HNF3β protein alone in hepatocytes is sufficient to trigger downstream signaling pathways and the abundance of HNF3β protein often correlates well with the expression of its target genes. Additionally, cooperation between LETFs and ligand-activated nuclear receptors can contribute to tissue-specific increase or decrease in expression of nuclear receptor target genes. For instance, overexpression of HNF4α facilitates CAR- and PXR-mediated induction of CYP2B6, CYP3A4, and CYP2C9, while HNF4α-mediated expression of CYP7A1, a key enzyme in cholesterol metabolism and bile acid synthesis, was down-regulated by competing with CAR for binding to the direct repeat 1 (DR1) motif in the promoter of CYP7A1. We have previously shown that a single nucleotide polymorphism introducing a functional C/EBPα binding site at the proximal region of CYP2B6 promoter synergistically enhanced PXR-mediated induction of this gene. In the current study, we found that overexpression of HNF3β increased the activity of CYP2B6 reporter construct, and the endogenous expression of CYP2B6 mRNA and protein in HepG2 cells. Notably, although overexpression of either CAR or HNF3β increased CYP2B6 promoter activity, their combination only produced additive effects on the transactivation of CYP2B6 in transfected HepG2 cells, suggesting CAR and HNF3β may influence CYP2B6 expression independently rather than collaboratively. Subsequent computer-based analysis revealed the presence of three potential HNF3β binding sites at -1887/-1871bp, -347/-331bp, and -428/-412bp regions of the first 2 kb of the CYP2B6 promoter. Luciferase reporter assays demonstrated that maximal activation of CYP2B6 reporter by HNF3β requires the presence of all three elements. Of importance, serial deletion and mutation assays further revealed the importance of the HNF3β-a and HNF3β-c motifs, while HNF3β-b only exhibited minimal activity. Intriguingly, although no consensus HNF3β binding site(s) was predicted from our *in silico* analysis in the -1.6k to -1.4k region of the CYP2B6 promoter, deletion of this region unexpectedly resulted in a significant reduction of CYP2B6 luciferase activity, indicating the presence of an unidentified HNF3β binding site which warrants more detailed investigation in the future. In characterizing the recruitment of HNF3β to the promoter of CYP2B6 in a more physiologically relevant system, results from our ChIP assays in HPH confirmed the interaction between HNF3β protein and chromatin regions harboring the HNF3β-a and HNF3β-c motifs of *CYP2B6* gene. It is worth noting that HNF3β was also co-precipitated with the DNA fragment covering the -1.6/-1.4bp region, where an unidentified HNF3β binding motif was speculated. However, given the proximity between the -1.6/-1.4bp and the HNF3β-a containing (-1887/-1871bp) areas, such recruitment may only represent a HNF3β-a dependent phenomenon. Quantitative kinetics for binding affinities between HNF3β and the three predicted binding sites were further estimated by a SPR approach. Consistent with results of the luciferase reporter and ChIP assays, our SPR affinity assays revealed that the HNF3β-a and HNF3β-c elements were efficiently recruited to the HNF3β protein, while the HNF3β-b containing region appeared not to be directly associated with HNF3β. In conclusion, our data show that HNF3β represents a novel LETF that regulates the transcription of *CYP2B6* gene, in addition to HNF4α and C/EBPα. Ectopic expression of HNF3β in HepG2 cells was associated with increased CYP2B6 transactivation and endogenous expression. Such enhancement can be at least partly explained by the identification and functional characterization of two consensus HNF3β binding sites located in the 5’-flank of CYP2B6 upstream. These findings reveal additional mechanistic bases in our understanding of *CYP2B6* gene transcription, and suggest that expression of CYP2B6 is governed by a complex regulatory network including genes such as, CAR, PXR, HNF4α, C/EBPα, and HNF3β. It is worth noting that multiple cis-acting responsive elements for C/EBPα and HNF4α have also been localized within the first 2 kb of the CYP2B6 promoter. Future studies to elucidate the mechanisms by which these LETFs work in concert with CAR/PXR to confer the optimal expression of CYP2B6 are warranted. Given that expression of HNF3β can be disturbed by the fluctuation of steroids and hormones, altered expression of HNF3β may contribute to the large interindividual variations in hepatic CYP2B6 expression. The authors thank Dr. Yinghua Zhang (Department of Physiology, University of Maryland School of Medicine, Baltimore, MD) for assistance in the Plasmon Resonance Binding analysis. We are also grateful to members of the Wang laboratory for discussions and comments on the manuscript. [^1]: The commercial affiliation of SH with Bioreclamation IVT does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: HW. Performed the experiments: LL DL SH. Analyzed the data: LL DL. Contributed reagents/materials/analysis tools: SH. Wrote the paper: LL SH HW.

# Introduction Neurons are specialized cells responsible for exchanging information with other neurons or cells through synapses. During development, differentiating neurons explore the surrounding environment in order to form the correct contacts and they use highly motile structures called growth cones (GCs) located at the tip of their neurites. GCs consist of a flat extension, named ‘lamellipodium’ with varying width from which finger-like submicron diameter structures called filopodia emerge. The process of polymerization of actin filaments is the main source of GC protrusion, which is regulated and controlled by several proteins such as Arp2/3, cofilin, formin and molecular motors, such as myosin, dynein, controlling different features of cellular motility. Actin related protein 2/3 complex (Arp2/3) is widely studied for its involvement in lamellipodia formation and protrusion. Arp2/3 consists of seven subunits and promotes the formation of branched actin filament networks. Arp2/3 not only regulates the branching of actin filaments but it is also involved in the formation and dynamics of filopodia. Inhibition of Arp2/3 causes lamellipodia retraction and an increase of the actin retrograde flow rate. Arp2/3 is inactive in its native state and the members of the Wiskott-Aldrich syndrome protein (WASP) family, downstream of Rac and Cdc42 pathways activate the Arp2/3 complex to nucleate new filaments. Rac binds the WAVE (WASP family Verprolin Homology Domain-containing protein) complex to release active WAVE, which promotes actin polymerization through activation of Arp2/3. WASP and WIP (WASP-interacting protein), downstream effectors of Cdc42 interact directly with Arp 2/3 complex to promote filopodia formation. Recently a new protein called Arpin has been shown to be part of the Rac-Arpin-Arp2/3 inhibitory circuit playing a major role in steering during cell migration. Rac can both activate and inhibit Arp2/3-driven actin branching and polymerization to regulate speed, directionality and persistence of membrane protrusions. Rho family GTPase has distinct and specific roles in the regulation of growth, maintenance and retraction of GCs. The mammalian Rho GTPase family currently consists of three subfamilies, Rho (RhoA, RhoB and RhoC), Rac (Rac1, Rac2 and Rac3) and Cdc42 (Cell Division Cycle-42) (Cdc42Hs and G25K). RhoA, Rac1 and Cdc42 are well-studied members of Rho family GTPase controlling distinct cytoskeletal elements. Activation of Rac1 stimulates actin polymerization to form lamellipodia, Cdc42 induces the polymerization of actin to form filopodia or microspikes which are parallel actin bundles within the lamellipodium and Rho regulates the bundling of actin filaments into stress fibers and the formation of focal adhesion complexes. The Rho family of GTP-binding proteins are activated by a variety of growth factors, cytokines, adhesion molecules, hormones, integrins, G-proteins and other biologically active substances. Biochemical approaches or analyses of the morphology of fixed cells have shown that Rho GTPase also involves crosstalk. This may occur through the Rac/Cdc42 effecter PAK, which can negatively regulate Rho GEFs or other mechanisms including, via reactive oxygen species, phosphorylation and competitive binding of RhoGDI or binding of GEFs to actomyosin. Depending upon the concentration and localization of these Rho GTPase, mammalian cells show different morphology, movement and behavior. When the rate of actin polymerization overtakes the actin retrograde flow, the GC protrudes. Retrograde flow refers to the backward flow of the actin filament network away from the growth cone leading edge into the C-domain. This allows the addition of actin monomers/oligomers to actin filaments in close contact with the membrane, pushing the cellular membrane forward, leading to the protrusion. Mitchison and Kirschner proposed the ‘Molecular Clutch Hypothesis’, which postulates that an intracellular molecular clutch, formed by interactions between GC membrane adhesive receptors and the extracellular environment, couple to the overlying flow of actin filaments to slow down their retrograde rate. Formation of these ‘clutches’ together with myosin II contractile activity, provides a traction to pull and move the central region of the GC closer to the peripheral region, leading to axonal lengthening. Therefore, substrate adhesion decreases the actin retrograde flow. The decrease in the actin retrograde flow together with actin polymerization, results in the leading edge protrusion. The traction force is essential for cell migration, cell shape maintenance, mechanical signal generation and other cellular functions. There are different methods to quantify the cellular traction forces. Traction force microscopy measures the stress of a cell on an elastic gel substrate by detecting the movement of fluorescent beads embedded at the surface of the gel. With Optical Tweezers the bead is attached to the cell membrane either to apply the tensile force or to measure the retrograde flow rate. We have independently developed a method to estimate the force exerted by the lamellipodia and filopodia by measuring the displacement of the bead using quadrant photo detector (QPD). In our case, the bead is not initially attached to the cell membrane and it is kept in the vicinity of the lamellipodia or filopodia, so that their spontaneous motion can displace the bead. We observed that the lamellipodia transiently retract and recover back after 5–8 min of Arp2/3 inhibition. In this study we have investigated the role of Rac1, in the recovery of lamellipodia in Arp2/3 depleted condition and also in GC motility, by using Optical Tweezers and specific inhibitors of Arp2/3 (CK-548) and Rac1 (EHop-016 and F56). Motility of lamellipodia and of filopodia was also followed and characterized by video imaging. By combining these techniques together with immunofluorescence we have explored the interaction between Rac1 and Arp2/3 complex and their role in the formation of lamellipodia and filopodia of Dorsal root Ganglion (DRG) GCs. Here we show that Rac1 acts as a switch and activates upon inhibition of Arp2/3. # Results After 6–8 hours of culture, differentiating DRG neurons have neurites emerging from their soma. At the tip of the protruding neurites, GCs lamellipodia and filopodia explore the environment and their motion continues for 1–3 days. The motility of lamellipodia and filopodia slows down when appropriate connections are established and the neuronal network is formed; the leading edge of these lamellipodia can move with a speed 30–100 nm/s exerting a force exceeding 20 pN. The effect of the inhibitors of specific proteins involved in the regulation of GC motility was analyzed after 24–48 hours of culture, when the motility of filopodia and lamellipodia was more pronounced. We focused on the analysis of inhibitors of small GTPases and of the Arp2/3 complex. We used the small molecules CK-636, CK-548, CK-666 and CK-869 as inhibitors of the Arp2/3 complex. All these compounds at a high concentration, i.e. above 100 μM, abolished GC motility completely and in the experiments here described we used extensively CK548 (CK) as the Arp2/3 inhibitor, since CK decreases the affinity of rhodamine-N-WASP-VCA for BtArp2/3 complex approximately twofold. Furthermore, we tested two inhibitors of Rac1 namely, EHop-016 (EH) and F56 and the Cdc42 inhibitor ZCL-278 (ZCL). In addition to these drugs, CT04 (CT) and GSK 269962 (GSK) were also used as inhibitors of the RhoA and Rock pathways respectively. In order to check if the effect of the inhibitors was a side effect of toxicity we also checked their reversibility after washout (WO), as shown in. ## The effect of partial inhibition of Rac1 and Arp2/3 in lamellipodia motility The involvement of Rac1 and Arp2/3 in lamellipodia motility of DRG GCs was studied by analyzing the effect of their inhibitors EHop-016 (EH) and CK-548 (CK) respectively and by quantifying lamellipodia motility using the two algorithms as described in the Materials and Methods section, based on the analysis of Z-stack phase contrast video imaging. From the image sequences, kymographs were obtained by using algorithm I. The ability of lamellipodia to lift up vertically was quantified by computing the fraction of pixels in focus at 5 μm above the coverslip obtained by using algorithm II. When Rac1 activity was inhibited by 20 μM EH lamellipodia still exhibited protrusion retraction cycles (upper panel) and could lift up in the axial direction (lower panel). EH effects were reversible and period, persistence length, retrograde flow rate of lamellipodia returned to control level after washout. Interestingly, lamellipodia of DRG GCs, treated with 50 μM CK showed a transient retraction and were not able to lift up vertically in a significant manner. However, treated lamellipodia recovered their usual motility in 5–8 min (upper panel) and were able to lift up in the axial direction as in control conditions (lower panel). Following 50 μM CK treatment, period, persistence length and retrograde flow rate of lamellipodia were quantified during lamellipodia recovery period. The average period of protrusion/retraction cycles of lamellipodia increased significantly, both in the presence of 20 μM EH (129.6±5.2 s) and of 50 μM CK (115.1±4.2 s) respectively compared to control conditions (86.5±3.2 s) and after washout of 20 μM EH (90.3±4.1 s). The persistence length of lamellipodia i.e the maximum extension reached by the lamellipodia after which they start to retract, increased when Arp 2/3 was inhibited by 50 μM CK (1.90±0.09 μm) compared to control conditions (1.48±0.07 μm) and after washout of 20 μM EH (1.42±0.1). However, there was no significant change in the persistence length of lamellipodia when Rac1 was inhibited (1.56±0.09 μm), but the lamellipodia retrograde flow rate decreased when Rac1 was inhibited (0.05±0.007 μm/s) compared to what observed in control conditions (0.08±0.005 μm/s), after washout of 20 μM EH (0.07±0.008 μm/s) and in the presence of Arp2/3 inhibitors (0.09±0.009 μm/s). ## Rac1 activates when Arp2/3 is inhibited When the activity of Arp2/3 was inhibited by 100 μM of CK lamellipodia shrank and their motility was completely and permanently suppressed. The growth cone also lost the adhesion to the substrate and retracted towards the soma (data not shown). Remarkably, when DRG neurons were treated with 50 μM CK, lamellipodia showed a transient retraction that continued for 5–8 minutes, but then lamellipodia recovered their usual motility restoring protrusion and retraction cycles and were also able to lift up vertically almost as under control conditions. The results of these experiments suggest that following a partial inhibition of Arp2/3 another pathway is activated rescuing—to some extent—the usual GC motility. To test this possibility and to identify the origin of the recovery of motility in treated lamellipodia, we considered the Rho GTPase pathways, known to regulate many aspects of intracellular actin dynamics and GC metabolism. The most extensively studied members of Rho GTPase family are Rho A, Rac1 and Cdc42. Rac can not only regulate actin polymerization but it can also increase the availability of free actin-barbed ends by the removal of capping proteins and it can also increase the availability of actin monomers by regulating cofilin. These roles of Rac1 could help in the formation and protrusion of lamellipodia by polymerizing the pre-existing branched actin filaments at the leading edge of the lamellipodia, in Arp2/3 depleted condition. In addition, the newly formed actin branches generated by the remaining Arp2/3 can contribute to the lamellipodia protrusion. Therefore, we hypothesized that Rac1 could mediate the recovery of motility observed in. Lamellipodia that were first treated with 20 μM EH exhibited an increase in the period of protrusion/retraction cycles and could move up in the axial direction. Then, the same lamellipodia were treated also with 50 μM CK: in this case, as expected, lamellipodia shrank but could not recover their motility even after 10–20 minutes of exposure to these inhibitors. We tested also the simultaneous application of 20 μM EH and of 50 μM CK, which were mixed and added to the medium bathing of the neuronal culture at the same time. Lamellipodia exposed simultaneously to the two inhibitors retracted and did not show any sign of motility even after 10–20 minutes. The above results indicate that Rac1 could be behind the recovery of lamellipodia which were transiently retracted after Arp2/3 inhibition. However, the Rac1 inhibitor EHop-016 inhibits both Rac3 as well as Cdc42 above the concentration of 3μM. Therefore, in order to examine the possible role of the Cdc42 pathway, we used ZCL as a selective inhibitor which is known to target the binding site of the Cdc42 guanine nucleotide exchange factor, intersectin (ITSN) and to hinder Cdc42 activation. When 50μM ZCL was added lamellipodia did not show significant changes in their motility. Subsequent exposure of 50 μM CK to the same lamellipodia shrank the lamellipodia as usual, but then lamellipodia did recover after approximately 8 minutes of exposure. To dispose the possibility of Rac3, in the recovery of lamellipodia, in Arp2/3 inhibited condition, we used F56 as another specific Rac1 inhibitor. It is a control peptide version of Rac1 Inhibitor W56; comprises residues 45–60 of Rac1 with Trp56 replaced by Phe, which does not affect GEF-Rac1 interaction. When lamellipodia were treated with 100μM F56, lamellipodia did not show significant changes in their motility. The same lamellipodia were then exposed to 50 μM CK, the lamellipodia shrank as usual, but then lamellipodia did not recover even after 10–20 minutes of exposure. We tested also the simultaneous application of 100 μM F56 and of 50 μM CK, which were mixed and added to the medium bathing of the neuronal culture at the same time. Lamellipodia exposed simultaneously to the two inhibitors retracted and did not show any sign of motility after 10–20 minutes (data not shown). In the Arp2/3 depleted situation, in order to see the role of RhoA and Rock in the lamellipodia recovery, lamellipodia were exposed to CT (Rho A inhibitor) and GSK (ROCK inhibitor) (data not shown) independently, before the treatment with CK. In both situations lamellipodia recovered after 8 minutes of exposure and, at the end of their retraction, they were also able to reach the same height as in control conditions. We analyzed in detail the growth cone dynamics in the presence of the inhibitors of CDC42, RhoA and ROCK signaling pathways. The period of lamellipodia protrusion/retraction cycles in the presence of ZCL (101.7±2.6 s), CT (113.6±2.9 s) and GSK (107.8±2.9 s) increased when compared with the control conditions (86.5±3.2 s). The lamellipodia persistence length also increased in the presence of CT (1.92±0.13 μm) and GSK (1.8±0.11 μm) but remained constant in the presence of ZCL (1.35±0.07 μm) when compared to control conditions (1.48±0.07 μm). The retrograde flow rate decreased in the presence of ZCL (0.05±0.006 μm/s) but remained constant in CT (0.08±0.01 μm/s) and GSK (0.09±0.01 μm/s) when compared to control conditions (0.08±0.005 μm/s). These results discard the involvement of the Rac3, Cdc42, RhoA and Rock pathways in the recovery of lamellipodia which were transiently retracted after the inhibition of Arp2/3. It also indicates that Rac1 is crucial for the recovery of the transient retraction of lamellipodia due to inhibition of Arp2/3. To confirm the role of Rac1 during the recovery of the transient lamellipodia retraction due to 50 μM CK, we examined the Rac1-GTP level (the activated form of Rac1) in cultured DRG neurons with different exposure conditions of CK by using the G-LISA Rac 1 activation assay kit (Cytoskeleton, Inc., Denver, Colo.). The Rac1-GTP level, in the presence of 25 and 50 μM CK for 2 minutes did not show any significant change compared to the control conditions. However, the Rac1-GTP level significantly increased in the presence of 50 μM CK for 8 minutes (P\<0.005) i.e. the time during which lamellipodia recovered motility after the transient retraction. These results confirm that Rac1 activates upon inhibition of Arp2/3. ## Effect of Arp2/3 and Rac1 inhibitors on the force exerted by lamellipodia Optical Tweezers was used to investigate the effect of the partial inhibition of Rac1 and Arp2/3 on the force exerted by lamellipodia. Lamellipodia in control conditions pushed the trapped beads with a force up to 10–20 pN as previously described and often beads could be displaced out of the optical trap. The forces were measured from the same lamellipodia in control conditions and in the presence of the inhibitors. Exerted forces were analyzed according to four different stereotyped behaviors previously described, depending on the direction in which the lamellipodia were exerting the force on the bead: vertical push (VP), vertical retraction (VR), lateral push (LP) and lateral retraction (LR). Lamellipodia of DRG treated with a moderate concentration of Rac1 and Arp2/3 inhibitors were able to pull and push a trapped bead, but with a lower force compared to the force observed in control conditions. In the lateral direction: in case of LP the lamellipodia force decreased by 30–40% with an increase in the inhibition of Rac1, however it decreased by 50–65% when Arp2/3 was inhibited compared to control conditions. The retractile force LR decreased by 40% when Rac1 was inhibited by 10μM EH, inhibition of Rac1 by 20 μM EH decreased the LR force more than 70% probably due to a decrease in the lamellipodia retrograde flow rate. The retractile force LR decreased by 65% when Arp2/3 was inhibited. In the axial direction: when Rac1 was inhibited by 10 μM EH, the lamellipodia force in VP and VR decreased more than 60%. Besides, it decreased more than 75% in all the other VP and VR cases. These results suggest that lamellipodia were not able to explore the surrounding environment with an equal force when Rac1 and Arp2/3 were inhibited when compared to control conditions. In addition, lamellipodia were not able to exert a larger force in the axial direction than in the lateral direction, when compared with the control conditions state. ## The effect of Rac1 inhibitors on the rate of lamellipodia protrusion Lamellipodia in the presence of 10–20 μM EH exerted a lower force but were still able to extend. In order to measure their rate of protrusion, we used the Nanopositioner feedback which allows a precise and continuous measurement of the bead position by employing Optical Tweezers. All the measurements obtained using the nanopositioner feedback mechanism were compensated as explained in the Materials and Methods section. In each case, the total displacement of the bead in the lateral direction was computed. In each case—control conditions, 10 μM EH and 20 μM EH- 5–6 of such traces were averaged and plotted against time. Before averaging, traces were aligned so that their rising phase matched each other. The slopes of these traces were calculated to determine the lamellipodia protrusion rate. In control conditions, the speed of protrusion of lamellipodia could reach 100 nm/s (see black trace) and was reduced to 30–50 nm/s in the presence of 10 μM EH (blue trace in) and to 10–20 nm/s in the presence of 20 μM EH (magenta trace). These results indicate that inhibition of Rac1 has a similar effect on the amplitude of the force exerted by lamellipodia and on their protrusion rate. ## Effect of Arp2/3 and Rac1 inhibitors on the force exerted by filopodia and their motility The filopodia motility and the force exerted by them were quantified by video imaging, immunocytochemistry and Optical Tweezers. The protruding filopodia tips were followed in different frames to calculate the filopodia protrusion rate, and the maximum length of the filopodia was measured as described in the Materials and Methods section. In DRG GC the length of the filopodia increased by 60 to 80% when Arp2/3 was inhibited by 25 and 50 μM CK respectively. When Rac1 was inhibited by 10 μM EH the length of the filopodia increased by 20%. Remarkably, the filopodia length increased more than the double when Rac1 was inhibited by 20 μM EH compared to control conditions. The GCs were then fixed and stained with Alexa 488 phalloidin and imaged to observe the actin localization. The longer filopodia protruded from the GCs after the inhibition of Rac1 with 20 μM EH and showed an increase in the total F-actin compared to the controlled filopodia. The protrusion rate of filopodia did not change when Rac1 and Arp2/3 were suppressed by their respective inhibitors with a lower concentration. However, it increased by 30% when Rac1 was inhibited by 20 μM EH. In this case, the extension of the filopodia length could be the effect of this increase in the filopodia protrusion rate together with the decrease of the lamellipodia retrograde flow rate. Surprisingly, the filopodia protrusion rate decreased by 30% when Arp2/3 was inhibited by 50 μM CK. Inhibition of Rac1 and Arp2/3 significantly decreased the force exerted by lamellipodia; however, the force exerted by filopodia did not change when Rac1 was inhibited and, with a lower concentration of its inhibitor, Arp2/3 was suppressed, compared to control conditions. Very rarely filopodia emerged from lamellipodia exerted a force that is larger than 4 pN in control conditions. The forces exerted by filopodia were measured in the same neuron before and after the addition of inhibitors of Rac1 or Arp2/3. In each case collected data from 10 neurons showed that the filopodia force did not changed when Rac1 was inhibited by 10–20 μM EH and when the Arp2/3 was inhibited by 25 μM CK. Inhibition of Arp2/3 with 50 μM CK decreased the filopodia force by 20% when compared to control conditions. # Discussion In this study we have characterized the role of Rac1 and Arp2/3 in the motility and force exerted by lamellipodia and filopodia of DRG GCs. Our results suggest that in neuronal growth cones, Rac1 acts as a switch that activates following the inhibition of Arp2/3. Moreover, Arp2/3 and Rac1 not only control the force exerted by lamellipodia but also the dynamics of filopodia. ## The effect of the inhibition of Rac1 and Arp2/3 on lamellipodia motility We followed and quantified the protrusion/retraction cycles of DRG lamellipodia by measuring their period, persistence length and retrograde flow rate using kymographs (see in the section). Lamellipodia treated with a small amount of Rac1 and Arp2/3 inhibitors increased the period of their protrusion/retraction cycles. When Rac1 was inhibited, the lamellipodia retrograde flow rate decreased, leading to a longer retraction time and overall cycle period. These effects were then returned to the control level after the washout of the Rac1 inhibitor, 20 μM EH indicating that the inhibitor was not toxic to the neuronal growth cone. When Arp2/3 was inhibited, the lamellipodia retrograde flow rate remained constant but the persistence length increased. The combination of these two effects increased the period of protrusion/retraction cycle. The actin retrograde flow level decreased after the Arp2/3 complex was knocked down with siRNA in primary cultured hippocampal neurons and neuroblastoma cells but increased when the Arp2/3 complex was inhibited by CK666 and CK869. We found that partial inhibition of Arp2/3 with 50 μM CK548, did not affect the lamellipodia retrograde flow rate after the recovery of lamellipodium motility. ## Recovery of motility following partial inhibition of Arp2/3 When Arp2/3 was inhibited by 100 μM CK548 the growth cone dynamics was completely abolished presumably because of the loss of adhesion. However, when Arp2/3 was partially inhibited by 50 μM CK548, lamellipodia transiently shrank for 5–8 minutes but then recovered their usual motility. The Rho family of GTPase signaling proteins plays a pivotal role in regulating actin cytoskeleton and could be involved in the observed recovery of lamellipodia motility. The best characterized small GTPases of the Rho family are Rac1, Cdc42 and RhoA which act as molecular switches, cycling between an active GTP-bound state and an inactive GDP-bound state. To determine the possible role of Rho GTPase signaling pathways, in the transient retraction and recovery of lamellipodia when Arp2/3 was inhibited, we used selective inhibitors of Rac1, Cdc42 and RhoA. Lamellipodia treated with the inhibitor of Rac1, EH showed an increase in their period of protrusion/retraction cycle and could move in the axial direction. The lamellipodia regained their motility after the washout of the EH, which show that the inhibitor effect on the neuronal growth cone was not due by its toxicity. When the same lamellipodia were later treated also with 50 μM CK548, lamellipodia showed the usual retraction but did not recover even after 10–20 minutes. Moreover, when treated with both Rac1 and Arp2/3 inhibitors together, lamellipodia shrank as usual but again they did not recover after 10–20 minutes of exposure. The higher concentration of EH not only inhibit Rac1 but also Rac3 and Cdc42 in the MDA-MB-435 metastatic cancer cells. To rule out the possibility that the Rac3 and Cdc42 could be behind the recovery of transiently retracted lamellipodia in the Arp2/3 inhibited condition we tested the effect of the specific inhibitor of Cdc42, ZCL and of another specific inhibitor of Rac1, F56. Transiently retracted lamellipodia following partial Arp2/3 inhibition condition did recover also in the presence of the Cdc42 inhibitor, ZCL. Transiently retracted lamellipodia following Arp2/3 inhibition did not recover when treated also with 100 μM F56. The level of activated Rac1 following an exposure to 50 μM CK for 8 minutes significantly increased compared to what observed in control conditions, 25 μM CK(2 min) and 50 μM CK(2 min). These results indicate the specific role of Rac1 in the observed recovery of motility following partial inhibition of Arp2/3. A possible mechanism could be mediated by the Integrin pathways. Jacquemet, G. *et al*. suggested that the engagement of integrin followed by filamin-A, IQGAP1 and RacGAP1 enrollment, deactivates Rac1. Ilić, D. *et* al. and Saunders, R. M. *et al*. reported that Arp2/3 is recruited to nascent integrin adhesions through interaction with FAK and vinculin, where it reinforces the link between integrin and the cytoskeleton. Furthermore, Beckham et al. reported that Arp2/3 inhibition impairs integrin, an extracellular membrane attachment resulting in either a translocation or treadmilling of mature adhesions. Therefore, it is possible that inhibition of Arp2/3 could reduce the ligation and clustering of integrins and further suppress filamin-A, IQGAP1 and RacGAP1 recruitment, leading to an enhancement of Rac1 activity. Rac1 not only regulates actin polymerization but also increases the free actin-barbed ends and actin monomers. Therefore, the enhanced Rac1 activity could promote the formation and protrusion of lamellipodia, stimulating Arp2/3 by activating the WASP/WAVE family proteins.To study the role of other Rho GTPase pathways in the transient retraction and recovery of lamellipodia upon Arp2/3 inactivation, inhibitors of the respective pathways were used. Detailed quantification showed that the lamellipodia persistence length significantly increased after CT and GSK treatment but it remained constant after ZCL treatment, which is also consistent with what previously reported. The increase in the lamellipodia persistence length upon CT and GSK treatment is probably due to the crosstalk between RhoA and Rac1. The lamellipodia retrograde flow rate significantly decreased after ZCL treatment; however, it remained constant when treated with CT and GSK. As previously shown, CDC42 promotes retrograde flow rate thus the observed significant decrease in the lamellipodia retrograde flow rate is the direct result of CDC42 inhibition. Lamellipodia treated with CDC42 inhibitor increased the period of their protrusion/retraction cycles. When CDC42 was inhibited, the lamellipodia retrograde flow rate decreased, leading to a longer retraction time and overall cycle period. When RhoA and ROCK were inhibited, the lamellipodia retrograde flow rate remained constant but the persistence length increased. The combination of these two effects increased the period of protrusion/retraction cycle. These results show that inhibitors of CDC42, RhoA and ROCK were functioning appropriately. In addition to that, in all these cases lamellipodia showed recovery when treated with Cdc42, RhoA and ROCK inhibitors before treatment with Arp 2/3 inhibitor. Therefore, the involvement of these pathways in the recovery of lamellipodia motility can be discarded. ## Arp2/3 controls the formation and dynamics of filopodia In the active states Rac1, Cdc42 and RhoA interact not only with their specific downstream targets but also cross talk. Specifically, activation of Cdc42 triggers a localized activation of Rac1, initiating the filopodia formation. In our experiments the presence of actin was confirmed in the filopodia before and after the inhibition of Rac1 by using immunocytochemistry. Inhibition of Rac1 remarkably increased the protrusion speed as well as the maximum length of the filopodia. Since Rac1 inhibition reduces the activation of Arp2/3, it is possible that Rac1 inhibition decreases the formation and protrusion of lamellipodia, leaving filopodia behind. In addition, a decrease in the lamellipodia retrograde flow rate and a stable persistence length due to Rac1 inhibition are expected to cause an accumulation of actin at the peripheral region of the GC, from where the filopodia emerge. A higher concentration of actin at the base of the filopodia enables the growth of substantially longer filopodia. Korobova et al. found that inhibition of Arp2/3 reduced the lamellipodia protrusion as well as filopodia formation and dynamics. In our case we found that Arp 2/3 inhibition decreased the protrusion speed of filopodia but it increased their maximum length. Inhibition of Arp2/3, increased the lamellipodia persistence length and the retrograde flow rate, which will accumulate less actin at the periphery of the GC. This may possibly lead to a decrease in the protrusion speed of filopodia. Moreover, Arp2/3 is required for the formation of filopodia and inhibition of Arp2/3 could decrease the formation of new filopodia. The actin accumulated at the periphery of the growth cone upon Arp2/3 inhibition can be utilized by the remaining filopodia to form longer filopodia. We also found that when Arp2/3 was inhibited, the force exerted by filopodia decreased compared to control conditions. The above results indicate the direct involvement of the Arp2/3 in the formation and dynamics of filopodia. On the other hand, Rac1 inhibition increased the length of filopodia but it did not change the force they exerted. This indicates that, like Arp2/3, Rac1 may not directly take part in the formation and dynamics of filopodia. In conclusion, we show here that Rac1 activates when Arp2/3 is inhibited possibly through the Integrin pathways acting as a feedback. Besides its role in lamellipodia formation, Arp2/3 is directly involved in the formation and dynamics of filopodia, while Rac1 is not involved in the activity of the force generation of filopodia. # Materials and Methods ## Neuron preparation Wistar rats at postnatal days 10–12 (P10-P12) were sacrificed by decapitation after anesthesia with CO₂ in accordance with the Italian Animal Welfare Act. The Ethics Committee of the International School for Advanced Studies (SISSA-ISAS) has approved the protocol (Prot.n. 289-II/7). After dissection, Dorsal Root Ganglia (DRG) were incubated with trypsin (0.5 mg/ml; Sigma-Aldrich, Milan, Italy), collagenase (1mg/ml; Sigma-Aldrich) and DNase (0.1 mg/ml; Sigma-Aldrich) in 5 ml Neurobasal medium (Gibco, Invitrogen, Milan, Italy) in a shaking bath (37°C, 35–40 min). After mechanical dissociation, they were centrifuged at 300 rpm, resuspended in the culture medium and plated on poly-L-lysine-coated (0.5 μg/ml; Sigma-Aldrich) coverslips. Neurons were incubated for 24–48 h and nerve growth factor (50 ng/ml; Alomone Labs, Jerusalem, Israel) was added before performing the measurements. ## Quantification of lamellipodia and filopodia motility Neurons were maintained at 37°C in the sample holder of the microscope stage capable of moving in X and Y directions with nanometer precision and imaged through 100 X oil immersed, 1.4 NA objective lens mounted on an inverted microscope (IX80, Olympus). Stacks of phase contrast images of neurons from DRG ganglia were obtained by Charge couple device (CCD) camera (Olympus Megaview) and by moving the objective lens vertically. Each stack contains images obtained in the focal plane of the objective, focused on the coverslip where neurons were cultured i.e. at height 0 and at 1, 2, 3, 4, 5 and 6 micron above the coverslip. Stacks of images were acquired with 0.1–1 Hz frequency to quantify the 3D motion of lamellipodia. Then, for a further analysis, the time lapse image sequence for each height was extracted by using Xcellence software (Olympus) to create videos of different height. Two algorithms were developed to quantify the dynamics of lamellipodia. Algorithm I was designed to quantify in a semi-automatic way the time course of protrusion/retraction cycles by using an improved version of the Kymograph. Algorithm II was designed to quantify the vertical motion of lamellipodia during these cycles. ## Algorithm I The images at height ‘0’ i.e. the cover slip where neurons were cultured- were focused and were used to analyze the protrusion/retraction cycles of lamellipodia. The lamellipodia edges were extracted from each image of the video by using the difference of Gaussian filter. Lamellipodia edges were tracked and followed during the entire duration of the video. A profile of the temporal movement of the lamellipodium edge was obtained. These profiles allowed to follow and quantify lamellipodia cycles of protrusion and retraction. Then the regions of interest of each line were cut and lined up with the time course, to obtain separate kymographs corresponding to each line. The white dotted line in the kymograph shows the lamellipodia leading edge. The changes in the grey values show lamellipodia movements. Mainly the ascending white dotted parts of the dotted line show the protrusion of lamellipodia (white line showing single protrusion) while the descending white dotted parts of the line represent the retraction of lamellipodia. The time to complete one protrusion and retraction by the lamellipodia was considered as a period (T) of protrusion/retraction cycle of lamellipodia. The maximum protrusion length after which lamellipodia starts retracting (white line, dl; micrometers) was defined as the persistence length of lamellipodia. The dark appearances in the kymograph during each retraction of lamellipodia represent the retrogradely moving lamellipodia features (black line showing single lamellipodium retrograde flow). The slope of the lines drawn on these dark appearances was calculated to find out the lamellipodium retrograde flow rate (dx/dt; micrometer per second). Each parameter, the lamellipodia period of the protrusion/retraction cycles, the persistence length and the retrograde flow rate, were calculated by extracting these features from many kymographs and averaged over for statistical significance. ## Algorithm II Lamellipodia not only show periodic motion of protrusion and retraction but, during retraction, they also lift up and ruffle. To study the axial motion of GC lamellipodia, image sequences taken at different heights i.e. 0, 1, 2…6 were acquired and analysed. Algorithm II was based on the theory of defocusing, in which a pixel is assumed to be in focus at a specific height when its intensity equalises with the background intensity of the image of that height. The background intensity of the image for each height was computed as the median of pixel intensities of the image for that height. The number of pixels in focus at a specific height was obtained and normalized by dividing it by the total number of pixels in focus at all the given heights. In this way, the fraction of pixels of the lamellipodium in focus at different heights, was extracted and plotted against time. In this manner it was possible to study the maximal height reached by the lamellipodia edge during retraction in different conditions. Usually lamellipodia lift up high around the maximal retraction, so, in our experiments, their cyclic motility could be characterized both by the kymograph and by the fractional height that was reached (Figs). In order to quantify the motility of filopodia, phase contrast time lapse image sequences acquired at height ‘0’ were analyzed. An Imagej (Image processing and analysis in Java) software was used to measure the maximum length of the filopodia and plug-in, ‘manual tracking’ was used to identify the protrusion rate of the filopodia. ## Force Measurements The force exerted by lamellipodia and filopodia was calculated by measuring the displacement of the optically trapped bead and the known trap stiffness. Unlike traction force microscopy or other similar measurement methods, initially, the bead was not in contact with the cell membrane but was kept in the vicinity of the motile lamellipodia and filopodia. In this way the lamellipodia and filopodia can displace the bead in a spontaneous manner. The Optical Tweezers (OT) set-up used for force measurements was as previously described. The optical tweezers set-up was built as described in Ref. 31. ## Nanopositioner feedback In the OT setup, the detection of the position of the bead was based on the interference signal in the back focal plane, monitored with Quadrant Photo Detector (QPD). Often lamellipodia were able to push the bead out of the linear range–typically 200 nm—in which the QPD could provide a reliable measurement. To overcome this situation, a feedback mechanism, based on a nanopositioner stage- Nanodrive (Mad City Labs, USA) was used. To summarize, from the detected x and y coordinates of the bead the displacement ‘r’ of the bead position from the centre of the trap was computed as sqrt (x^2+y^2). The nanodrive stage brings back the bead into the centre of the optical trap when r is larger than the threshold (which is usually set to be equal to 200 nm). By using the information of the displacement of the nanodrive stage (lower panel) and the bead position determined by the QPD (X, Y axis original) we recovered the x-y axis of the compensated displacement. ## Immunostaining Cells were fixed in 4% paraformaldehyde containing 0.15% picric acid in phosphate-buffered saline (PBS), saturated with 0.1 M glycine, permeabilized with 0.1% Triton X-100, saturated with 0.5% BSA in PBS (all from Sigma-Aldrich, St.Louis, MO) and then incubated for 1h with primary antibodies. The secondary antibodies were goat anti-rabbit 594 Alexa (Invitrogen, Life Technologies, Gaithersburg, MD, USA) and anti-mouse IgG_2a biotynilated (Santa Cruz Biotechnology, Santa Cruz, CA) and the incubation time was 30 min. F-actin was marked with Alexa Fluor 488 phalloidin, whereas biotin was identified by Marina Blue-Streptavidin (Invitrogen, Life Technologies, Gaithersburg, MD, USA) and incubated for 30 min. All the incubations were performed at room temperature (20–22°C). Cells were examined using a Leica DMIRE2 confocal microscope (Leica Microsystems GmbH, Germany) equipped with DIC and fluorescence optics, diode laser 405nm, Ar/ArKr 488nm and He/Ne 543/594nm lasers. The fluorescence images (1024x1024 pixels) were collected with a 63X magnification and 1.3 NA oil- immersion objective. Leica LCS Lite and Image J by W. Rasband (developed at the U.S. National Institutes of Health and available at <http://rsbweb.nih.gov/ij/>) were used for image processing. ## Rac1 activity assay The Rac1-GTP level (the activated form of Rac1) was determined in DRG neurons in control conditions, 25 μM CK (2 min), 50 μM CK (2 min) and 50 μM CK (8 min) using the G-LISA Rac 1 activation assay kit (Cytoskeleton, Inc., Denver, CO, catalog number BK128) according to the manufacturer’s instructions. After experimental treatment, neurons were washed with ice-cold (4°C) PBS and then lysed in ice-cold lysis buffer. The lysate was clarified at 10000 x g at 4°C for 1 min, a 20 μl aliquot was taken for a protein assay, and the remaining lysate was separated into at least two aliquots, snap frozen in liquid nitrogen, and stored at −70°C until the start of the ELISA portion of the assay. Protein concentrations were determined using the Precision Red Advanced Protein Assay that came with the kit. Absorption of the ELISA wells was determined with a Multiskan™ GO Microplate Spectrophotometer (Thermoscientific, USA). We acknowledge the financial support of the following projects within the Seventh Framework Programme for Research of the European Commission: the FOCUS Project n. FP7-ICT-270483, the NEUROSCAFFOLDS Project n. 604263. The author, Wasim A. Sayyad undertook this work with the support of ‘ICTP TRIL Programme’, Trieste, Italy. We thank M. Lough for editing the paper, Andrea Raffin for developing the LabView based algorithm and Priyadharishini V. for helping in protein estimation. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: VT. Performed the experiments: WAS. Analyzed the data: WAS. Contributed reagents/materials/analysis tools: WAS PF. Wrote the paper: VT WAS PF. Designed the software used in analysis: PF. [^3]: Current address: Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, Connecticut, United States of America

# Introduction Fibromyalgia (FM) is a syndrome primarily characterised by chronic, widespread musculoskeletal pain. Its prevalence is estimated to be 3–6% of the world population, and it occurs predominantly in women, with a female to male ratio of 10:1. FM symptoms are not restricted to pain, but often include a heterogeneous group of other conditions, such as hyperalgesia and/or allodynia, physical and mental fatigue, disrupted or non-restorative sleep, headache, irritable bowel, psychiatric disorders, cognitive impairment, and other functional complaints. The aetiology of this syndrome is not completely understood, but a crucial role seems to be played by complex interactions among biological, genetic, psychological, and socio-cultural factors, such as medical illness, neuroendocrine disturbances, stress, and psychiatric disorders. In particular, high levels of stress and psychiatric symptoms may negatively influence the perception of disease severity, functional ability, and the threshold and tolerance for pain. Some authors have indeed suggested that the development of FM might stem from stress-induced disruption of the hypothalamic-pituitary- adrenal (HPA) axis. Exposure to prolonged stressful conditions can alter the function of the HPA axis, with a consequent increase in the production of corticotropin-releasing factor and potentially amplified pain perception. For this reason, FM is often defined as a *central sensitisation syndrome*, caused by increased sensitivity of the central nervous system to pain signals. Among psychological factors, the high prevalence of depression (20–80%) and anxiety disorders (13–64%) has been widely reported. Only recently, researchers have also started to explore emotional functioning in FM syndrome, with their attention focused on alexithymia, a multifaceted personality construct that affects the regulation of a person’s own emotions. Alexithymia is characterised by difficulty in identifying and describing subjective feelings, difficulty in distinguishing between feelings and bodily sensations of emotional arousal, restricted imagination processes, and a stimulus-bound, externally oriented cognitive style. Most of these studies have reported high levels of alexithymia in patients with FM, suggesting the presence of a deficit in emotional self- awareness. On the other hand, the detection and experience of emotional signals from other people have only been sparsely investigated in FM syndrome. A link between alexithymic traits and deficits in the processing of other people’s emotions has been highlighted in both healthy individuals and specific clinical populations, e.g. affective disorders, eating disorders, borderline and psychopathic personality disorders, schizophrenia, somatoform disorders. To the best of our knowledge, only one study to date has examined the ability to identify other people’s emotions in FM syndrome. The results of this study highlighted the fact that patients with FM had reduced performance in a face-recognition task, with a higher percentage of misclassifications of emotional expressions compared with healthy controls. In addition, pain intensity, alexithymia, depression, and anxiety were inversely related to recognition performance, while psychiatric co- morbidity and medication had no impact on performance. The ability to decipher information about the intentions and affective states of social partners is crucial for the implementation of appropriate behaviour during social interactions. This complex process is part of the so-called *social cognition domain*, defined as the ability to construct mental representations of the relations that exist between oneself and others and to flexibly use these representations to function effectively in one’s social environment. Examples of social cognition abilities are the capacity to represent other people’s intentions and beliefs (i.e. Theory of Mind, ToM), and the ability to share and recognise the emotions and sensations of others. From a neurological point of view, ToM and emotional processing abilities are associated with overlapping but distinct brain networks. Common areas of activation are the prefrontal cortex, the superior temporal sulcus, and the temporo-parietal junctions. These areas form the basis for making inferences about mental states. However, these areas are not sufficient for the evaluation of one’s own and other’s emotions, and it is necessary to recruit the additional involvement of emotional networks, in particular those of the amygdala. In fact, while the amygdala is not involved in ToM processes per se, its role is crucial in processing basic and social emotions, related both to the self and to others. The social cognition domain includes a series of different abilities, which gradually evolve throughout the lifetime. An open issue concerns the relationship between these capacities, particularly ToM processes, and the higher-level cognitive skills known as executive functions (EF). Growing numbers of studies are trying to address this relationship in patients with different psychiatric and neurological conditions. EF refers to all of the skills that people use to control and coordinate their cognitive abilities and behavior. These are essential for independent everyday functioning in life, and for the establishment of adaptive social relations. In the last decade, evidence for the multifaceted nature of EF has replaced the idea of a unitary function. Among the several classifications that have been proposed to distinguish EF subcomponents, the model of Miyake et al. identified three separate types of operations: Updating, Shifting, and Inhibition. *Updating* is related to working memory and requires monitoring and coding information as well as replacing old non-relevant information with new relevant information. *Shifting* implies the ability to shift attention between different sub-tasks or different elements of the same task. *Inhibition* is concerned with the individual’s ability to withhold dominant, automatic or prepotent responses when they are inappropriate, and is considered to be a key component in planning abilities. Fisk and Sharp later added a fourth subcomponent, and revised the model of Miyake and colleagues. The factor, called *Access*, refers to the processes involved in verbal fluency tasks, which are believed to mediate access to representations in long-term memory. Currently, there are two opposing views about the relationship between ToM abilities and EF. Some authors believe that ToM is a circumscribed cognitive process, independent of general intellectual functioning and other cognitive domains, included EF. Basing their ideas on theoretical and experimental data, others have suggested instead that lower-level perceptual abilities (e.g. detection of gaze direction and voice recognition) that are required for an appropriate implementation of ToM skills, may be related to specific and circumscribed cognitive domains, while higher-order ToM processes, involving interpreting and associating information as well as hypothesising, would be the result of a more general ability regarding metarepresentation and EF. The present study is based on both clinical evidence, which has highlighted high levels of emotional distress (depression, anxiety, and in particular, alexithymia) in patients with FM, and neuroimaging and neuropsychological data that has reported functional and structural alterations in brain areas crucial for ToM and emotional processing abilities (i.e. the prefrontal cortex and amygdala) in these patients, as well as cognitive deficits in the EF domain. On these bases, the present study aimed to address two main objectives. The first goal was to evaluate the social-cognitive profile of patients with FM, and analyse ToM and emotional processing abilities. In particular, four different areas of the social cognition domain were examined: (1) regulation of one’s own emotions; (2) empathic capacities; (3) recognition of other’s emotions; (4) representation of other people’s affective mental states (i.e., affective ToM). The second goal was to explore the possible relationships between the performance on executive-function tasks and performance on social-cognition tasks in patients with FM. Furthermore, relationships between social-cognitive abilities and demographic, clinical and psychological variables were also investigated for explorative purposes. # Materials and Methods ## Ethics Statement The study was approved by the San Giovanni Battista University Hospital’s ethics committee and was conducted in accordance with the Declaration of Helsinki. All the participants gave their written informed consent to participate in the study. ## Participants and procedure Forty female participants with FM (51.75 ± 7.76 years of age) were consecutively recruited from the Fibromyalgia Integrated Outpatient Unit (FIOU), a multidisciplinary unit based on the collaboration between rheumatologists, psychologists, and psychiatrists at the San Giovanni Battista University Hospital of Turin. All patients had a main diagnosis of fibromyalgia, made by rheumatologists who are experts in the field. In addition, a psychiatric interview based on DSM IV-TR axis II criteria (Diagnostic and Statistical Manual of Mental Disorders, 4th edition, Text Revision) was performed by an expert psychiatrist, in order to exclude FM patients with personality disorders. Exclusion criteria were as follows: less than 18 years old, low education level (\<5 years), and the presence or history of a neurological or a severe psychiatric disorder. Forty-one healthy women (51.83 ± 7.78 years of age) were recruited to the HC group. Exclusion criteria for the HC group were the presence of rheumatic diseases or chronic pain, as well as the presence or history of a neurological or psychiatric disorder. ## Pain evaluation As an index of pain intensity, the item “Pain” of the Italian version of the Fibromyalgia Impact Questionnaire (FIQ) was used to assess the average intensity of pain in the previous week on a scale ranging between 0 and 10. ## Psychological assessment The presence of symptoms of depression and anxiety was assessed using the Italian version of the Hospital Anxiety and Depression Scale (HADS). It consists of 14 items on a 0 to 3 range, and is divided into two subscales, one for depression (HADS-D) and one for anxiety (HADS-A). Each subscale score ranges from 0 to 21 and a score of 8 (cut-off) or more suggests a level of depression/anxiety symptoms that is clinically relevant. ## Neuropsychological assessment Neuropsychological assessment was performed, using tests for short-term memory (Digit Span-Forward—DS F), learning (Rey auditory-verbal learning test—AVLT), and attention (Trail-making test—TMT—A-B). For the investigation of executive functions, four different tests were used, specific for each one of the four subcomponents into which EF has been divided according to the models of Miyake et al. and Fisk and Sharp. Specifically, the Digit Span-Backward (DS B) was employed for evaluating the Updating component, the TMT B for Shifting, the Tower of London (ToL) for Inhibition, and the verbal fluency (FAS) for Access. ## Social cognition assessment **Twenty-Item Toronto Alexithymia Scale (TAS-20)**. Alexithymia was assessed using the Italian version of the Toronto Alexithymia Scale (TAS-20). Subjects were asked to indicate the extent to which they agreed or disagreed with each statement on a five-point Likert scale. The results provide a TAS-20 total score, and three subscale scores that measure different aspects of alexithymia: *difficulty identifying feelings* (Factor 1), which measures the inability to distinguish specific emotions and between emotions and the bodily sensations of emotional arousal; *difficulty describing feelings* (Factor 2), which assesses the inability to verbalise one’s emotions to other people; and *externally- oriented thinking* (Factor 3), which evaluates the tendency of individuals to focus their attention externally and not on the inner emotional experience. The TAS-20 cut-off scores are as follows: ≤51 no alexithymia, 52–60 borderline alexithymia, ≥61 alexithymia. This scale has shown good internal consistency and test-retest reliability, as well as convergent, discriminant and concurrent validity, and it is currently one of the most utilised instruments in studies of alexithymia and emotion. **Empathy Quotient (EQ)**. The EQ is a validated self-report questionnaire, employed to assess the capacity to empathise with another, i.e. to recognise another’s affective state and to respond to this with an appropriate emotion. The EQ comprises 60 items, broken down into two types: 40 items assessing empathy and 20 filler/control items, included to distract the participant from a relentless focus on empathy. For each empathy item, a person can score 2, 1, or 0, so the EQ has a maximum score of 80 (higher scores indicate greater empathy). The EQ is able to detect considerable individual, gender, and group differences, in both general-population and clinical samples. **Ekman 60 Faces**. The Italian version of this test was used to assess the recognition of facial expressions pertaining to basic emotions. The Ekman 60 Faces Test uses photographs of the faces of 10 actors (six female and four male) selected from the Ekman and Friesen series. Each actor displays one of the six basic emotions investigated (happiness, sadness, disgust, fear, surprise, and anger). The subject is required to respond verbally, deciding which of the six labels for basic emotions that are placed below each photograph can best describe the facial expression shown. The maximum test score (indicating best performance) is 60 for all six emotions and 10 for each basic emotion. **Reading the Mind in the Eyes Test (RME)**. The RME was employed to assess the ability to represent other people’s affective mental states. In the test, the experimenter presents a set of 36 photographs of the eye region of various human faces. Participants are required to choose among four words that are printed on the page that the picture appears on, using the criterion of which word best describes the mental state of the person depicted in the photograph. Participants have unlimited time to decide, and a glossary is provided. Participants have to put themselves into the mind of another person and recognise his or her complex mental state. In the gender-recognition control task, participants are asked to judge the gender of the person in each of the 36 photographs. For both the experimental (mental state attribution) and control (gender attribution) conditions, the maximum score indicating the best performance is 36. ## Statistical analyses All the statistical analyses were conducted using IBM SPSS Statistics, version 20.0. Normal distribution was assessed using indices of asymmetry and kurtosis. Non-parametric equivalent tests were performed on data that violated this assumption. For normally distributed variables, independent *t*-tests were used. In order to evaluate the possible relationships between variables, Spearman or Pearson correlations were computed, as appropriate. # Results ## Demographic, clinical and psychological data Data on the demographic and psychological variables are presented in. The two groups were matched for age and education. For psychological assessment we used data from the HADS total score and the scores for subscales HADS-A and HADS-D. The results showed significantly higher scores in patients with FM both for the total score (*p*\< 0.001) and for each of the two subscales evaluating anxiety (*p*\< 0.001) and depression (*p*\< 0.001). According to the cut-off scores of the HADS, 67.5% (27/40) of the patients with FM showed a clinically relevant level of both anxiety and depression, compared with 34% (14/41) for anxiety (*Χ*²(1) = 9.011, *p*\< 0.001) and 19.5% (8/41) for depression (*Χ*²(1) = 19.000, *p*\< 0.001) in the HC group. Concerning the clinical characteristics of the FM group, patients reported 6.47 (± 5.81) years of duration of illness and a high rate of pain intensity (7 ± 2.55 to the item “Pain” of the FIQ). ## Neuropsychological assessment The comparisons between the neuropsychological scores of the two groups are shown in. Patients with FM performed worse than the HC group on all the four tasks evaluating EF (*p* values ranging from \< 0.001–0.011). Furthermore, a poorer performance in the FM group also emerged in the DS F (*p* = 0.005), in the AVLT-Delayed recall (*p* = 0.006) and in the TMT B-A (*p* = 0.004). No statistically significant differences were found on the other neuropsychological measures. In order to bring out the individual differences that could be flattened by group analyses, the individual scores were analysed comparing for each test the number of subjects with impaired or borderline performance according to the age- and education-corrected scores (equivalent score ≤1). The results showed that a significantly higher number of patients with FM compared with HC had a deficient performance in the DS F (short-term memory) and B (working memory), and in the Delayed recall task of AVLT (episodic memory). ## Social cognition tasks Data from social cognition tasks are reported in. **Regulation of own emotions**. Concerning alexithymia, statistical analyses revealed the presence of significant differences between FM and HC on the TAS-20 total score (*p*\< 0.001), and on the F1 (*p*\< 0.001) and F2 subscales (*p* = 0.011); in all these comparisons patients with FM scored higher than HC. According to the TAS-20 cut-off scores, 27.5% (11/40) of the patients with FM were alexithymic and 45% (18/40) were borderline, compared with 7% (3/41) and 19.5% (8/41), respectively, in the HC group. **Empathic capacity**. No significant difference between the two groups was found in the EQ score. **Recognition of others emotions**. Concerning the Ekman 60 Faces Test, independent *t*-tests revealed the presence of significant differences between the two groups on total score (*p* = 0.010), and on two of the six emotions investigated by means of the test, i.e. anger (*p* = 0.049) and disgust (*p* = 0.016). Once again, patients with FM showed significantly lower scores, indicating a reduced ability to recognise other people’s emotions, especially anger and disgust. **Representation of other people’s affective mental states**. No significant difference between FM and HC was found in the control task for RME. In the experimental condition, patients with FM evidenced a significantly lower performance on the mental states attribution task (*p* = 0.007). ## Correlations The second aim of this study was to investigate the possible relationships between social cognition tasks and EF measures in patients with FM. Moreover, we also evaluated the correlations between social cognition tasks and demographic, clinical and psychological data. To do that, we only considered the variables that showed a significant difference in the comparison between FM and HC, i.e. TAS-20 F1, F2, and total score; Ekman anger, disgust and total score; and RME experimental task. Correlations between social cognition measures and EF tasks are listed in. As shown, no significant correlation was found, with the only exception of a low positive correlation between the Ekman total score and the DS B. In addition, we verified the possible relationships between social cognition measures, EF tasks and duration of illness (DI) in FM group and no significant correlation emerged (DI and FAS: *r* = -0.276, *p*: ns; DI and DS B: *r* = -0.132, *p*: ns; DI and ToL: *r* = 0.196, *p*: ns; DI and TMT B: *r_s* = 0.185, *p*: ns; DI and TAS-20 F1: *r* = 0.319, *p*: ns; DI and TAS-20 F2: *r* = 0.333, *p*: ns; DI and TAS-20 Total: *r* = 0.342, *p*: ns; DI and Ekman Anger: *r* = 0.087, *p*: ns; DI and Ekman Disgust: *r_s* = -0.175, *p*: ns; DI and Ekman Total: *r* = -0.036, *p*: ns; DI and RME Experimental: *r* = -0.053, *p*: ns). Regarding the relationship between social cognition measures and demographic, clinical and psychological variables, no significant correlations were detected between age, HADS-A, HADS-D, FIQ-pain, and the RME experimental or the Ekman anger, disgust and total score; a positive correlation was only found between the RME experimental and the level of education (*r* = 0.359, *p* = 0.023). However, significant correlations were detected between demographic, clinical and psychological variables and the TAS-20 scores. In particular, positive correlations were found between the TAS-20 total score and the HADS-A (*r* = 0.334, *p* = 0.035), the HADS-D (*r* = 0.630, *p* \< 0.001), and the FIQ-pain (*r* = 0.518, *p* = 0.001). Likewise, positive correlations were found between the TAS-20 F1 and the HADS-A (*r* = 0.462, *p* = 0.003), the HADS-D (*r* = 0.476, *p* = 0.002) and the FIQ-pain (*r* = 0.442, *p* = 0.004). Finally, the TAS-20 F2 was positively correlated only with the HADS-D (*r* = 0.537, *p* \< 0.001) and the FIQ-Pain (*r* = 0.344, *p* = 0.030). # Discussion The present study aimed to address two main objectives. Firstly, we evaluated the social-cognitive profile of patients with FM, investigating ToM and emotional processing abilities. Secondly, we analysed the relationship between EF deficits and social cognition tasks in patients with FM. Correlations between demographic, clinical and psychological variables, and measures of social cognition were also evaluated. The results highlighted a significant difference between patients with FM and the HC group in most of the social cognition tasks employed. In particular, the FM group showed significantly higher levels of alexithymia, especially in the subscales “Difficulty in identifying feelings” and “Difficulty describing feelings” of TAS-20, compared to the control sample. These data are in line with most of the studies that have evaluated the prevalence of alexithymia in patients with FM. A similar significant difference was found for the experimental task of the RME, while no significant difference was observed for the control task. Patients with FM experienced specific difficulties in representing other people’s affective mental states that cannot be attributed to a basic sensory deficit. The Ekman 60 Faces Test results also showed the presence of significant differences (lower performance of patients with FM) between the two groups both for the total score and for two of the six emotions investigated by means of the test, i.e. anger and disgust; no significant differences were found for other emotions. These data are in line with the only study that has investigated the ability to recognise another’s facial emotions in patients with FM. As mentioned above, these authors showed that patients with FM had reduced performance in the facial affect recognition task, with a higher percentage of misclassifications of emotional expressions compared with the HC group. The only social cognition task, in which no differences between the two groups were found, was the EQ. In this case, the FM group didn’t report a lower capacity for empathy compared to the control group. Concerning the general cognitive profile, neuropsychological assessment revealed the presence of significant differences in most of the measures. In particular, the FM group displayed significantly lower performance on the verbal fluency (FAS), the DS B and F, the AVLT-Delayed recall, and the TMT B and B-A, compared to HC. These data are consistent with previous studies that have reported cognitive deficits in attention, memory, and EF domains in patients with FM. In particular, Park et al. found that patients with FM demonstrated lower performance on measures of working memory, free recall, verbal fluency, and verbal knowledge, but showed intact speed of processing, compared with age- and education-matched controls. Significantly, patients with FM in that study performed no differently from controls who were 20 years older on most cognitive tasks, with the exception of speed of processing and vocabulary. Only self- reported pain on the Arthritis Impact Measurement Scales predicted poor cognitive performance in the FM group. Measures of depression, anxiety, and the McGill Pain Questionnaire scores were all unrelated to poor cognitive performance. Verdejo-Garcia et al. observed that in the Wisconsin Card Sorting Test, women with FM showed poorer performance than healthy women on the number of categories and non-perseverative errors, but not on perseverative errors. Patients with FM also exhibited an altered learning curve in the original Iowa Gambling Task (IGT) (where reward is immediate and punishment is delayed), suggesting compromised emotion-based decision-making. This was not the case in the variant IGT (where punishment is immediate but reward is delayed), suggesting hypersensitivity to reward. Self-reported pain intensity and pain interference were significantly associated with task performance. In contrast, cognitive performance was not associated with measures of negative mood (i.e. affective distress) or duration of pharmacological treatment, and was very mildly associated with personality characteristics. Finally, from a psychological standpoint, our group of patients with FM presented with significantly higher levels of depressive and anxiety symptoms (67.5% in both cases) compared with the HC group. These results corroborate, once again, the high prevalence of psychological distress reported in previous studies of patients with FM. To the best of our knowledge, this is the first study to investigate social- cognitive abilities in a homogenous sample of patients with FM. The results show that patients with FM have impairments both in the regulation of their own affect and in the recognition of other’s emotions, as well as in representing other people’s affective mental states. There is evidence that appropriate behaviour in social interactions is determined by the ability to decipher information about the intentions and affective states of social partners. Thus, impairments in facial affect recognition and difficulties in accurately inferring other people’s affective mental states may lead to substantial difficulties in interpersonal contacts (e.g. interaction problems with family and friends, or social isolation), which have been already reported in patients with FM. Furthermore, poor psychosocial functioning and unsatisfactory relationships might contribute to the genesis and maintenance of chronic pain, intensifying the symptomatology in individuals with FM. From a neurological standpoint, the brain networks relevant for pain and emotional processing partially overlap. The amygdala plays a crucial role in sharing emotional experiences and in recognising emotions in oneself and others. In particular, this structure is involved in the decoding of emotional expressions, and modulates the activity of the fusiform gyrus, which constitutes the most prominent face-selective area of the brain. Neuroimaging studies have indeed shown sensitivity of the amygdala to the kind and valence of facially expressed emotions. The insular cortex may also be relevant in this context. It has been reported, for instance, that impaired disgust recognition is associated with reduced insula activity. In addition to their prominent role in emotional functioning, the amygdala and the insula are integral parts of the neural network underlying pain. Specifically, both structures are involved in transmitting the affective dimension of pain perception and are altered in patients with FM. The hyperactivity of the pain network due to central nervous system sensitisation, may lead to an increased demand on structures such as the amygdala and insula, reducing the available resources for other functions such as emotional processing. As far as the second goal of this study is concerned, we investigated whether EF measures in the FM group were related to the different social cognition tasks that were used. In addition, we also analysed the possible relationships between demographic, clinical and psychological variables, and measures of social cognition. Concerning EF tasks, no correlations were found between social cognition tasks and each of the four subcomponents of the EF domain that were analysed. The only exception was represented by a low positive correlation between the Ekman total score and the DS B. Regarding the demographic, clinical and psychological variables, correlation analyses showed no relationship between Ekman total score, anger and disgust, RME experimental, on the one hand, and anxiety, depression and pain intensity, on the other hand. Positive correlations were only detected between the latter measures (anxiety, depression, and pain) and the total score and scores for the F1 and F2 subscales of the TAS-20. This result is consistent with previous studies that have investigated the presence of alexithymia in FM patients. In particular, Steinweg et al. found higher levels of alexithymia in FM patients compared with either general medical or rheumatoid arthritis patients. However, they also revealed that alexithymia was strongly associated with moderate-to-severe depression, but no group differences were detected when mood disturbance was controlled for. Taken together, as far as the first aim of this study is concerned, the results show the presence of several impairments in social cognition skills in patients with FM. As for the second aim, i.e. to explore the possible relationships between the performance on executive-function tasks and the performance on social-cognition tasks in patients with FM, we found that the latter are largely independent of both EF deficits and symptoms of psychological distress. The only exception seems to be represented by alexithymia; in fact, psychological disorders, but not EF deficits, seem to play a role in explaining the high levels of alexithymia found in the FM sample. Concerning other measures of social cognition, no relationship was found with EF deficits or symptoms of psychological distress. In our sample, impairments in ToM and emotional processing ability appeared to be independent of the EF domain. This study also has some limitations. Firstly, even though we enrolled an adequate number of patients with FM, our study is still limited by a relatively small sample size. Secondly, the self-reported measures we used might have elicited a bias towards social desirability, masking the real profile of some individuals. Thirdly, although in patients with FM there is evidence of structural and functional alterations in brain areas crucial for ToM and emotional processing abilities (i.e. the prefrontal cortex and the amygdala), we didn’t directly measure the activity in these brain structures. Future studies should include neuroimaging evaluations and use performance-based instruments for the analysis of both empathic capacity and alexithymia, in addition to traditional self-reported tests. In spite of these limitations, the findings reported in the present study represent the first contribution towards understanding the complex social- cognitive profile of patients with FM. The impairments reported in tasks that evaluate ToM and emotional processing abilities highlight the importance of adequately assessing these abilities in clinical practice. In this way, it could be possible for clinicians to plan better pharmacological and/or psychological treatment based on each patient’s needs. The authors would like to thank Valentina Tesio, Stefania Brighenti and Francesca Monoli for their help in collecting data. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MDT LC MA. Performed the experiments: MDT FC EF RT. Analyzed the data: MDT LC RBA MA. Contributed reagents/materials/analysis tools: LC RBA MA. Wrote the paper: MDT LC RBA MA.

# Introduction Recently, genome editing by the CRISPR/Cas9 system has became a widely used strategy to study gene functions in cells and *in vivo*. In the widely-used CRISPR/Cas9 system derived from *Streptococcus pyogenes*, the Cas9 nuclease creates a double strand break in the genome at a site complementary to the guide RNA, which is then repaired by various repair machineries of the cell. This often leads to the introduction of indels, resulting in an efficient loss of function of gene. However, even with the most efficient delivery strategies, gene disruption does not reach 100% efficiency in cell culture. For this reason, researchers have often performed isolation of mutant clones before doing their assay of interest. Therefore, a strategy to identify mutant clones quickly and accurately is very important. The indels caused by CRISPR/Cas9 editing are often small, thus standard PCR with genomic DNA (gDNA) is not useful to detect the mutants since the amplicon sizes are not different enough. The usage of Cas9 nuclease with two guide RNAs targeting relatively distant sites enables an easy discrimination by inducing a bigger deletion detectable by PCR (not to be confounded with the double-nickase strategy where the offset should be small enough to get a double strand break). However, with this strategy only clones having the large deletion in all alleles are discriminated, and the potentially useful mutant clones having only small indels are ignored, reducing the success rate of obtaining mutants. Also, this strategy doubles the effort of plasmid preparation as well as the chance of having off-target effects. Multiple methods to identify CRISPR/Cas9-induced indels have been published. A strategy that is often used is the mismatch cleavage assay. This method uses enzymes that cleave DNA heteroduplexes at mismatch sites. Although being easy to perform, this method does not discriminate wild type clones and mutant clones harbouring identical mutations in both alleles. Also, heterozygous mutants are not discriminated from homozygous mutants with different mutations between alleles. These issues can be avoided by mixing wild type PCR amplicon with the amplicons of the analysed samples. However, this adds another step for the experiment, and DNA concentrations of PCR products have to be quantified in order to have a correct ratio of mixed amplicons, since this affects the cleavage efficiency. An ideal method should discriminate completely mutated cells from the others without the need of too many steps, while being cost effective, and not requiring special laboratory equipment. In case of difficult- to-transfect cells or when doing multiplex gene disruption, hundreds of clones might have to be analysed. Therefore, it is also desirable to reduce any steps like template DNA quantification and PCR product purification. None of the existing methods fulfil all the above requirements. For example, high-resolution melting analysis is easy and quick, but requires specific instruments. Also, since different type of indels will affect the melting behaviour of PCR products, unambiguous identification of homozygous mutants might be difficult. Capillary electrophoresis of PCR products enables accurate detection of indels, but requires specific instruments. Quantitative PCR efficiently identifies mutations, but the highly quantitative nature of this method gives rise to the need to quantify and normalize template DNA amount. Heteroduplex mobility assay is simple to perform, but has similar drawbacks to mismatch cleavage assays when applied to clone analysis. Restriction fragment length polymorphism and RNA-guided engineered nuclease restriction fragment length polymorphism are almost ideal. However, the former requires a convenient restriction site at the cleavage site, and for the latter one needs to generate *in vitro* the nuclease complex used for restriction analysis, which might be laborious. In addition, enzyme restriction is another step added to PCR, which one might want to avoid if possible. Most of the methods stated above have the benefit of being able to estimate the mutation efficiency (with more or less accuracy) in bulk cells. This degree of quantitation is not required when the only requirement is to detect fully mutated clones. Thus, we wanted to establish a method that is easy to perform without the need to normalize template DNA amount, and requiring only minimal laboratory instruments, even if it might be less quantitative than the ones stated above. Here, we show that when primers are appropriately designed, a simple PCR using a set of primers flanking the target site of interest, together with an additional one overlapping with the site cleaved by Cas9, enables the identification of homozygous mutant clones. # Materials and methods ## Cell culture McA-RH7777 cells and HeLa MZ cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM high glucose, GlutaMAX, pyruvate), supplemented with 10% fetal bovine serum and 100 U/mL penicillin-streptomycin (all from Thermo Fisher Scientific, Waltham, MA, USA). ## Plasmid construction for genome editing Plasmid backbones for CRISPR/Cas9 experiments were either pX330, which contains expression cassettes for both Cas9 and single guide RNAs (sgRNAs), or an in- house constructed plasmid (pUC-U6-sg) for expressing only sgRNAs. The latter was constructed by incorporating the sgRNA expression cassette from pX330 into a pUC19 backbone. The insert and the pUC19 backbone were amplified by PCR using PrimeSTAR GXL DNA Polymerase (TAKARA Clonthech, Otsu, Shiga, Japan) with the primers listed in, and assembled using the Gibson assembly system (New England Biolabs, Ipswich, MA, USA). Target sites for genome editing were selected using either the Zhang laboratory algorithm (<http://crispr.mit.edu>) or CRISPOR (<http://crispor.tefor.net>), choosing those having low off-target scores and high on-target scores. The pairs of oligo DNA were synthesized at Microsynth AG (Balgach, St. Gallen, Switzerland). Plasmids were constructed by Golden Gate assembly, based on the Yamamoto laboratory protocol with slight modifications. Restriction and ligation was performed in a single tube using FastDigest Bpi I (Thermo Fisher Scientific) and quick ligase (New England Biolabs, Ipswich, MA, USA) in quick ligase buffer. Tubes were incubated at 37°C and 25°C for 5 minutes each, and this cycle was repeated three times. Then, an additional Bpi I digestion was done for 1 hour. Products were transformed into STBL3 chemical competent cells (Thermo Fisher Scientific), plasmids were sequence-verified by Sanger sequencing by Fasteris SA (Plan-les-Ouates, Geneva, Switzerland), and obtained in large scale using either QIAFilter plasmid midiprep kit (QIAGEN, Hilden, Germany) or PureLink HiPure midiprep kit (Thermo Fisher Scientific). ## Genome editing in cultured cells Genome editing in McA-RH7777 cells was performed based on a previously described Hprt co-targeting strategy, which was optimized for our cell lines. In this method, mutant cells are strongly enriched by selecting cells having a mutation in the co-targeted Hprt1 gene, which confers resistance to 6-thioguanine (6-TG, Sigma-Aldrich, St. Louis, MO, USA). Transfection of plasmids (450 ng of pX330-based plasmid for the target gene, and 50 ng of pUC-U6-sg-based plasmid for rat Hprt1 gene) was done using lipofectamine 3000 (Thermo Fisher Scientific), by reverse transfection; the mixed transfection reagents were put in an empty 24 well plate, and 100,000 cells were seeded on them in a volume of 1 mL. Cells were maintained in culture with occasional passages. Nine days post- transfection, cells were detached and diluted in a medium containing 4 μg/mL 6-TG. Selection was done for one week, with a medium change at day 4. To isolate triple knockout clones, cells were transfected with equal amounts of the three pX330-based plasmids to target Sgpl1, Sgpp1, and Sgpp2, and subjected to limiting dilution 5 days post-transfection on a type I collagen (Sigma- Aldrich)-coated 96 well plate. Conditioned medium was added at 20% to promote the growth of clonal populations. Genome editing in HeLa MZ cells was done without co-targeting (empty pX330 plus pUC-U6-sg-based plasmid for human HPRT1 gene), using the same transfection method. Clones of HeLa MZ cells were obtained by limiting dilution into 96 well plates. To analyse polyclonal HPRT1-mutated HeLa MZ cells, the transfectants were selected with 4 μg/mL 6-TG 5 days post- transfection. ## Primer design Primer sequences for the amplification of genomic DNA were designed using primer-BLAST (<http://www.ncbi.nlm.nih.gov/tools/primer-blast/>). Melting temperature was calculated using the same program. The primers were typically designed to amplify 700 to 900 bp, with the Cas9-cleaved site located between 150 to 300 bp from one end. For competition-based PCR, the inner primers were designed only after amplification of the outer amplicon was confirmed. Except for the initial experiments during optimization, the inner primers were designed to have a melting temperature not exceeding 60°C, with the 3’ end extending 3 bases beyond the cleavage site. If this design resulted a primer shorter than a 16-mer, the primer was designed as a 16-mer regardless of its melting temperature. Then, the outer primer having the same orientation with the inner one was redesigned (when required) to have a lower melting temperature than it. Details are provided in. ## Isolation of genome DNA and confirmation of mutation rates Genomic DNA was isolated as following. Cells were lysed in lysis buffer (20 mM Tris-HCl, pH 8.0, 1 mM EDTA, 0.67% (w/v) SDS, and 124 μg/mL proteinase K (Roche Life Science, Rotkreuz, Zug, Switzerland)) at 55°C for at least 4 hours. DNA was precipitated with isopropanol, pellets were washed with 70% ethanol, and were dissolved in TE buffer (20 mM Tris-HCl, pH 8.0, 1 mM EDTA). TIDE (Tracking of Indels by DEcomposition) analysis was performed to confirm that samples used in the optimization steps are mutants. For this, target regions were amplified by PCR (primers are listed), amplicons were treated with Exonuclease I and FastAP alkaline phosphatase (both from Thermo Fisher Scientific), followed by heat denaturation, and directly used for Sanger sequencing. Competition-based PCR was performed using a mixture of 3 primers at 0.2 μM each. PCR reactions were carried out using ExTaq polymerase (Takara Clontech) or PrimeSTAR GXL polymerase. For ExTaq reactions, 5% (v/v) DMSO was added. PCR reactions with the following temperature cycles were performed using a T Professional TRIO Thermocycler (Biometra, Göttingen, Germany): 30 seconds at 94°C, 30 seconds at 60°C, and 50 seconds at 72°C (total 35 cycles). For PrimeSTAR GXL reactions, the conditions were: 10 seconds at 98°C, 15 seconds at 60°C, and 50 seconds at 68°C (35 cycles). PCR products were analysed by agarose gel electrophoresis, typically using 1.5% (w/v) gels in TBE buffer (89 mM Tris-borate, 2 mM EDTA). Bands were visualized using EZ-VISION DNA dye (amresco, Solon, OH, USA) or ethidium bromide (Sigma-Aldrich). Primers are listed in. ## Image processing and quantification of band signals For clarity on printed pages, signals of some images were adjusted using photoshop (Adobe, San Jose, CA, USA), while maintaining the linearity of signals. For quantification of signals, images without adjustment were used. The digital gel images were analysed using ImageJ. We obtained images using different exposure times, and did quantification with those that did not have saturating signals. ## Analysis of 6-TG sensitivity Individual HeLa MZ clones were placed in different wells of a 96 well plate with approximately 20% confluency, and incubated with 6 μg/mL 6-TG (day 0). Medium change (still containing 6-TG) was done at day 4, and cells were fixed with methanol at day 6. Cells were stained with 0.5% crystal violet (in 20% methanol) for visualization. ## Analysis of mutant clones by sequencing To analyse genome sequencing results of clones that were selected by competition-based PCR, TIDE analysis was done for each locus. When TIDE analysis gave sufficient explanation for the mixed sequencing chromatograms, it was concluded that the clone contained the indels proposed by the algorithm. When an indel larger than 50 bases was present, this allele could not be calculated by TIDE, so chromatogram deconvolution was done manually. Control and mutant chromatograms were compared by eye, and the aberrant sequence was estimated from their difference, which was matched with downstream sequence in case of large deletions. If a large stretch of aberrant sequence was found to be not corresponding to any sequence of the locus, this was interpreted as an insertion of a relatively random sequence. When different alleles of mutant clones had to be analysed, target regions were amplified by PCR using PrimeSTAR GXL (primers at), amplicons were cloned using Zero Blunt TOPO PCR Cloning Kit (Thermo Fisher Scientific), and transformed into STBL3 chemical competent cells. Bacterial colonies containing plasmids with inserts were selected by colony-direct PCR, and these PCR products were used for direct Sanger sequencing after Exonuclease I and FastAP alkaline phosphatase treatment as written above. # Results ## Competition-based PCR detects mutant cells A previous study showed that quantitative PCR with one of the primers overlapping with the Cas9 cleavage site enables the detection of mutations in zebrafish. In the same study, endpoint PCR was used only to detect mutant fragments after isolation into plasmids. This motivated us to develop an endpoint PCR-based method to detect mutant clones of cultured cells, without the need to isolate fragments into plasmids. To optimize such a method, we used gDNA from mutant cells that were established for another unpublished study. Since this manuscript focuses on the detection strategy, details about the generation and characterization of these mutant cells will be described elsewhere. In this study, the PCR strategy is designed to detect mutations generated by wild type *Streptococcus pyogenes* Cas9 nuclease (hereafter simply described as Cas9). The potential use of this strategy to detect mutations induced by other genome editing methods will be addressed in the Discussion. We used McA-RH7777 cells that have mutations in Sgpl1 gene for initial optimization. First, we performed a standard endpoint PCR with one primer overlapping with the Cas9 cleavage site, placing the cleavage site between the third and fourth bases from the 3' end of the primer. As can be expected, the amount of PCR products did not differ when using gDNA from wild type or mutant cells, nor did it when using different template DNA amounts. Thus, as is well known, amplification reaches a plateau in conventional end point PCR, and even if the affinity of the primer with template DNA from mutant cells is reduced, the amount of final products is unchanged. This shows that mutants cannot be discriminated using this design of primers, unless using quantitative PCR as in the previous study. Next, we designed a PCR strategy where one pair of primers is designed to flank the Cas9 cleavage site (the outer primers), and another primer (the inner primer) overlaps with the cleavage site. Among the outer primers, the one proximal to the Cas9 cleavage site will be called the “F-out” primer, and the other one the “R-out” primer. The inner primer will be called “F-in” when it directs to the same direction as F-out, and “R-in” in the other case (see also for the R-in orientation). The amplicon of outer primers will be called “out-amplicon”, and the one from the inner primer and the corresponding outer primer will be called “in-amplicon”. Under such a PCR condition, competition between amplicons should occur. First, since the amount of substrates (dNTPs and primers) is not infinite, the amplification of one amplicon reduces the amount of substrates that can be used for the other. Second, if the inner primer is present on a template DNA, the elongation from the outer primer that is located upstream will be blocked (unless using a polymerase with strand-displacing activity). Thus, when an amplicon is strongly amplified, the other one is reduced. The fact that out-amplicons can be used as a template for PCR of in-amplicon, but not the inverse, makes this competition even more complex (not depicted in the Figure). When gDNA from a mutant is utilized, the efficiency of inner primer binding should be lower, shifting the balance of amplification toward the out-amplicon. We thought that this design would lead to a different ratio between out- and in-amplicons when using mutant gDNA, enabling their discrimination from wild type. This design was inspired by the fact that before the appearance of real-time PCR, quantitative PCR was done based on competition of primers caused by the addition of a competitor DNA. The difference of the current design is that it does not require the preparation of a competitor DNA. Based on this design, we tested whether competition between individual amplicons indeed occurs. We did PCR of the Sgpl1 locus from wild type gDNA using individual primer pairs (F-out and R-out or F-in and R-out), or with the three primers mixed (upper). The signal of the out-amplicon was strongly inhibited when the inner primer was present (lanes 1 versus 3). On the other hand, the signal of the in-amplicon was unaffected by the presence of F-out primer (lanes 2 versus 3). We did the same experiment using genomic DNA from Sgpl1 mutant cells. This time, the signals of both of the amplicons were weaker when the three primers were used than when doing PCR individually (bottom). Thus, as assumed, each amplicon has a negative effect on the other. In wild type cells, the out-amplicon was so strongly inhibited that it could not affect the amount of the in-amplicon. The balance of inhibition was different in mutant cells due to the lower affinity of the inner primer. Therefore, wild type and mutant DNA could be discriminated by calculating the ratio of out- and in-amplicon when doing PCR with the three primers. We will call this PCR method “competition- based PCR (cbPCR)”. ## Optimization of cbPCR Having shown that cbPCR can discriminate mutant DNA from wild type, we next investigated the factors that affect the efficiency of this method in order to reduce the chance of potential identification errors. This was important since CRISPR/Cas9 leads to various indel sizes, thus some might be more or less easily discriminated (due to differences in inner primer affinity), and we thought that optimization was required to enable discrimination of all of them from wild type. We first tested whether there is an optimal orientation of the inner primers. Based on the nomenclature defined above, the F-in primer provides a larger in-amplicon, which utilizes more dNTPs. We did cbPCR using F-in or R-in as the inner primer. The use of F-in led to a stronger inhibition of out- amplicons than when R-in was used. Both primers enabled the discrimination of mutant and wild type DNA using the ratio of out- and in-amplicons, but the difference was bigger when F-in was used. This enabled a better genotype discrimination due to a stronger inhibition of out-amplicon when using wild type DNA. The same pattern was seen when another gene, Sphk1, was tested. In this case, in-amplicons were absent when cbPCR was done with mutant DNA regardless of inner primer orientation, but the use of F-in led to a stronger inhibition of out-amplicon when using wild type DNA. Thus, when primer design allows it, the inner primer should be designed in the direction of F-in. We next tested the effect of F-in primer length. A longer F-in primer might increase its affinity with template DNA, leading to a stronger inhibition of out-amplicon when using wild type DNA, thus improving mutant discrimination. However, a longer primer might also increase the tolerance to mutation in the template, reducing mutant discrimination. Therefore, it was difficult to predict whether the F-in primer should be long or short. We did cbPCR using F-in primers of varying lengths. As expected, a longer F-in primer leads to a smaller out- to in-amplicon ratio in wild type DNA, but at the same time increased the tolerance for mutations (see the appearance of in-amplicons in PCR products of mutant gDNA). Therefore, the best discrimination was achieved when the F-in primer was short, as long as it enabled efficient detection of in-amplicon in wild type DNA (21 mer). In our PCR conditions, we usually achieved this condition when the F-in primer was designed to be the longest while not exceeding a melting temperature of 60°C. We also routinely designed the F-out primer with a melting temperature lower than the F-in primer, in order to increase the inhibition of out-amplicon when doing cbPCR with wild type DNA. We used this primer design as a default. We next tested whether by using this default setting, we could discriminate mutants of different genes by cbPCR. For this, we used gDNA from 8 mutant cell lines that were generated a priori. Without further optimization of primer design, we could clearly discriminate 7 out of 8 mutant lines from wild type using cbPCR, by the absence of in-amplicons. The exception was the Plpp2 gene, where the in-amplicon was also absent in cbPCR product of wild type gDNA. Except for this, using the default primer settings, most of mutants could be discriminated by cbPCR. ## Guidelines for primer refinement If we could overcome the problem of Plpp2 mutant detection, we should be able to determine how a failed primer design could be improved. We first did conventional PCR to test whether the F-in primer could bind to the template or not. When used without competition, the F-in and R-out primer set enabled the amplification of a detectable in-amplicon. However, in cbPCR settings, the out- amplicon outcompeted the in-amplicon even in wild type DNA. This result shows that by doing PCR with individual primers, we can define whether the amplification of in-amplicon just failed or whether there was a too strong competition. If the inner primer does not work, we should either use the one with the other direction (R-in primer), or change PCR conditions (annealing temperature and/or enzyme) to make it work. If the inner primer does work, then adjustment of the balance of PCR amplification efficiency could be done. Since the inner primer did work for Plpp2, we did the latter. We tested F-out primers of different lengths, and found that slightly shorter ones (17 or 18 bp) enabled the detection of in-amplicons in cbPCR using wild type, but not mutant DNA. Therefore, by refining primer design in failed experiments, we could design primer sets that discriminate mutants. As far as we tested, whenever an in- amplicon is efficiently detected by cbPCR using wild type gDNA, mutants are discriminated (when using the default primer design). Since heterozygous mutant cells should also be present in screening experiments, we tested whether completely mutant cells can be correctly discriminated from them. For this experiment, we selected targets for which we had weak signals of in-amplicons when doing cbPCR (Figs), because we thought that these targets might be more difficult to discriminate. We mixed 1:1 wild type and mutant DNA to obtain artificial heterozygous mutant DNA, and did cbPCR. Since we pre- selected targets with weak in-amplicon signals, some bands were difficult to visualize by eye, but a complete absence of in-amplicon was seen only when completely mutant DNA was used. The difference was clearer when we analysed the signals from gel images by Image J. The latter analysis revealed that the signals of in-amplicons tend to be reduced in heterozygous mutants, but less obviously than in completely mutant clones. Thus, the calculated ratios between out- and in-amplicons for heterozygotes were higher than for wild type, but the differences were not as pronounced as when using knockouts. This shows that the increase of amplicon ratios is not linear according to wild type copy number, and that a change in wild type copy number from one to zero causes a bigger shift than a change from two to one. Therefore, the results suggest that this strategy can be used for an easy screening to discriminate completely mutant clones from wild type or partially mutant clones. ## Testing the efficiency of mutant detection Finally, we evaluated the error rate of cbPCR for mutant clone screening. For this, we mutated the HPRT1 gene in HeLa MZ cells using CRISPR/Cas9 and isolated clones. HPRT1-deficient cells are resistant to the toxicity of a nucleotide analog, 6-thioguanine (6-TG). Thus, by comparing 6-TG resistance and cbPCR results, we were able to evaluate how efficiently this method discriminates mutants from the others. This experiment was done using two different guide RNAs targeting different sites for HPRT1, and samples from one experiment served as negative controls for the other experiment (since the mutation sites were distant enough). In one experiment, all 6-TG resistant clones had a complete absence of in-amplicons when analysed with cbPCR. When out- to in-amplicon ratios were calculated and ranked, only 6-TG resistant clones had infinite values due to absent in-amplicons. One 6-TG sensitive clone (clone 19) had an out- to in-amplicon ratio which was intermediate between controls and 6-TG resistant clones. Sequencing analysis showed that this clone had a 2 bp deletion and a wild type allele. In addition, we selected six 6-TG sensitive clones that had ratios comparable to controls for sequencing, and found that all were completely wild type. Thus, for this target, the accuracy of mutant discrimination was 100%, and sequencing also suggested that partial mutants could be detected to some degree. When we did the same experiment with another target site of HPRT1, some 6-TG resistant mutants had detectable in-amplicons, but calculation of the out- to in-amplicon ratio enabled us to correctly discriminate them from the 6-TG sensitive clones. This shows that mutants are not always detected as a complete lack of in-amplicon, but can be discriminated using the ratios between out- and in-amplicons. One exception was a 6-TG sensitive clone (#10) that was ranked as a mutant based on quantification of amplicon ratios. By doing cbPCR with a shorter F-in primer, we could obtain a complete absence of in-amplicon in all the mutants. However, the 6-TG sensitive clone \#10 was still classified as a mutant even with this optimized primer design. From this result, we speculated that clone \#10 was indeed a mutant, but retained an active HPRT enzyme. We did sequencing to detect the pattern of mutations in clone \#10, and found that one allele contained a point mutation without indels. This mutation was silent, leading to a protein with wild type primary sequence. Therefore, clone \#10 was indeed a mutant, but was 6-TG sensitive due to a wild type HPRT protein. We further performed sequencing of all the clones having intermediate out- to in- amplicon ratios (between zero and infinite), as well as four clones each from samples having either zero or infinite ratios. All the analysed clones with zero values were complete wild type. Heterozygous clones (8 and 23) had slightly increased ratios but at a lower degree than for homozygous mutant clones. As a tendency, mutants with bigger indels had bigger changes in ratios, and all mutants with infinite ratios had only large (\>4) indels. Importantly, homozygous mutants with the smallest indels had bigger changes in ratios than the heterozygous mutant clone 23 that had a large deletion, meaning that the size of indels affect less the results than the wild type copy number. As a conclusion, cbPCR had also 100% accuracy (at least for the 6-TG resistant clones and the sequenced clones) for discriminating mutants of this locus, even without the further optimization that we did to obtain a complete disappearance of in- amplicons when doing cbPCR with mutant gDNA. This also shows that cbPCR is sensitive enough to detect even a point mutant. The sequencing results suggested that although the size of indels affects the quantification results, the presence or absence of wild type alleles have a bigger impact, thus enabling the discrimination of complete mutants. # Discussion In this manuscript, we described competition-based PCR (cbPCR) as a method to detect mutant clones after genome editing experiments. The competitive factor in this experiment is very important. If the discrimination was based only on the lack of binding between a primer and its target, the primer (and the annealing temperature) should be designed to give a marginal binding to the target, which is lost even by the smallest indel. When competition is present, this strict optimum is no longer needed, since a decreased affinity is sufficient to give a different outcome. The advantage of this method is that it requires only equipment for PCR and agarose gel electrophoresis (or any other alternatives), thus can be performed in almost all laboratories, and at low cost. The method is so simple that we speculate that similar methods might already be used, even without recognizing that the detection is based on competition between amplicons, and not solely based on loss of inner primer binding. However, existing reviews or online experimental guides (<http://blog.addgene.org/crispr-101-validating-your-genome-edit>) did not describe such a method for clone identification, thus we thought that it would be useful to share the theory and design optimization steps for cbPCR. We provide a protocol to explain the steps for establishing cbPCR for different targets. Although optimization of additional factors such as extension time or cycling numbers (to test whether we need to reach an amplification plateau or not for efficient mutant detection) might be done, these issues were not investigated in this manuscript since the detection efficacy was already satisfactory with the current design. The present method can discriminate homozygous mutants from the other undesired clones, even when the mutation is the same in both alleles. This is extremely important, since mutations induced by CRISPR/Cas9 experiments are not completely random. Indeed, we have found cases where \~60% of the mutants had the same indel at the target site. In such biased cases, many of the mutant clones will have the same mutation in all alleles, and screening of mutant clones with mismatch cleavage assays or heteroduplex mobility assay will result in the misidentification of many mutant clones, unless wild type PCR amplicon is added to all samples. Therefore, cbPCR is easier and might have a higher detection power than these other methods. The success of mutant detection relies on the design of primers that can identify efficiently the mutants. The default setting in our laboratory for primer design was good enough to detect mutants of 7 out of 8 genes. Furthermore, we could easily optimize primer design for the remaining one. Therefore, it is not difficult to obtain primer sets that work efficiently. One constraint for primer design is that the inner primer should overlap with the Cas9 cleavage site. It can be imagined that for some genes, this constraint will lead to the design of primers that are non-specific or that do not bind to the target, due to secondary structure or other reasons. We speculate that the non- specificity issue is not a problem, since we usually design guide RNAs that avoid off-target effects, thus the target should be unique in the genome. For the second issue, as far as we have tested, we did not find an inner primer that could not bind to its target. We routinely use two polymerases (ExTaq or PrimeSTAR GXL) for PCR, and all of the targets have been amplified by at least one of them. The freedom to design two orientations of the inner primer (although F-in orientation is preferable than the other) also reduces the probability to fail in the design of inner primers. In this manuscript, we often used mutant DNA that was generated a priori. From these experiments, we found that whenever the primers are designed based on the default setting and two bands are seen in cbPCR using wild type DNA, mutants can be discriminated. Therefore, even if one does not have mutant DNA a priori (as should be in most cases), a successful primer set could be designed based on cbPCR results using wild type DNA. If the in-amplicon is seen in wild type (and preferably also the out-amplicon, which should demonstrate that it is not overcompeted by the in- amplicon), then the primers should work. It should be noted that even if the primer is not fully optimized, the power of mutant detection by cbPCR is still very high (compare). In this manuscript, only random mutations induced by the widely-used wild type *Streptococcus pyogenes* Cas9 nuclease were analysed. However other genome editing methods exist to induce random mutations, such as Transcription Activator-Like Effector Nucleases, double-nickase using single-mutant Cas9, or CRISPR systems from other species. We believe that cbPCR can detect mutations induced by other methods, as far as the mutation site is predictable. Since some of these genome editing strategies can induce staggered (instead of blunt) double strand breaks, it is possible that in such cases the prediction of where indels happen is more difficult, thus the placement of the inner primer might be more complicated. Indeed, indels generated by a double-nickase strategy could happen in two different sites separated from each other by more than 20 bp, making the design of an inner primer that can overlap with all the mutations impossible. Therefore, the type of double strand break should be considered before designing a cbPCR approach to analyse clones. Another application of genome editing is precise editing based on homology-directed repair using donor DNA. We speculate that for such experiments, we can detect mutant clones with the desired mutation by designing an inner primer having a perfect match with the desired mutant site (and not the wild type site). By doing so, correctly mutated DNA should be detected by an increase in the ratio between in- and out- amplicons, since only correct mutants have efficient amplification of the in- amplicon, which is the inverse case of what is done in this manuscript. This hypothesis still has to be examined. It should be emphasized that the experiment was specifically designed for detection of fully mutated clones. We do not expect this method to be quantitative enough for estimation of mutation rate in polyclonal cells, or for the detection of heterozygous mutant cells, although the results still suggest some degree of quantitation. Indeed, heterozygous mutants could be discriminated by cbPCR at least in all the experiments done in this manuscript, but only a limited number of targets were analysed until now. It should also be noted that some variability in signal ratios is seen between experiments (for example, compare), so it is appropriate to include at least one wild type sample in all experiments as a control, while avoiding comparison of ratio values between experiments done in different days. In addition, we recommend doing sequencing after the screening by cbPCR. Although the experiment with HPRT1 mutants suggests a very low misidentification rate, we have not yet evaluated the misidentification rate in large numbers of target genes. The ease of cbPCR is especially useful when large numbers of clones have to be analysed. Such a requirement might arise when multiplex genome editing is done. In such experiments, one can do cbPCR for all targets first, rank clones based on the ratio of out- and in-amplicons of each target, and sequence only genomic DNA of those cells that are expected to have mutations in all the targets. This will reduce drastically the time and cost needed to obtain the correct clones needed. We could obtain triple knockout cells easily using this strategy. We do not recommend to directly sequence cbPCR products, since the in-amplicon might serve as a megaprimer for the out-amplicon, leading to sequencing reads that are partially wild type, even in completely mutated clones. To conclude, we established a method to discriminate fully mutated clones from the other undesired ones, based on a simple PCR experiment. The method is cost- and time effective, can be performed with standard equipment while being at least equally sensitive to existing methods. # Supporting information McA-RH7777 cells and HeLa MZ cells were kindly provided by Christian Toso (University of Geneva) and Marino Zerial (Max Planck Institute), respectively. The plasmid pX330 was deposited to Addgene (plasmid \#42230) by Feng Zhang (Broad Institute). We thank all the members of the Riezman laboratory for valuable comments. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** TH. **Data curation:** TH. **Formal analysis:** TH. **Funding acquisition:** TH HR. **Investigation:** TH. **Methodology:** TH. **Project administration:** TH HR. **Resources:** TH HR. **Supervision:** HR. **Validation:** TH HR. **Visualization:** TH HR. **Writing – original draft:** TH. **Writing – review & editing:** TH HR.

# Introduction The riding posture is very important because an awkward riding posture may cause discomfort to the riders and increase the risk of developing musculoskeletal disorders (MSD) in long-term exposure. De Cássia et al. stated that body discomfort is an indicator of many MSD due to prolonged discomfort. A sustained flexion of lumbar lordosis (low back) during driving usually is the main factor in the generation of low back pain. According to Makhsous et al., sitting decreases lumbar lordosis compared to the standing posture, resulting in increased disc pressure and low back muscle activity. This is because while sitting, the ischial tuberosity mainly supports the majority of the upper body weight. Increased pressure in this area is significantly associated with increased spinal load. Riding a motorcycle exposes the officers to excessive physical demand, especially for prolonged periods. They will tend to feel discomfort and possibly fatigue while sitting in the same position with restricted movements. Thus, to improve spinal posture and comfort in transportation settings, seat design has a significant role and efficiency. A well-designed seat would decrease pressure within the spinal discs, spinal ligaments, and gluteal muscle. Lumbar support is the solution for the problem, whereby the weight and pressure of the trunk are taken by the back support and increased lumbar lordosis. The use of a vehicle seat (lumbar support) has been significantly associated with the reduction of muscle discomfort and low back pain during prolonged riding and driving journey. Adjustable lumbar support with extra cushioning, which provided a massage-like comfort, has been proved to reduce transmission of vibration, which indirectly increases muscle comfort by improving oxygen and blood flow to the tissue. According to Mansfield et al., the seat shape, user suitability, seat material, duration of sitting in the same position, vibration, and posture changes are among the factors that influence seat comfort. Traffic police is a police officer who serves to enforce the rules related to traffic. They have to perform various work tasks, such as escorting important person to any events, patrol selected locations, and finding any road offenders. All this work task requires them to use a motorcycle as their vehicle. At present, the officer riding a motorcycle takes at least five hours per day during duties. Therefore, traffic police riders have to endure prolonged riding duration. Rashid et al. believed that continuous riding of a standard motorcycle for an extended time would result in a high level of postural fatigue and health problems. From a previous research study on the CBX 750P21 motorcycle, it was found that 88.3% of traffic police riders suffered from MSD with 34.3% of them suffering from lower back pain due to static posture and prolonged sitting while riding the motorcycle. Another previous study also revealed that more than half (54.7%) of the traffic police riders rode high-powered motorcycles for an average of 5.64 hours per day with a fixed posture leading to increased discomfort from prolonged sitting that enhanced muscle fatigue. This showed that most of their working time involves riding motorcycles. However, to the best of the authors’ knowledge, no research has been done on the spinal riding posture during a motorcycle ride, especially among police officers in Malaysia who use a motorcycle to perform their duties daily. To solve this problem, ergonomics is the best solution for scientific research in man-machine interaction at the workplace since this field involves fitting machine to workers comfortability to improve their working performance, reduce fatigue, and stress. The ergonomic application is very significant in areas involving prolonged and static riding activities that directly affect the riders’ healthy spinal posture and reduce muscle fatigue. Lumbar support with a built-in massager system is an intervention in this study to solve the ergonomic issue among motorcycle riders. The use of a back (lumbar) support has been proved to help maintain the natural spine lordotic lumbar curvature of the person while sitting compared to those without back support. In a clinical and laboratory setting, automobile seat massage is utilised widely, and it has been proved efficacious in recovering from postural fatigue. However, research and data on the effectiveness of this intervention seat, lumbar support with a built-in massager system, in an in-field setting (on-the-road) are insufficient. Thus, the present study integrates a massager system and lumbar support with the existing seat which an on-road test was conducted to maintain and support the spinal posture of the riders more efficiently as the test involved prolonged and static riding. In this study, the spinal angle profile was measured and assessed to determine the effectiveness of theprototype seat in supporting the spinal body posture of the riders throughout 20 minutes motorcycle ride. Thus, the present study seeks to evaluate the effect of lumbar support with a built-in massager system (prototype seat) among traffic police riders. # Materials and methods ## Study design A randomised controlled trial, pretest-posttest control group design was conducted among 24 traffic police riders who ride a high-powered motorcycle (Honda CBX 750). A simple random sampling was used in this study which the subjects were randomly assigned to the control group (12 riders) and experimental group (12 groups). Data collection commenced in March 2020 and finished in July 2020. ## Eligibility criteria Eligible subjects were male riders only because almost 90% of traffic police riders are male. BMI was between 18.5 to 29.9 kg/m². The age group recruitment was between 20 and 39 years old because more than 40 years old are usually prone to experiencing low back pain due to the age factor and changes in spinal posture. At least one year of experience in high-powered motorcycle riding. Exclusion criteria were the presence of any injury (under treatment or taking any medication related to muscle pain), especially lower back pain, in the 12 months preceding this survey. ## Sample size To calculate the appropriate sample size, we use the formula of group comparison guidelines in a study done by Donnelly et al. as a reference. The estimated SD (30.2), estimated larger mean (57.2) and estimated lower mean (19.0) were used in this formula. In this study, the desired power is 80% and the significant level is 95%. Thus, each experimental and control group would have 10 subjects. However, an additional 20% dropout rate was added in the sample size calculation. Thus, each experimental and control group will have 12 subjects. Therefore, a total of 24 subjects was selected in this study. ## Participant recruitment The recruitment strategy involved taking the name list of all officers working at the Kuala Lumpur Traffic Police Station. The name lists were obtained from the Human Resources Department. However, 97 officers from the escort department only were chosen to ensure the consistency of data in which different department have different task or job and the duration of riding police’s motorcycle. Only 24 respondent who fulfilled the criteria were selected for the study. For qualification criteria, respondents were asked to fill up the pre-survey form and the BMI were measured by the main researcher. After the eligible subjects were selected, then the main researcher randomly assigned 24 of them into control and experimental group. This study’s sampling method was simple random sampling, in which all the subjects’ names were numbered into pieces of paper and put in a container. Then, the subjects were chosen randomly into the experimental or control group by Fishbowl Technique. Each subject in the experimental group received lumbar support with a built-in massager system during post-test. The pre-test and post-test were conducted at a one-week interval. Detail about study flow was illustrated in the CONSORT diagram. Subjects (n = 12) allocated into the control group had undergone post-test session without the lumbar support with a built-in massager system (existing seat) during 20 minutes riding session. On the other hand, the experimental group (n = 12) had undergone a post-test session with the lumbar support with a built-in massager system (prototype seat) for 20 minutes riding session. Before the experimental session began, subjects were allowed to choose the Truposture smart shirt that fits their body. This was done to ensure the point of each sensor would be placed in the correct position of the spine. There were no cases of unable to follow up or incomplete experimental testing as planned in this study. All subjects completed the follow-up stage (n = 24) in both intervention groups. This study had no excluded analysis cases as all the data were successfully recorded without any missing data. ## A prototype The lumbar support on the prototype seat can be adjusted either upward and downward according to the lumbar height position for the comfortability of the riders. There is also a massage fitted inside the lumbar support with two rotating balls. The lumbar massage was set at a ratio of 1 minute on and 4 minutes off. It was used for 20 minutes in the experimental riding session. It can support up to 150 kg body weight. ## TruPosture A spinal posture angle was obtained using the TruPosture Smart Shirt connected with the TruPosture mobile app among the police riders while riding a motorcycle with an existing seat for both groups; control and experimental during the pre- test. Then, during the post-test, the control group used the same motorcycle seat, and the experimental group used a prototype seat containing adjustable lumbar support with a build-in massager system. The TruPosture Smart Shirt covered different regions of the spinal alignment, with five nano-sensor technology attached undershirt. The five sensors cover the thoracic vertebrae, lumbar, and the pelvic regions (T1, T8, L1, L3, and Pelvis). The technology helped the researcher monitor the entire spinal’s alignment and curvature in several modes, including standing and sitting. The TruPosture mobile apps interface and Windows software were utilised as these tools help the researcher track the posture and record the movement of posture in real-time. The sensors detect the real-time changes of spinal posture angle based on the spinal movement. Based on, the blue curve represents the spinal reference posture (ideal posture) at 0 minutes, and the orange curve indicates the actual spinal riding posture throughout 20 minutes. The number on the left and right side of the interface represents the angle reading for each spinal point sensor. The positive and negative values of angle indicate the spinal position, which is forward and backward respectively. The validity of the equipment had been tested and approved by the previous study which was suitable for monitoring the posture of the spine. The reason for choosing this method is that the spinal sitting pattern studies were limited to laboratory and clinical settings only. According to Ma’arof et al., the research field provides vital outcomes and real-world motorcycling information. Thus, TruPosture Smart Shirt is the most suitable method to monitor the spinal change pattern among motorcyclists on-road with the application of real working conditions. ## Data collection In this study, the spinal reference posture at the 0 minutes riding session was set first. The subjects wore the TruPosture Smart Shirt during the monitoring, according to the ideal riding posture. An upright riding posture is the best posture in riding this type of motorcycle based on the standard operating procedure (SOP) of Honda CBX750P motorcycle riding. According to Ma’arof et al., a good riding posture for the upright posture is that the arms should bend slightly when gripping the throttle and sitting in the upright position. The shoulders and elbows are held easily on the holds without exceeding or over- expanding the elbows. Elbows are flexed, and lower arms are parallel to the ground. Furthermore, the legs are held near the fuel tank while relaxing the hip and pelvis. This posture was set as a spinal reference posture for riding, and the angle for each sensor was recorded. However, this is the real problem for the riders, as they could not maintain the ideal posture in medium- and long- distance trips. Thus, the subjects were asked to ride a motorcycle for only 20 minutes. This is because the average commute time of traffic police riders for one task is less than 30 minutes. Moreover, According to Deros et al., 82.5% of the discomfort variance divergence is accounted for after being seated for 20 minutes and it takes less than five minutes in postural changes without back support. The riding posture was assessed by the same person (an ergonomist) in the pretest-posttest session. This study was conducted at the MEX highway route (the route between the commercial centre of Kuala Lumpur to the Federal Administrative capital of Putrajaya and Kuala Lumpur International Airport), with good road condition and maintenance. The route was chosen as it was commonly used in their work task as escort riders. The pre-post test results were then compared between the two groups. ## Quality control A pre-test was conducted approximately 10% of the sample size. Two respondents who had undergone pre-test were not included in the real experimental testing as it was not to prove the superiority of the treatment but to test the procedures and processes and estimate parameters for the main trial sample size calculation. Other than that, the pre-test could ensure that all the instruments were working properly and in good order. The other purpose of this study was to familiarise the researcher with the placement of sensors of Truposture smart shirt to avoid any mistakes during the data collection process. To ensure a good quality control, the supplier had demonstrated the handling of Truposture smart shirt to the researcher and the possible errors that might occur. The measurement procedure used was obtained from the Truposture smart shirt Manual Book. The respondents must ensure that the size of the shirt was fitted with their body to make sure the sensor placed aligned their spinal correctly. Then, the respondents need to ride a motorcycle for 20 minutes. Each test was taken around 10:00 am to 10:20 am, to ensure consistency of traffic conditions. The measurement was taken for 20 minutes because the average time taken by traffic police riders in riding motorcycle was approximately 20 minute per task. ## Statistical analysis The collected data from the questionnaire and measurement were analysed using the IBM SPSS Software (Version 26). The Shapiro-wilk test was used in this study to determine the normality of data distribution for variables. The normality of data distribution was assumed when the p-value was more than 0.05. The present study found that the data variables were not normally distributed. Thus, a non- parametric test was used in this study. The data collected were analysed using the Mann-Whitney test and Wilcoxon signed-ranked test to verify a statistically significant difference between and within the control and experimental groups. The dependent variable of this study was the spinal angle (0^th and 20^th minutes). The measurement of the spinal angle was done by using Truposture smart shirt. The study was conducted using a 95% confidence level, 80% of power, and the results of p≤0.05 were considered significant. ## Ethics approval This study was submitted and approved by the Ethics Committee, Universiti Putra Malaysia (reference number: UPM/TNCPI/RMC/JKEUPM/1.4.18.2 (JKEUPM)). Permission from the subjects of this study was obtained with their written consents before the study was conducted. Privacy of the information and confidentiality of the subjects were and are always protected. # Results ## Subjects In total, 24 traffic police riders were included in the study, and there were no drops out. All subjects enrolled completed the study protocol as planned. The subjects’ demographic data from both the control group and the experimental group are tabulated in ; no significant differences were observed between the two groups. ## Distribution of five different sensors for spinal posture angle The mean spinal posture angle in each sensor for the control and experimental groups is presented in. The graph showed that the experimental group (post-test) values are successfully maintained with minimal changes to the spinal posture angle compared to the other groups. The positive and negative values of angle indicate the position of the spinal, which is forward and backwards, respectively. Meanwhile, the line graph in show the trend comparison of spinal posture angle deviation throughout the 20 minutes riding session. The line graph compares the level of spinal angle deviation in four groups (control pre-test, experimental pre-test, control post-test, and experimental post-test). Based on the graph line comparison, a deviation of spinal angle increased in each sensor, and Sensor 1 experienced the greatest deviation from the other sensors. From the graph, a deviation of spinal angle on each group shows an upward trend. However, the spinal angle deviation in the experimental (post-test) group is always at a lower level than the other groups. It showed that the experimental group intervention has a huge improvement in maintaining posture, which only involves slight spinal angle deviation changes from the spinal reference angle throughout the 20 minutes ride. The details of the spinal posture angle profile throughout 20 minutes are available as supporting information;. ## Comparison of five different sensors compares the spinal angle (0^th and 20^th minutes) between pre-test and post-test study for five different sensors within two groups. As depicted in this table, a Wilcoxon signed-rank test revealed a statistically significant difference between pre-test and post-test session in all sensors at the 0^th and 20^th minutes in the experimental group (p\<0.05). Meanwhile, no statistically significant difference is observed among the control group in all five sensors (p\>0.05). The results obtained from a Mann-Whitney test showed that there was no statistically significant difference in spinal angle between the control group and experimental group in the pre-test session as shown in. However, there were statistically significant differences in the spinal angle between control and experimental groups for all sensor (p\<0.05) throughout 20 minutes of a motorcycle ride. # Discussion The present study found that the existing seat contributes to the huge changes in Sensor 1 (Thoracic 1) throughout 20 minutes of riding for both groups. This could be explained when the riders applied a slumped posture at the end of the 20 minutes of riding due to muscle fatigue while adopting an upright posture without back support. Kwon et al. explained that the thoracic angle is usually related to the outcome of a slumped posture. These findings were supported by Ma’arof et al., who reported that an upright riding posture involved higher muscular activity compared to the forward-lean posture, which makes riders muscle less comfortable, leading to fatigue. Shoulder stiffness is frequently encountered by motorcyclists, including young and healthy riders. Therefore, officers using existing seat proved that they could not maintain an ideal and good posture while riding a motorcycle, indicating the leading cause of concern for this problem. The application of lumbar support with a built-in massager system showed a lower deviation angle in the trend compared to the group without intervention. This happened because the upright posture without back support (exiting seat) leads to the rider’s unnatural spinal curve. This can be explained by the fact that this prototype helps correct the spinal posture (upright) of motorcyclists better than the absence of lumbar support. These findings are parallel with the previous research that reported the lumbar support on a motorcycle seat was capable of providing good posture and reduce muscle discomfort of riders during the riding process. This is supported by Ceunen et al., who stated that the upright sitting posture with lumbar support is efficient to change pelvis and spinal structure to their natural curve during sitting, which reduces the load on the ischial tuberosity and lower spinal, reduce muscular activity, maintained lumbar lordosis; thus, potentially reduce the risk of developing low back pain. Alyami and Albarrati also found that the workstation would be safer when the posture’s biomechanical risk could be reduced by applying ergonomic rules in design that would support healthy (ideal) body posture. A statistically significant difference was noted throughout this research in all sensors at the 0^th, and 20^th minutes (p\<0.05) between pre-test and post-test measurement within the experimental group. However, there was no statistically significant difference in spinal angle within the control group. This proved that the intervention of the prototype seat in the current study could effectively maintain spinal posture in an upright riding position throughout 20 minutes riding session. The value of spinal angle deviation between the pre-test and post-test in the experimental group throughout 20 minutes of riding also showed that the slumped spinal posture could be prevented with lumbar support and a built-in massager system. The analyses between-group differences also found that there was a significant difference in all spinal angle between control and experimental groups throughout 20 minutes of a motorcycle ride. The findings in the current study proved that the prototype seat provides a positive effect on the spinal posture of motorcyclists in adjusted and maintaining their spinal posture which can serve as a channel for the distribution of the force as well as static loading by the body. In this case, the intradiscal pressure could be reduced and the supporting back muscle could be enhanced as shown in. Another crucial finding in this study involved applying the massager system (4 minutes off to 1 minute on), whereby the results of the experimental group showed relatively not much deviation compared to the other groups. The value of spinal angle deviation in the experimental group was reduced after every four minutes. This could be possibly due to the massager in the lumbar system improves spinal posture from an awkward posture (slump) over time. These results proved the theory that the massager in the lumbar area could reduce muscle fatigue, increase blood flow and oxygenation, and maintain body posture effectively. Contrarily, Tanaka et al. reported differences that suggested the lumbar massage failed to exert significant changes in any electromyographic (EMG) measurement. Besides that, Van Poppel et al. reported that lumbar support was not useful in preventing low back pain. The outcome of this study contradicted the study by Kolich et al., where they pointed out that one minute of lumbar massage in every five minutes of the driving session has a beneficial effect on the lower back muscle activity. A study conducted by Franz et al. suggested that lumbar support usage with lightweight massager in automobile seats has been reported to decrease muscle discomfort during the driving process and improve lumbar lordosis. There were several limitations in the present study. Only male traffic police riders were recruited. Thus, it cannot be used to generalise the whole population. A future study might include female traffic police riders and other occupations such as food delivery. Even though we found a positive outcome of this prototype in maintaining spinal posture, since this study was conducted in an on-the-road setting, there was a limited objective measurement that can be done such as the measurement of muscle fatigue using EMG electrode (clinical testing) due to environmental factors such as road condition and vibration. This study was unable to blind the respondents to the condition that they were exposed to, which might impact the posture riding in this study. This is because traffic police riders were aware of which seat they were using, and as such, there was no way to prevent a bias toward the prototype seat. However, it is assumed that any bias towards the motorcycle seat prototype would disappear through the riding duration if the seat did not truly adjust and maintain riding posture throughout the riding process. Due to time constraints (pandemic outbreak) and limited resources such as budget, it was not possible to include all aspects of the problem with a bigger sample size in this research. Besides that, there was only one prototype available for this research. # Conclusion In summary, the present study proved that lumbar support with a built-in massager system successfully maintains and improves the spinal posture angle ergonomically throughout 20 minutes of riding. Although there is a positive outcome in this research, there is still a lack of evidence (i.e. muscle fatigue, discomfort, muscle activity) for this study. Hence, further study in the artificial laboratory and clinical settings as well as virtual testing is needed to support these findings. # Supporting information The author would like to thank all of the people who were involved in this study especially the Royal Malaysian Police (RMP). 10.1371/journal.pone.0258796.r001 Decision Letter 0 Mosa Ahmed Mancy Academic Editor 2021 Ahmed Mancy Mosa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 27 May 2021 PONE-D-21-13337 Assessment of spinal angle profiles among high-powered traffic police motorcycle riders PLOS ONE Dear Dr. Karuppiah, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please consider all the comments of all reviewers including the comments of Reviewer 3 Please submit your revised manuscript by 30 May 2021. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Ahmed Mancy Mosa, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: No Reviewer \#3: No Reviewer \#4: No Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No Reviewer \#4: No Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes Reviewer \#4: No Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: Yes Reviewer \#3: No Reviewer \#4: No Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Dear Authors, This is an interesting research area. While the study has some merits to consider for publication several clarifications are needed to assist in the decision. My comments and questions are as follow: 1\. What is the new contribution of this study? What was previously known about the ergonomic function and effect of the prototype seat, and how will this study add to the existing knowledge? Why is it important to measure the spinal angle profile? 2\. The problem statement highlighted in the introduction does not match the aim of the study which is an experimental study comparing between an intervention and control group. 3\. Please clarify whether there is any sample size calculation undertaken prior to the study, and provide the relevant references for the sample size calculation. What was the power of the study based on the sample size calculation? 4\. Please provide adequate references for the prototype and the TruPosture app. 5\. Please provide a detailed explanation on the recruitment process, and the allocation to control and intervention group. 6\. What was the dependent variable in the statistical analysis. please provide the operational definition of the dependent variable. 7\. Please provide the general description of your study participants in the results section, and describe the characteristics between the experimental and control group to show that they are comparable. 8\. Do you expect any confounding factors in your final analysis to look at the difference in outcome between the intervention and control group? What are the appropriate statistical tests to be used to control for the confounders, if there is any? 9\. please explain clearly the within and between groups comparison in the final analysis of the study outcome. 10\. Please align the discussion with the aim of the manuscript. What is it that you plan to highlight? is it about the usability of the TruPosture smart Shirt in measuring posture, or it is about showing the effect of the prototype seat on posture of the police riders? 11\. What is the study limitation? 12\. Please revise the manuscript title to reflect on the aim and experimental design. Reviewer \#2: 1. SUMMARY OF THE RESEARCH AND OVERALL IMPRESSION – The manuscript is well written with concise language, but I have major concerns on the methodology that prevent me to endorse acceptance at current stage. I see that sound methodology is lacking and that compromise the data analyses and could have led to potentially wrong results and misleading conclusion. Recommended course of action is to re-look at the methods and try to conform to standard reporting guidelines such as CONSORT guideline. The authors can expand the study sample size, with proper calculation and reference to previous study to ensure its adequate to draw to solid conclusion. Current study can serve as preliminary or pilot study to come up with better technical standard of experimental methodology and sufficient description on the details. 2\. DISCUSSION OF SPECIFIC AREAS FOR IMPROVEMENT – a\) Methodology (major issues) There are several issues with the methodology and analysis that need to be clarified, address and described in great detail. The methodology section should be expanded and clarified to support the conclusion and validate the findings. As the methods section is lacking, it might be premature to draw sound and valid conclusion. • There is no mention on the specific study design and how the randomisation is done (authors only mentioned randomisation once in Figure 2). Is this a randomised controlled trial? Why is it not mentioned in the title? Has it been registered in any trial’s registry? • There is no mention of how the sample size is calculated. How is the sample size being determined? Is the sample size adequate to represent the study population? Is the sample size adequate to meet the assumption of statistical analyses and come to conclusion? • Eligibility criteria is not clear (authors only mention no history of MSD and low back pain, but what about baseline age, height, weight, BMI, fat percentage, gender, ethnicity, years of service, and other factors that might influence the spinal angle profile?). • There is no mention on how the recruitment is done. Is it convenience? Can it be representative of the study population? Who enrolled the participants? • How is the allocation to the control and intervention group being done? Is there any blinding involved? What are the methods used to do randomisation and type of randomisation (when allocating participants to control or intervention group)? The authors did not mention who did the randomisation and who assigned the participants interventions. • There is no mention of the time period of data collection defining recruitment, pre and post-test. How long is the gap between pre and post-test? What specific time of the day the experiment being conducted? (is the timing and duration of work on that particular experiment day influence the spinal angle profiles?) What is the justification for 20 minutes duration of the experimental riding session? (authors mention riders spend at least 5 hours per day on motorbikes). • The author did not mention anything on the instrument validation and calibration. How are the sensors being placed? Is it done by the same person for both control and intervention groups? Has the apps being tested and validated before? How accurate is the reading? • The author did not mention whether both groups are being assessed by the same person (e.g. posture for riding). If it’s not done by the same person, do the researchers take into account inter-rater reliability? b\) Discussion There is no mention or any discussion on the experiment’s limitations in the discussion part. The authors did not discuss potential source bias and confounder, threat to validity that might compromise the findings. c\) Results • The authors did not report the sociodemographic and baseline characteristics of both control and experimental group. How does the researcher ensure that both groups are similar at baseline? How can the researcher then conclude that the outcome is due to their intervention, rather than existing differences? Are there any additional methods of analyses such as subgroup analyses to account for confounder or differences in the baseline? • There is no justification on why non-parametric statistical analyses was chosen. What are the assumptions and limitation of the analysis? Have all the criteria being met? • Too much data being presented that can be summarised in a sentence or in simpler table. For example, Table 1 presents a range of minimum and maximum value for each reading throughout 20 minutes – can be summarised by providing the mean/median value. Easier to interpret and understand, rather than the reader have to go thru each range. d\) Others • Figure 1 – no legend to help reader understand the figure. Which line belongs to control and intervention group? • Figure 2 – suggest authors to follow CONSORT flow diagram format (more details) 3\. OTHER POINTS Authors have provided adequate literature review to justify the problem statement, significance of the study and burden of the disease. The authors also adequately described previous research and gap of the study. The data presented has potential to be published if its extended and properly developed. Current study can serve as preliminary findings or pilot to come up with better protocols and larger sample size to test the same hypotheses and draw more concrete results and conclusion. Reviewer \#3: Thank you for the opportunity to review your manuscript entitled “Assessment of spinal angle profiles among high-powered traffic police motorcycle riders”. The topic is interesting; however, I believe the study was not appropriately designed to address the research question. The analysis and reporting should not lead to the conclusion made by the authors. The introduction needs to be strengthened. I have highlighted some points that I believe would improve the quality of the manuscript. Abstract Line 30: the purpose of the study described in the abstract does not match with what was done in the methods. This was an intervention study and I suggest reformatting the purpose of the study. Line 40: Are we really interested in a pre-test VS post-test analysis (within group analysis)? Reporting the between group comparison is more insightful… Introduction Provide some statistics of MSD/low back in the population Line 51: provide a full meaning of MSD Line 59: I believe it was meant to be “been” instead of “seen” Line 61: has instead of have Line 67: delete “the” Line 81: The purpose of the study needs to be reformulated The introduction lacks content. The consequences of increased lumbar lordosis are not fully described. There are some grammatical errors that need to be corrected. Materials and Methods More information on the eligibility criteria (inclusion and exclusion criteria) is needed Any reference for the TruePosture mobile app? Has this app been used before in any study? What are the validity and reliability properties of the app? How were the groups defined? Were the participants randomly assigned to the groups? What are the baseline characteristics of the control and experimental groups? Why did the authors not perform a between-group analysis? That comparison is more interesting than all those pre-test vs post-tests performed. Results Data on the participants missing The data reported does not indicate whether the spinal change pattern in the intervention group is superior to the control group. Discussion and conclusion Not sustained by the analysis and results presented Reviewer \#4: Abstract Background Context and research gap were not indicated Materials and Methods Type of study, sampling method and data collection method were not specified Major statistical analysis was not stated Main body Materials and Methods It has major methodological defect. Type of study, sampling method and data collection method were not specified Gold standard of experimental study was not succinctly stated. Outcome was not assessed Major statistical analysis was not indicated. Why Wilcoxon signed rank test and median was used? Reviewer \#5: Line 51: Please indicate the full meaning of MSD when been used for the first time in write up Line 178: Consider replacing "respondents" with "subjects" which best suits the study and its concept Please give reasons for the sample size choice and indicate the exclusion and inclusion criteria for participating in the study Line 243: "The value of spinal angle deviation between the pre-test and post- test....." Indicate the angle of deviation to make your point clear There are some typographical errors indicated in the attached document. Please revise them accordingly \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No Reviewer \#4: No Reviewer \#5: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0258796.r002 Author response to Decision Letter 0 4 Jun 2021 Dear Dr Ahmed Mancy Mosa (Academic Editor), Thank you for giving me the opportunity to submit a revised draft of my manuscript titled Assessment of spinal angle profiles among high-powered traffic police motorcycle riders. We appreciate the time and effort that you and the reviewers have dedicated to providing your valuable feedback on my manuscript. We are grateful to the reviewers for their insightful comments on our paper. We have been able to incorporate changes to reflect most of the suggestions provided by the reviewers. We have highlighted the revisions within the manuscript. The line mention in the author response. Here is a point-by-point response to the reviewers’ comments and concerns. Reviewer 1 1\. What is the new contribution of this study? Thank you for pointing this out. The new contribution of this study is: • The application of the method (Truposture smart shirt) in spinal profile measurement was evaluated in an occupational setting (on-the-road) which riding in a real working condition was applied. • In motorcycling industry, new invention for an ergonomic motorcycle seat had been developed and tested. 2\. What was previously known about the ergonomic function and effect of the prototype seat, and how will this study add to the existing knowledge? Why is it important to measure the spinal angle profile? We agree with this comment. Therefore, the relevant justification has been made in the introduction section (line 92- 99). 2\. The problem statement highlighted in the introduction does not match the aim of the study which is an experimental study comparing between an intervention and control group. We agree with this comment and has made changes accordingly in introduction section (line 94-96). 3\. Please clarify whether there is any sample size calculation undertaken prior to the study, and provide the relevant references for the sample size calculation. What was the power of the study based on the sample size calculation? You have raised an important point here. We agree with this comment and the sample size calculation has been added in sample size section (line 122-129). 4\. Please provide adequate references for the prototype and the TruPosture app. Adequate references for the prototype and Truposture app have made (line 186-189). 5\. Please provide a detailed explanation on the recruitment process, and the allocation to control and intervention group. Thank you for pointing this out. The recruitment process has been explained accordingly (line 131-158). 6\. What was the dependent variable in the statistical analysis. please provide the operational definition of the dependent variable. The dependent variable and the operational definition have been added in the statistical analysis as suggested (line 223-225). 7\. Please provide the general description of your study participants in the results section, and describe the characteristics between the experimental and control group to show that they are comparable. Thank you for pointing this out. The general description of the participants has been discussed and tabulated in Table 1 (line 234-242). 8\. Do you expect any confounding factors in your final analysis to look at the difference in outcome between the intervention and control group? What are the appropriate statistical tests to be used to control for the confounders, if there is any? The confounding factors in this study was expected with the difference age, gender, BMI and present of low back pain. However, these criteria were controlled based on inclusion and exclusion criteria during sampling (line 144-121). 9\. please explain clearly the within and between groups comparison in the final analysis of the study outcome. Thank you for pointing this out. We agree with this comment and has made changes accordingly in results and discussion section (line 316-330). 10\. Please align the discussion with the aim of the manuscript. What is it that you plan to highlight? is it about the usability of the TruPosture smart Shirt in measuring posture, or it is about showing the effect of the prototype seat on posture of the police riders? Agree. We have, accordingly, revised the discussion part and removed the unrelated part. 11\. What is the study limitation? Thank you for pointing this out. We have added the study limitation accordingly in the discussion section (line 350-359). 12\. Please revise the manuscript title to reflect on the aim and experimental design. We agree with this comment. Therefore, we have revised and changed the manuscript title into “Effectiveness of lumbar support with built-in massager system on spinal angle profiles among high-powered traffic police motorcycle riders: A randomised controlled trial” Reviewer 2 1\. • There is no mention on the specific study design and how the randomisation is done (authors only mentioned randomisation once in Figure 2). Is this a randomised controlled trial? Why is it not mentioned in the title? Has it been registered in any trial’s registry? Thank you for the input. Yes, it is randomised controlled trial. However, we apologise for overlooking this matter. Thus, the title and study design have been revised as suggested (line 108-112). It has not been registered in any trial’s registry. We are sorry for this and hope for your understanding. Nevertheless, the sampling method is approved and supported by the Ethical Committee of Universiti Putra Malaysia after consulting their expertise. 2\. • There is no mention of how the sample size is calculated. How is the sample size being determined? Is the sample size adequate to represent the study population? Is the sample size adequate to meet the assumption of statistical analyses and come to conclusion? You have raised an important point here. We agree with this comment and the sample size calculation has been added (line 122-129). 3\. • Eligibility criteria is not clear (authors only mention no history of MSD and low back pain, but what about baseline age, height, weight, BMI, fat percentage, gender, ethnicity, years of service, and other factors that might influence the spinal angle profile?). You have raised an important point here. We agree with this comment and has included relevant information and explanation accordingly (line 114-121). However, due to time (pandemic outbreak) and budget constraints (e.g: fat percentage analyser), we are unable to include all aspects of the problem and other factors. Thank you for pointing this out. For future studies, this will be a great insight and we will be sure to use it. Also, we will highlight this matter as the limitation in this study. 4\. • There is no mention on how the recruitment is done. Is it convenience? Can it be representative of the study population? Who enrolled the participants? Thank you for pointing this out. The participant recruitment has been included and explained accordingly (131-158). 5\. • How is the allocation to the control and intervention group being done? Is there any blinding involved? What are the methods used to do randomisation and type of randomisation (when allocating participants to control or intervention group)? The authors did not mention who did the randomisation and who assigned the participants interventions. Thank you for pointing this out. The participant recruitment has been included and explained accordingly (131-158). 6\. There is no mention of the time period of data collection defining recruitment, pre and post-test. How long is the gap between pre and post-test? Thank you for pointing this out. The participant recruitment has been included and explained accordingly (line 144). 7\. What specific time of the day the experiment being conducted? (is the timing and duration of work on that particular experiment day influence the spinal angle profiles?) What is the justification for 20 minutes duration of the experimental riding session? (authors mention riders spend at least 5 hours per day on motorbikes). You have raised an important point here. We agree with this comment and have included the relevant information and justification in data collection section (line 205-209). 8\. • The author did not mention anything on the instrument validation and calibration. How are the sensors being placed? Is it done by the same person for both control and intervention groups? Has the apps being tested and validated before? How accurate is the reading? We agree with this comment. Therefore, we have discussed this information in TruPosture section (line 186-192). 9\. • The author did not mention whether both groups are being assessed by the same person (e.g. posture for riding). If it’s not done by the same person, do the researchers take into account inter-rater reliability? You have raised an important point here. Thus, the relevant information has been added accordingly (line 209-210). 10\. b) Discussion There is no mention or any discussion on the experiment’s limitations in the discussion part. The authors did not discuss potential source bias and confounder, threat to validity that might compromise the findings. Thank you for pointing this out. We have added the study limitation accordingly (line 350-364). 11\. c) Results • The authors did not report the sociodemographic and baseline characteristics of both control and experimental group. How does the researcher ensure that both groups are similar at baseline? How can the researcher then conclude that the outcome is due to their intervention, rather than existing differences? Are there any additional methods of analyses such as subgroup analyses to account for confounder or differences in the baseline? Thank you for pointing this out. The general description of the participants has been discussed and tabulated in Table 1 (line 234-242). 12\. • There is no justification on why non-parametric statistical analyses was chosen. What are the assumptions and limitation of the analysis? Have all the criteria being met? We agree with this comment and has made changes accordingly in statistical analysis section (line 218-223). 13\. • Too much data being presented that can be summarised in a sentence or in simpler table. For example, Table 1 presents a range of minimum and maximum value for each reading throughout 20 minutes – can be summarised by providing the mean/median value. Easier to interpret and understand, rather than the reader have to go thru each range. Agree. We have removed this table accordingly and the mean value of spinal posture angle changes throughout 20 minutes riding for both groups had already presented in Figure 3 (line 250). 15\. d) Others • Figure 1 – no legend to help reader understand the figure. Which line belongs to control and intervention group? Actually Figure 1 is an example of Truposture mobile apps interface and not related with the control and intervention group results. • Figure 2 – suggest authors to follow CONSORT flow diagram format (more details) Thank you for your suggestion. Thus, Figure 2 has been changed and followed CONSORT format (Fig1: Line 146). Reviewer 3 1\. Abstract Line 30: the purpose of the study described in the abstract does not match with what was done in the methods. This was an intervention study and I suggest reformatting the purpose of the study. The purpose of the study has been reformulated accordingly (line 30-31). 2\. Line 40: Are we really interested in a pre-test VS post-test analysis (within group analysis)? Reporting the between group comparison is more insightful… Thank you for pointing this out. Thus, results and discussion related to comparison between groups have been added accordingly 3\. Introduction Provide some statistics of MSD/low back in the population The statistics of MSD and low back pain among traffic police riders have been provided as suggested. (line 78-85) 4\. Line 51: provide a full meaning of MSD The full meaning of MSD has been provided as suggested (line 50). 5\. Line 59: I believe it was meant to be “been” instead of “seen” Thank you for the comment. The relevant word has been changed accordingly (line 104-106). 6\. Line 61: has instead of have Thank you for the comment. The relevant word has been changed accordingly. 7\. Line 67: delete “the” Thank you for the comment. The relevant word has been deleted accordingly. 8\. Line 81: The purpose of the study needs to be reformulated The purpose of the study has been reformulated accordingly. 9\. The introduction lacks content. The consequences of increased lumbar lordosis are not fully described. There are some grammatical errors that need to be corrected. Thank you for pointing this out. The consequence of decreased lumbar lordosis has been explained accordingly (line 52-60). 10\. Materials and Methods More information on the eligibility criteria (inclusion and exclusion criteria) is needed You have raised an important point here. We agree with this comment and has included relevant information and explanation accordingly (114-121). 10\. Any reference for the TruePosture mobile app? Has this app been used before in any study? What are the validity and reliability properties of the app? Adequate references for the Truposture app have been added and discussed (line 186-189). 11\. How were the groups defined? Were the participants randomly assigned to the groups? What are the baseline characteristics of the control and experimental groups? We agree with this comment. Thus, the relevant information has been added and explained accordingly. 12\. Why did the authors not perform a between-group analysis? That comparison is more interesting than all those pre-test vs post-tests performed. Thank you for pointing this out. Thus, results and discussion related to the comparison between groups have been added accordingly. 13 Results The data reported does not indicate whether the spinal change pattern in the intervention group is superior to the control group. Thank you for pointing this out. Thus, results and discussion related to the spinal change pattern between groups have been added accordingly. 14\. Discussion and conclusion Not sustained by the analysis and results presented The analysis and results presented have been revised. Reviewer 4 1\. Context and research gap were not indicated. The research gap was explained in line 85. 2\. Type of study, sampling method and data collection method were not specified. It has major methodological defect. Type of study, sampling method and data collection method were not specified Gold standard of experimental study was not succinctly stated. Thank you for pointing this out. We agree with this and have incorporated your suggestion throughout the manuscript accordingly. 3\. Major statistical analysis was not indicated. Why Wilcoxon signed rank test and median was used? The reason the analysis was chosen has been discussed in details (line 218-223). Reviewer 5 1\. Reviewer \#5: Line 51: Please indicate the full meaning of MSD when been used for the first time in write up. Thank you for pointing this out. The change has been made accordingly (line 50). 2\. Line 178: Consider replacing "respondents" with "subjects" which best suits the study and its concept Thank you for pointing this out. The change has been made throughout the manuscript accordingly. 3\. Please give reasons for the sample size choice and indicate the exclusion and inclusion criteria for participating in the study You have raised an important point here. We agree with this comment and has included relevant information and explanation accordingly (line 114-129). 4\. Line 243: "The value of spinal angle deviation between the pre-test and post-test....." Indicate the angle of deviation to make your point clear Thank you for the comment. Actually, we want to highlight the effect in the experimental group. However, we have made some changes based on your suggestion “The value of spinal angle deviation in the experimental group….” 5\. There are some typographical errors indicated in the attached document. Please revise them accordingly Thank you for pointing this out. The typological error has been revised accordingly. 10.1371/journal.pone.0258796.r003 Decision Letter 1 Mosa Ahmed Mancy Academic Editor 2021 Ahmed Mancy Mosa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Jul 2021 PONE-D-21-13337R1 Effectiveness of lumbar support with built-in massager system on spinal angle profiles among high-powered traffic police motorcycle riders: A randomised controlled trial PLOS ONE Dear Dr. Karuppiah, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please, carefully, consider all the comments of all reviewers Please submit your revised manuscript by 27 June 2021. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Ahmed Mancy Mosa, Ph.D. Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#3: All comments have been addressed Reviewer \#4: (No Response) Reviewer \#5: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#3: Partly Reviewer \#4: Yes Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#3: No Reviewer \#4: Yes Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#3: Yes Reviewer \#4: Yes Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#3: Yes Reviewer \#4: Yes Reviewer \#5: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#3: The authors appropriately revised the manuscript and made the corresponding adjustments. The amount of time devoted to this paper is evident. I have some more comments. Additional comments: How was the randomization performed? Any use of software or what was used for the randomization? Table 3 is not understandable and needs to be reformatted. The formatting of this Table is fine for a within group difference analysis like the authors did for Table 2 but not for a between group difference analysis. A Table should be self-explanatory and this one is not. What does the p-value in the control column indicate? Difference between baseline data (Precont VS Preexp)? Similarly, what does p-value in the experimental column indicate? Difference between after intervention data (Postcont VS Postexp)? I would even suggest comparing the median of changes between control and experimental groups. Line 277: The sentence about no significant difference at baseline should be presented before the results of after intervention. This is a good sign for a between group comparison after intervention. Reviewer \#4: Abstract Background Context and research gap were not stated. methods major statistical analysis was not succinctly stated Main body Methods Sampling technique was not clear data quality control was not indicated Major statistical analysis was not indicated Reviewer \#5: The manuscript is well written and technically sound. All the issues have been addressed by the author. However, there are a couple of technical and grammatical errors that have been noticed: Line 132: “ the recruitment strategy used by taken the name list of all officers” It should rather read “ the recruitment strategy involved taking the name list of all the officers” Line 139: “….then a main researcher was randomly assigned them …….” It should rather read “………, then the main researcher randomly assigned them into control and experimental Line 189: “this equipment had been tested validity….” It should rather read “the validity of the equipment had been tested and approved……” Other comments have been made in the attached document \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#3: **Yes: **Libak Abou Reviewer \#4: No Reviewer \#5: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0258796.r004 Author response to Decision Letter 1 15 Jul 2021 Dear Dr Ahmed Mancy Mosa (Academic Editor), Thank you for giving me the opportunity to submit a second revised draft of my manuscript titled effectiveness of lumbar support with built-in massager system on spinal angle profiles among high-powered traffic police motorcycle riders: A randomised controlled trial. We appreciate the time and effort that you and the reviewers have dedicated to providing your valuable feedback on my manuscript. We are grateful to the reviewers for their insightful comments on our paper. We have been able to incorporate changes to reflect most of the suggestions provided by the reviewers. We have highlighted the revisions within the manuscript. Here is a point-by-point response to the reviewers’ comments and concerns. Reviewer 3 1\. How was the randomization performed? Any use of software or what was used for the randomization? The randomization of this study was explained using the list name and Fishbowl technique (line 140-142). 2\. Table 3 is not understandable and needs to be reformatted. The formatting of this Table is fine for a within group difference analysis like the authors did for Table 2 but not for a between group difference analysis. A Table should be self-explanatory and this one is not. What does the p-value in the control column indicate? Difference between baseline data (Precont VS Preexp)? Similarly, what does p-value in the experimental column indicate? Difference between after intervention data (Postcont VS Postexp)? I would even suggest comparing the median of changes between control and experimental groups. We are sorry for the confusion, and for our understanding we did this as mentioned in the earlier comments (i. Line 40: Are we really interested in a pre-test VS post-test analysis (within group analysis)? Reporting the between group comparison is more insightful…ii. Please explain within and between groups comparison in the analysis of the study outcome.). Thus, we created Table 2 for within-group analysis and Table 3 reporting between-group comparisons as suggested. The p-value in Table 2 represents a value for a significant/not significant between pre-test and post-test studies within two groups and p-value in Table 3 explain a value for a significant/not significant between control and experimental groups. 3\. Line 277: The sentence about no significant difference at baseline should be presented before the results of after intervention. This is a good sign for a between group comparison after intervention. Thank you for your suggestion. We agree with this comment. The revision has been made on lines 293-297. Reviewer 4 1\. Background:Context and research gap were not stated. i\. The context and research gap of traffic police riders suffered low back pain, and MSD which no research has been done on the spinal riding posture during a motorcycle ride was explained in line 72-87. ii\. The context and research gap in which there is a lack of data on the effectiveness of this intervention seat, lumbar support with a built-in massager system, in an in-field setting (on-the-road) was explained in line 96-99. 2\. methods: major statistical analysis was not succinctly stated. The major statistical analysis used in this study was the Mann-Whitney test and Wilcoxon signed-ranked test. The details of the statistical analysis were explained in line 233-243. 3\. Methods: Sampling technique was not clear. The sampling technique was explained in Participant Recruitment (line 132-158). 4\. data quality control was not indicated. Thank you for pointing this out. We agree with this comment. The data quality control had been added in line 216-231. Reviewer 5 1\. Line 132: “ the recruitment strategy used by taken the name list of all officers” It should rather read “ the recruitment strategy involved taking the name list of all the officers” Thank you for pointing this out. The change has been made accordingly (line 132). 2\. Line 139: “….then a main researcher was randomly assigned them …….” It should rather read “………, then the main researcher randomly assigned them into control and experimental Thank you for pointing this out. The change has been made accordingly (line 139). 3\. Line 189: “this equipment had been tested validity….” It should rather read “the validity of the equipment had been tested and approved……” Thank you for pointing this out. The change has been made accordingly (line 189). 4\. Other comments have been made in the attached document Thank you for the comments. We agree with this and have incorporated your suggestion throughout the manuscript accordingly. 10.1371/journal.pone.0258796.r005 Decision Letter 2 Mosa Ahmed Mancy Academic Editor 2021 Ahmed Mancy Mosa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Sep 2021 PONE-D-21-13337R2Effectiveness of lumbar support with built-in massager system on spinal angle profiles among high-powered traffic police motorcycle riders: A randomised controlled trialPLOS ONE Dear Dr. Karuppiah, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please, consider all the comments Please submit your revised manuscript by 16 September 2021. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Ahmed Mancy Mosa, Ph.D. Academic Editor PLOS ONE Journal Requirements: Additional Editor Comments (if provided): \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#3: (No Response) Reviewer \#4: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#3: Yes Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#3: No Reviewer \#4: No \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#3: Yes Reviewer \#4: No \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#3: Yes Reviewer \#4: No \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#3: Thank you for responding to my comments. I believe the authors dedicated a lot of time going through all the reviews. Congrats on that! I still believe the authors did not explain the randomization method. Mentioning that randomization was performed is not enough, how was it done? They are several randomization methods? Which one did you use? Also, the authors compared pre and post in exp and control group (fine); compared pre (cont) vs pre (int) and post (cont) vs (pos int), which is also fine but the most important comparison was not performed. The mean (median) difference between group is the one that is really informative and should be compared to the MCID. Reviewer \#4: Abstract Background Research gap was not stated method major statistical analysis was not specified Main body major statistical analysis was not specified results The major statistical model output were not presented \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#3: No Reviewer \#4: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0258796.r006 Author response to Decision Letter 2 13 Sep 2021 Here is a point-by-point response to the reviewers’ comments and concerns. No. Comments from Reviewer 3 1\. Thank you for responding to my comments. I believe the authors dedicated a lot of time going through all the reviews. Congrats on that! I still believe the authors did not explain the randomization method. Mentioning that randomization was performed is not enough, how was it done? They are several randomization methods? Which one did you use? Firstly, we want to apologize because we are not clear on the type of randomization intended by the reviewer. However, this is the best explanation that we can give for this remark. A randomised controlled trial, pretest-posttest control group design was conducted among 24 traffic police riders who ride a high-powered motorcycle (Honda CBX 750). A simple random sampling was used in this study which the subjects were randomly assigned to the control group (12 riders) and experimental group (12 groups). Data collection commenced in March 2020 and finished in July 2020. Further explanation had been explained in line 138-165. 2\. Also, the authors compared pre and post in exp and control group (fine); compared pre (cont) vs pre (int) and post (cont) vs (pos int), which is also fine but the most important comparison was not performed. The mean (median) difference between group is the one that is really informative and should be compared to the MCID. Thank you for pointing this out. However, this MCID actually will be discussed in details in our future paper. The main objective for this work is to evaluate the effect of lumbar support with a built-in massager system on spinal angle profiles among traffic police riders which focus on comparison of pretest-posttesst and between control-experimental groups. Nevertheless, for future studies, this will be a great insight and we will be sure to use it. No. Comments from Reviewer 4 1\. Reviewer \#4: Abstract Background Research gap was not stated method major statistical analysis was not specified Main body major statistical analysis was not specified results The major statistical model output were not presented We really appreciate your valuable comments here. We find it really useful to improve our manuscript. After discussion with our co-authors, we have improved our manuscript (in the 2nd revision) based on the given comments as below: - i\. The context and research gap of traffic police riders suffered low back pain, and MSD which no research has been done on the spinal riding posture during a motorcycle ride was explained in line 72-87. ii\. The context and research gap in which there is a lack of data on the effectiveness of this intervention seat, lumbar support with a built-in massager system, in an in-field setting (on-the-road) was explained in line 96-99. The major statistical analysis used in this study was the Mann-Whitney test and Wilcoxon signed-ranked test. The details of the statistical analysis were explained in line 233-243. The sampling technique was explained in Participant Recruitment (line 132-158). However, the similar comments are asked again in this 3rd revision. Thus, we decided to explained this in the abstract since it has a word abstract in the beginning of the comments. Traffic police riders are exposed to prolonged static postures causing significant angular deviation of the musculoskeletal, including the lumbar angle (L1-L5). This postural alteration contributes to awkward posture, musculoskeletal disorders and spinal injury, especially in the lower back area, as it is one of the most severe modern diseases nowadays. Thus, the study aimed to evaluate the effect of lumbar support with a built-in massager system on spinal angle profiles among traffic police riders. A randomised controlled trial (pre-testpost-test control design) was used to assess spinal angle pattern while riding the high-powered motorcycle for 20 minutes. Twenty-four traffic police riders were randomly selected to participate and 12 riders were assigned to the control group and 12 riders to the experimental group. The pre-test and post- test were conducted at a one-week interval. Each participant was required to wear a TruPosture Smart Shirt (to monitor spinal posture). The TruPosture Apps recorded the spinal angle pattern. The data indicated that the police riders using motorcycle seat with lumbar support and built-in massager system showed a huge improvement in maintaining posture which only involves slight spinal angle deviation changes from the spinal reference angle throughout the 20 minutes ride. The data collected then were analysed using the Mann-Whitney test and Wilcoxon signed-ranked test to verify a statistically significant difference between and within the control and experimental groups. There were significant differences in all sensors between the control group and experimental groups (p\<0.05) and within the experimental group. According to the findings, it can be said that the ergonomic intervention prototype (lumbar support with built-in massager system) successfully helps to maintain and improve the natural curve of the spinal posture. This indirectly would reduce the risk of developing musculoskeletal disorders and spinal injury among traffic police riders. We do provide the justification as requested as based on our understanding. Please do advise us in details if we need further revision as we feel your valuable comments will improve further our manuscript. Thank you. 10.1371/journal.pone.0258796.r007 Decision Letter 3 Mosa Ahmed Mancy Academic Editor 2021 Ahmed Mancy Mosa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Oct 2021 Effectiveness of lumbar support with built-in massager system on spinal angle profiles among high-powered traffic police motorcycle riders: A randomised controlled trial PONE-D-21-13337R3 Dear Dr. Karuppiah, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Ahmed Mancy Mosa, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#3: (No Response) Reviewer \#4: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#3: (No Response) Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#3: (No Response) Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#3: (No Response) Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#3: (No Response) Reviewer \#4: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#3: (No Response) Reviewer \#4: Remove" The dependent variable of this study was the spinal angle (0th and 20th minutes). Indicate how outcome of interest is assessed before data analysis? \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#3: No Reviewer \#4: No 10.1371/journal.pone.0258796.r008 Acceptance letter Mosa Ahmed Mancy Academic Editor 2021 Ahmed Mancy Mosa This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Oct 2021 PONE-D-21-13337R3 Effectiveness of lumbar support with built-in massager system on spinal angle profiles among high-powered traffic police motorcycle riders: A randomised controlled trial Dear Dr. Karupiah: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Ahmed Mancy Mosa Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction The enterohemorrhagic *Escherichia coli* (EHEC) O157:H7 causes acute gastroenteritis, hemorrhagic colitis and hemolytic-uremic syndrome (HUS) in humans. The latter is a severe infection sequelae characterized by thrombotic microangiopathy, hemolytic anaemia and acute renal failure which can lead to long-term kidney damage or fatal outcome. *E.coli* O157:H7 was first recognized as an important human pathogen in 1982 during an investigation of a food-borne disease outbreak in the United States and has caused many outbreaks in the past three decades, with a wide range of clinical illness. In 2006, an outbreak associated with spinach caused high rates of bloody diarrhea (22/23), hospitalization (13/23) and HUS (7/23), suggesting that the outbreak strain, TW14359, has evolved to higher virulence. Further studies have shown that TW14359 expresses higher levels of many virulence genes and a range of other chromosomal and pO157-encoded genes. It has also been shown that TW14359 has better adherence to epithelial cells than *E.coli* O157:H7 strain Sakai. An outbreak of *E. coli* O157:H7 occurred in Xuzhou, China in 1999, causing about 20,326 infections, 195 hospitalized HUS patients and 177 deaths. Our group has recently sequenced the outbreak strain Xuzhou21. We also discovered that a novel conjugative plasmid pO157_Sal was present in Xuzhou21 and was found nearly unique in the outbreak isolates in China. The pO157_Sal contains 52 ORFs and has a full set of genes for the type IV secretion system (T4SS), but no known virulence-related genes were identified. Several genes found on pO157_Sal are homologous to transcriptional regulatory genes such as *stpA* and *hha* ¸ both of which have been reported to be implicated in virulence or environmental adaptation. In addition, the pO157_Sal *mpr* gene is homologous to *stcE* on pO157 which encodes a zinc metalloproteinase and may play a role in adherence. We further showed that Xuzhou21 has the capacity to provoke elevated proinflammatory responses with levels of IL-6 and IL-8 induction being significantly higher than that induced by EDL933. Xuzhou21 also carries a highly inducible Stx2 prophage. RNA-Seq (whole transcriptome shotgun sequencing) is a next generation sequencing platform based assay of genome-wide bacterial gene expression. RNA-Seq can reveal the entire transcriptional landscape and has less systematic bias compared to microarray technology. To further enhance our understanding of the Xuzhou outbreak strain and to investigate the global effects of pO157_Sal on the gene expression, virulence and adaptation of Xuzhou21, we cured the pO157_Sal plasmid from Xuzhou21 and compared the transcriptomic differences by RNA-Seq between the wild-type strain Xuzhou21 and the pO157_Sal cured strain Xuzhou21m. # Results ## Plasmid curing and complementation Xuzhou21 was treated with SDS and high temperature to cure the pO157_Sal plasmid and curing was confirmed by PCR using primer pairs, ehxA-F/ehxA-R and p247-F/p247-R, targeting *ehxA* and *traL* specific for amplification of pO157 and pO157_Sal respectively. Xuzhou21 was positive for these two genes while the pO157_Sal cured strain Xuzhou21m was only positive for *ehxA*. We found that the efficiency of curing pO157_Sal plasmid was 1%∼3%. We further sequenced Xuzhou21m using Illumina sequencing to confirm that no other changes occurred in the genome during plasmid curing. Reads were mapped to Xuzhou21 genome with an average 114-fold coverage. No insertion or deletion was found and plasmid pO157 was intact in Xuzhou21m. No DNA fragment from plasmid pO157_Sal was detected in Xuzhou21m ****. Fifty-eight tentative SNPs were identified. However, using RNA- Seq data, only two non-synonymous SNPs, one each from two genes of unknown function (CDCO157_2530 \[C to A\] and CDCO157_2534 \[G to T\]), were confirmed to be genuine. We further created a pO157_Sal complemented strain, Xuzhou21c. The plasmid pO157_Sal was first marked with the kanamycin resistance gene using one-step gene inactivation and was transformed into Xuzhou21m successfully. The presence of the pO157_Sal in this complemented strain Xuzhou21c was confirmed by PCR using pO157_Sal specific primer pair, p247-F and p247-R. Q-PCR was used to determine the relative copy number of pO157 and pO157_Sal in Xuzhou21, Xuzhou21m using 2^−ΔΔCT method. The target genes on pO157 and pO157_Sal were *espP* and *traL*. The reference chromosomal gene was g*apA*. Q-PCR showed that the two strains had 2.35±0.49 and 2.21±0.24 copies of pO157 respectively. The copy number of pO157_Sal in Xuzhou21 was 2.04±0.18. These results suggested that the two plasmids have similar copy numbers and the curing of pO157_Sal has no effect on the copy number of pO157. ## Transcriptional profiling of Xuzhou21 and Xuzhou21m To determine the effect of pO157_Sal at transcriptomic level, the transcriptomes of Xuzhou21 and Xuzhou21m were compared using RNA-Seq. Cells were cultured in LB broth to exponential phase and mRNA were sequenced using Illumina sequencing ****. About 6.5 million reads were obtained for both Xuzhou21 and Xuzhou21m, giving a 212 fold coverage respectively. About 1.9 and 1.4 million “clean reads” were obtained for Xuzhou21 and Xuzhou21m respectively after filtering out any reads with more than 10% uncalled bases (Ns) or with more than 50% of bases with a base quality less than 5.0. The reads were mapped to Xuzhou21 reference genome. Allowing for up to 2 mismatches for a read, 67.9% of Xuzhou21 and 63.35% Xuzhou21m reads were mapped to the Xuzhou21 genome. The reads covered 70.13% and 65.6% gene regions for Xuzhou21 and Xuzhou21m respectively. Sequence reads located in intergenic regions account for 28.28% and 29.89% of all sequence reads obtained for Xuzhou21 and Xuzhou21m respectively. Although expression of intergenic regions may have important information, these reads were not analyzed further. As a result, a high-resolution transcriptomic map of the Xuzhou21 genome was generated. The level of gene expression was calculated using reads per kilobase per million reads (RPKM) method. Since the reproducibility was relatively low in genes with low RPKM value, only genes with RPKM≥10 were considered expressed. We found that 3991 of 5214 genes from Xuzhou21 had a RPKM greater than ten, 963 genes have a RPKM greater than zero but less than ten, and 260 genes had no reads (RPKM = 0). Ninety-five percent of the genes of Xuzhou21 genome had an RPKM\>0 and were thus transcribed. One hundred and sixteen genes (2.22%) had a RPKM greater than 1,000 representing highly expressed protein coding genes, among which, 42 and 24 encode ribosomal proteins and hypothetical proteins respectively. Other highly transcribed genes including genes encoding elongation factor Tu, MokW, cold shock protein CspE and flagellin. MokW is only 51 amino acids long and little is known of its function. Of the 1223 genes with RPKM less than ten, 432 (35.32%) were genes of unknown function and a further 369 (30.17%) had putative functions. Of the 5070 chromosomal genes and 92 pO157 plasmid genes, 4814 (94.95%) and 88 (95.65%) were transcribed respectively while all pO157_Sal genes were transcribed in Xuzhou21. In contrast, 94.62% (4749/5070) and 95.65% (88/92) of the chromosomal and pO157 genes were expressed In Xuzhou21m respectively. Interestingly, pO157_Sal genes showed much higher level of expression on average than genes on the chromosome and pO157 ****. ## Validation of RNA-Seq results using RT-qPCR To validate RNA-Seq results, 167 genes were selected to validate their level of expression using RT-qPCR. We selected 51 highly (RPKM\>500), 72 moderately (500\>RPKM\>100) and 44 poorly (100\>RPKM\>10) expressed genes to encompass the wide spectrum of variation in expression levels. As shown in, RT-qPCR data is consistent with the RNA-Seq data (R² = 0.5505, *P*\<0.01). The slope of the trend line as shown in is less than 45 degrees, indicating that RNA-Seq is generally less sensitive than RT-qPCR for quantification of gene expression. Thus RNA-Seq gave a conservative estimate of the number of differentially expressed genes. ## Differentially expressed genes between Xuzhou21 and Xuzhou21m There were 3733 genes with RPKM ≥10 in both strains as candidates for differential expression analysis. To provide a reliable comparison, genes with RPKM less than10 in both strains were excluded. However genes that had an RPKM\<10 in one strain but were highly expressed (RPKM ≥10) in the other (231 in Xuzhou21 and 265 in Xuzhou21m) were included as candidates. Consequently, 168 genes with 2-fold change or greater and *P*\<0.05 were identified as significantly differentially expressed genes. These genes are listed in along with their functions and their levels of expression. Excluding the 52 genes located on pO157_Sal, 67 genes were up-regulated and 101 genes were down- regulated in Xuzhou21. In the 67 up-regulated genes, the expression level of 24 genes in Xuzhou21m was zero; and in the 101 down-regulated genes, the expression level of 12 genes in Xuzhou21 was zero. The vast majority of the differentially expressed genes were chromosomal. Five genes were from pO157 including three down-regulated and two up-regulated genes. Sixteen of the up-regulated genes were hypothetical proteins with unknown functions, while 28 of the down- regulated genes were hypothetical proteins. The up-regulated genes in Xuzhou21 were in 15 gene function classes based on cluster of orthologous genes (COG) classification ****. Eleven up-regulated genes belong to the COG classes of transport and metabolism of amino acids, carbohydrates, inorganic ions, lipids and nucleotides and six in the class of energy production and conversion. Two genes encoding pilus/fimbrial assembly proteins, FimA and PilN, and a gene involved in curli production (CsgG) were up- regulated in Xuzhou21. Curli is associated with biofilm formation, adhesion and invasion. It is interesting to note that several genes are related to stress response including heat shock chaperone gene *ibpB*, cold shock protein gene *cspE*, hyperosmotically inducible periplasmic protein gene *osmY*, peripheral inner membrane phage-shock protein gene *pspD*, multidrug efflux system subunit gene *mdtA*, adenine-specific DNA-methyltransferase gene *dam* and oxidation- reduction related genes, suggesting that pO157_Sal enhances expression of these genes which play a role in stress response. The down-regulated genes also encompass a range of COG gene function classes. Twenty three down-regulated genes belong to the COG classes of transport and metabolism of amino acids, carbohydrates, inorganic ions, lipids and nucleotides. Five genes are related to virulence including *ler*, *ehxD*, 3 LEE effector genes and 1 non-LEE effector gene. Five genes were in the class of cell motility and are involved in pilus biogenesis and assembly, which may also affect virulence. Thus, the pO157_Sal has a repression effect on the expression of these metabolic and virulence associated genes. ## pO157_Sal enhances hyperosmosis and bile salt resistance Since the RNA-Seq data showed that several genes related to stress responses were more highly expressed in Xuzhou21, which suggests that pO157_Sal plays a role in stress response, we tested resistance to bile salt and osmotic pressure. The growth rates of Xuzhou21, Xuzhou21m and Xuzhou21c in the M9 basal medium with 0.2% glucose were the same. In contrast, the growth rates of these three strains in the M9 medium containing an additional 0.37% NaCl or 0.67% NaDC were different. The generation time of Xuzhou21 and Xuzhou21m when cultured in M9 containing the additional NaCl, were 48.8±1.25 minutes and 56.3±3.21 minutes (*t* test, *P* = 0.048) respectively, while the generation time of Xuzhou21 and Xuzhou21m when cultured in M9 containing NaDC were 86.3±3.40 minutes and 100.0±7.50 minutes respectively (*t* test, *P* = 0.016) respectively. The generation times of Xuzhou21c when cultured in M9 containing NaCl and NaDC were 51.3±5.51 minutes and 88.3±1.44 minutes respectively and were similar to those of Xuzhou21 ****. Therefore, pO157_Sal promoted the growth of Xuzhou21 significantly under both conditions. ## The effect of pO157_Sal encoded global gene regulator Hha Analysis of the genes on pO157_Sal found that only two known homologs of global gene regulators *stpA* and *hha* that may play the role in hyperosmosis and bile resistance in Xuzhou21. We attempted to test this by constructing a plasmid carrying a pO157_Sal *stpA* or *hha*. We initially used a constitute expression vector pMD®20 T (TAKARA). However, the empty T vector has a high background effect on Xuzhou21m. We then used an arabinose inducible vector pBAD30, which was found to has no effect on Xuzhou21m, to create pBAD30 constructs carrying the pO157_Sal *stpA* and *hha* separately and transformed to Xuzhou21m as Xuzhou21m+*stpA* and Xuzhou21m+*hha* respectively. However, Xuzhou21m+*stpA* did not grow under arabinose induction. We then tested resistance to osmotic pressure and bile salt of Xuzhou21m+*hha*. As shown in, Xuzhou21, Xuzhou21c and Xuzhou21+*hha* grew similarly while Xuzhou21m grew significantly slower than the other three strains (t test, *P*\<0.05). We further tested whether *hha* affects the expression of the LEE associated T3SS genes, six of which *ler*, *ehxD*, 3 LEE effector genes (*espH, espF, espG*) and 1 non-LEE effector gene (*espY3*) were differentially expressed based on RNA-Seq data. The expression of *espF* and *espH* in Xuzhou21m+*hha* grown in LB to late exponential phase (OD₆₀₀ = 3.0) and under induction of 0.05% arabinose were measured by qPCR. The transcription levels of *espF* and *espH* in Xuzhou21m were 3.2±0.44 (*t* test, *P* = 0.001) and 1.9±0.27 times (*t* test, *P* = 0.003) higher than Xuzhou21. The transcription levels of *espF* and *espH* in Xuzhou21m+*hha* were reduced to 0.5±0.06 (*t* test, *P* = 0.003) and 0.3±0.04 times (*t* test, *P* = 0.001) of those of Xuzhou21. # Discussion To understand further the role of the novel plasmid pO157_Sal, we created a pO157_Sal cured Xuzhou21m strain for comparative studies. Using RNA-Seq, a high- resolution transcriptomic map of Xuzhou21 was constructed which showed that 95% of the genes were transcribed (RPKM\>0). The expression of over 168 genes including 163 genes from the chromosome and 5 genes from the pO157 plasmid were affected by plasmid pO157_Sal. Thus the presence of pO157_Sal has a far reaching effect on the biology of Xuzhou21. pO157_Sal specific genes in Xuzhou21 were all transcribed. Notably, pO157_Sal genes showed much higher expression level on average than other genes in the chromosome and pO157. pO157_Sal affects bile salt resistance since Xuzhou21m grew 16% slower than Xuzhou21 in the presence of bile salt. The growth rate of the complemented strain Xuzhou21c was restored to the same level as the wild type. Seven genes/systems that were up-regulated by pO157_Sal can be associated with this role including multidrug efflux system subunit gene *mdtA*, the adenine-specific DNA-methyltransferase gene *dam*, hyperosmotically inducible periplasmic protein gene *osmY* and oxidation-reduction related genes ****. The multidrug efflux system encoded by *mdtABCD* locus could expel bile salt from the cytoplasm after they breach the cell membrane and thus increase resistance to bile salt. The adenine-specific DNA-methyltransferase coded by *dam* is important for bile resistance by controlling integrity of the envelope as shown in *Salmonella enterica*. *dam* mutants are sensitive to bile salt mostly in exponential-growth phase. Four genes related to oxidation-reduction may also affect bile salt resistance since proteins involved in oxidation-reduction reactions are differentially expressed after exposure to bile salt. pO157_Sal also affects resistance to osmotic pressure as shown by growth rate difference. Differentially expression of two genes (*osmY* and *pspD*) related to osmotic resistance was observed. *osmY* encodes a periplasmic protein OsmY which is commonly involved in osmotic stress and resistance to bile salt. *pspD* and the other 4 genes in the *psp* operon was up-regulated by pO157_Sal although only *pspD* reached the cutoff. The major function of the *psp* regulon is to stabilize and maintain proton motive force within a stressed cell and thus may also promote resistance to osmotic pressure. In addition, heat shock chaperone gene *lpbB* and cold shock protein gene *cspE* related to heat and cold resistance respectively were up-regulated in Xuzhou21. *E. coli* cold shock protein CspA family consists of nine proteins (CspA to CspI). CspE and CspC are constitutively produced at 37°C and up-regulate the expression of the gene encoding global stress response regulator RpoS through *rpoS* message stabilization. In addition, CspE functions as a negative regulator for *cspA*. CspA alters the secondary structure of RNA, making it more susceptible to degradation. Very few virulence-associated genes were found to be differentially expressed between Xuzhou21 and Xuzhou21m ****, which may be related to the exponential growth phase and the medium we used. Previous studies show that LEE1-3 genes usually express much higher than exponential phase in the transition from the late exponential phase to the stationary phase. In the stationary phase, the expression of LEE-encoded genes are down-regulated. For examples, the *ler* promoter activity increased greatly from the mid-exponential phase to the stationary phase. Ler directly regulates genes within the LEE PAI as well as genes elsewhere in the genome. Therefore, genes located on LEE were up-regulated in late exponential phase via *ler*. Interestingly pO157_Sal represses *ler* and four LEE and non-LEE encoded effectors as described above, consistent with repression of virulence genes during late exponential growth reported previously. It would be interesting to examine the effect of pO157_Sal on these genes in the stationary phase. Previous studies have shown that virulence genes are up-regulated when O157:H7 is grown in minimal medium. The expression of virulence genes of O157:H7 have also used DMEM and LB medium previously. Future studies will be conducted using other media to compare with expression in LB to shed further light on the effects of pO157_Sal on the expression of virulence genes. Additionally, there are 7 genes related to bacteria motility including 5 down- regulated (*hofC, yadN, papD, fimL* and major tail protein V gene) and 2 up- regulated (*pilN* and *fimA*) genes by pO157_Sal. But no difference in motility between Xuzhou21 and Xuzhou21m was found when conventional soft agar and U-shape tube motility assay was performed (data not shown). pO157_Sal clearly has a major effect on the expression of chromosomal and pO157 genes, in particular genes associated with cell metabolism. Since a large number of genes of different functions were differentially expressed and these genes are distributed across the chromosome, the effects of pO157_Sal on these genes must be exerted through regulation by global regulatory genes present on pO157_Sal. There are two genes homologous to global regulatory genes, *stpA* and *hha* on the pO157_Sal plasmid. Previous studies showed that the chromosomal homolog of *stpA* plays a similar but minor role to H-NS, both of which bind DNA nonspecifically as global gene regulators with certain differential effect. The differential effect may be associated with their intracellular concentration of which H-NS is higher than StpA. There are 2 *stpA* homologs on the chromosome (CDCO157_3291 and CDCO157_1668). Both homologs had no transcriptional difference between Xuzhou21 and Xuzhou21m with RPKM values of 1035.06 and 898.659 for CDCO157_1668 and 283.562 and 275.738 for CDCO157_3291 respectively. The pO157_Sal plasmid copy of *stpA* is homologous to the two chromosomal copies (the pO157_Sal *stpA* shares 46% and 43% DNA identity with CDCO157_3291 and CDCO157_1668 respectively). The RPKM value of pO157_Sal *stpA* is much higher (2818.8) and may play a functionally similar but larger role. Studies have also shown that the expression of the chromosomal *stpA* in *E. coli* K-12 is up-regulated by high osmolarity during cell growth, further pointing to a regulatory role of StpA in the resistance to osmotic pressure. Unfortunately, Xuzhou21m+*stpA* did not grow in M9 when *stpA* expression was induced using arabinose and thus the function role of *stpA* could not be confirmed. The failure to grow may be associated with overexpression of *stpA*which can orgnize DNA into compact higher order structure via the magnesium in M9 medium. Hha could form heteromeric complexes with H-NS and StpA and have multiple effects on several virulence or adaptation associated genes,. Hha enhances the ability of H-NS to repress the *ehx* operon on the pO157, and mediates repression of *ler* leading to reduced expression of the *esp* operon. Hha also controls biofilm formation by repressing the transcription of rare codon tRNAs and repressing transcription of fimbrial genes. RNA-Seq showed that both *ler* and *ehxD* were repressed by pO157_Sal. The transcription of *espF* and *espH* was repressed by Hha in Xuzhou21m+*hha*. Therefore we conclude that the *hha* located on pO157_Sal represses the transcription of the LEE encoded T3SS and its effectors. Both the whole pO157_Sal plasmid and single gene (*hha*) complementations of Xuzhou21m showed that *hha* plays a role in the resistance to NaCl and bile salt. It is interesting to note that there is also a *hha* homolog present in pO157 as well as on the chromosome. The pO157 *hha* homolog (pO157_0071) shares 37% amino acid identity with the pO157_Sal *hha*, although the two homologs have no significant similarity at DNA sequence level. The RPKM values of the chromosomal copy, CDCO157_0501, were 845.206 and 1022.92 in Xuzhou21 and Xuzhou21m respectively, indicating no transcriptional difference. In contrast, the RPKM values of the pO157 copy, CDCpO157_0071, were 117.155 and 0 in Xuzhou21 and Xuzhou21m respectively, and it seems that pO157_Sal turns on the expression of pO157_0071. Therefore, even if pO157_0071 plays a similar role to the pO157_Sal *hha* it is conditional upon the activation by pO157_Sal. Previous study have shown that *hha* regulates the expression of *csgD* which is a positive regulator of curli genes encoded by *csgDEFG* and *csgBAC* operons. However, the RNA-Seq data showed some opposing effects of the pO157_Sal *hha* on the curli genes, *csgG* of the *csgDEFG* operon was up-regulated in Xuzhou21 while *csgA* of the *csgBAC* operon was down-regulated in Xuzhou21. The RPKM values of *csgD* had no difference between Xuzhou21 and Xuzhou21m, suggesting that the Hha regulation is independent of CsgD. Overall the expression of both operons was low with RPKM values on average of 8.42 and 8.87 in Xuzhou21 and Xuzhou21m respectively. This may be due to the growth phase of the cells we used for RNA-Seq as curli fimbriae are usually expressed higher during the stationary phase. The pO157_Sal *mpr* gene encodes a putative zinc metalloproteinase, which is homologous to *stcE* on pO157. StcE plays a role in intimate adherence of *E.coli* O157:H7 to HEp-2 cells by cleavage of glycoproteins from the cell surface. Thus, the Mpr homolog encoded on pO157_Sal may also be involved in adherence. Future studies will assess the role of the *mpr* gene in virulence. In conclusion, plasmid pO157_Sal affected the expression of 168 genes involved in a range of functions under the conditions tested and may contribute to the resistance to NaCl and NaDC. The wide-ranging effect we observed is likely to be a result of gene regulation by global regulators encoded on the plasmid. The *stpA* and *hha* homologs on pO157_Sal are the most likely regulators playing this role and we have confirmed the effect of *hha* on resistance to NaCl and NaDC. Further studies will be needed to determine the effects by these regulators specifically and their effects on virulence. # Materials and Methods ## Bacterial strains, plasmids and growth conditions The *E. coli* O157:H7 strain Xuzhou21 was isolated from feces of a HUS patient from an outbreak in China in 1999.The nonpathogenic *E.coli* K12 MG1655 was used as negative control. The bacteria were routinely grown in Luria-Bertani (LB) broth or on LB agar plates (pH 7.2). Chloramphenicol (35 µg/ml), kanamycin (40 µg/ml), ampicillin (100 µg/ml) and arabinose (0.05%) were added as required. For growth assays, the bacteria were incubated overnight in 5 ml LB broth at 37°C, and were collected and washed with PBS (137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄ and 2 mM KH₂PO₄, pH 7.2–7.4) for three times. The bacteria were adjusted to about OD₆₀₀ = 0.90 with PBS and 40 µl were inoculated into 4 ml M9 minimal medium (12.8 g/L Na₂HPO₄, 3.0 g/L KH₂PO₄, 0.5 g/L NaCl, 1.0 g/L NH₄Cl, 2 mM MgSO₄, 0.1 mM CaCl₂ and 0.001% thiamine), with the addition of 0.2% glucose or 0.2% arabinose (for gene complemented strains and pBAD30 negative control). Additional differential compositions were NaDC (0.67%) to test bile salt resistance or NaCl (0.37%) to test osmotic stress resistance. We tested a range of NaDC and NaCl concentrations initially. The concentration gave the best difference and thus were selected for the final growth assays. The growth was measured as optical density at 600 nm at hourly intervals. All assays were performed in triplicate, and the results were expressed as mean± standard deviation (SD). ## Curing of plasmid pO157_Sal The plasmid pO157_Sal was removed from Xuzhou21 using sodium dodecyl sulfate (SDS) and high temperature treatment. Briefly, steps were (i) Xuzhou21 was inoculated in 5 ml LB broth at 37°C with shaking for 16 h. (ii) 50 µl of the above culture was inoculated in 5 ml fresh LB broth with 0.05% SDS at 37°C with shaking for 16 h. (iii) 50 µl of the above culture was inoculated in 5 ml fresh LB broth at 42°C with shaking for 16 h. Repeat steps (ii) and (iii), and then 50 µl of the culture was inoculated in 5 ml fresh LB broth at 37°C with shaking for 16 h. The culture was then 10-fold serial diluted and spread on LB plates in order to obtain single colonies. The PCR primers used to identify the plasmid pO157_Sal were p247-F (5′-AGCGCATCGCTACAAGCACA-3′) and p247-R (5′-ACGACAACCCCACCGAGGCT-3′). The PCR primers used to identify the pO157 were ehxA-F (5′-AGCTGCAAGTGCGGGTCTG-3′) and ehxA-R (5′-TACGGGTTATGCCTGCAAGTTCAC-3′). The plasmid and chromosomal DNA integrity were confirmed by genome sequencing. Xuzhou21 cured of the pO157_Sal plasmid was named as Xuzhou21m. ## Plasmid and single gene complementation of Xuzhou21m Vector pRS551 was used as PCR template for kanamycin resistant (Km^r) gene.The Km^r gene was inserted between pO157_Sal_36 and *nikB* gene without disrupting any open reading frame (ORF) using a one-step gene inactivation method as described by Datsenko and Wanner. The km^r marked pO157_Sal was extracted using SDS alkaline lysis method and introduced into Xuzhou21m by electroporation using gene pulse II apparatus (Bio-Rad, USA). The complemented strain was named as Xuzhou21c. The single gene complementation (*stpA* and *hha*) strains were constructed using pBAD30 (Ap^r) due to its lower copy number in this study. The *stpA* and *hha* genes were amplified from purified Xuzhou21 DNA template by PCR. The primers used were listed in. The *stpA* and *hha* products were cloned into the *Eco*RI-*Hin*dIII site of pBAD30 and transformed by electroporation into Xuzhou21m, creating Xuzhou21m+*stpA* and Xuzhou21m+*hha* respectively. Empty pBAD30 was transferred into Xuzhou21m as negative control. ## DNA and RNA isolation Genomic DNA were extracted from both Xuzhou21 and Xuzhou21m using Wizard Genomic DNA purification kit (Promega, Madison, WI, USA) according to the manufacturer's protocols. For total RNA isolation, the bacteria were inoculated in 5 ml LB broth at 37°C with shaking for 16 h. 50 µl of the above culture was inoculated in 5 ml fresh LB broth and the culture was shaken at 37°C for about 2.5 h until the OD₆₀₀ reached 0.6. 500 µl of the culture were mixed with 1 ml RNA protect bacterial reagent (Qiagen, Hilden, Germany) to stabilize RNA according to the manufacturer's instructions and was centrifuged at 5000×g for 10 minutes to pellet the cells, which were then resuspended in 100 µl TE buffer (30 mMTris·HCl,1 mM EDTA, pH 8.0) containing 15 mg/ml lysozyme and 1 mg/ml proteinase K. Total RNA were then isolated according to the standard protocol using an RNeasy mini kit (Qiagen). The RNA quantity and integrity were analyzed using Agilent 2100 Bioanalyzer (Agilent Technologies). RNA was stored at −80°C for the use of transcriptome analysis. We extracted RNA from 3 separate biological experiments and these RNA samples were pooled for RNA-seq. We did not perform RNA-Seq for each biological replicate separately partly due to the cost of RNAseq at that time. Further Illumina RNA-seq data are highly replicable with relatively low technical variation. ## Plasmid copy number and major virulence genes transcriptional level The copy number of pO157 was determined using real-time PCR with the Rotor-gene Q machine (Qiagen). The primer sequences and annealing temperatures are listed in. Expression level of major virulence genes was measured by reverse- transcription quantitative PCR (RT-qPCR) using One Step SYBR PrimeScript RT-PCR Kit II (Takara, Dalian, China) in the Rotor-Gene Q machine (Qiagen). Primer sequences and annealing temperatures are listed in. The relative expression level of target genes and major virulence genes, genes copy numbers were calculated as 2^−ΔΔCT. The mRNA expression level of each target gene was normalized to the expression level of *gapA*. Each assay was performed in triplicate. ## Genome sequencing, RNA sequencing and mapping The Xuzhou21m genome was sequenced using Illumina paired end sequencing. All reads were mapped to the complete genome in previous study (Genbank accession no. CP001925, CP001926 and CP001927) by SOAP pipeline. The total RNA extracted from Xuzhou21 and Xuzhou21m were first treated with Ribo-Zero™ rRNA Removal kit to remove rRNA. The mRNA was fragmented and produced cDNA libraries primed with random hexamers. cDNA was selected by size, amplificated using PCR and then sent to sequencing using Illumina Hiseq™ 2000 commercially. The RNA-Seq data have been submitted to GEO database and the GEO accession number is GSE44846 (<http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE44846>). ## Bioinformatics analysis and transcriptome analysis Images generated by sequencers were converted by base calling into nucleotide sequences, which are called raw data or raw reads and are stored in FASTQ format. Reads were discarded if containing only adaptors, unknown bases more than 10% of a read, or more than half the bases of a read with quality less than 5. Reads after filtering were called clean reads, on which all following analyses were based. Xuzhou21 genome sequence was used as the reference for RNA sequencing mapping and functional analysis. Clean reads were mapped to reference genome and genes sequences respectively using SOAP2. Mismatches no more than 2 were allowed in the alignment. The distribution of reads was plotted by its location in the reference genome, and then divided into gene region and intergenic region. Genome and gene coverage was calculated by counting the number of reads mapped to the genome and individual genes respectively. The gene expression was calculated using the RPKM method. RPKM ratio and p value were used to evaluate the difference between two samples. Differential expression genes (DEGs) were chosen with a P\<0.05 and a ratio of 2 or greater. ## Validation of the transcriptome results by RT-qPCR To confirm the results of the gene expression data from RNA-Seq, the expression levels of 167 selected genes that maintain higher, moderate and lower expression levels in RNA-Seq were measured using RT-qPCR. Primer sequences and annealing temperatures for these genes examined are in. The *gapA* gene was used for within sample normalization. ## Statistical analysis The results were analyzed using the statistical software package SPSS 15.0 for Windows (IBM SPSS). Statistical analysis was performed using the *t* test. Values of *P* ≤0.05 were considered statistically significant. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HZ CC YX CY RL JX. Performed the experiments: HZ YX XZ HS QC YC QM XNB XL. Analyzed the data: CC HW XY HZ YX XX. Contributed reagents/materials/analysis tools: AZ XMB. Wrote the paper: HZ YX CC XX RL JX.

# 1 Introduction Experimental hardware for quantum computing has been steadily improving in the past twenty years, indicating that a useful quantum computer that outperforms a classical computer may eventually be built. However, until a large-scale and viable quantum computer has been realized, numerically simulating quantum circuits on a classical computer will be necessary for predicting the behavior of quantum computers. Such simulations can play an important role in the development of quantum computing by (1) numerically verifying the correctness and characterizing the performance of quantum algorithms, (2) simulating error and decoherence due to the interaction between the quantum computer and its environment, and (3) improving our understanding of the boundary between classical and quantum computing in terms of computational power, for which recent efforts for characterizing the advantage of quantum computers over classical computers serve as an example of this direction. In this work, we consider the problem of quantum circuit simulation as one where we are given a quantum circuit and an initial state, with the goal of determining the probability of a given output state. Several approaches are possible for such simulation tasks. The most general method is to represent the state vector of an *N*-qubit state by a complex unit vector of dimension 2^*N* and apply the quantum gates by performing matrix-vector multiplications. This is essentially the approach adopted in, for example. This method has the advantage that full information of the quantum state is represented at any point during the circuit propagation. However, the exponential cost of storing and updating the state vector renders it prohibitive for simulating circuits larger than ∼45 qubits. On the other hand, for a wide class of circuits with restricted gate sets and input states, efficient classical simulation algorithms are available. For example, the numerical package Quipu has been developed for taking advantage of prior results on the stabilizer formalism to speed up general quantum circuit simulation. Finally, path integral-based methods have also been proposed—though they do not improve the simulation cost, they lead to reduced memory storage requirements. Other than considering the gate sets involved, an alternative perspective of viewing a quantum circuit is through its geometry or topology. Under this view, a quantum circuit is simulated via tensor network contractions. An advantage of viewing quantum circuits as tensor networks is that one can afford to ignore the particular kinds of quantum gates used in a circuit, and instead only focus on the graph theoretic properties. While general quantum circuits involving universal sets of elementary gates are likely hard to simulate on a classical computer, this geometric perspective sometimes allows for the efficient simulation of a quantum circuit with a universal gate set, provided that it satisfies certain graph theoretic properties. We note that at least one open source implementation of tensor network simulation for quantum circuits already exists, called TNQVM, which can simulate tensor networks but also focuses on integrating algorithms with real quantum hardware. Aside from the field of quantum computation, tensor networks and related methods are an important and active area of research in the simulation of quantum mechanical problems in theoretical physics. Among others, *treewidth* is an important graph theoretic parameter that determines the efficiency of contracting a tensor network of quantum gates. A property of graphs that is actively studied in the graph theory literature, the treewidth provides important structural information about a quantum circuit. Namely, if the circuit’s underlying tensor network has treewidth *T*, it is shown in that the cost of simulating the circuit is *O*(exp(*T*)). In treewidth is also used for estimating the classical resource needed for simulating certain quantum circuits. Motivated by the importance of tensor networks in quantum circuit simulation in general (and for example quantum computational supremacy tests in particular), it is useful to have a circuit simulation platform singularly dedicated to tensor network contraction. One immediate challenge in contracting tensor networks is to find an efficient contraction ordering, which relies on explicitly or implicitly finding a reasonable *tree decomposition* of the underlying graph (definitions are further discussed in Section 2). However, finding the optimal contraction ordering (or equivalently finding the minimum- size tree decomposition, or finding the treewidth of a graph) is NP-complete. Therefore one must typically resort to heuristic methods when finding this decomposition. For this study, we implemented a plug-and-play tensor network (TN) contraction code with two contraction schemes. Other schemes were attempted, but were significantly inferior to those that became part of the software package. However, there are likely other heuristic schemes that outperform our stochastic algorithm, and this is an avenue worth pursuing. For a large set of quantum circuits, our tensor network based methods are shown to be less costly than simulation of the full Hilbert space, by comparing to simulations using the LIQ*Ui*\|\> software package. We emphasize again that the tests in this report give timing data for finding the expectation value of a measurement performed after implementing a quantum circuit, not for fully characterizing a circuit’s final state. The remainder of the paper is organized as the follows. Section 2 sets up the definitions and notations used in the paper. Section 3 describes the heuristic methods used for contracting the quantum circuit tensor networks, along with other relevant details of the code implementation. Section 4 presents the example quantum circuits used as benchmarks for demonstrating the performance of our contraction algorithms. Section 5 gives results of comparisons between the qTorch contraction methods, and between qTorch simulations and LIQ*Ui*\|\>’s Hilbert space simulations. # 2 Preliminaries In this section, we provide an overview of relevant concepts and definitions. All graphs that we consider in this paper are undirected. We denote a graph as *G*(*V*, *E*), consisting of the set of nodes *V* = {*v*₁, *v*₂, ⋯, *v*_*n*} and edges *E* ⊆ *V* × *V*. Two relevant concepts are a graph’s tree decomposition and treewidth. A *tree decomposition* of a graph *G*(*V*, *E*) is a pair (*S*, *T*(*I*, *F*)), where *S* = {*X*_*i*\|*i* ∈ *I*} is a collection of subsets *X*_*i* ⊆ *V* and *T* is a tree (with edge set *F* and node set *I*), such that ∪_*i*∈*I**X*_*i* = *V*. Two nodes *X*_*i* and *X*_*j* are connected by an edge only if the intersection between *X*_*i* and *X*_*j* is not null. The *width* of a tree decomposition (*S*, *T*) is max_*i*∈*I*\|*X*_*i*\| − 1. The *treewidth* of a graph *G* is the minimum width among all tree decompositions of *G*. Another important concept in tensor network methods is the *linegraph* of graph *G*, denoted by *L*(*G*). *L*(*G*) is itself an undirected graph, with each edge in *G* corresponding to a node in *L*(*G*). Two nodes in *L*(*G*) are connected if and only if these two nodes’ corresponding edges in *G* are connected to the same node in *G*. There exists an optimal tree decomposition of *L*(*G*) that provides the optimal contraction ordering of *G*. In the context of this work, a *tensor* is defined as a data structure with rank *k* and dimension *m*. More specifically, each tensor is a multidimensional array with *m*^*k* complex entries. Each index may have a different dimension, though in this work each index has the same dimension *m* = 4. A tensor $A_{i_{1},i_{2},i_{3}...i_{k}}$ has *k* indices, which take values from 0 to *m* − 1. A *tensor contraction* is a generalized tensor-tensor multiplication. Here a rank (*x* + *y*) dimension *m* tensor *A* and a rank (*y* + *z*) dimension *m* tensor *B* are contracted into *C*, a rank (*x* + *z*) dimension *m* tensor. $$\begin{array}{r} {C_{i_{1},i_{2}...,i_{x},k_{1},k_{2}...,k_{z}} = \sum\limits_{j_{1},j_{2}...,j_{y} \in {\{ 0..,m - 1\}}}{A_{i_{1},i_{2}...,i_{x}, j_{1}...,j_{y}}B_{j_{1},j_{2}...,j_{y},k_{1}...,k_{z}}}} \\ \end{array}$$ Note that the number of floating point operations performed is *m*^*x*+*y*+*z*, exponential in the number of indices contracted on *y* and the rank of the resulting tensor (*x* + *z*). It is important to point out that pairwise contractions are always optimal. In other words, a function that contracts three or more nodes at a time will not achieve an improvement in scaling. A *tensor network* is a graph *G* = (*V*, *E*) with tensors as vertices, and edges labeled by a set of indices. The rank of each tensor is given by the number of edges connected to it. An edge from one tensor to another indicates a contraction between the two tensors, and multiple connected edges indicate a contraction on multiple indices. shows an example of a tensor network. Note that a tensor may have open edges, *i.e*. edges that do not connect to any other tensor, though this possibility is not allowed in the current version of qTorch. A contraction ordering or contraction scheme determines the order in which the tensor network is contracted. The ordering chosen for the contraction will greatly affect the computation and memory requirements, because some contraction orderings can result in much larger intermediate tensors than others. Although in this work we focus on contracting the tensor network to a scalar that equals the expectation value of the quantum circuit’s measurement, the goal of a tensor network algorithm is often *not* to contract it to a scalar. An example of this is the infinite tensor networks used to study periodic systems in condensed matter physics. An important goal is to avoid tensors of large intermediate rank when contracting the network, as floating point operations grow exponentially with tensor rank. However, it is often the case that increasing the tensor rank is unavoidable. A simple example of this issue is a tetrahedral graph of rank 3 tensors, which cannot be contracted without forming intermediate tensors of rank greater than 3. The larger the treewidth of *L*(*G*) is, the more one will be forced to create new tensors of higher rank as the network is contracted, greatly increasing the computational cost. We note that tensor network methods are commonly used to efficiently find approximate solutions—indeed this is often the main source of a TN’s utility. In approximate TN methods, the space of the smaller values of the eigenspectrum are removed, after e.g. performing a singular value decomposition on the tensor. This allows one to contract to form a larger tensor, then renormalize its size before continuing to contract the network. Though this strategy is useful in the context of quantum circuits as well, the current version of qTorch is limited to numerically exact contractions of the tensor network. Before contracting, the tensor network graph must first be created from a quantum circuit, a procedure that has been summarized in previous work. Each node on the graph represents one of the following: An initial state of the qubit (usually \|0〉〈0\|), a gate operation, or a measurement. The initial density matrix is represented as a rank 1 dimension 4 tensor (i.e. a vector), \[*ρ*_\|0〉〈0\|;*ρ*_\|0〉〈1\|;*ρ*_\|1〉〈0\|;*ρ*<sub>\|1〉〈1\|</s ub>\]. Measurement nodes are rank 1 as well. All indices in the graph are dimension 4, regardless of rank. Nodes corresponding to quantum gates are represented in superoperator form. Hence a gate *U* which acts on the quantum state as *ρ* → *UρU*^† is represented by the superoperator $\widetilde{U}$. The same operation can be expressed as $\left. \widetilde{\rho}\rightarrow\widetilde{U}\widetilde{\rho} \right.$, where $\widetilde{\rho}$ is the Lindblad representation of the density operator. Single qubit gates correspond to rank 2 tensors and two-qubit gates correspond to rank 4 tensors. The graph’s connectivity is identical to the connectivity of the original quantum circuit. We end this section with explicit examples of tensors for standard quantum circuit components. Tensors for other circuit components can be viewed in the source code for qTorch. The initial state \|0〉〈0\| corresponds to the tensor $$\begin{array}{r} {{\widetilde{\rho}}_{0} = \lbrack 1,0,0,0\rbrack.} \\ \end{array}$$ Superoperator tensors for the Pauli matrices are $$\begin{array}{r} {\widetilde{X} = \begin{bmatrix} 0 & 0 & 0 & 1 \\ 0 & 0 & 1 & 0 \\ 0 & 1 & 0 & 0 \\ 1 & 0 & 0 & 0 \\ \end{bmatrix};\widetilde{Y} = \begin{bmatrix} 0 & 0 & 0 & 1 \\ 0 & 0 & {- 1} & 0 \\ 0 & {- 1} & 0 & 0 \\ 1 & 0 & 0 & 0 \\ \end{bmatrix};\widetilde{Z} = \begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 \\ 0 & 0 & 1 & 0 \\ 0 & 0 & 0 & 1 \\ \end{bmatrix}.} \\ \end{array}$$ The CNOT gate is represented as a sparse rank 4 tensor for which only the following entries are nonzero: $$\begin{array}{r} {{\widetilde{U}}_{CNOT}:{\widetilde{U}}_{0000} = {\widetilde{U}}_{0101} = {\widetilde{U}}_{0202} = {\widetilde{U}}_{0303} = {\widetilde{U}}_{1011} = {\widetilde{U}}_{1110} = {\widetilde{U}}_{1213} = {\widetilde{U}}_{1312} =} \\ {{\widetilde{U}}_{2022} = {\widetilde{U}}_{2123} = {\widetilde{U}}_{2220} = {\widetilde{U}}_{2321} = {\widetilde{U}}_{3033} = {\widetilde{U}}_{3132} = {\widetilde{U}}_{3231} = {\widetilde{U}}_{3330} = 1} \\ \end{array}$$ Finally, the nodes for measurement are rank 1 tensors. ${\widetilde{M}}_{X}$, ${\widetilde{M}}_{Y}$, and ${\widetilde{M}}_{Z}$ correspond respectively to determining expectation values for measurements in the *X*, *Y*, and *Z* bases. Note that using ${\widetilde{M}}_{X}$, ${\widetilde{M}}_{Y}$, or ${\widetilde{M}}_{Z}$ is equivalent to inserting a Pauli gate at the end of the circuit before implementing ${\widetilde{M}}_{Trace}$. $$\begin{array}{r} {{\widetilde{M}}_{Trace} = \begin{bmatrix} 1 \\ 0 \\ 0 \\ 1 \\ \end{bmatrix};{\widetilde{M}}_{X} = \begin{bmatrix} 0 \\ 1 \\ 1 \\ 0 \\ \end{bmatrix};{\widetilde{M}}_{Y} = \begin{bmatrix} 0 \\ i \\ {- i} \\ 0 \\ \end{bmatrix};{\widetilde{M}}_{Z} = \begin{bmatrix} 1 \\ 0 \\ 0 \\ {- 1} \\ \end{bmatrix}} \\ \end{array}$$ # 3 Contraction schemes and implementation details For many problems in quantum physics to which matrix product states (MPS) or other tensor network methods have been applied, an efficient contraction scheme is often obvious from the underlying structure of the Hamiltonian. However, efficient contraction schemes are not available for arbitrary tensor networks. A general heuristic contraction scheme is important for the simulation of general quantum circuits, when one does not know a priori the topological properties of the underlying tensor network problem. ## 3.1 Contraction schemes qTorch implements two algorithms for determining the contraction ordering. For what we call the line graph (*LG*) method, outlined in Algorithm 1, we first create the line graph of the quantum circuit’s graph. Then, the software package QuickBB is used to determine an approximately optimal tree decomposition of this linegraph. The resulting tree decomposition is used to define the order of contraction. This linegraph-based approach was first described by Markov and Shi. QuickBB uses a so-called anytime algorithm, meaning that it can be run for an arbitrary amount of time, such that when the program is stopped it provides the best solution found thus far. The algorithm is based on the branch and bound (B&B) method, though the authors use several techniques based on modern graph theory to improve efficiency in the pruning and propagation steps, making QuickBB faster at finding low-width tree decompositions than vanilla B&B. The second contraction scheme is a simple stochastic procedure we refer to as *Stoch* (Algorithm 2). First, a wire is randomly chosen. If the rank of the contracted tensor is higher than the highest rank of the two nodes, plus a given threshold, the contraction is rejected. After a fixed number of rejected contraction attempts, the threshold is relaxed. **Algorithm 1** Contraction via TD of L(G) 1: Create line graph L(G) of graph G 2: *π* ← (Calculate approx. optimal elimination ordering of L(G)) 3: Eliminate wires of G in order *π* **Algorithm 2** Simple stochastic contraction 1: Define *G* ← The tensor network Graph 2: *Threshold* ← −1 3: Define *MaxRejections* ← Maximum Number of Rejections 4: **repeat** 5:  Choose a random wire *w* 6:  (*A*, *B*) ← (Nodes of *w*) 7:  Cost ← rank(*C*) − max(rank(*A*), rank(*B*)) 8:  **if** Cost ≤ Threshold **then** 9:   Contract *w* to form node *C* 10:   *Rejections* ← 0 11:   *Threshold* ← -1 12:   Update *G* 13:  **else** 14:   *Rejections* ← *Rejections* + 1 15:   **if** *Rejections* \> *MaxRejections* **then** 16:    *Threshold* ← *Threshold* + 1 17:    *Rejections* ← 0 18:    Continue 19: **until** Graph completely contracted ## 3.2 Threading The tensor contractions are parallelized using the C++ standard library’s `std::thread` class. A particular tensor-tensor contraction is parallelized if the cost of contracting a pair is higher than a provided threshold. We implement other parallelization schemes at the network level, i.e. splitting up the nodes into separate groups to compute on different threads, but the vast majority of the parallelization speedup comes from threading the tensor–tensor contractions. Currently, qTorch does not support parallelization across multiple compute nodes within a cluster, but it allows the user to specify the number of threads (default of 8). ## 3.3 Estimating the answer string qTorch computes expectation values of the form 〈*ψ*\|*M*\|*ψ*〉, where *M* is a measurement operator such as a Pauli string, and \|*ψ*〉 is the quantum state produced by the circuit. If instead one wishes to capture all the information of this final state of *n* qubits, it generally requires *O*(2^*n*) repetitions of the algorithm. However, many quantities of interest may be calculated efficiently. For instance, the probability that one measurement operator (e.g. a Pauli string) will provide a particular outcome can be estimated in just one contraction of the tensor network, a result essential to simulating the variational quantum eigensolver (VQE). qTorch provides a heuristic scheme to output a high-probability answer string from the circuit, which we summarize here. Though this scheme is not used for the results presented in Section 5, it may be useful in the future for simulating algorithms (like QAOA) where the goal is to estimate a most likely bit string. The scheme is implemented as follows. First we run one simulation, and measure in the computational basis to project the first qubit into 0 or 1. Based on the resulting expectation value from the simulation, we choose the value for the first qubit that has the greater probability. If the 0 and 1 are equally likely, one is chosen randomly. In the next full contraction, we set the resulting binary value as the measurement outcome for the first qubit in the next simulation, and repeat with a projective measurement on the second qubit. We continue this process for the rest of the qubits. As we show below, this method often gives a good approximation of the most likely final computational basis state. In original tests on 3-regular graphs of 30 vertices, the scheme (used on Quantum Approximate Optimization Algorithm \[QAOA\] circuits) gave bit strings that provided good estimates to the solution of the Max-Cut problem (average approximation ratio of 94% compared to the exact brute force solution). As a way to test the general applicability of this scheme, we performed some tests on more general circuits than the QAOA problem. These tests are meant to provide some early insight into how useful this heuristic would be for estimating the most likely bit string of a quantum algorithm, for the users who are interested in running this string estimation subroutine. We note that it is abundantly clear that in many cases this scheme does not produce a string closest to the most likely string—indeed, if it was a generally accurate scheme then we would have no need for a quantum computer. In the remainder of this section, we consider the most likely bit string of the final state \|*ψ*〉 = ∑_*i* *ψ*_*i*\|*i*〉, which we define as argmax_*i*\|*ψ*_*i*\|², where the vectors {\|*i*〉} are in the computational basis. We apply a unitary of the form $$\begin{array}{r} {U_{p}\left( \mathbf{\beta},\mathbf{\gamma} \right) = \prod\limits_{j = 1}^{p}\text{exp}\left( i\beta_{j}\sum\limits_{i = 1}^{n}X_{i} \right)\text{exp}\left( i\gamma_{j}D_{j} \right)} \\ \end{array}$$ where the matrix *D*_*j* is a diagonal matrix with entries chosen randomly from the integers {1, ⋯, *nm*}. Here *m* is a parameter that could be interpreted as the number of clauses, if this were a QAOA problem. The elements of the *p*-dimensional vectors ***β*** and ***γ*** are drawn uniformly from \[0, *π*\] and \[0, 2*π*\] respectively. We use the construction of *U*_*p*(***β***, ***γ***) to emulate the form of parametrized unitary operations used in QAOA with the same *p*. Starting from the uniform superposition over all 2^*n* bit strings \|*s*〉, we apply *U*_*p* to compute the final state $\left| \Psi \right\rangle = U_{p}\left| s \right\rangle = \sum_{i = 0}^{2^{n} - 1}\psi_{i}\left| i \right\rangle$. Let *p*_*i* = \|*ψ*_*i*\|² denote the probability distribution associated to the QAOA-like output state \|Ψ〉. We ran 10,000 trials (with n = 6, m = 10, and p = 2) using, and ranked the result by how many bit strings in the true state were more likely than our outputted bit string. Conceptually, our likely string estimation algorithm can be thought of as falsely assuming that the output state is a product state. Suppose we apply our algorithm to the state \|Ψ〉. The product state then reads $$\begin{array}{r} {|\Psi^{\prime}{\rangle =}\left( \alpha_{1} \middle| 0 \right\rangle + \beta_{1}{\left| 1 \right\rangle)} \otimes \left( \alpha_{2} \middle| 0 \right\rangle + \beta_{2}{\left| 1 \right\rangle)} \otimes \cdots\left( \alpha_{n} \middle| 0 \right\rangle + \beta_{n}{\left| 1 \right\rangle)}} \\ \end{array}$$ where \|*α*_*k*\|² is the probability of \|0〉 that the algorithm obtains at the *k*^th step, with an analogous definition for *β*_*k*. With this conceptual framing, we also numerically study the 1-norm distance ∥**p**′ − **p**∥₁ between the approximate distribution **p**′ which the algorithm effectively assumes and the actual distribution **p**. The results are shown in. Here we use the number of qubits *n* = 6, with parameters *m* = 10 and *p* = 2. shows that most of the time our algorithm produces a high ranking bit string—roughly 90% of the time the output of the algorithm is among the top 10% most likely bit strings. shows that the 1-norm distance between the approximate and exact distributions is less than 0.1 for nearly all of the data points. These results suggest that our heuristic for estimating an output bit string will produce acceptable estimates for some circumstances—in QAOA for instance, where one might be interested in a good approximate (as opposed to exact) solution to the constraint satisfaction problem. ## 3.4 A note on noise It is possible to simulate noise within the quantum circuit model, by mapping a noise model onto a set of one-qubit or multi-qubit operators. Note that any quantum operation can be expressed in terms of Kraus operators {*E*_*j*} $$\begin{array}{r} \left. \rho\rightarrow\sum\limits_{j}E_{j}\rho E_{j}^{\dagger} \right. \\ \end{array}$$ where {*E*_*j*} are called Kraus operators, and $\sum_{j}E_{j}E_{j}^{\dagger} = 1$ because for our purposes the noise process is assumed to be trace-preserving. A noise model can be expressed in terms of such Kraus operators, which can in turn be expressed as superoperators for insertion into the quantum circuit’s tensor network. The most commonly used approximations assume that noise on different qubits is uncorrelated, which allows for single-qubit “noise gates” to be used. In this case, because rank 2 tensors can always be contracted without increasing the rank of the resulting tensors, the cost of simulating the resulting “noisy” quantum circuit would not substantially increase. One common and easily implementable approximation is the Pauli twirl approximation, which approximates a noise process purely in terms of Pauli rotations, and therefore can be implemented with the built-in quantum gates of qTorch. A more physically realistic noise model would assume correlated noise, which necessitates the insertion of noise gates that operate on at least two qubits. In this case, the tree width of the circuit’s underlying line graph, and hence the complexity of the problem, would increase in all but the most trivial cases. qTorch does not incorporate built-in noise gates. Instead, we include functionality that allows for user-defined gates. # 4 Circuit simulations Here we describe the classes of quantum circuits that were simulated for this work. ## 4.1 QAOA / Max-Cut The quantum approximate optimization algorithm (QAOA) was recently developed, for the purpose of demonstrating quantum speedup for combinatorial problems on low-depth quantum circuits. Given a constraint satisfaction problem (CSP), a QAOA quantum circuit produces an output that provides approximate solutions. Several aspects of QAOA have been studied since its introduction, including its application to different classes of CSP, implementations of different classical optimization routines, and numerical and analytical comparisons to classical algorithms. We use qTorch to simulate QAOA for the Max-Cut problem, a combinatorial problem that has been a focus of QAOA. Given an arbitrary undirected graph, the goal of Max-Cut is to assign one of two colors to each node, so as to maximize the number of cuts. A cut is any edge that connects two nodes of different color. A more detailed explanation of QAOA is given in the Appendix, though we summarize important aspects of the algorithm here. In QAOA, a set of constraints is mapped to a an objective function represented by a set of operators. Specifically for the Max-Cut problem, the object function is $$\begin{array}{r} {C = \sum\limits_{\langle ij\rangle}C_{\langle ij\rangle},} \\ \end{array}$$ with $$C_{\langle ij\rangle} = \frac{1}{2}(1 - \sigma_{i}^{z}\sigma_{j}^{z}),$$ where 〈*ij*〉 represents the edge between nodes *i* and *j*, $\sigma_{k}^{z}$ is the Pauli-Z operator on qubit *k*, and each node in the underlying Max-Cut graph (which is related to but not the same as the quantum circuit’s graph) corresponds to one qubit in the quantum circuit. Define two operators *U*(*C*, *γ*) and *U*(*B*, *β*) as $$\begin{array}{r} {U(C,\gamma) = e^{- i\gamma C} = \prod\limits_{m = 1}^{n}e^{- i\gamma C_{m}}} \\ \end{array}$$ and $$\begin{array}{r} {U(B,\beta) = e^{- i\beta B} = \prod\limits_{k = 1}^{q}e^{- i\beta\sigma_{k}^{x}}.} \\ \end{array}$$ where $B = \sum_{k = 1}^{q}\sigma_{k}^{x}$, *σ*_*x* is the Pauli-X operator, *q* is the number of qubits, and *n* is the number of clauses (for Max-Cut this is the number of edges). These two operators are applied *p* times (with different *γ* and *β* allowed for each step), with a larger *p* providing a better approximation. The *γ* and *β* parameters are modified with a classical optimization routine to maximize the cost function. The cost function is evaluated after each measurement, with the bit string that resulted from the measurement. To generate the graphs for the underlying Max-Cut problem, we made random *k*-regular graphs by placing edges randomly throughout a given vertex set to satisfy a given regularity, before ensuring that disconnected graphs are rejected. QAOA/Max-Cut Quantum circuits based on these graphs are then constructed. In the numerical results of this paper, we report only the timing for a single contraction of each quantum circuit. A full analysis of QAOA is beyond the scope of this work. However, we note that once the graphs have been created, qTorch currently has the functionality to optimize the QAOA angles using the classical optimization library NLopt. Finally, one can use qTorch to estimate a Max-Cut for the randomly-generated graph, using the most-likely bit string estimation method described above. ## 4.2 Hubbard model Quantum simulation of fermionic systems is one of the most relevant applications of quantum computers, with direct impact on chemistry and materials science, including for the design of new drugs and materials. Among all the algorithms proposed for quantum simulation of fermions, the quantum variational algorithm (VQE) and related approaches are arguably the most promising for near-term hardware because they have lower circuit depth requirements. We note that many types of chemistry-related circuits can be prepared with the software package OpenFermion. In the VQE algorithm, a quantum computer is employed to prepare and measure the energy of quantum states associated with a parameterized quantum circuit. The approximate ground state of a Hamiltonian is obtained by variationally minimizing the cost function (corresponding to e.g. the molecular energy) with respect to the circuit parameters using a classical optimization routine. This hybrid quantum-classical approach offers a good compromise between classical and quantum resources. Classical simulations of the VQE algorithm for tens of qubits could provide insights into the complexity of the circuits used for state preparation and help design better ansatzes for the quantum simulation of fermions. As an example of a VQE simulation, we used qTorch to classically simulate variational circuits employed for the quantum simulation of 1D Hubbard lattices. We consider half-filled Hubbard models on *N* sites, with periodic boundary conditions. To construct variational circuits for these systems, we considered the variational ansatz introduced by Wecker et al. In this case, the Hubbard Hamiltonian is divided as *H* = *h*_*h* + *h*_*U*, where *h*_*h* is the sum of hopping terms in the horizontal dimension and *h*_*U* is the repulsion term. The variational circuit is constructed as a sequence of unitary rotations by terms in the Hamiltonian with different variational parameters, with the sequence being repeated *S* times. In each step, there are two variational parameters, $\theta_{U}^{b}$ and $\theta_{h}^{b}$, where *b* = 1, ⋯, *N* such that $$\begin{array}{r} {U = \prod\limits_{b = 1}^{S}\left\lbrack U_{U}\left( \frac{\sigma_{U}^{b}}{2} \right)U_{h}\left( \theta_{h}^{b} \right)U_{U}\left( \frac{\theta_{U}^{b}}{2} \right) \right\rbrack} \\ \end{array}$$ where *U*_*X*(*θ*_*X*) denotes a Trotter approximation to exp(*iθ*_*X* *h*_*X*), and *X* can be *U* or *h*. For our numerical simulations, we employed the variational circuit of with *S* = 1 using a 1-step Trotter formula for all the *U*_*X*(*θ*_*X*) terms. Notice that this is approximate for the *h*_*h* term, which comprises a sum of non-commuting terms. We also assigned the value of 1 to all variational amplitudes. The corresponding unitary was mapped to a quantum circuit using the Jordan-Wigner transformation and the circuit was generated using a decomposition into CNOT gates and single-qubit rotations. # 5 Results Simulations were performed on the NERSC Cori supercomputer, using one “Knights Landing” (KNL) node per simulation, each of which contains 68 cores and 96 GB of memory. Each LIQ*Ui*\|\> simulation was run on a full node as well, using Docker (computational details in the Appendix). The free version of LIQ*Ui*\|\> allows for the simulation of 24 qubits. Because full Hilbert space simulation scales exponentially regardless of the quantum algorithm’s complexity, we would not have been able to simulate more than ∼31 qubits on one of these compute nodes. For each set of parameters (regularity and number of vertices/qubits) 50 instances of Max-Cut/QAOA circuit were created. For higher qubit counts and higher regularities, only a subset of these circuits were completed, since many simulations exceeded memory capacity. In this section, *LG* or *qTorch-LG* refer to the use of qTorch with the linegraph-based contraction, *Stoch* or *qTorch- Stoch* refer to qTorch with stochastic contraction. To determine a qTorch-LG contraction ordering, QuickBB simulations were run for an arbitrary time of 3000 seconds for each quantum circuit. The plotted qTorch results do not include the QuickBB run time. We note that LIQ*Ui*\|\> implements a thorough set of important optimizations, which makes it a fair benchmark against which to compare qTorch. For example, LIQ*Ui*\|\> fuses many gates together before acting on the state vector, and uses sparse operations. qTorch, on the other hand, does not yet use sparsity at all (even when the circuit consists primarily of sparse CNOT gates), which is one of several optimizations that we expect would further improve performance. LIQ*Ui*\|\> is the fastest simulation method to use for the Hubbard simulations, as shown in. This is because the treewidth of the circuit’s graph increases substantially with the number of qubits, even for these short-depth circuits. The result is not surprising—if the algorithm were easy to simulate with a tensor network on a classical computer, then it would not have been worth proposing as a candidate for a quantum computer. Simulation timing results for 3-, 4-, and 5-regular Max-Cut/QAOA circuits are shown using Tukey boxplots in Figs and. Stoch and LG simulation times are of similar order of magnitude for these circuits, though LG is generally faster. The exception is the 3-regular graph problems, where Stoch often appears to find a more efficient contraction than the 3000-second run of QuickBB does. We note that if the QuickBB algorithm were run for infinite time before beginning the contraction, then qTorch-LG should always (except in very simple graphs) contract the circuit faster than qTorch-Stoch. This is because, while the Stoch search is purely local (considering only individual wires), the tree decomposition approach of QuickBB implicitly considers the effects of multiple contraction steps. Note that actual search time of Stoch is negligible compared to the tensor contraction time. Note that LIQ*Ui*\|\> begins to outperform tensor contraction methods once the algorithm is run on 5-regular graphs, because the increased circuit complexity leads to larger intermediate tensors in qTorch. Note that in principle, Hilbert space simulation can be considered a subset of TN contraction, where the state vector is simply a large tensor. Hence one might expect that there would not be a crossover point at all, *i.e*. that in the worst case TN contraction would not ever be slower than Hilbert space simulation. However, because our implementation considers density matrices instead of state vectors, one would in fact expect this crossover point to exist. The largest tensor in qTorch would have 4^*N* entries, while the state vector has just 2^*N* entries. The various choices made in software implementations for qTorch and LIQ*Ui*\|\> would also affect the position of this crossover point. Using a single Cori NERSC node, we were able to contract quantum circuits of 90 qubits for a very small subset of the simulated graphs, though not on enough graphs to report statistics. Full Hilbert space methods would be limited to ∼30 qubits on these nodes, and indeed previous simulation packages have not yet surpassed 46 qubits, using thousands of nodes. Interesting trends appear when the simulation time is plotted against regularity of the Max-Cut problem’s graph. It is notable that the LG method runs out of memory before the Stoch method does. As previously mentioned, the LG method contracts more efficiently the longer QuickBB has been run, and we chose 3000 seconds as an arbitrary QuickBB limit for all circuits. There is a trade-off between running a longer QuickBB simulation and instead immediately using the Stoch method. Even with few qubits, at higher regularities the full Hilbert space simulation (using LIQ*Ui*\|\>) performs better. This is expected, since as the complexity of the quantum circuit increases, higher-rank intermediate tensors appear. shows simulation time as the estimated upper bound for the treewidth increases, for Max-Cut/QAOA circuits of 18 qubits. These include 3- through 7-regular graphs. This treewidth upper bound is simply the treewidth of the tree decomposition that defines the contraction ordering. The plot shows the expected general trend of an increase simulation time with increased treewidth, regardless of contraction scheme. Finally, we note that we were easily able to perform simulations of 100 qubits for less complex graphs. To report one such example, we produced a random 3-regular graph with a slightly different procedure from that given in of Section 4.1. Beginning with a 2-regular graph (i.e. a ring) of 100 vertices, we added edges between random pairs of vertices until all vertices were of 3 degrees. Contracting this graph’s Max-Cut/QAOA circuit took ∼150 seconds. # 6 Conclusions We have implemented a tensor contraction code for the efficient simulation of quantum circuits. We compared a stochastic contraction scheme to one based on the line graph of the quantum circuit’s graph, showing that the latter is more efficient in most circuits simulated herein. However, there is a subset of cases for which calculating an efficient approximate optimal tree decomposition of the line graph takes longer than contracting the circuit stochastically, in which case the stochastic scheme is superior. For the circuits studied in this work, our simulations suggest that the point at which qTorch is no longer faster than LIQ*Ui*\| \> occurs in QAOA/Max-Cut approximately when the Max-Cut graph has a regularity of five. In the future, qTorch may be used to estimate these points of equivalent computational cost in other classes of circuits, which may help to determine which simulation method to use in simulations. Several immediate algorithmic improvements are possible for this software. The use of sparse tensors would reduce the number of floating point operations for some relevant circuits. Tensor decompositions (such as the singular value decomposition) along with trimming can be added as intermediate steps, as has been done in tensor network based simulations of physical systems. Additionally, more advanced parallelization methods would allow for faster calculation of a tree decomposition as well as faster contractions. # Supporting information We are grateful to Edward Farhi and Aram Harrow for discussions about QAOA, to Salvatore Mandrà for general discussions, and to Dave Wecker for helpful advice on using LIQ*Ui*\| \>. N.S. acknowledges the use of resources of the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. A.A.-G. and E.F. were supported by the Office of Naval Research under grant N00014-16-1-2008 (Vannevar Bush Faculty Fellowship). A. A.-G. and Y.C. were supported by NSF grant CHE-1655187.

# Introduction Initiation of inflammation following infection requires recognition of the invading microbe by innate immune pattern recognition receptors (PRRs) that signal in response to pathogen-associated molecular patterns (PAMPs). PRRs recognize self- and microbe-associated molecules. Members of the Toll-like receptor (TLR) family of PRRs are transmembrane receptors that are expressed either on the cell surface or within the endosomal compartment and respond to a variety of PAMPs. Murine TLR3, TLR7 and TLR9 are expressed in the endolysosome and are implicated in recognition of viral dsRNA, ssRNA and dsDNA, respectively,. Ligation of the nucleic acid-sensing TLRs results in transcription of antiviral genes including type I IFNs (IFN-α/β) and proinflammatory cytokines. TLR3 responses require signaling through the adaptor molecule Toll/IL-1R domain-containing adapter-inducing interferon-β (TRIF), while TLR7 and TLR9 are dependent on the adaptor molecule myeloid differentiation primary response gene 88 (MyD88) to activate transcription factors and induce gene transcription,. Murine cytomegalovirus (MCMV) is a betaherpesvirus that can establish acute infection in multiple organs including the liver. Acute MCMV infection induces an early systemic proinflammatory cytokine response including high levels of type I IFNs, IFN-γ, IL-12 and TNF-α. Infection in the liver induces early production of IFN-α, predominantly by plasmacytoid dendritic cells (pDCs), by 40 h post-infection. Type I IFN production mediates downstream responses including chemokine and cytokine production as well as monocyte/macrophage, natural killer (NK) cell and T cell recruitment. Early type I IFN signaling is necessary for NK cell recruitment to the liver, where they deliver the antiviral cytokine IFN-γ within the first 48 h post-MCMV infection. The NK cell IFN-γ response is an important early step in the control of liver infection. This response induces IFN-γ-dependent chemokines, which contribute to the recruitment of CD8+ T cells to the liver. Liver CD8+ T cell responses occur by days 5 and 7 post-MCMV infection and are an important source of cytokines late in acute infection that contribute to resistance against MCMV. While early responses to MCMV infection in the liver are well understood, it remains unclear how the virus is sensed in this compartment. This is in contrast with other sites, namely the spleen, in which studies by our group and others have definitively shown a role for TLR9 and MyD88 signaling in IFN-α, proinflammatory cytokine and cellular responses in addition to restriction of virus replication. Although TLR7 alone does not appear to have a strong role in MCMV recognition, TLR7 and TLR9 combined deficiency was shown to severely impair pDC responses against MCMV in the spleen. A significant but minor role for TLR3 signaling in the spleen has also been suggested in response to MCMV infection. In the liver, however, studies by our group have demonstrated that early innate responses are TLR9-independent but MyD88-dependent. Liver pDCs from mice genetically deficient in TLR9 produce wild-type (WT) levels of IFN-α at 40 h post-MCMV infection, with intact downstream cellular and proinflammatory cytokine responses. Further, TLR9-deficient mice do not exhibit elevated liver viral titers. Conversely, MyD88-deficient mice have severely impaired liver cytokine and cellular responses, and are unable to control virus replication in this compartment. MyD88 is a common adaptor molecule for TLR9 and TLR7 signaling; however, evaluation of TLR7-deficient mice also demonstrated that TLR7 signals alone were not required to initiate liver antiviral defense. These TLR-independent but MyD88-dependent antiviral responses suggested possible redundancies among TLR signals in the liver compartment in response to MCMV infection. To investigate this possibility, we utilized mice containing an H412R missense mutation in the endoplasmic reticulum protein UNC93B1 to address the combined function of nucleic acid-sensing TLRs in the liver during acute MCMV infection. The UNC93B1 mutation (known as ‘triple d’ or ‘3d’) impairs signaling through TLR3, TLR7 and TLR9 due to improper trafficking of these receptors to the endosomal compartment, and has been shown to affect exogenous antigen presentation. Our studies show that proinflammatory cytokine production after early infection with MCMV is dependent on UNC93B1. Further, UNC93B1 deficiency exacerbates liver disease and increases viral burden, although MCMV-specific CD8+ T cell responses are not impaired. Collectively, these results suggest a level of redundancy within the liver to promote viral recognition by demonstrating that a combination of nucleic acid-sensing TLRs contributes to innate inflammatory responses during MCMV infection. # Results ## Systemic cytokine production is impaired in 3d mice during MCMV infection Considering the potential of endosomal TLR signals to induce proinflammatory cytokine expression, UNC93B1 deficient 3d mice were first assessed for systemic IFN-α, IFN-γ, IL-12 and TNF-α production during early infection with a moderate (5×10⁴ PFU) dose of MCMV. C57BL/6 (WT) and 3d mice were uninfected or MCMV-infected for 40 h or 48 h. Serum was collected at indicated time points and IFN-α, IFN-γ, IL-12p70 and TNF-α were measured by enzyme-linked immunosorbent assay (ELISA). In WT mice, maximal production of IFN-α, IFN-γ and IL-12p70 was detected at 40 h post-MCMV infection before declining by 48 h post-infection (A–C). In contrast, 3d mice exhibited lower serum levels of these cytokines in response to MCMV infection. Specifically, while serum IFN-α reached 1300±420 pg/mL at 40 h post-MCMV infection in WT mice, IFN-α production was reduced by three-fold at this infection time point in 3d mice (450±300 pg/mL), with comparable levels maintained at 48 h post-infection. Likewise, while average IFN-γ concentrations in WT mice reached maximal levels of 530±245 pg/mL at 40 h post-MCMV infection, 3d mice failed to induce detectable levels of this cytokine . IL-12p70 production similarly peaked at 40 h post-MCMV infection in WT mice, with levels reaching 1000±600 pg/mL. 3d mice, however, produced 12-fold less IL-12p70 at this infection time point. Serum TNF-α levels were elevated in response to MCMV infection at both 40 h and 48 h post-infection in WT mice (69±10 and 63±17 pg/mL, respectively). In 3d mice, average concentrations of TNF-α were 6-fold lower than WT at 40 h and 2.5-fold lower than WT at 48 h after infection. These results indicate a requirement for endosomal TLR signals for early systemic proinflammatory cytokine production, and concur with previous reports. ## Liver cytokine production is impaired in MCMV-infected 3d mice Having observed a reduction in the level of proinflammatory cytokines in the serum of 3d mice, cytokine responses in liver cells from 3d mice infected with MCMV were evaluated to address the impact of endosomal TLR signaling in a localized tissue site of infection. The best characterized liver cytokine responses are IFN-α and IFN-γ. Therefore, to determine whether combined endosomal TLR signaling contributes to the production of these cytokines during MCMV infection, IFN-α and IFN-γ in liver homogenates and in individual cell populations were measured in WT and 3d mice uninfected or infected with MCMV for 40 h and 48 h. As shown in, WT mice displayed four-fold higher levels of IFN-α than 3d mice at 40 h post-infection, with increased levels still evident at 48 h following infection. Since pDCs expressing the marker PDCA-1 have been shown to produce the majority of liver IFN-α at 40 h post-MCMV infection, this cell type was examined in WT and 3d livers during early infection. There was evidence of pDC accumulation in the livers of both WT and 3d mice. However, liver pDCs from 3d mice were impaired in their ability to express IFN-α. There were 4-fold fewer PDCA-1+ pDCs expressing intracellular IFN-α at 40 h, and 3-fold fewer at 48 h post-infection, in 3d mice as compared to WT. This trend was also reflected in the proportion of PDCA-1+ IFN-α+ pDCs at 40 h and 48 h after infection in 3d mice (0.8%±0.4% and 2%±1%) compared with WT (2%±1% and 6%±1%). 3d mice similarly demonstrated a defect in liver IFN-γ production in response to MCMV infection. In WT mice, IFN-γ reached maximal levels of 3500±1200 pg/g liver at 40 h before contracting by approximately half at 48 h post-infection. In contrast, at 40 h post-infection, 3d mice induced 5-fold less IFN-γ than WT. NK cells are an important early source of IFN-γ in the liver during MCMV infection, and accumulated at this site in both WT and 3d mice. Using intracellular cytokine staining, results shown in demonstrate a 7-fold reduction in the absolute numbers of NK1.1+ TCRβ- liver cells expressing IFN-γ in 3d mice at 40 h post-infection as compared to WT. There were also fewer IFN-γ-expressing liver NK cells in 3d mice by proportion (0.8%±0.3% and 1%±0.4%) when compared to WT (6%±2% and 4%±0.4%) at 40 h and 48 h, respectively. Together, these results demonstrate that, in addition to an effect on systemic cytokine production, combined endosomal TLR signals can affect the expression of critical proinflammatory cytokines in the liver during MCMV infection. ## Liver T cell responses are not impaired in 3d mice during MCMV infection The innate immune response is important both in establishing early control of virus replication and in coordinating downstream adaptive responses. Following MCMV infection, virus-specific CD8+ T cells are recruited to the liver within 5 days and control viral replication at this site through release of cytotoxic molecules and production of cytokines such as IFN-γ and TNF-α. Given the abated liver cytokine responses observed in 3d mice, the effect of endosomal TLR signaling on liver CD8+ T cell responses was examined at late time points during acute MCMV infection. The results shown in demonstrate comparable absolute numbers of CD8+ T cells in WT and 3d mice at days 5 and 7 post-MCMV infection, a trend that was also reflected in proportion (data not shown). To determine whether CD8+ T cells in 3d mice were properly activated against MCMV infection, intracellular expression of IFN-γ and TNF-α in CD8+ T cells was examined following *ex vivo* restimulation with H-2D^b M45 viral peptide, an immunodominant epitope of MCMV. CD8+ T cells from 3d mice expressed these two cytokines at day 5 and day 7 post-MCMV infection by proportion and absolute numbers at levels that were comparable or slightly increased over WT (B–E). As an indication of degranulation, surface expression of CD107a was also examined on liver CD8+ T cells from mice infected with MCMV; however, no differences in CD107a expression were detected between 3d and WT mice (data not shown). These results suggest that MCMV-specific CD8+ T cell responses in the liver are not compromised in the absence of endosomal TLR signals. ## Increased sensitivity to virus-induced liver disease in 3d mice during MCMV infection Previous studies have demonstrated resolution of virus-induced liver disease after 5 days of MCMV infection in WT mice. Therefore, given the impaired inflammatory responses observed in the absence of endosomal TLR signaling, liver sections prepared from 3d and WT mice that were uninfected or infected for 3, 5, or 7 days were hematoxylin and eosin (H&E) stained to evaluate pathology. The histological appearance of liver sections from uninfected 3d and WT mice appeared comparable (A and B). By day 3 post-MCMV infection, clusters of infiltrating cells or inflammatory foci, which have been shown to coincide with sites of MCMV antigen expression, were present in WT mice and persisted through day 5 before inflammation was resolved by day 7 post-infection (C, E and G). While the inflammatory foci per area of liver were equally apparent in liver sections from 3d mice infected for 3 days, there was an increased presence of cytomegalic inclusion bodies characteristic of MCMV-infected cells that were not readily apparent in WT mice. By day 5 post-infection, livers from 3d mice were characterized by widespread areas of inflammation compared to the more punctate foci in WT livers (E and F). Moreover, the inflammatory foci per area of liver in 3d mice at day 5 post-infection were significantly more numerous and contained a greater number of nucleated cells compared to WT. However, by day 7 post-infection, inflammation in 3d mice showed signs of resolution that were similar to WT (G and H). To further evaluate the effects of endosomal TLR responses on overall liver function, the liver enzyme alanine aminotransferase (ALT) was measured in serum samples from WT and 3d mice that were uninfected or infected with MCMV for 3, 5 or 7 days. Uninfected mice had comparable baseline levels of systemic ALT. By day 3 of infection, similar elevations in ALT levels were detected in both groups. In contrast, by day 5 post-infection, the levels of systemic ALT in 3d mice reached values of 900±400 U/L and were three-fold higher than the ALT levels measured in WT mice (270±100 U/L;), indicating augmented liver disease. In contrast, by day 7 post-infection, 3d mice demonstrated a sharp decline in systemic ALT levels to values of 117±63 U/L, which were comparable to ALT levels exhibited in WT mice (108±85 U/L). Thus, at this time point of acute infection, both groups of mice exhibited ALT levels near baseline levels, consistent with the resolution of virus-induced liver pathology observed in 3d and WT mice (G and H;). Taken together, these results suggest that in the absence of endosomal TLR signals, there is pronounced liver inflammation accompanied by a transient increase in liver damage during acute MCMV infection. ## Impairment of viral clearance in 3d mice Given the diminished early cytokine responses and enhanced liver disease observed in 3d mice, the contribution of endosomal TLR responses to control of virus replication in the liver was assessed in WT and 3d mice infected with MCMV. Compared with those in WT mice, viral titers were elevated by 1 log on day 3 of infection in 3d mice, and remained significantly higher at day 5 post- infection when compared to WT. Ultimately, while WT livers showed evidence of viral clearance at days 7 and 9 post-infection, 3d liver virus titers remained approximately 2 logs higher at these times of infection as compared to WT mice. Thus, endosomal TLR signaling contributes to the control of MCMV replication in the liver during acute infection. # Discussion The aim of these studies was to identify the TLR signaling pathways required for the innate recognition of virus infection in the liver, a common target organ of many viruses that significantly contributes to innate immune defenses. Moreover, the liver contains various innate immune cells that express TLRs, ; however, the role of TLRs in host defense against infection at this site remains largely unclear. Because responses in the liver do not appear dependent on individual TLRs, we utilized 3d mice, which lack endosomal TLR3, TLR7 and TLR9 signaling due to a mutation in the endoplasmic reticulum-resident protein UNC93B1 , to address the contribution of endosomal TLRs in liver antiviral defenses against acute MCMV infection. The results demonstrated impaired production of proinflammatory cytokines by NK cells and pDCs in livers from MCMV-infected 3d mice. Additionally, 3d mice had elevated viral titers in the liver that coincided with transient but exacerbated liver disease, although virus-specific CD8+ T cell responses were not affected. Interestingly, a previous study demonstrated that TLR3 was not required for the generation of adaptive antiviral responses to MCMV, although there is evidence that TLR3 signaling contributes in part to the early control of MCMV infection by the systemic induction of type I IFN. Other studies have implicated a synergistic role for TLR7 and TLR9 in promoting MCMV recognition and immune defense in the spleen. The impaired liver responses observed in 3d mice that were not apparent in TLR9 or TLR7-deficient mice suggests that a level of redundancy unique to innate immunity is in place within infected tissue sites to rapidly respond to viral infection. Overall, these studies advance our understanding of the process of viral recognition in the complex liver environment and suggest that UNC93B1 is a critical intermediate factor in innate virus sensing activated by MCMV. Previous studies have demonstrated impaired serum cytokine production and increased susceptibility to infection in 3d mice infected with a high lethal dose of MCMV ; however, our study is the first report to document the contribution of the 3d mutation to impaired MCMV defense in the liver. In agreement with the study by Tabeta *et al.*, our results, using a moderate dose of MCMV, demonstrated a diminished serum cytokine response, reduced splenic IFN-α production and elevated viral titers in the spleen (data not shown). These results were not unexpected given the known role for the nucleic acid-sensing TLRs in MCMV recognition and the subsequent production of proinflammatory cytokines and type I IFN by splenic pDCs. Production of type I IFN is a critical early step in antiviral defense, and we further demonstrated diminished levels of liver IFN-α in MCMV-infected 3d mice. This defect was in part due to an impaired ability of liver pDCs to express this cytokine and is consistent with previous reports identifying pDCs as an important early source of IFN-α in response to TLR7 and TLR9 ligands,. Altogether, these results concur with previous studies indicating that pDCs are the predominant leukocyte producer of type I IFN in the liver during early MCMV infection, and demonstrate that production of these cytokines by pDCs is influenced by the 3d mutation. Despite impairments in liver IFN-α production in 3d mice, it is notable that this response was not totally abrogated. This suggests the presence of alternative pathways to type I IFN production in the liver, and may be the reason that IFNα/βR-deficient mice die by day 5 in response to infection with a moderate dose of MCMV, as reported previously, while 3d mice do not. Studies have also demonstrated production of type I interferon from cells other than pDCs at 44–48 hours following MCMV infection. In addition, the liver contains a variety of parenchymal and non-parenchymal cells that express TLRs and are capable of type I IFN production,. There may also be TLR-independent pathways leading to the production of these cytokines in response to MCMV, including cytoplasmic RNA- and DNA-sensing receptors, as have been demonstrated with other virus models but have yet to be examined in the context of MCMV infection,. It has been established that NK cell inflammatory responses and production of IFN-γ are essential to defense against MCMV in the liver –. In 3d mice, NK cell production of IFN-γ in the liver was severely impaired during early MCMV infection and likely contributed to increased viral burden and liver pathology. The reduced levels of IFN-γ in NK cells may reflect the deficiency of serum IL-12 seen in 3d mice. These results concur with previous reports that type I IFN regulates IL-12 production by conventional DCs and consequently the production of IFN-γ by NK cells. The defect in liver cytokine production in 3d mice is reminiscent of that reported for MyD88-deficient mice. Notably, however, MyD88-deficient mice exhibited more severe liver pathology and greater elevations in viral titers when compared to WT than we observed in 3d mice. Interestingly, mice deficient in TLR7 and TLR9 exhibited decreased levels of systemic IFN-α/β and increased susceptibility to MCMV infection. Taken together, these observations further support the notion that the liver possesses compensatory mechanisms to combat viral infection in the combined absence of endosomal TLR signals. Accordingly, despite the early effects of endosomal TLR deficiency, 3d mice were able to mount robust CD8+ T cell cytokine responses. It should be noted that the 3d defect has previously been shown to impair exogenous antigen presentation, including cross presentation, which has a reported role in priming CD8+ T cell responses during MCMV infection. However, we detected no overt defect in CD8+ T cell responses within the first seven days of MCMV infection in the liver. Further, examination of activation markers suggested that a similar proportion of CD8+ T cells from 3d mice were more highly activated when compared to WT (data not shown). Several studies have demonstrated the contribution of activated virus-specific CD8+ T cells to effective hepatic immunity against MCMV infection. In addition, the normal development of adaptive responses despite impaired innate responses is well documented during MCMV infection. Studies have shown that reduced levels of type I IFN do not affect the accumulation or activation of antigen-specific CD8+ T cells in response to low or moderate MCMV inoculums. Likewise, while IL-12 is critical in inducing NK cell IFN-γ expression, T cell responses can occur in an IL-12-independent manner,. NK cells have the potential to negatively regulate CD8+ T cell responses during MCMV infection –; thus, it is probable that impaired NK cell function in the absence of endosomal TLR signals contributes to inflated T cell responses. The presence of increased virus in the liver may also contribute to the robust T cell recruitment and cytokine production observed at late infection time points in 3d mice. Interestingly, although the severity of viral liver pathology was more pronounced in 3d mice, inflammation and liver injury subsided late in infection. These observations suggest that CD8+ T cells in the liver can respond to limited amounts of type I interferon for activation in the presence of compromised innate responses, emphasizing the prevalence of compensatory mechanisms in place within the liver to deal with infection and promote adaptive immunity. In contrast, IFN-α/βR deficiency negatively impacts innate inflammatory responses and resistance to MCMV infection in the liver,. In conclusion, this study indicates that UNC93B1, which is essential for combined endosomal TLR signaling, contributes to development of effective innate immune responses to an acute virus infection in the liver. Our results show that this contribution involves modulation of early innate proinflammatory cytokine production from liver pDCs and NK cells and subsequent control of MCMV replication and pathology before activation of an adaptive immune response. Altogether, these results highlight a process of virus recognition with multiple pathways in place to promote host resistance to infection in the liver microenvironment. # Methods ## Mice Pathogen-free C57BL/6J mice were obtained from the Jackson Laboratory (Bar Harbor, ME). C57BL/6 *Unc93b1^3d/3d* mice were a kind gift from Dr. Bruce Beutler (The Scripps Research Institute, La Jolla, CA) and were generated as described. C57BL/6J mice were housed and *UNC93b1^3d/3d* mice were bred in pathogen-free mouse facilities at Brown University. Male and female mice aged 8–10 weeks were used in experiments. This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All animal work was approved by the Brown University Institutional Animal Care and Use Committee (Protocol Number: 0909082 and 0903035). ## Virus Infection and viral titer determination MCMV Smith strain was used in all experiments. This strain was prepared as a salivary gland-passaged stock from CD1 mice. Moderate dose infection (5×10⁴ PFU per mouse) was initiated on day 0 by intraperitoneal injection. In vivo responses were measured at indicated time points. For infectious viral titer quantification, organs were weighed, homogenized in cold supplemented DMEM (Invitrogen Life Technologies) and supernatants were collected following centrifugation. Serially diluted samples were used to inoculate confluent monolayers of bone marrow stromal cells (ATCC M2-10B4) in 24-well tissue culture plates and incubated for one hour at 37°C, 5% CO₂. Following incubation, inoculums were removed and monolayers were overlaid with a 1×DMEM/0.5% low-melt agarose solution. Cells were incubated for 7 days at 37°C, 5% CO₂, then fixed in 10% buffered formalin and stained with crystal violet. Plaques were counted to determine viral titer as previously described. ## Sample preparation Liver leukocytes were prepared as previously described. Briefly, following mechanical dissociation of tissue, red blood cells were removed by lysis with ammonium chloride and leukocytes separated by Percoll density gradient. To generate homogenates for cytokine analysis, the liver caudate lobe was homogenized in RPMI 1640 (Invitrogen Life Technologies) and supernatants collected following centrifugation. Serum was isolated from whole blood by centrifugation in the presence of heparin and stored at −80°C until further use in cytokine analyses or ALT assays. ## Cytokine analysis Liver homogenates and serum were tested for cytokines by standard sandwich ELISA. IFN-γ, IL-12p70 and TNF-α were assayed by DuoSet (R&D Systems). IFN-α was measured by VeriKine mouse ELISA kit (R&D Systems). Limits of detection were 15 pg/mL for DuoSets and 12.5 pg/mL for VeriKine ELISA kits. ## Flow cytometric analysis The following fluorochrome-conjugated mAbs were used in flow cytometric analyses: NK1.1-PE and TCRβ-APC to distinguish NK cells; CD8α-PECy7 and TCRβ- FITC to distinguish CD8+ T cells; and PDCA-1-APC (Miltenyi Biotec) to identify pDCs. Prior to surface staining, cells were incubated with anti-CD16/CD32 mAb to block nonspecific binding of Abs to Fcγ III/II receptors (clone 2.4G2). Unless otherwise noted, antibodies were obtained from BD Biosciences or eBioscience. For intracellular staining of cytokines, cells were treated with Brefeldin A (eBioscience) for 4 hours at 37°C, 5% CO₂ and permeabilized prior to labeling with IFN-α-FITC (R&D Systems), IFN-γ-PE, or TNF-α-APC (BD Biosciences). When indicated, leukocytes were stimulated for 5 hours with 100 ng/mL MCMV M45 peptide in addition to Brefeldin A treatment. Samples were acquired using a FACSCalibur and analyzed with BD Cell Quest software. For analysis, viable cells were gated by FSC and SSC. Isotype controls were used to set positive analysis gates. ## Liver histology and serum ALT analysis Portions of the median liver lobes were isolated, fixed in 10% neutral buffered formalin, and paraffin embedded. Tissue sections (5 µm) were stained with H&E and analyzed microscopically. Images shown were photographed at the indicated magnification with a DP70 digital camera and software (Optical Analysis Corporation). Inflammatory foci, defined as discrete clusters containing between 6–60 nucleated cells, were quantitated as described previously. In brief, inflammatory foci were identified, at a magnification of 200, as clusters of cells in totals of 8×50 µm² areas per representative tissue. In some cases, liver sections had larger areas of inflammation with \>60 cells with less defined foci, and are indicated as such. Numbers of nucleated cells per inflammatory foci were counted in 20 individual foci per representative tissue at a magnification of 400. Liver alanine aminotransferase (ALT) was measured in serum samples by Marshfield Labs (Marshfield, WI). ## Statistical analysis Student's *t* test was used to determine statistical significance of experimental results when indicated (p≤0.05). We acknowledge Paula Weston from the Molecular Pathology Core Facility at Brown University for assistance with histology sample preparations. We would also like to thank Dr. Bruce Beutler of the Scripps Research Institute, La Jolla, CA, for generously providing UNC93B1*^3d/3d* mice for these studies. [^1]: Conceived and designed the experiments: MJC TSM. Performed the experiments: MJC PJG. Analyzed the data: MJC TSM. Wrote the paper: MJC TSM. [^2]: The authors have declared that no competing interests exist.

# Introduction Since the 1980s, lithic researchers have worked to develop a series of methods to measure reduction intensity, particularly as a tool for interpreting curation or re-sharpening of tools and morphological variability, and they have applied them to both experimental and archaeological assemblages. Among them, the quantitative reduction indices proposed by Dibble (the ratio of the remaining surface area to platform area), and Kuhn (the ratio of flake thickness at the point where retouch scars terminate to maximum medial thickness, also well-known as a Geometric Index of Reduction) are the two most influential indices. These two indices and their modified versions are still frequently applied in current research, which confirms their usefulness. Generally speaking, these methods for measuring reduction mainly focus on morphological attributes to estimate reduction intensity. In other words, as reduction continues, the corresponding size, shape and other morphological properties change as well. Shott and Weedman have summarized them as three specific methods, namely, a simple size measurements method, a geometric measurements method, and an allometric method that relates shape and other attributes to size. Allometry has also influenced the reduction study of handaxes. Based on the reduction hypothesis, McPherron reanalyzed 38 British handaxe assemblages classified by Roe into either pointed or ovate groups. He argued that these shapes actually reflect different reduction intensities, with pointed handaxes in an initial stage of reduction and ovate handaxes in a later stage. For quantitatively measuring the reduction of handaxes, McPherron employed linear measurements, namely the tip length, overall length and width in his study and assumed that pointed handaxes would have both a long tip length and a long overall length. As reduction continued, tip length and length decreased quickly, but width decreased at a lower rate, finally leading to the formation of ovate shaped handaxes with lower elongation values (Length/Width). Thus, through comparing the tip length, length and elongation ratio, McPherron argued that it is possible to estimate the reduction intensity of different handaxe assemblages. This allometric method provided a new perspective in interpretation of the morphological variability of handaxes, and more importantly, it shifted attention from the final form of handaxes to their reduction process and flaking strategies. Consistent with this allometric method, most current reduction intensity analyses of handaxes are now integrated into studies of morphological variability. In contrast to McPherron’s analysis of the size and shape of handaxes in relation to re-sharpening, McNabb et al. proposed a method for the technological study of ‘shaping’, without reference to re-sharpening. By recording the extent of secondary flaking (flake scars \> 1.5cm in length) and the degree of edge trimming (flake scars \< 1.5cm in length), the authors identified five subcategories of secondary flaking for each face of a handaxe and five ordinal scales of edge trimming for each section of a handaxe (see McNabb et al., Figs). The frequencies of these attributes are then classed by the extent of reduction as light, moderate or extensive. This is a detailed and useful approach to reduction analysis of handaxes involving technological attributes. However, a potential problem may lie in the subjective divisions of different types of secondary flaking and different scales of edge trimming, which can vary according to the observer. Another easily applied approach used by some researchers is the direct counting of the number of scars on a handaxe. Scar numbers are relatively easier to quantify at different stages of reduction for Early Stone Age (Lower Palaeolithic) handaxes than for the generally small-sized tools in the Middle Stone Age (Middle Palaeolithic) and Later Stone Age (Upper Palaeolithic). The assumption here is that extensively retouched handaxes would have more flake scars than less reduced handaxes. Using this concept, Hou et al. compared the number of flake scars on the Bose Large Cutting Tools (LCTs; \~0.803Ma) with the number of scars on LCTs in two western Acheulean assemblages of similar age (0.99Ma-0.7Ma; Olorgesailie Members 1 through 7 in Kenya and Bed IV of Olduvai in Tanzania). They concluded that the Bose LCTs have a similar number of scars as the western Acheulean, and therefore they support the proposal that there is no technological difference between handaxes in the East and West. In addition, in Sharon’s comparative study of handaxes from Africa, West Asia and India, the number of scars was also regarded as an important attribute for the analysis of reduction extent. Although the counting of flake scars is a useful and easily applied approach, it also has one limitation. As mentioned by Sharon, the visible number of flake scars on discarded handaxes is likely lower than the flake scars generated during manufacture, as a portion of the piece is lost in the process. For example, a handaxe with 20 flake scars but of a smaller size is not definitely less retouched than a handaxe of larger size with 30 flake scars. Considering the number of scars in conjunction with the size of handaxes would make this attribute size-independent and improve its value. Coincidentally, in the analyses of core reduction intensity of some East African Oldowan sites, Braun et al. also suggested that flake scar number divided by mass of the piece is a more appropriate measure of reduction intensity. For exploring the use-life and implied human behaviors of handaxes (e.g., raw material transport), Shipton proposed a flake scar density index in his analysis of Indian and East African material. Specifically, the scar number on a handaxe is divided by the product of the handaxe length and width as an indication of the surface area. The principle here is that a handaxe will start off with a low flake scar density, and as the reduction progresses, the value of flake scar density will steadily increase. This is a size-independent method which addresses the limitation of the scar number approach discussed above. Due to the imprecision in measuring surface area, Shipton et al. then applied a 3D technique to capture the area more accurately, producing a 3D surface area. Simultaneously, Clarkson used a similar method to measure the reduction intensity of different types of cores (with bifaces included as one core type), and he introduced the Scar Density Index (SDI, or the ratio of flake scar number to 3D surface area). Moreover, Clarkson, Clarkson et al. and Shipton and Clarkson have used both experimental and archaeological materials to reinforce the reliability of this index. The purpose of this paper is to present a 3D quantitative analysis of reduction intensity of a handaxe assemblage from the Danjiangkou Reservoir Region (DRR), central China. Here, reduction of the DRR handaxes includes both shaping and the probable re-sharpening process, as these two aspects cannot be objectively distinguished, especially in cases where the reduction intensity is generally low, as in DRR. In addition to the 3D Scar Density Index (SDI), a Flaked Area Index (FAI) which can quantify the reduction extent in different parts (i.e. tip, medial and base) of a handaxe, will also be used. # Materials and Methods ## Materials The handaxes analysed in this paper are from both surface collections and excavations on the third terrace (T3) of the Danjiangkou Reservoir Region (DRR), central China. Systematic investigations and excavations over the last two decades in this area have revealed it as another important handaxe-bearing region in China, along with the well-known regions of Dingcun, Bose and Luonan. To ensure accurate measurements of surface area and volume for handaxes, only complete specimens are employed in our study, namely 92 handaxes in total (see for raw data of each specimen). Of these, 76 are surface collected and 16 are excavated from the third terrace of the Han River, the longest tributary of the Yangtze River and the main feeder of the Danjiangkou Reservoir. The surface- collected material has been statistically confirmed to be consistent with excavated specimens in both morphology and technology. In terms of the type of shaping of the DRR T3 handaxes, with the exception of one indeterminate specimen, 37 (40.2%) were bifacially shaped, 38 (41.3%) were partly bifacially shaped and only 16 (17.4%) were shaped unifacially (See). The ESR, OSL and palaeomagnetic dating of the third terrace at the Shuangshu and Maling 2A sites indicate that these two handaxe-bearing sites formed in the Middle Pleistocene. In addition, 25 surface collected handaxes from the second terrace (T2) of the Han River were also used in a comparative study of differences in reduction intensity through time in the DRR. The T2 Dishuiyan site is dated to ca 100–50 ka by the OSL and TT-OSL methods. Here more than 20 handaxes comparable to the 25 surface-collected specimens used in this study were excavated. ## Ethics statement The surface-collected handaxes (N = 101) were retrieved during field investigations carried out by one of the co-authors, Chaorong Li, and permission to study these materials was issued by the Institute of Vertebrate Paleontology and Paleoanthropology (Beijing), Chinese Academy of Sciences, in which these specimens are stored. The excavated handaxes (N = 16) are stored in the Danjiangkou Museum in Danjiangkou City, Hubei Province, and study of these materials was permitted by this museum. We ensure that the fieldwork did not involve endangered or protected species. ## 3D scanning and measuring For capturing the 3D image of each handaxe, we used two types of laser scanners. The NextEngine 3D Laser Scanner was used in the field, as it is light and portable. To scan the whole surface of a handaxe, we conducted two separate scans in vertical and horizontal views which were subsequently merged into one complete 3D image. In the laboratory, the Range 7 3D Laser Scanner was used, as it gives excellent resolution but is difficult to carry in the field. Handaxes were rotated manually to obtain a complete 3D image. The mean value of triangles representing the degree of resolution is two times higher with the Range 7 scanner than with the NextEngine scanner. Holes on images were filled using the Geomagic Studio software, regardless of the scanner used. After attaining the 3D images, we then imported them into the Avizo Fire 3D Imaging Software (version 8.0) to accurately calculate the surface area and volume of the handaxes. The segmentation function of this software was also applied to divide the handaxes into three portions with the piece aligned along the long axis, using the distal end as the guide. The scar coverage of each portion was then accurately extracted using this software. An example is given in, which shows the proportional division of the surface into distal, medial and proximal parts based on the length, for which the area of flake scar coverage is then calculated in each sector. ## Indices of SDI and FAI Flake scar density (equal to the Scar Density Index, SDI) has been used as an effective indicator of reduction intensity of Indian and East African handaxes. In an experimental test of the reliability of SDI, the regression analysis of transformed log SDI and log % Mass Remaining of bifaces produced a very strong relationship (*R*^*2* = 0.916). In addition, a recent study of the experimentally reduced handaxes also indicates a strong relationship (*R*^*2* = 0.803) between increasing SDI and decreasing % original mass. For this reason, the 3D SDI was adopted in our analysis of reduction extent of the DRR handaxes. All visible flake scars on a handaxe were counted in this study, regardless of their interpretation as shaping vs. refining scars. Because of the fresh condition of the DRR handaxes, it is easy to count the scar numbers. And because most of these handaxes were made on primary flakes or cobble opening flakes, dorsal scars on flake handaxes were mainly formed by the subsequent flaking. Therefore, all dorsal scars were counted as flaked area. However, the flaked area does not include the ventral surface of handaxes made on flakes if no scars are present. As a result of the use of 3D technology, a new index, Flaked Area Index (FAI, flaked area divided by the total surface area), is now possible. A reasonable assumption for this index is that the flake scars area on handaxes increases with reduction. The unretouched blank for a handaxe would have a FAI value of 0, while a completely retouched handaxe would reach a FAI value of 1. Through the accurate measuring of the flaked area in different sectors of the handaxes, we can estimate not only the overall reduction intensity, but also the reduction intensity of the different parts, which is a benefit of FAI. We need to bear in mind that the flaked area does not necessarily relate to the number of flake scars. This is particularly applicable to hard hammer percussed handaxes, in which a small number of large scars can produce a large area of scar coverage, and conversely, a large number of small scars can produce a small area of scar coverage. Despite this, the FAI index can reflect the general pattern of the reduction extent of handaxes. In addition, it should be noted that both the SDI and FAI indices will reveal the relative extent of reduction, but not the actual mass lost during the reduction. In order to investigate quantitatively how much mass has been lost in the reduction process, it is necessary to conduct knapping experiments in future research. # Results ## The application of SDI to reduction intensity Correlation analysis between volume and SDI shows that the relationship between volume and SDI is significant (*r* = 0.523, *p* \< 0.001), although there is considerable variation. In addition, to test the effect of outliers, we exclude specimens (N = 9) whose SDI values are larger than 0.1. Results show that the correlations between volume and SDI with and without outliers are very close to each other (*r* = 0.523 vs. *r* = 0.520). Therefore, the DRR handaxes can be confirmed to be made from similar size cobbles, and we suggest that the size- independent SDI used in this study is appropriate for measuring the reduction intensity of handaxes. shows three handaxes (left side of the figure) that are low in volume and weight but high in SDI, and three handaxes (right side of the figure) that are high in volume and weight but low in SDI. Detailed information for each specimen is provided in. Although the indices of SDI used here cannot show how much mass is lost during reduction, they do provide us with information about the relative intensity of handaxe reduction. Adapting statistical models used in demography, Shott and colleagues suggest that different distributional patterns of reduction could correspond to different cumulative-survivorship curves. In other words, the fit between the length of handaxe use and SDI has the ability to reveal the underlying use-lives of tools and the related human behaviours. In, we can see that most of the DRR handaxes possess lower SDI values and locate on the left side of the histogram, with 83.5% (N = 76) of them lower than 0.09 on the SDI value. Only a small number of handaxes has relatively higher SDI values and locate on the right side of this diagram. Therefore, it is reasonable to infer from that most of the DRR handaxes were less extensively reduced and generally had short use-life before discard. This situation may relate to the use of locally available raw materials close to the site and brief occupation periods. For visualizing the reduction intensity of the DRR handaxes, six handaxes with the values of SDI from 0.08 to 0.09 are presented in (see for information on individual specimens). ## The application of FAI to reduction intensity Correlation analysis between SDI and FAI shows that these two indices are significantly related (*r* = 0.424, *p* \< 0.001; see), although there is considerable variation. This indicates the validity of FAI in estimating the reduction intensity of DRR handaxes. Based on the 3D segmentation of handaxes into three proportional parts, namely, tip, middle and base, we can examine the pattern of reduction intensity in each sector. and show that the values of FAI for tips are high, with 34.8% of them scoring in the range 0.75–0.99 and 41.3% of them fully covered by flake scars. The mean value of FAI for tips is 0.87. For the middle sections, 59.7% are concentrated in the range 0.26–0.75 and 30.4% in the range 0.76–0.99, and the mean value for middle sections is 0.69. The FAI values for the bases are generally low, with 32.6% scoring in the range 0.01–0.25, 34.8% in the range 0.26–0.50 and 23.9% in the range 0.51–0.75, and a mean value of 0.37. Therefore, we can conclude that most of the shaping was invested in the tips of the DRR handaxes, while the bases usually have the least reduction, with an intermediate degree of reduction in the middle sectors. The reason for this pattern is likely related to the functional differences for each part: the tip is the most utilised part, while the middle may be related either to use or to shaping of the tip; and the smooth cobble surface is often left on the base for holding comfort. This conclusion is supported by using the sector method which records shaping status and calculates the proportion of shaping or cortex covered in each sector in the whole handaxe assemblage. Analysis of 36 handaxes found from 2004 onwards shows that shaping was mainly concentrated in the distal ends of DRR handaxes (86.8%), while the proximal ends were mainly occupied by cortex (50.0%), with a smaller percentage of shaping (38.9%; see Li et al.). The overall extent of reduction for the whole body of handaxes ranges from 0.26–0.99, with over 45.7% in the range of 0.51–0.75, and the mean value of FAI is 0.60. If a cut-off point of 0.75 is used to represent the boundary between extensive (\> 0.75) and moderate reduction (≤ 0.75), we can see that more than three quarters (78.3%) of handaxes in DRR were only moderately reduced. This result is consistent with our analyses of the SDI, where results show that the DRR handaxes generally show a low extent of reduction. ## Comparing reduction intensity for different types of raw materials, blanks and shaping Quartz phyllite (N = 67) and trachyte (N = 15) were the most frequently used raw materials for DRR handaxes, although the number of trachyte pieces is much lower than the number of quartz phyllite in the current study sample. Both raw materials are abundant and locally available in the nearby gravel layers of the Han River. The comparisons of SDI and FAI by raw material show substantial overlap (left side of Figs). This observation is supported by the *t*-test (*t* = -0.305, *p* = 0.761 for SDI; *t* = 0.478, *p* = 0.634 for FAI), which shows there is no statistically significant difference in the levels of reduction between the two raw materials. We can, therefore, further infer that a consistent reduction strategy was employed despite raw material differences. In contrast, there is a clear trend of lower SDI and FAI with cobble blanks (N = 32) compared with flake blanks (N = 45). This is visible on the middle part of Figs and. The *t*-test also shows significant differences in the levels of reduction between the two blank groups (*t* = 2.438, *p* \< 0.05 for SDI; *t* = 2.708, *p* \< 0.01 for FAI). During the experimental test of the effects of blank type on handaxe reduction, Shipton and Clarkson also noticed that cobble blanks tend to have lower SDI compared to flake blanks for a given percentage of mass lost. This difference in DRR T3 handaxes probably relates to the large flat ventral surface provided by flake blanks and their relative thinness compared to cobbles, both of which facilitate reduction. In terms of the effect of types of shaping on the reduction intensity, the Kruskall-Wallis test shows that there are considerable differences for both SDI (ChiSquare = 25.6, *p* \< 0.0001; see) and FAI (ChiSquare = 39.0, *p* \< 0.0001; see) among the three shaping groups, namely, bifacial, partly bifacial and unifacial. Bifacially shaped handaxes have the greatest mean values of SDI (0.087) and FAI (0.743), while the unifacially shaped handaxes have the least mean values of SDI (0.042) and FAI (0.458), with partly bifacially shaped handaxes being intermediate (mean SDI = 0.058; mean FAI = 0.532). Therefore, it is clear that different types of shaping have a strong influence on the degree of reduction for the DRR T3 handaxes. ## Comparing the reduction intensity of handaxes from T2 and T3 of the DRR The 3D quantitative method provides an objective way to compare reduction intensity through time. Here, the 25 handaxes from the second terrace of the DRR are compared with the 92 handaxes from the third terrace analysed in the foregoing sections. The preliminary age of handaxes from the second terrace is the early Late Pleistocene (100–50 ka), while the handaxes on the third terrace are dated to the Middle Pleistocene. Therefore, handaxes in the DRR provide an opportunity to examine the regional change in reduction intensity from the Middle to the early Late Pleistocene. Because of the relatively small sample size from T2, we do not conduct a statistical analysis according to different types of raw materials, blanks and shaping, as was done for the sample from T3. Attributes used in this analysis include the number of scars, surface area, volume, total flaked area, and the indices of SDI and FAI. The *t*-test shows that there are no statistically significant differences (*p* \> 0.05) between handaxes from T2 (the second terrace) and T3 (the third terrace) in any of these attributes and thus no technological change in reduction intensity through time in the DRR is presented. # Summary and Conclusion The extensive application of reduction intensity indices in the past three decades has remarkably improved the ability of lithic analysts to interpret human behaviour. With the aid of 3D scanning technology, in this paper we applied two quantitative reduction indices, the Scar Density Index (SDI) and the Flaked Area Index (FAI), to the analysis of reduction intensity of the T3 DRR handaxes. The SDI in this study shows that most of the handaxes in DRR have a relatively low intensity of reduction, which also indicates a generally short use-life as argued by Shott and Sillitoe’s reduction distribution model. In addition, the short use-lives of these handaxes may suggest that open-air sites along the river terrace were not occupied by hominids for a long time. The analysis of FAI also shows that the overall reduction intensity of the DRR handaxes represents a least-effort reduction strategy, with 78.3% of handaxes only moderately reduced (FAI ≤ 0.75). The detailed FAI analysis of the different parts of the DRR handaxes shows that tips generally show the most reduction, while the bases show the least, a pattern which is relatively common in some Chinese handaxe assemblages. It is not surprising that the handaxe tip would receive most attention in shaping. The middle section could be functional if the edge were used, but it relates also to shaping of the piece overall. The base of a handaxe was the holding unit, with only limited or no shaping present. The two raw materials used in DRR, both locally available, did not have an influence on the reduction intensity, suggesting that the behavioural interpretation of short-term use is correct. In addition and because of the abundance of raw materials, the DRR handaxe knappers seem to have employed an expedient exploitation strategy. However, the type of blanks and the shaping types did play a role in the reduction extent of the DRR handaxes. Those made on flake blanks generally show a higher level of reduction than those made on cobble blanks, presumably because flake blanks were thinner than cobbles, and they provided a large flat surface which made reduction easier. In terms of shaping, bifacially shaped handaxes show a greater degree of reduction than partly bifacial handaxes, with the unifacial handaxes showing the least reduction. The preliminary comparison of handaxes from T2 and T3 of the DRR suggests that there is no technological change in the reduction intensity from the Middle Pleistocene to the early Late Pleistocene in this region, although more specimens from terrace two need to be analysed. The results presented here demonstrate that the application of quantitative technological indices is necessary and useful in estimating the reduction intensity of handaxes. According to this estimation, we can further investigate the behaviour of handaxe makers in the DRR, such as their adaptation to the local raw materials, their energy investment in making handaxes, and the use- life of handaxes. The potential of the indices used in this paper has been confirmed; however, as we have mentioned already, these indices can only indicate the relative extent of handaxe reduction and they still need to be further tested. In future research, experiments will be conducted to further evaluate the mass lost at different levels of the index values, and to support the validity of the current indices. Additionally, owing to the long lasting and widespread use of handaxe technology in the Pleistocene, the reduction intensity of handaxes at different developmental stages and in different regions will be further examined to address the technological evolution and adaptive behaviour of Acheulean hominids. Finally, this study has provided detailed information on the nature of handaxes in the DRR, which will serve as a comparative sample for a better overall understanding of these industries in China, in comparison with the western Acheulean. # Supporting Information We would like to thank Kristian J. Carlson, Tea Jashashvili, Matt Caruana, and Raymond Couzens for their help and suggestions on 3D scanning and the statistics in this research. And we thank Shannon McPherron and one anonymous reviewer for their invaluable and insightful suggestions for improving this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HL KK CRL. Performed the experiments: HL. Analyzed the data: HL. Contributed reagents/materials/analysis tools: KK. Wrote the paper: HL KK CRL.

# Introduction DNA double-strand breaks (DSBs) are among the most serious types of DNA damage in cells and can lead to genetic instability and tumorigenesis. DSB may be induced by exogenous genotoxic insults, such as ionizing radiation, but also occur spontaneously in various cellular processes including DNA replication and V(D)J recombination. There are two major pathways for DSB repair: non-homologous end joining (NHEJ) and homologous recombination (HR). NHEJ directly ligates the two broken ends of a DSB and is accessible throughout the cell cycle. In contrast, HR is an error-free repair mechanism that primarily uses the intact sister chromatid as a template for repair and predominates in S-phase cells. The physiological importance of HR is underscored by the association of cancer predisposition and developmental defects with mutations in HR genes. The tumor suppressors BRCA1 and BRCA2, key players at different stages of HR, are frequently mutated in familial breast and ovarian cancers. Other HR proteins, including PALB2 and RAD51 paralogs, have also been identified as tumor suppressors. On the other hand, tumor cells with defective HR repair show increased sensitivity to chemotherapeutic reagents that act via the induction of DNA damage, including platinum-containing agents and PARP inhibitors. These observations suggest that HR-proficient tumor cells might be sensitized to chemotherapeutics if HR repair could be therapeutically inactivated. In genetic engineering, HR represents a powerful tool to precisely manipulate the genome for experimental uses. Its use revolutionized the genetic approach to study biological processes in mice by the late 1980s with the generation of the first gene targeted knockout mice. Since then, thousands of genes have been modified in mouse embryonic stem cells by HR with gene-targeting vectors. However, the low efficiency of HR in vertebrate somatic cells limits the utility of this approach. Following the discovery that induction of a DSB increases the frequency of homology directed repair (HDR) by several orders of magnitude, targeted nucleases have emerged as the method of choice for improving the efficiency of HDR-mediated genetic alterations. ZFNs (zinc finger nucleases), TALENs (transcription activator-like effector nucleases) and CRISPR/Cas9 (clustered regularly interspaced short palindromic repeats/CRISPR-associated system 9) are all engineered endonucleases that can introduce DSBs at desired locations in the genome. Once a targeted DSB has been made, HDR may reconstruct the cleaved DNA using an exogenous DNA template. Various cell-based assays have been developed for the accurate measurement of HR events. The use of fluorescence has proved to be an effective approach for detecting HR and the most commonly used assay is the direct repeat green fluorescent protein (DR-GFP) assay. The DR-GFP reporter consists of two tandem, inactive copies of the GFP gene, which can be integrated into the cellular genome or expressed transiently. The first GFP copy (SceGFP) has a promoter and contains the I-SceI restriction site with an in-frame stop codon and the second copy (iGFP) harbours truncations at both ends. After cleavage within SceGFP by I-SceI or a Cas9 nuclease, HR uses iGFP as a template to restore the GFP gene to its functional form. Then, GFP fluorescence can be measured using flow cytometry. Here, we report the development of a novel fluorescence-based assay for evaluating cellular HR activity. As opposed to the DR-GFP assay, in our system HR reconstitutes the full-length coding sequence of the enhanced green fluorescent protein (eGFP), which fluoresces much more intensely than GFP, thus facilitating the detection of HR events. Our assay also differs from the DR-GFP assay in that our two non-functional copies of eGFP are not in tandem. Rather, a single copy of the HR substrate is integrated as a single copy into the AAVS1 safe harbor locus and the template donor is delivered exogenously via AAV particles, mimicking the scenario of the HR-mediated correction of a monogenic disorder. Following this procedure, we have detected changes in HR rates resulting from the altered expression of DNA repair genes, similar to those previously reported. We provide a well-characterized tool that will help to further investigate the mechanisms that control HR and analyze the effects of exogenous manipulations, with i.e., drugs, miRNAs, and genes on HR mediated correction. # Materials and methods ## Cells Human colon cancer cell line HCT116 was obtained from the American Type Culture Collection (Manassas, VA). Human embryonic kidney cell line HEK-293 optimized for the packaging of AAV virions (AAV-293) was obtained from Agilent Technologies (Santa Clara, CA). Both cell lines were handled according to each manufacturer’s recommendations. ## Plasmids construction The sequences of the primers used in this study are shown in. The pAAV-MCS- eGFPΔ3’ plasmid (also named as peGFPΔ3’) used to generate the HR reporter cell line was obtained through the following steps. Firstly, a fragment containing the 5’AAVS1 region homology arm (832 bp) followed by a puromycin cassette (1014 bp) was amplified from pAAV-CAGGS-eGFP plasmid with primers 1F and 1R. pAAV- CAGGS-eGFP plasmid was a gift from Rudolf Jaenisch (Addgene plasmid \# 22212; <http://n2t.net/addgene:22212>; RRID:Addgene_22212). The puromycin cassette (SA-T2A-Puro-pA) is promoterless and contains a splice acceptor site followed by the coding sequence of the T2A peptide, the puromycin resistance gene and a polyadenylation signal. Secondly, a 1290 bp fragment containing human cytomegalovirus (CMV) immediate early (IE) promoter and eGFP lacking the last 73 bp was amplified from peGFP-C1 plasmid (Clontech) with primers 2F and 2R. Thirdly, the 3’AAVS1 region homology arm (1300 bp) was amplified from HCT116 genomic DNA using 3F and 3R. The three fragments were digested and cloned into the NotI site of pAAV-MCS (Stratagene). With respect to the designation of the primers used in this work, F stands for forward, and R for reverse. The HR donor plasmid, pAAV-MCS-eGFPΔ5’ (also named as peGFPΔ5’), was obtained through the following steps. Firstly, a 929 bp fragment containing eGFP lacking the first 38 bp at 5’ end was amplified from peGFP-N1 plasmid (Clontech) with primers 4F and 4R. Secondly, an EcoRI-SalI fragment (1582 bp) containing a blasticidin resistance cassette flanked by loxP sequences was obtained from a pBluescript II based plasmid previously generated in our lab. Thirdly, the 3’AAVS1 region homology arm (1300 bp) was obtained as described for pAAV-MCS- eGFPΔ3’ plasmid. The three fragments were digested and cloned into the NotI site of pAAV-MCS (Stratagene). AAV2 particles were produced by co-transfecting 293 c18 cells (ATCC CRL-10852) with the pAAV-MCS- eGFPΔ5’ donor plasmid, pHelper and pAAV-RC (at a 1:1:1 ratio). Three days later cells were harvested and AAV were released by 4 freeze-thaw cycles. Viral titers were determined by SybrGreen based real time qPCR using ITR_F and ITR_R primers as previously described. pHelper and pAAV-RC plasmids were obtained from Stratagene. The RAD52 gene of *Saccharomyces cerevisiae* (ScRAD52) was cloned into the expression vector pET15b (Novagen) for His-tagged production of TAT-NLS-RAD52 where TAT peptide (GRKKRRQRRR) promotes cell permeability and NLS peptide (KKKRKV) is a nuclear localization signal. Wild-type RAD52 sequence was amplified by PCR from the genomic DNA of *Saccharomyces cerevisiae* with primers ScRAD52_F1 and ScRAD52_R1 and cloned into HindIII/XhoI digested pTriEx-HTNC vector immediately downstream the His-TAT-NLS sequence. pTriEx-HTNC was a gift from Klaus Rajewsky (Addgene plasmid \# 13763). The resultant construct was digested with NcoI and XhoI and the His-TAT-NLS-ScRAD52 fragment was cloned into NcoI/XhoI digested pET15b, thus obtaining pET15b-TAT-NLS-ScRAD52. pET15b-TAT- NLS-ScRAD52 was transformed into BL21 (DE3) and the selected bacteria were grown. His-TAT-NLS-ScRAD52 expression was induced with 1 mM IPTG for 3 h and the recombinant protein was purified using Nickel-Sepharose beads from the soluble fraction of the bacterial extracts. Recombinant protein was stored in a solution containing 50% (v/v) glycerol, 20 mM HEPES (pH = 7.4) and 500 mM NaCl. Several concentrations of TAT-NLS-ScRAD52 ranging from 0.02 to 2 μM were tested for their capacity to increase the HR frequency. The maximum frequencies were obtained with concentrations equal to or greater than 0.2 μM, and a significant level of cytotoxicity was observed only at concentrations higher than 1.8 μM. The TAT-NLS-ScRAD52 experiments shown in this work were performed using the fusion protein at a concentration of 0.2 μM. DNA fragments encoding ScRAD52, RAD51, and RAD52 Flag-tagged at the N-termini were generated by PCR and cloned into mammalian expression vector pcDNA3 (Invitrogen). The Flag sequence was added to the forward primers. The restriction sites used in the cloning step are shown in. ScRAD52 was amplified with the primer pair ScRAD52_F2/ ScRAD52_R2 using pET15b-TAT-NLS-ScRAD52 plasmid as template. Human RAD51 was amplified from the plasmid CMV-hRad51 using primers hRAD51_F/hRAD51_R. CMV-hRad51 was a gift from David Liu (Addgene plasmid \# 125570; <http://n2t.net/addgene:125570>; RRID:Addgene_125570). Human RAD52 was amplified from the plasmid pMM1413-SUMO-RAD52 using primers hRAD52_F/hRAD52_R. pMM1413-SUMO-RAD52 was a gift from Mauro Modesti (Cancer Research Center of Marseille). The resultant constructs were named pScRAD52, phRAD51, and phRAD52. The plasmid encoding for Flag-PALB2 was identified as phPALB2 in this work and corresponds to the expression plasmid pDEST-FRT/T0-Flag-PALB2. pDEST- FRT/T0-Flag-PALB2 was a gift from Daniel Durocher (Addgene plasmid \# 71114; <http://n2t.net/addgene:71114>; RRID:Addgene_71114). The constructs were transfected into the reporter cell line when indicated, and the expression of the Flag-tagged HR promoters was analyzed by Western Blot using mouse monoclonal antibodies against Flag peptide (clone M2, Sigma-Aldrich) and β-actin (AC-40; Sigma-Aldrich) as the loading control. ## Generation of the HCT116-eGFPΔ3’ reporter cell line and HR-mediated rescue of eGFP expression HCT116 cells were nucleofected with pAAV-MCS-eGFPΔ3’ plasmid and AAVS1 ZFN mRNA (Sigma-Aldrich). AAVS1 ZFN mRNA encodes a pair of ZFNs that target the genomic integration site of AAVS1. Targeted integration of pAAV-MCS-eGFPΔ3’ in puromycin-resistant individual clones was verified by PCR using the following primers: P1F and P1R for analysis of 5’-arm recombination; P2F and P2R for analysis of 3’-arm recombination. Homo- and heterozygosity of the eGFPΔ3’ transgene at the AVVS1 locus was explored by PCR using primers P1F and P2R located outside the homology arms. The single copy integration of eGFPΔ3’ into the AAVS1 locus was verified by Multiplex Ligation-Dependent Probe Amplification (MLPA) and droplet digital PCR (ddPCR) (see below). The resultant cell line was named HCT116-eGFPΔ3’. HCT116-eGFPΔ3’ cells were transduced with AAV particles containing pAAV-MCS-eGFPΔ5’ donor plasmid (MOI of 10³). HR leads to reconstitution of full-length eGFP coding sequence and the appearance of green fluorescent cells 48 hours post-transduction. Individual clones were obtained by limiting dilution in the presence of blasticidin (5 μg/ml) and were analyzed by PCR with primers P3F and P3R. The following primers against human SDHA were used for the genomic DNA loading control PCR: SDHA_F and SDHA_R. The restored full length eGFP cassette was also sequenced and its expression analyzed by Western Blot using mouse monoclonal antibodies against GFP (clone B34, Biolegend) and β-actin (AC-40; Sigma-Aldrich) as the loading control. ## Multiplex ligation-dependent probe amplification (MLPA) MLPA reactions were performed according to the manufacturer’s general recommendations (MRC-Holland) with the use of the probes designed and generated according to the strategy developed and described in detail before. Briefly, 100 ng DNA in 5 μl from parental HCT116 and reporter cell line were denaturated for 5 min at 98°C, cooled to room temperature and mixed with 1.5 μl of probes mix (containing 1.5 fmol of each probe), and 1.5 μl of SALSA hybridization buffer. The reaction was then denatured at 95°C for 1 min and probes were hybridized to their respective targets at 60°C for 16–20 h. After the addition of 32 μl of ligation mixture, the hybridized probes were ligated together at 54°C for 15 min. After heat inactivation, ligation reaction was cooled to room temperature, mixed with 10 μl of PCR mixture (polymerase, dNTPs, and universal primers, one of which was labeled with fluorescein) and subjected to PCR amplification for 35 cycles. The MLPA products were subsequently diluted 20x in HiDi formamide containing GS Liz600, which was used as a DNA size standard and separated by capillary electrophoresis (POP7 polymer) using an ABI Prism 3130XL apparatus (Applied Biosystems). The electropherograms were visualized and analyzed using GeneMarker software v1.91 (2.4.0). The eGFP-specific MLPA probes and 7 control MLPA probes are listed in. For each probe, the sequences of the 5’ and 3’ half- probes, universal sequences to which PCR primers are targeted, the “stuffer” sequence, and the target specific sequence are provided. Details on gene loci and chromosome locations for control probes are also provided. ## Droplet digital PCR (ddPCR) The ddPCR was performed using the QX200 system and EvaGreen Supermix (BIO-RAD), according to the manufacturer’s general recommendations, as generally described and used before. To determine the exact copy number of the eGFP transgene in the reporter cell line, we designed two test-amplicons entirely located in the transgene (eGFP1 and eGFP2) and two control-amplicons \[C1 (up) and C2 (down)\], located upstream (2 Kb) and downstream (overlapping exon 2 of *PPP1R12C*) of the transgene site (AAVS1), respectively. The following sets of primers were used: (i) T1F and T1R for test amplicon eGFP1 (187 bp in length); (ii) T2F and T2R for test-amplicon eFGP2 (163 bp in length); (iii) C1F and C1R for control-amplicon C1 (up) (170 bp in length); and (iv) C2F and C2R for control-amplicon C2 (down) (238 bp in length). Of note, genomic DNA from the reporter cell line was digested with the *CviQ*I restriction enzyme (New England Biolabs) prior to ddPCR. *CviQ*I was chosen because it cuts the transgene between the sequences corresponding to amplicons eGFP1 and eGFP2, which prevents tandemly repeated transgenes from being incorrectly identified as a single copy. Briefly, ddPCR procedure was as follow: the PCR mixture containing 10 μl of EvaGreen Supermix (Bio-Rad), 1 μl of 4 μM forward primer, 1 μl of 4 μM reverse primer, 4 μl (80 ng) of DNA from the tested cell line and 6 μl of water was partitioned into 20,000 droplets with the use of a QX200 ddPCR droplet generator (Bio-Rad). The generated droplets were transferred to a 96-well plate and amplified in a T100 Thermal Cycler (BioRad) under the following conditions: 5 min at 95°C, followed by 40 cycles of 30 s at 95°C, 30 s at 60°C, and 45 s at 72°C, followed by 2 min at 72°C, 5 min at 4°C, 5 min at 90°C for enzyme inactivation and holding at 12°C. The amplified products were analyzed using a QX200 droplet reader (Bio- Rad), and the number of positive and negative droplets were counted using the QuantaSoft version 1.7.4.019 software (Bio-Rad). The copy number of eGFP was calculated based on the number of positive droplets in eGFP-specific PCR reactions (either eGFP1 and eGFP2) normalized against the average number of positive droplets in control reactions \[C1 (up) and C2 (down)\]. Control amplicons were selected in diploid regions and their copy numbers set to 2. ## HR assay and screening protocol In all the assays, HCT116-eGFPΔ3’ cells were sequentially treated with: (i) the selected siRNA oligoes, or expression plasmids encoding HR promoters, (ii) a TALEN pair designed to facilitate HR, and (iii) AAV particles containing the donor plasmid. In brief, HCT116-eGFPΔ3’ were plated on 12 well plates at 125,000 cells/well. Forty-eight hours later, cells were transfected with siRNAs using TransIT-X2 (Mirus Bio). The siRNAs used in this work are listed in. Silencer Select highly potent and chemically modified siRNAs were used at a final concentration of 5 nM, whereas classical siRNAs were used at a final concentration of 40 nM. Alternatively, cells were transfected with 400 ng of the expression plasmids encoding HR promoters using TransIT-X2. Twenty-four hours after the first treatment, cells were transfected with a TALEN pair designed to facilitate HR. The TALEN pair (named eGFP) targeted a region within the AAVS1 locus adjacent to the 3’end of eGFPΔ3’ in the HCT116-eGFPΔ3’ cell line. We used the fuzznuc software provided by the EMBOSS package to identify the targeting sequences: `5’-TGCCAGAACCTCTAAGGTTT-3’` (sense component) and `5’-TCCCTCCCAGGATCCTCTCT-3’` (antisense component). TALEN expression vectors were constructed using the LIC (ligation-independent cloning) TAL Effector Assembly Kit (Addgene \#1000000023). The kit provides a TALE repeat unit library of 2-mer fragments that were assembled into an expression ready TALEN construct in 2 hierarchical assembly steps. Six hours after the transfection of HCT116-eGFPΔ3’ cells with 1 μg of each of the TALEN expression plasmids, cells were transduced with AAV particles containing pAAV-MCS-eGFPΔ5’ donor plasmid (MOI of 10³). When indicated, reporter cells were treated with TAT- NLS-ScRAD52 (at a final concentration of 0.2 μM) 5 hours after the transfection of the TALEN plasmids and 1 hour before the infection with the AAV particles. After 24 hours, medium was replaced with fresh medium, and after the next 24 hours (48 hours post-transduction), the recombination frequency was determined as the percentage of cells expressing eGFP protein following flow cytometry analysis using a Beckman Coulter Gallios (Beckman Coulter). Data were analyzed using Kaluza software (Beckman Coulter) and were subsequently corrected for transfection and transduction efficiencies ****. Dead cells and debris were excluded based on scatter signals and propidium iodide fluorescence. When indicated, cells were fixed, stained with Hoechst 33342 (5 μg/ml) and imaged using a Nikon Eclipse 90i fluorescent microscope coupled to a Nikon DS-Ri1 CCD camera. The transfection efficiency of the eGFP-TALEN plasmids and siRNAs using TransIT- was \>85% in all the HR experiments shown in this work. Transfection efficiency was measured in parallel to HR experiments by cotransfecting HCT116-eGFPΔ3’ with either eGFP-TALEN plasmids, expression plasmids or siRNAs under the same conditions as in the HR experiments but in combination with 1 μg of peGFP-C1. Twenty-four hours later, cells were analyzed for eGFP by flow cytometry analysis and the percentage of eGFP positive (eGFP+) cells was determined to estimate the transfection efficiency. The efficacy of the siRNAs in gene silencing was validated by a SYBR Green based real-time PCR assay using the primers listed in. HPRT1 and β-actin were used as housekeeping genes. All siRNAs tested showed over 70% target knockdown at 48 hours post-transfection. Transduction efficiency of AAV2 was tested using particles carrying the entire eGFP gene. AAV2 particles were prepared in 293 c18 cells using the plasmids pAAV-MCS-eGFP, pHelper and pAAV-RC (at a 1:1:1 ratio), as described above. HCT116 were transduced with different doses (from MOI 10 to 10⁴). The highest transduction efficiency (measured as frequency of eGFP+ cells) achieved with AAV2 in HCT116 was 10±1.5% (mean ± SEM) at MOI 10³, and corroborate previous findings. ## Statistics All analyses were performed using Prism software (GraphPad). Data are expressed as mean ± SD and were analyzed by using the unpaired Student’s t-test. In all graphs, \* p \< 0.05, \*\* p \< 0.01, \*\*\* p \< 0.001. # Results ## Development of a novel eGFP-based HR reporter system In our assay, HR leads to the restoration of the full-length eGFP coding sequence from two different truncated eGFP copies. eGFP is a variant of the wild-type GFP with higher-intensity emission than GFP, thus facilitating the detection of HR events. In order to generate the two truncated eGFP copies, DNA sequences for amino acid residues shown to be essential for eGFP fluorescence were deleted, such that the coding sequence lacked 72 bp at the 3’ end in the eGFPΔ3’ copy, and 42 bp at the 5’ end in eGFPΔ5’ (including the start codon). The assay was designed for the fragment eGFPΔ3’ (the recombination substrate) to be integrated in the genome, while eGFPΔ5’ (the donor template) is delivered exogenously via AAV particles. We optimized the assay system in the human colon cancer cell line HCT116, which is highly proficient in gene targeting and has been widely used to study the molecular mechanisms of HR. To avoid position effects, we aimed to integrate eGFPΔ3’ into the adeno-associated virus integration site 1 (AAVS1) within the intron 1 of PPP1R12C on human chromosome 19. We chose the AAVS1 locus as the target because it is considered a safe harbor locus for integrating transgenes, since it is constitutively expressed across a variety of cell types, including HCT116 cells, and biallelic disruption of PPP1R12C results in no discernible phenotype. The targeting construct (peGFPΔ3’) contained a left and right homology arm to the AAVS1 genomic integration site flanking the eGFPΔ3’ transgene and a puromycin cassette. The targeting construct design strategy is shown in. The eGFPΔ3’ transgene was under the control of the CMV promoter. The puromycin selection cassette was preceded by a splice acceptor site. Therefore, its expression relied on splicing and was driven by the endogenous PPP1R12C promoter. Finally, HCT116 were nucleofected with the targeting construct peGFPΔ3’ and a well-validated pair of ZFNs engineered to target the AAVS1 locus and increase the targeting rate. Puromycin-resistant clones were screened for successful integration of the transgene by PCR analysis of genomic DNA using different combinations of primers located in the transgene and outside the 5’ and 3’ homology arms as illustrated in shows representative results of PCR analysis of 4 correctly targeted clones which were named HCT116-eGFPΔ3’ followed by a serial number. PCR using primers outside the arms showed all clones were heterozygotes for the eGFPΔ3’ transgene. We randomly selected the cell clone HCT116-eGFPΔ3’ \#1 for further experiments. The presence of the transgene eGFPΔ3’ in the reporter cell line was confirmed by MLPA using two pairs of probes specific for eGFP (eGFP1 and eGFP2) and seven probes specific for diploid control regions (C1 to C7). shows electropherograms of MLPA reactions performed on genomic DNA extracted from parental HCT116 and HCT116-eGFPΔ3’ \#1 cell lines. eGFP-specific signals were present in the HCT116-eGFPΔ3’ \#1 electropherogram (red arrowheads), but not in the parental HCT116 electropherogram, which confirmed the presence of the transgene in the HCT116-eGFPΔ3’ \#1 cells. The peak heights for eGFP were approximately half those of the control probes suggesting that HCT116-eGFPΔ3’ \#1 cells were heterozygous for the transgene. Next, ddPCR was performed to precisely determine the eGFPΔ3’ gene copy number and rule out integration at other off-target sites. Pairs of primers were designed to amplify two test-amplicons (eGFP1 and eGFP2) located entirely within the eGFPΔ3’ transgene and two control-amplicons (C1 and C2) specific for the diploid flanking regions. Of note, genomic DNA from the reporter cell line HCT116-eGFPΔ3’ \#1 was digested with the *CviQ*I restriction enzyme prior to ddPCR. *CviQ*I was chosen because it cuts the transgene between the sequences corresponding to amplicons eGFP1 and eGFP2, thus preventing tandemly repeated transgenes from being misidentified as a single copy. As shown in, the fluorescence intensity of the HCT116-eGFPΔ3’ \#1 DNA droplets when eGFP1 and eGFP2 assays were used was half of that obtained with the control assays, indicating the HCT116-eGFPΔ3’ \#1 reporter cell line contains a single copy of the eGFPΔ3’ transgene. Next, we produced AAV particles harboring the HR donor template that contained a truncated eGFP lacking the first 42 bp, a blasticidin cassette and the 3’AAVS1 homology arm (pAAV-MCS-eGFPΔ5’). AAV particles were added onto HCT116-eGFPΔ3’ \#1 cells at a MOI of 10³ viral particles/cell. Two days after transduction, eGFP+ cells (also named HCT116-rec-eGFP) were detected (**)**. The frequency of recombination was 4.3x10^-3 ± 1.6x10^-3. To formally determine whether or not fluorescence cells had undergone HR, we isolated fluorescent cells to assess if they harbored full-length eGFP coding sequences. Single HCT116-rec-eGFP cell clones were obtained by limiting dilution and then expanded to clonal cell lines in the presence of blasticidin. HR- induced restoration of an intact eGFP gene in fluorescent cells was assessed by PCR and sequencing. Western blot analysis confirmed the rescue of eGFP expression in HCT116-rec-eGFP cells. The absence of a truncated form in HCT116-eGFPΔ3’ \#1 cells is commented in the discussion section. ## Validation of the HR system Next, we evaluated the ability of our assay to identify factors affecting HR. To measure HR activity, eGFP+ cells were counted by flow cytometry and the raw data was subsequently corrected for transfection and transduction efficiencies as described in Materials and Methods and in. In all the experiments that will be described below, HCT116-eGFPΔ3’ \#1 cells were pre-treated with a pair of TALENs designed to induce DSB in the region adjacent to the 3’end of eGFPΔ3’ (the eGFP TALEN pair). The design and generation of the TALEN expression vectors are described in Materials and Methods. The eGFP TALEN pair was added 6 hours prior the transduction with AAV particles containing the donor template. As expected, adding the eGFP TALEN pair led to a significant increase of the basal frequency of specific HR events (from 3.6x10^-3 ± 1.6x10^-3 to 28x10^-3 ± 6.5x10^-3). First, we tested how adding TAT-NLS-ScRAD52 impacted the frequency of HR. RAD52 is an important HR protein that has the strongest effect on *Saccharomyces cerevisiae*. In this regard, the yeast protein, ScRAD52, has been demonstrated to be more effective than its human homologue in promoting HR. The TAT-NLS- ScRAD52 fusion protein was generated as described in the Materials and Methods section. The TAT peptide made ScRAD52 cell permeable and the NLS targeted the protein to the nucleus. TAT-NLS-ScRAD52 was added to the HCT116-eGFPΔ3’ \#1 reporter cell line 5 hours after the transfection of the eGFP TALEN vectors and 1 hour before the infection with the AAV particles containing the donor plasmid. As expected, the addition of TAT-NLS-ScRAD52 resulted in an approximate 3-fold increase in HR frequency ****. We also explored the effect of the overexpression of the following human genes known to promote HR: RAD51, RAD52, and PALB2. The HCT116-eGFPΔ3’ \#1 reporter cells were transfected with expression plasmids encoding the different Flag-tagged HR promoters and 24 hours later they were transfected with the eGFP TALEN vectors. Six hours after the transfection of the TALEN vectors, the reporter cells were transduced with AAV particles containing the donor plasmid. As expected, the overexpression of PALB2 and ScRAD52 led to an increase in HR frequency ranging from approximately 2.5-fold increase in the case of ScRAD52 to 2.8-fold increase when PALB2 was overexpressed ****. It is of note that the plasmid encoded ScRAD52 increased HR frequency approximately as much as TAT-NLS-ScRAD52. In contrast, the overexpression of RAD51 and RAD52 had no significant effect on HR frequency. We comment on this result in the discussion section. The expression of the Flag-tagged proteins was confirmed by Western Blot ****. Second, we explored the effect of various siRNAs targeting well-characterized DNA repair-related genes on the frequency of HR. In these assays, the HCT116-eGFPΔ3’ \#1 reporter cells were transfected with different siRNAs and then treated with the eGFP TALEN and donor plasmid at the same times as those of the overexpression experiments described above. We used two siRNAs for each of the following genes: RAD51, RAD52, PALB2, CTDP1, XRCC6, LIG4, CBP and SMCHD1. As indicated above, RAD51, RAD52, and PALB2 play a major role in HR-mediated DNA repair, as does CTDP1. The efficacy of the siRNAs in gene silencing was validated by real-time PCR assay ****. As expected, the silencing of RAD51, PALB2 and CTDP1 led to a significant decrease in the HR rate ****. The silencing of RAD52 had no effect, which we comment on in the discussion section. The XRCC6 and LIG4 genes encode protein Ku70 and human ATP-dependent DNA ligase respectively, which are the components of the NHEJ repair pathway. It has been reported that a decrease in NHEJ leads to a compensatory increase in HR. Accordingly, XRCC6- and LIG4-silenced reporter cells exhibited increased frequency of HR. Similarly, reporter cells silenced for CBP and SMCHD1 expression exhibited increased frequency of HR. It is important to note that both CBP and SMCHD1 facilitate the recruitment of NHEJ factors at DSB sites, so their depletion was also expected to lead to an increase in HR rates. Of note, the silencing of XRCC6 and CBP led to the greatest increase in HR. Similar results were obtained between different siRNAs against the same gene in all cases. # Discussion In this study, we present the development and validation of a novel assay in which HR reconstitutes the full length eGFP from a 5’ and a 3’ truncated copies of eGFP. The eGFPΔ3’ fragment, the HR substrate, consisted of a eGFP copy lacking the nucleotides encoding the last 24 amino acids, which are required for the protein’s fluorescence. This transgene was integrated into the AAVS1 locus of HCT116 cells, which are commonly used for studying HR mechanisms. Although AAVS1 is considered a well validated safe harbor in the human genome, it is important to recall that transgene integration at this locus disrupts the gene phosphatase 1 regulatory subunit 12C (PPP1R12C). As described above, the transgene contained a puromycin cassette preceded by a splice acceptor sequence in addition to the CMV promoter-driven eGFPΔ3’. As reported for other transgenes containing splice acceptor sites, the insertion of the puromycin cassette into the AAVS1 locus led to a downregulation of PPP1R12C mRNA levels ****. In this regard, it is important to mention that although the haploinsufficiency or complete inactivation of PPP1R12C does not lead to a discernible phenotype, the consequences of this disruption should be further investigated. Another point we want to highlight is that the eGFPΔ3’ transgene was integrated as a single copy, and this is relevant when calculating the frequency of recombination because the appearance of a fluorescent signal in a cell is the result of a single HR event. Finally, it should be noted that the expression of the truncated eGFPΔ3’ copy could not be detected by western blot using an antibody against full length protein, suggesting the degradation of aberrant transcript and protein. Interestingly, the transgene has persisted over several passages (over 40) in our reporter cell line, which is likely due to the chromosomal stability of HCT116. The donor template consisted of a copy of eGFP lacking 42 bp at the 5’ end (including the start codon) and was delivered into the reporter cells via AAV particles. AAVs are on the rise as a powerful tool in gene therapy thanks to their lack of pathogenicity, wide range of cell tropism and long-term gene expression among other properties. It is also well known that recombinant AAV vectors facilitate homologous recombination in mammalian cells at high efficiencies (even up to 10x10^-3) and this is thought to be due, at least in part, to its single stranded nature. The recombination frequency in our system was 3.6x10^-3 ± 1.6x10^-3, which falls within the range of values obtained by others using recombinant AAVs. Finally, our study also validates the ability of our assay to identify factors affecting HR. Of note, all validation experiments were performed in the presence of a pair of TALENs designed to induce DSB in the region adjacent to the 3’ end of eGFPΔ3’. As expected, the addition of the eGFP TALEN pair led to a significant increase of the basal frequency of specific HR events (by approximately 8-fold), which helped improve the system. ScRAD52 is more effective than its human homologue in promoting HR both in vitro and in vivo. We observed that overexpression of ScRAD52 resulted in an approximate 3-fold increase in HR frequency when either the recombinant protein or the plasmid encoded protein were used. Similarly, the overexpression of PALB2 led to an increase in the HR frequency as previously reported by others using the DR-GFP reporter system. In our assay neither RAD51 nor RAD52 had any significant effect on HR frequency. In this regard, it is important to mention that it has been previously reported that the overexpression of RAD51 does not have a significant effect on HR rates using DR-GFP system, while the overexpression of its dominant negative forms which are ATPase mutants results in a 8-fold decrease in HR frequency. On the other hand, Kim et al. have shown that the overexpression of RAD51 and/or RAD52 reduces DSB-induced HR while enhances spontaneous HR. The distinct results obtained by us and others may reflect differences in recombination substrate structures or different levels of overexpression. The impact of siRNA-mediated silencing of RAD51, RAD52, PALB2, CTDP1, XRCC6, LIG4, CBP and SMCHD1 on the HR frequency that we report here is similar to that observed in previous genome wide HR screens and single gene approaches by others. Thus, as expected, RAD51, PALB2, and CTDP1 depletion resulted in a marked reduction of eGFP+ cells whereas depletion of NHEJ-related genes led to an increase in their number. This last result highlights the tight regulation of the balance between NHEJ and HR. In the case of RAD52, its silencing had a negligible effect on HR frequency. This apparently unexpected result has been reported by others, indicating that RAD52 is not essential to HR. In conclusion, we have established a sensitive assay for measuring the efficiency of HR in the context of the DNA repair of an integrated transgene using an exogenous DNA template. This novel tool will be useful to accurately test the impact of continually emerging tools for HR-mediated genome editing, drugs, miRNAs and genes on HR and DNA repair. We anticipate that refining the precision of repair during in vitro culture will contribute to developing innovative gene therapy products. # Supporting information We thank Roberto Cantalapiedra and Raquel Carretero for technical support. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Institute of Pharmacological Research (ININFA), University of Buenos Aires–National Scientific and Technical Research Council (CONICET), Buenos Aires, Argentina [^3]: ‡ These senior authors contributed equally to this work.

# Introduction The renin-angiotensin system (RAS) plays a central role in the development and regulation of blood pressure response. Angiotensin II (Ang II), the effector peptide of the pro-hypertensive axis of the RAS that also includes angiotensin converting enzyme (ACE) and the Angiotensin -type 1 receptor (AT₁R), exerts diverse physiological actions in both the peripheral and central neural systems. The anti-hypertensive counterbalance to these mediators includes ACE2, Ang-(1–7) and the Mas receptor. Importantly, all these essential components of the RAS, including renin and angiotensinogen, as well as various cardiovascular- modulatory aminopeptidases, are synthesized within the brain, suggesting the existence of a comprehensive intrinsic brain RAS. Recent evidence suggests that dysregulation of the individual brain RAS axes may play a critical role in the development and maintenance of hypertension. Ang II, acting through the AT₁R, plays a prominent role in the central regulation of blood pressure by activating the sympathetic nervous system, regulating fluid and salt balance and the secretion of aldosterone, amongst other actions. Previous studies suggest that systemically delivered Ang II likely acts upon the circumventricular organs, where the blood brain barrier is weak or absent, and subsequently activates hypothalamic and brain stem sites such as the paraventricular nucleus (PVN) and ventrolateral medulla, contributing to sympathoexcitation and hypertensive response. Experimental evidence indicates that the hypothalamic PVN is an important center for integrating Ang II-induced neural outflow signals for the pressor response and sympathetic vasomotor tone. Recent findings from our lab and others suggest that the RAS, in addition to inducing neurohumoral excitation, also increases the production of proinflammatory cytokines (PICs), such as tumor necrosis factor-alpha (TNF), in brain cardiovascular regulatory centers, and has been shown to contribute to the neurogenic component of hypertension, both through direct actions and through modulating reactive oxygen species (ROS) signaling pathways. A chronic increase in peripheral Ang II levels is proposed to initiate a cascade of signaling events involving PICs and ROS in brain cardioregulatory sites raising sympathetic activity, hypertension and end organ damage. A study by Marvar et al. showed that Ang II-mediated hypertension is caused by central mechanisms and described a feed-forward process in which the central pressor effects of Ang II lead to activation of T cells, which in turn, promote vascular inflammation and further raise blood pressure, leading to severe hypertension. In addition, PICs can be produced locally in the brain by glia and neurons, thereby contributing to the neuroinflammatory response implicated in the pathogenesis of hypertension. These observations, coupled with the emerging role of PICs and the little known role of the anti-hypertensive axis of the RAS in hypertension, led to hypothesize that the central effects of Ang II are, at least in part, mediated by the activation of PICs, especially TNF. The resultant of these actions are the differentially dysregulated RAS axes within cardiovascular relevant brain regions, including the PVN, ultimately enhancing the neurogenic hypertensive response. In the current study, we investigate this hypothesis by examining central TNF inhibition via intracerebroventricular (ICV) etanercept infusion, a soluble TNF receptor fusion protein, on pro- and anti-hypertensive RAS components in the PVN in Ang II-induced hypertension. # Materials and Methods ## Ethics Statement All animal and experimental procedures in this study were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) at Louisiana State University in compliance with National Institutes of Health Guide for the Care and Use of Laboratory Animals. ## Experimental Design Male Sprague-Dawley rats (10–12 weeks old) were used in this study. Animals were housed in a temperature-controlled room (25±1°C) and maintained on a 12∶12 hour light:dark cycle with free access to food and water. The rats were implanted with radio-telemetry transmitters to measure blood pressure, and subjected to ICV infusion of etanercept (ETN; 10 µg/kg/day) or artificial cerebrospinal fluid (aCSF) (Alzet, model 1004; 0.11 µl/hr), with and without subcutaneous infusion of Ang II (200 ng/kg/min) for 4 weeks. Osmotic minipumps (Alzet, model 2004; 0.25 µl/hr) were filled with Ang II dissolved in 0.9% saline or saline alone, and were implanted subcutaneously in the retroscapular area. The rats were divided into 4 groups: 1) Control group: saline minipumps+ICV aCSF, 2) ETN group: saline minipumps+ICV etanercept, 3) Ang II group: Ang II minipump+ICV aCSF, and 4) Ang II+ETN group: Ang II minipump+ICV etanercept. At the end of the study, rats were euthanized; the hearts and brains were collected and stored at −80°C until further analysis. ## Blood Pressure Measurement Blood pressure was measured continuously in conscious rats implanted with radio- telemetry transmitters (Model TA11PA-C40, Data Sciences International, St. Paul, MN). Rats were anesthetized with a ketamine (90 mg/kg) and xylazine (10 mg/kg) mixture (i.p.) and placed dorsally on a heated surgical table. The adequacy of anesthesia was monitored by limb withdrawal response to toe pinching. An incision was made on the ventral surface of the left leg, and the femoral artery and vein were exposed and dissected apart. The femoral artery was ligated distally, and a small clamp was used to temporarily interrupt the blood flow. The catheter tip was introduced through a small incision in the femoral artery, advanced into the abdominal aorta such that the catheter tip was distal to the origin of the renal arteries, and secured into place. The body of the transmitter was placed into the abdominal cavity and secured to the abdominal wall. The abdominal musculature was sutured and the skin layer was closed following implantation. Rats received benzathine penicillin (30000 U, i.m.) and buprenorphine (0.1 mg/kg, s.c.) immediately following surgery and 12 h postoperatively and allowed to recover for seven days. ## ICV Cannula Implantation Following the transmitter recovery period, the rats were implanted with ICV cannula for infusion of etanercept or aCSF. The rats were anesthetized with ketamine (90 mg/kg) and xylazine (10 mg/kg) mixture (i.p.) and the head was positioned in a Kopf stereotaxic apparatus. An ICV cannula was implanted into the right lateral cerebroventricle (1.3 mm caudal to bregma, 1.5 mm lateral to the midline, and 3.5 mm ventral to the dura) according to Paxinos and Watson, and fixed to the cranium using small screws and dental cement. A 4-week osmotic minipump was implanted subcutaneously and connected to the infusion cannula via the catheter tube to deliver etanercept or aCSF into the brain. ## Measurement of Plasma IL-10 At the end of the study, blood was collected in chilled EDTA tubes; plasma was separated and stored at −80°C until assayed. Circulating levels of IL-10 were quantified in the plasma using a commercially available rat IL-10 ELISA kit (Invitrogen) according to manufacturer’s instructions. ## Real Time RT-PCR The PVN punches were made from frozen brain sections using a Stoelting brain punch (Stoelting). Total RNA was isolated from PVN tissue using RNeasy plus micro kit (Qiagen) and cDNA was synthesized using iScript cDNA synthesis kit (Bio-Rad). Real Time PCR amplification reactions were performed with iQ SYBR Green Super mix with ROX (Bio-Rad) using the ABI Prism 7900 Real time PCR machine (Applied Biosystems). Data were normalized to GAPDH expression by the ΔΔC_T comparative method. ## Western Blot Analysis Western blot analysis was performed according to standard protocols. The PVN tissue was homogenized with RIPA lysis buffer. The protein concentration was measured using a bicinchioninic acid protein assay kit (Pierce). Equal amounts of protein (5 µg) were separated by SDS-PAGE on 10% or 12% gels, transferred on to PVDF membrane (Immobilon-P, Millipore), and blocked with 1% BSA in TBS-T at room temperature for 60 min. The membranes were subjected to immunoblot analyses with anti-ACE (Santa Cruz, 1∶500), anti-AT₁R (Santa Cruz, 1∶500), anti-ACE2 (Santa Cruz, 1∶500), anti-AT₂R (Santa Cruz, 1∶200), anti- Mas (Alomone Labs, 1∶500), anti-NOX-2 (BD Biosciences, 1∶500), anti-NOX-4 (Santa Cruz, 1∶1000), anti-iNOS (Santa Cruz, 1∶500), anti-nNOS (Santa Cruz, 1∶500), and anti-GAPDH (Santa Cruz, 1∶1000) antibodies. The membranes were washed and incubated with anti-rabbit or anti-goat secondary antibodies (Santa Cruz, 1∶5000) for 1 hour at room temperature. Specific bands were detected using an enhanced chemiluminescence kit (Amersham). The bands were quantified by densitometry using Chemidoc XRS system and Quantity-One software (Bio-Rad) and were normalized to GAPDH expression. ## Determination of NAD(P)H Oxidase Activity in the PVN Homogenates were prepared from PVN samples and total protein concentration was determined using a bicinchoninic acid protein assay kit (Pierce). The NAD(P)H oxidase activity was measured using lucigenin enhanced chemiluminiscence detection of superoxide as previously described. In brief, the homogenates were diluted in modified HEPES buffer containing 140 mM NaCl, 5 mM KCl, 0.8 mM MgCl₂, 1.8 mM CaCl₂, 1 mM Na₂ HPO₄, 25 mM HEPES, and 1% glucose (pH 7.0). The reaction was started by addition of NAD(P)H (100 µM) and dark adapted lucigenin (5 µM). Light emission was recorded and expressed as mean light unit (MLU) per minute per milligram of protein over 10 min. Using this method, the superoxide anion production also represents NAD(P)H oxidase activity. The specificity of superoxide measured was confirmed either by adding superoxide dismutase (200 units/ml) or apocynin (1 mM). ## *In situ* Detection of Superoxide Production in the PVN Dihydroethidium, an oxidative fluorescent dye, was used to detect in situ superoxide in the PVN of rats as previously described. At the end of the experiment, the brains were removed, quickly frozen, embedded into OCT, and cryostat sectioned (30 µm, coronal) directly onto chilled microscope slides. Sections were then incubated in a light protected humidified chamber at 37°C for 30 minutes with 1 µmol/L dihydroethidium (Molecular Probes). After washing with phosphate-buffered saline, red fluorescence was visualized by confocal laser scanning microscopy using an excitation wavelength of 490 nm and emission wavelength of 610 nm. ## Statistical Analysis All results are expressed as mean±SEM. For statistical analysis of the data, Student’s *t* test, one-way ANOVA or repeated measures ANOVA followed by Bonferroni’s *post hoc* test was performed using GraphPad Prism version 5.0 for Windows (Graph Pad Software, San Diego California, USA) to determine differences among groups. A value of *p*\<0.05 was considered statistically significant. # Results ## Effect of Central TNF Blockade on Mean Arterial Pressure and Cardiac Hypertrophy To assess the effect of central TNF blockade on the Ang II-induced hypertensive response, mean arterial pressure (MAP) was measured using a radio-telemetry system. After 28 days, chronic Ang II infusion significantly increased the MAP in rats when compared with control rats (165±5 mmHg vs 108±8 mmHg, respectively; *p*\<0.05). In contrast, ICV treatment with etanercept attenuated the Ang II- induced increase in MAP (126±21 mmHg vs 165±5 mmHg, respectively; *p*\<0.05), while etanercept treatment alone had no effect on MAP (104±4 mmHg). To evaluate Ang II-induced changes on cardiac hypertrophy in these rats, the hearts were harvested and weighed at the end of experimental period. The ratio of heart weight/body weight (HW/BW) was calculated as an indicator of cardiac hypertrophy. Chronic Ang II infusion lead to increased cardiac hypertrophy versus controls as indicated by the increased HW/BW ratio. ICV treatment with etanercept inhibited Ang II-mediated cardiac hypertrophy. Furthermore, mRNA expression of a molecular marker of cardiac hypertrophy, atrial natriuretic peptide (ANP), was measured in cardiac tissue using real time RT-PCR. Chronic Ang II infusion showed increased mRNA expression of ANP in the heart, which was decreased by ICV treatment with etanercept. These data suggests a role for TNF in the brain on Ang II-induced blood pressure regulation and cardiac hypertrophy in the hypertensive state. ## Effect of Central TNF Blockade on the Expression of Pro- and Anti-inflammatory Cytokines To determine the effect of Ang II on the production of PICs and chemokines, the mRNA expression of TNF, IL-6, IL-1β, and the chemokine MCP-1 were measured in the PVN by real time RT-PCR. Ang II infusion induced an increase in the gene expression of the PICs TNF, IL-6 and IL-1β, and the chemokine MCP-1, in the PVN, when compared to the control group. These PICs were attenuated in the PVN of Ang II-infused rats treated ICV with etanercept, demonstrating that chronic Ang II infusion increases the pro-inflammatory response within the PVN through TNF in Ang II-induced hypertension. We also examined the effect of TNF inhibition on the expression of the anti- inflammatory cytokine IL-10. Chronic subcutaneous Ang II infusion resulted in significant decrease in PVN mRNA and plasma protein levels of IL-10. However, Ang II infusion and simultaneous inhibition of TNF in the brain using etanercept restored the PVN mRNA expression of IL-10 and plasma levels of IL-10. ## Effect of Central TNF Blockade on the Expression of RAS Components in the PVN Both Ang II and TNF have been shown to modulate RAS component expression. To determine the manner by which Ang II infusion alters the expression of the pro- and anti-hypertensive components of the RAS in the PVN, we examined the mRNA and protein expression levels of ACE, ACE2, AT₁R, AT₂R, and the Mas receptor. The PVN mRNA and protein expression of the RAS pro- hypertensive components ACE and AT₁R were significantly increased in Ang II-infused rats when compared with control rats; this was prevented by ICV treatment with etanercept. Conversely, the anti-hypertensive components of the RAS (ACE2, Mas receptor and AT₂R) showed a decreased gene and protein expression in Ang II-treated rats. These levels were increased in the Ang II+ETN group. These data suggest that in Ang II-induced hypertension, the pro- and anti-hypertensive components of the RAS are differentially regulated within the PVN in a deleterious manner and TNF inhibition restores the balance between these RAS components. ## Effect of Central TNF Blockade on the Oxidative Stress Markers in the PVN Ang II and TNF have both been shown to act through oxidative stress mediated pathways, especially within the PVN, in inducing elevated sympathetic outflow and a progressive hypertensive response. Thus, several experiments were performed to test the effect of central TNF inhibition on oxidative stress markers in the PVN in response to the Ang II infusion hypertension model. First, NAD(P)H oxidase dependent superoxide production in the PVN was measured using a lucigenin enhanced chemiluminiscence method. Consistent with earlier reports, Ang II caused a significant increase in NAD(P)H oxidase dependent superoxide production in the PVN homogenates compared with control rats. This response was abolished by ICV inhibition of TNF using etanercept. Next, in situ dihydroethidium fluorescence was used to assess Ang II-induced superoxide production in the PVN of rats receiving systemic Ang II-infusions concomitant with ICV infusions of either etanercept or aCSF. Consistent with earlier reports, dihydroethidium fluorescence was increased in the PVN of Ang II-infused rats compared with control rats. This response was markedly inhibited in rats receiving ICV etanercept. Furthermore, Ang II infusion significantly increased the NOX-2 and NOX-4 mRNA and protein expression in the PVN of Ang II infused rats. These changes were attenuated in Ang II-infused rats treated centrally with etanercept. To further determine the effect of TNF blockade on Ang II- induced NO signaling, we analyzed mRNA and protein expression of inducible NOS (iNOS) and neuronal NOS (nNOS) within the PVN. Four weeks of Ang II infusion significantly increased iNOS mRNA and protein expression and decreased the nNOS mRNA and protein expression. These changes were prevented in rats with central infusion of etanercept. Overall, these data suggest that increased oxidative stress in Ang II-induced hypertension is potentially through a TNF-driven mechanism and the effects of central TNF blockade on attenuation of Ang II- induced hypertension, at least in part, are mediated by a decrease in oxidative stress. # Discussion In the present study, the role of central TNF on Ang II-induced hypertension and cardiac hypertrophy was investigated. These data suggest for the first time that inhibition of TNF in the brain using etanercept lowered blood pressure, reduced inflammation, decreased oxidative stress, and restored the balance between the pro- and anti-hypertensive axes of the RAS. Our findings demonstrate that the changes observed in Ang II-induced hypertension are regulated, at least in part, through the central actions of TNF and potentially via the dysregulation of components of the RAS within the hypothalamic PVN. Recent evidence suggests that hypertension is an inflammatory condition where various PICs such as TNF, IL-6 and IL-1β, both centrally and peripherally, have been shown to play an important role in the pathogenesis of hypertension. A recent study from our lab demonstrated that chronic peripheral Ang II infusion results in increased production of PICs within the PVN. Etanercept is a soluble recombinant fusion protein that inhibits TNF by posing as a TNF receptor decoy and acting through competitive inhibition of TNF and an overall reduction in free TNF to act on endogenous receptors. Blockade of TNF by etanercept has been shown to prevent renal damage in a genetic hypertensive rat model, as well as lower blood pressure in rats with Ang II- and salt-induced hypertension, suggesting a role for TNF in blood pressure regulation and renal injury. Another study also showed that mice treated with etanercept had an attenuated hypertension and a blunted increase in superoxide production in response to Ang II. In our study, using ICV etanercept infusion techniques, we inhibited the TNF levels specifically in the brain. Our present observations complement those prior findings and show that blockade of TNF by ICV administration of etanercept into the brain protects rats against Ang II-dependent cardiac hypertrophy and hypertension. Ang II can act as a potent proinflammatory agent and stimulate the production of chemokines such as MCP-1, and PICs, such as TNF, IL-6 and IL-1β in the brain. TNF is commonly considered as one of the initiators of the pro-inflammatory cascade, which can induce production of other cytokines, and inhibition of its action during inflammatory events abrogates many of the ensuing responses, including the production of IL-1β and IL-6. A recent study by Shi et al. demonstrated that Ang II-induced hypertension involves activation of microglia and increased expression of PICs within the PVN. A previous study from our group demonstrated that chronic Ang II infusion increases proinflammatory cytokines expression in the PVN of rats. In our study, TNF blockade with etanercept decreased the PVN expression not only of TNF, but of other PICs such as IL-6, IL-1β, and the chemokine MCP-1, supporting the hypothesis that PICs are involved in the Ang II-induced hypertensive response. It has been shown that central gene transfer of IL-10 reduces hypothalamic inflammation in heart failure rats after myocardial infarction. In addition, IL-10 overexpression in the PVN attenuates Ang II-induced hypertension. In our study, chronic Ang II infusion resulted in decreased mRNA and protein expression of anti-inflammatory cytokine IL-10, which was restored by central TNF blockade suggesting that some of beneficial effects of TNF blockade are mediated by restoring the levels of anti-inflammatory IL-10. Neurogenic hypertension is characterized by an overactive brain RAS,. In addition to locally generated Ang II within the brain, blood borne Ang II can enter the brain via circumventricular organs and modulate the pathogenesis of hypertension and fluid homeostasis. Elevated activity and expression of RAS components in central cardiovascular regulatory regions has been shown to be involved in the pathogenesis of hypertension in several genetic and experimental models. Treatment with ACE inhibitors and Ang receptor blockers has been shown to prevent this RAS overactivity and restore normal cardiovascular function. In the brain, the AT₁R mediates the central effects of Ang II, including vasopressin release, water and salt intake and balance, and increased sympathetic drive, all of which contribute to the development of hypertension. In many animal models of hypertension, the expression of the AT₁R is up-regulated in central cardiovascular regulatory centers, including the hypothalamic PVN. Both *in vitro* and *in vivo* studies have demonstrated a cross-talk between Ang II and TNF, a mechanism we have also shown using TNF knockout mice where Ang II-induced hypertension was attenuated via a decreased expression of AT₁R. Within the last decade, the discovery of an alternate set of components of the RAS which may act as a counterbalance to the actions of the ACE/Ang II/AT₁R pathway added complexity to the understanding of RAS regulation, especially within the brain. These components, termed anti-hypertensive due to their cardio-protective effects, along with all the components of the pro-hypertensive RAS axis, are known to be expressed throughout the various central cardio-regulatory regions, including the PVN. In various experimental hypertensive models, the components of anti-hypertensive RAS axis (ACE2, Ang (1–7), the Mas receptor and AT₂ receptors) are shown to be down-regulated, while the pro-hypertensive components are increased, . This dysregulation may be the lynch pin trigger towards developing the hypertensive state, but it remains poorly understood. It has been shown that the brain ACE2 activity was inhibited in a chronically hypertensive mouse model with high Ang II levels and this decrease was mediated by AT₁R. We previously showed that ACE2 overexpression within the PVN attenuated Ang II- induced hypertension by abolishing PIC production in the PVN in combination with restoring the balance between pro- and anti-hypertensive axes of the RAS. In the present study, ICV treatment with etanercept resulted in reduction of Ang II- induced pro-hypertensive RAS components expression in the PVN, including ACE and AT₁R, as well as the restoration of the anti-hypertensive RAS components ACE2, Mas receptor, and AT₂ receptors. These results suggest that this RAS dysregulation and perpetuation of the hypertensive state may be the result of a pro-inflammatory response through the actions of TNF. Excessive ROS production in brain cardio-regulatory centers such as the PVN can contribute to the neurogenic component of the hypertensive response by enhancing sympathetic activity and outflow. It has been previously shown that NAD(P)H oxidase is the primary source of Ang II-induced ROS in neurons and that treatment with Tempol, a cell permeable superoxide dismutase (SOD) mimetic, inhibits Ang II-mediated superoxide production and hypertension. Enhanced NAD(P)H oxidase activity associated with increased expression of several NAD(P)H oxidase subunits including NOX-2 and NOX-4, have been shown to be predominant homologues expressed in the forebrain, including the PVN. A recent study showed that in an aldosterone/salt induced hypertensive animal model, both NOX-2 and NOX-4 are necessary to generate functional NAD(P)H oxidase within the PVN. Furthermore, TNF can induce activation of NADPH oxidase leading to enhanced oxidative stress and decreased bioavailability of NO. In the present study, we found that central TNF inhibition abolished the Ang II-induced oxidative stress as indicated by decreased super oxide production, NAD(P)H oxidase activity as well as attenuated NOX-2 and NOX-4 subunit expression within the PVN. These results indicate that TNF mediates NAD(P)H oxidase-derived superoxide production during Ang II-induced hypertension. In addition, the decreased nNOS expression and increased iNOS expression in the PVN indicates NO signaling dysregulation in Ang II-induced hypertensive rats. Neuronal NOS is an inverse indirect indicator of sympathoexcitation, in that a decrease in nNOS correlates with an increase in sympathetic outflow. Therefore, a reduction in beneficial NO not only decreases with the reduction in nNOS expression and activity, but also due to the rapid interconversion of NO to peroxynitrite (ONOO⁻) due to the increased production of NADPH oxidase-derived superoxide. This is further complicated by the involvement of iNOS, which rapidly uses up available L-arginine for NO production and commences with superoxide production and thereby increasing free radical concentrations. This combination of factors can lead to an increased PIC response, sympathoexcitation and a continued propagation of neurogenic hypertension. The present study provides further support for these observations by showing that central TNF blockade with etanercept results in reduced oxidative stress within the PVN of hypertensive rats. In summary, chronic Ang II infusion resulted in cardiac hypertrophy and elevated MAP, and within the PVN, an increased expression of PICs and markers of oxidative stress. More importantly, Ang II-infused rats had an increased expression of the injurious pro-hypertensive RAS components ACE and AT₁R, and a decreased expression of the protective anti-hypertensive RAS components ACE2, the Mas receptor and AT₂ receptors. These findings suggest that elevated TNF levels by Ang II hypertension are associated with initiation of inflammatory cascade, which in turn, promote downstream events and further raise blood pressure, leading to severe hypertension. Central blockade of TNF with etanercept resulted in attenuation of hypertension, cardiac hypertrophy and PIC expression, decreased oxidative stress, as well as a restored the balance between the protective and deleterious axes of the RAS, within the hypothalamic PVN. The beneficial effects of central TNF blockade in Ang II-induced hypertensive responses appears to be mediated by the returned balance of the central RAS components, especially within the PVN. It is important to note, however, that due to the administration of etanercept ICV, the TNF inhibitory effects may have impacted additional cardio-regulatory regions in the brain and elicited a similar response as in the PVN, but in light of its central integrative function versus the other regions, the PVN was of utmost concern. Future studies should investigate these additional cardio- regulatory regions, as well as looking more specifically at the pathway between Ang II, TNF and the differential regulation of the RAS arms in the Ang II hypertension-induced animal model. Our findings provide further evidence and insight for the involvement of the RAS within the PVN and its interaction and mediation through TNF in the neurogenic component of hypertension. Further exploration of these system interactions within the brain may be beneficial towards the development of novel hypertensive therapeutics. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SS JF. Performed the experiments: SS JC. Analyzed the data: SS. Contributed reagents/materials/analysis tools: JF. Wrote the paper: SS. Reviewed the manuscript: JC JF.

# Introduction Oxygen supply is essential for physiological functioning and metabolism in human organs. Retinal neurons need more oxygen than other tissues, and abnormal retinal oxygenation is thought to play a pivotal role in the pathogenesis of various retinal diseases such as diabetic retinopathy, retinopathy of prematurity, and retinal vein or artery occlusion. *In vivo* measurement of retinal oxygen levels would be of help for the management of such diseases. Recently, a non-invasive retinal oximeter Oxymap T1 has become commercially available. Oxymap T1 is based on the principle that deoxyhemoglobin and oxyhemoglobin have different light absorbance at specific wavelengths of light (570 nm and 600 nm). Using Oxymap T1, various investigators have reported the oxygen levels in the major retinal vessels and studied the repeatability of the measurements in healthy populations, mainly in Caucasians. In addition, recent studies have reported the retinal oxygen concentration in eyes with retinal pigmentosa, glaucoma, high myopia, or diabetic retinopathy. Longitudinal studies on retinal oximetry would contribute to the early detection and management of such chronic diseases. In diabetic patients, the level of oxygen saturation is reported to be increased in the major retinal veins, as early as the stage of background retinopathy or even no retinopathy. In eyes with retinitis pigmentosa, the retinal venous oxygen saturation is correlated with the residual visual field. However, previous studies have indicated that retinal oximetry may be influenced by the flash intensity, face position, or the degree of fundus pigmentation. So far, limited information is available on either the normative data for the Japanese population or the reproducibility of retinal oximetry in the Japanese. In addition, only one previous report investigated the intervisit reproducibility of Oxymap T1, which is essential to set up and interpret in longitudinal studies. Thus, the purpose of the current study was to establish a normative database for retinal oximetry using Oxymap T1 in a healthy Japanese population, and study the reproducibility of the measurements in Japanese. # Patients and Methods The Ethics Committee at Kagawa University Faculty of Medicine approved this prospective study, which was conducted in accordance with the tenets of the Declaration of Helsinki. Written informed consent was obtained from each subject before any study procedures or examinations were performed. This study is registered in ClinicalTrials.gov (NCT02318641). ## Subjects This prospective study consisted of 252 eyes with no ocular disease of 252 healthy Japanese subjects. Oxygen saturation in the major retinal vessels was studied using Oxymap T1 (Oxymap ehf, Reykjavik, Iceland) at the Kagawa University Hospital Department of Ophthalmology between December 2014 and June 2015. Exclusion criteria were as follows: keratoconus, high myopia (more severe than -6 diopters), high astigmatism (more severe than ± 3 diopters), best- corrected visual acuity worse than 20/25, prior intraocular surgery, or any co- existing ocular disease (e.g., glaucoma, diabetic retinopathy, retinal vein or artery occlusion, hypertension retinopathy, age-related macular degeneration, retinal degenerative disease, or senile cataract that diminished image quality). Patients who had suspected adverse effects to pupil dilation, diabetes mellitus, severe cardiovascular or respiratory diseases, pregnancy, or collagen disease were also excluded from the current study. ## Image acquisition for retinal oximetry The principle of a commercially available retinal oximeter Oxymap T1 has been described previously by other researchers. In brief, Oxymap T1 is composed of two digital cameras, an image splitter, and two narrow band-pass filters and is attached to a fundus camera (TRC-50DX, Topcon, Tokyo, Japan). Oxymap T1 simultaneously captures two fundus images at two different wavelengths of light (570 nm and 600 nm). Light of 570 nm, which is isosorbetic to deoxyhemoglobin and oxyhemoglobin, is insensitive to oxygen saturation, and light of 600 nm, which is more sorbetic to oxyhemoglobin than deoxyhemoglobin, is sensitive to oxygen saturation. After a comprehensive medical interview, all subjects underwent a thorough ocular examination, including autorefractometer, best-corrected visual acuity measurement with a 5-m Landolt chart, slit lamp examination, intraocular pressure measurement, dilated funduscopy, 50° digital fundus photographs, and optical coherence tomography (Spectralis HRA+OCT; Heidelberg Engineering, Heidelberg, Germany). When both eyes met the inclusion criteria, one eye was selected randomly for the examination. After mydriasis with 0.5% of tropicamide and 0.5% of phenylephrine hydrochloride (Mydrin-P; Santen, Osaka, Japan), each included eye was examined with Oxymap T1 in a dark room. Setting of Oxymap T1 for image acquisition was as follows: flash intensity was 50 W and small aperture and large pupil settings were applied to the fundus camera. Fifty-degree fundus images were obtained centered at the optic disc by using Oxymap T1. At each setting, more than 2 images were obtained per eye. ## Oxygen saturation and vessel diameter measurement from acquired images Fundus images acquired with Oxymap T1 were processed using a built-in software Oxymap Analyzer (version 2.4.2, Oxymap Ehf). The software calculates optical density of retinal vessels at two wavelengths (570 nm and 600 nm). The ratio of the optical densities at these two wavelengths has an inverse and approximately linear relationship with oxygen saturation. The resolution of Oxymap T1 is 9 μm. In each acquired image centered at the optic disc, the oxygen saturation was calculated in major retinal arteries and veins measuring more than 6 pixels in vessel width in the measurement zone. The evaluators set analyzed vessel segments in a standardized manner. For the analysis, vessel selection was done with the disc centered in a 1.5-to 3-disc diameter area and 1 disc diameter away from the disc margin to avoid uneven retinal background reflections near the optic disc margin. All branches and vessel crossings within the measurement area were also manually excluded from the analysis. After detailed selection of the vessel section for the analysis, Oxymap Analyzer automatically measured the levels of oxygen saturation and vessel diameter in each selected vessel. ## Reproducibility of retinal oximetry with Oxymap T1 Reproducibility of retinal oximetry was examined in 20 eyes of 20 healthy subjects (15 men and 5 women; 5 right and 15 left eyes; mean age, 24.0 ± 1.7 years \[range, 23 to 29 years\]). To investigate interphotograph reproducibility of retinal oximetry measurements, an evaluator (YN) performed the calculation of the levels of oxygen saturation in each major retinal vessel with Oxymap Analyzer, from two different photographs obtained by Oxymap T1 on the same day. To investigate intervisit reproducibility of retinal oximetry measurements, 2 photographs were obtained by the same photographer at 2 different visits. Photography was performed 1 week apart. Calculation of the levels of oxygen saturation in each major retinal vessel was performed by the same evaluator (YN). To investigate interevaluator reproducibility of retinal oximetry measurements, two different evaluators (TS, YN) independently performed calculation of the levels of oxygen saturation in each major retinal vessel from a single photograph. The intraclass correlation coefficient (ICC) was used to determine the reproducibility using oxygen saturation in a major retinal vessel in each quadrant and the mean of all 4 major retinal arteries and veins. ## Statistical analysis All data were collected in an Excel database (Microsoft Office 2010; Microsoft, Redmond, WA). All statistical analyses were conducted using Software R (version 2.8.1; R Foundation for Statistical Computing, Vienna, Austria). The oxygen saturation was calculated in each major retinal artery and vein in 4 quadrants, the mean of values of 2 vessels in superior or inferior hemispheres, and the mean of values of 4 vessels in all quadrants. Pearson’s correlation coefficient was calculated to assess the relationship between age, refractive sphere, or vessel diameter and the level of oxygen saturation. Stepwise forward multivariate linear regression analyses were also performed to evaluate the contribution that each initially identifiable factor made to retinal oximetry. A p value \<0.05 was considered to be statistically significant. # Results In the current study, retinal oximetry was performed in 252 eyes of 252 healthy Japanese subjects (137 men and 115 women). Mean age was 61.1 ± 18.8 years and ranged between 20 and 93 years. Oxymap T1 allowed us to capture the monochromatic fundus images at two wavelengths, and Oxymap Analyzer calculated the oxygen saturation in each major retinal vessel. Mean oxygen saturation in 4 quadrants was 97.0 ± 6.9% in arteries and 52.8 ± 8.3% in veins. The mean arteriovenous (A-V) difference in mean oxygen saturation was 44.2 ± 9.2%. shows the oxygen saturation in each major retinal artery and vein in 4 quadrants, the mean of values of 2 vessels in each hemisphere, and the mean of values of 4 vessels in all quadrants. Measurements of both arterial and venous oxygen saturation were significantly lower in the temporal side of the retina, especially in the temporal-inferior vessels. However, there were limited differences in A-V difference of oxygen saturation in the 4 quadrants. shows the mean oxygen saturation and mean vessel diameter in major retinal vessels in all groups stratified by age. Reproducibility of retina oximetry was examined in 20 eyes of 20 healthy subjects. shows the ICCs in arterial and venous oximetry and A-V differences. Interphotograph, intervisit, and interevaluator ICCs of the retinal oximetry in one vessel were 0.891–0.970, 0.766–0.949, and 0.379–0.922, respectively. Although interphotograph ICC was high, interevaluator ICC was relatively low. However, the mean retinal oximetry in 4 quadrants had high ICCs between two photographs (0.936–0.979), two visits (0.809–0.837), or two evaluators (0.732–0.947). and show the associations between the retinal oxygen saturation and age in healthy subjects. In the major retinal arteries, oxygen saturation increased with age (r = 0.18, p\<0.01). Arterial oxygen saturation increased by 0.67% per 10 years. However, the venous oxygen saturation showed no correlation with age (r = -0.06). A-V difference in retinal oxygen saturation increased with age (r = 0.19, p\<0.01) by 0.92% per 10 years. shows the correlation between refractive sphere, age, or vessel diameter and oxygen saturation. There was no association between refractive sphere and oxygen saturation in the major retinal vessels. Multivariate analysis showed that arterial retinal saturation was significantly associated with age (β = 0.026, p = 0.002) and vessel diameter (β = -0.16, p = 0.013). The A-V difference was significantly associated with age (β = 0.19, p = 0.003). # Discussion In the current study, retinal oximetry was performed in 252 eyes of 252 healthy Japanese subjects, which is the largest normative database of Oxymap T1 to date. In a study on 120 Caucasians, Geirsdooier et al. reported that retinal oxygen saturation was 92.2 ± 3.7% in arterioles and 55.6 ± 6.3% in venules, and the A-V difference was 36.7 ± 5.4%. In a study on 118 Asians, Yip et al. reported that the oxygen saturation was 93.64 ± 6.9% in retinal arterioles and 54.22 ± 6.9% in venules, and the A-V difference was 39.43 ± 8.9%. In a recent study on 98 Indians, Mohan et al. reported that the oxygen saturation was 90.3 ± 6.6% in retinal arterioles and 56.9 ± 6.3% in venules, and the A-V difference was 33.2 ± 5.2%. Thus, it is important to note that the results of retinal oximetry with Oxymap T1 show variation across races, and this fact should be kept in mind while obtaining and interpreting measurements. We used the default formula of Oxymap Analyzer to calculate the retinal oxygen concentration, which uses the mean oxygen saturation values obtained from retinal vessels in healthy individuals by Schweitzer et al.. In their measurement, the mean oxygen saturation was 92.2 ± 4.1% in retinal arteries and 57.9 ± 9.9% in retinal veins. In a recent study of Oxymap T1 by Ueda-Consolve et al., the mean arterial and venous oxygen saturation values in 14 healthy Japanese were 99.9 ± 8.9% and 54.6 ± 6.3%, respectively. We can estimate that the A-V difference was 45.3%, which is relatively higher than that reported in previous reports, which were mainly on Caucasians. Two fundus images captured at two different wavelengths of light by Oxymap T1 may be influenced by the degree of fundus pigmentation. Because the fundus of Japanese subjects is more pigmented than that of Caucasians, this difference may account for the racial variations in the measurements of retinal oximetry. We may need to establish a calibration formula optimized for each race. In the current study, measurements of both arterial and venous oxygen saturation were significantly lower in the temporal hemisphere, especially in the temporal- inferior quadrant. Geirsdooier et al., Mohan et al., and Palsson et al. reported similar findings. The exact reason for this regional variation is unclear. Indeed, venous oxygen concentration may be influenced by retinal metabolism. However, arterial oxygen concentration should not show such a great variation because all retinal arteries originate from a single central retinal artery. Rather, this variation may be explained by the measurement error, perhaps on the photographs by Oxymap T1, because venous oxygen saturation was also lower in the temporal hemisphere and the A-V difference was almost canceled out. Recently, Mohan et al. reported a strong negative correlation between calculated oxygen saturations measured using Oxymap T1 and peripapillar retinal nerve fiber thickness, which is large in the temporal hemisphere. This finding would support our current finding that both arterial and venous oxygen saturation were low in the temporal hemisphere. We measured oxygen saturation in all 4 quadrants. To minimize the error in photography, it would be most appropriate to calculate the mean of the values in all quadrants. In the current study on a healthy Japanese population, interphotograph, intervisit, and interevaluator ICCs for retinal oximetry in one vessel ranged from 0.891 to 0.970, from 0.766 to 0.949, and from 0.379 to 0.922, respectively. Interphotograph ICC in our measurement was high and Palsson et al., Goharian et al., and Yip et al. also reported high interphotograph (intravisit) reproducibility. In addition, O'Connell et al. reported relatively high intervisit reproducibility of retinal oximetry with Oxymap T1. However, it is suggested that pupil size or face position may influence the measurements in retinal oximetry. This would account for the slightly lower intervisit ICC in the current population, compared with interphotograph ICC. In the current study, interevaluator ICC was lower than interphotograph or interevaluator ICCs. In a study by Yip et al., intergrader (interevaluator) ICC was 0.77–0.94. Although Oxymap Analyzer has a built-in, semiautomatic software to calculate the oxygen concentration, the selection of the vessel segment depends on the evaluator. In addition to the standardization of retinal image capture by Oxymap T1, the standardization of analysis with Oxymap Analyzer is also essential to maximize the reproducibility of retinal oximetry. In our subjects, while arterial oxygen saturation increased with age, venous oxygen saturation showed no correlation with age. The A-V difference increased with age. The age-related changes in retinal oxygen saturation are still controversial. In the Indian population, it has been reported that both arterial and venous retinal oxygen saturation increase with age. In a previous study on a multiethnic population, both arterial and venous retinal oxygen saturation decreased with age, resulting in an increase in A-V difference. Various factors including the decrease in pupil dilation, cataract formation, decrease in the vessel caliber, and atrophy of the retinal pigment epithelium or choroid may be involved in this discrepancy. Judging from the results of previous reports and those of the current study, longitudinal changes in individual retinal oximetry may be small. However, these changes may be crucial in the comparisons of retinal oximetry between different generations. One of the major limitations of our study is its cross-sectional nature. Based on a large healthy population, we investigated the age-dependent changes in retinal oximetry. However, a long follow-up study would be necessary to investigate the longitudinal changes in retinal oximetry. In addition, we excluded patients with severe cardiovascular or respiratory diseases but subjects with smoking history or systemic hypertension were included. History of controlled systemic hypertension is reported to be associated with an increased A-V difference in retinal oxygen saturation. In addition, the mean age of the subjects recruited in the reproducibility test was substantially lower than that of the total study population. Subject selection may have had some influence on the measurement value and its reproducibility. In spite of these limitations, we constructed a database of retinal oximetry for healthy Japanese. Although interphotograph repeatability was high, interevaluator and intervisit repeatability was relatively low. However, mean retinal oximetry in the 4 quadrants may contribute to higher reproducibility of measurements. Since the results of retinal oximetry show some variation across races, we might need to establish a calibration formula optimized for each race. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YN AT. Performed the experiments: YN TS NK YM AO MK CS KH AT. Analyzed the data: YN TS. Contributed reagents/materials/analysis tools: YN AT. Wrote the paper: YN AT.

# Introduction RNA silencing is an ancient network of highly related pathways that repress gene expression in eukaryotic organisms by means of small regulatory RNAs. The mechanism is triggered by dsRNA and specifically targets any related RNA (post- transcriptional pathways) or DNA (transcriptional gene silencing). Key steps in the process are: 1) the dsRNA trigger is cut into 21–24 nucleotide (nt) short- interfering (si) RNA duplexes by a ribonuclease III-like enzyme termed Dicer, 2) one strand of the siRNA duplex associates with an argonaute-like protein to form the core of a silencing effector complex, and 3) the siRNA directs the complex to complementary genetic elements. In post-transcriptional gene silencing, the complex targets mRNA, which is then cleaved by the ribonuclease-H activity of argonaute (AGO). Higher plants, *C. elegans*, and fungi additionally amplify silencing via transitivity, a pathway that produces dsRNA and additional siRNAs from the targeted mRNA. Production of dsRNA in transitive silencing depends on cellular RNA-dependent RNA polymerase (RDR) activity, and the siRNAs generated are termed secondary siRNAs, while those that derive from the initial dsRNA trigger are termed primary siRNAs. In plants, production of both primary and secondary siRNAs entails cleavage of dsRNA by the Dicer-like (DCL) family of enzymes. In plants, therefore, transitivity not only increases the siRNA population, but also degrades the target mRNA in the process of making secondary siRNAs, thereby amplifying silencing in two ways. Interestingly, secondary siRNA production itself is not a mechanism of target degradation in *C. elegans* because antisense secondary siRNAs are transcribed directly from the target mRNA template in that organism. The importance of transitive silencing has recently come to light in studies in *C. elegans* showing that the vast majority of siRNAs that accumulate during RNA silencing in that organism are secondary siRNAs. Transitive silencing is also important in plants, where it is essential for some types of transgene-induced silencing and thought to play a key role in defense against viruses. Although RNA silencing was initially considered simply a novel defense mechanism against viruses and other invading nucleic acids, the subsequent discovery of endogenous small regulatory RNAs led to the realization that it is a fundamental genetic regulatory mechanism in eukaryotic organisms. There are three major classes of endogenous small regulatory RNAs in plants: micro (mi) RNA, trans- acting (ta) siRNA, and heterochromatin-associated (hc) siRNA. The biogenesis and mechanism of action of the different silencing-associated small regulatory RNAs are highly related; however, hc-siRNAs suppress transcription as opposed to mediating RNA degradation and, in animal systems, miRNAs primarily repress translation. Many of the enzymes involved in RNA silencing are encoded by multigene families, allowing for the possibility of diverse, specialized pathways. In *Arabidopsis*, there are four *DCL*, ten *AGO*, and six *RDR* genes. The roles of the four *Arabidopsis DCL* genes in small RNA biogenesis and silencing have been the subject of intense study, and the picture that has emerged is one of primary roles for many of the enzymes, plus functional redundancy. Thus, *DCL1* is required for the biogenesis of 21-nt miRNAs, and *DCL3* is responsible for the biogenesis of 24-nt hc-siRNAs. *DCL4* seems the most versatile, being required for production of 21-nt ta-siRNAs, which derive from *RDR6*-generated dsRNA, as well as for production of 21-nt *RDR6*-independent siRNAs from hairpin transgenes, in which case dsRNA can be produced directly by transcription of an engineered inverted repeat. *DCL4* also plays a primary role in production of siRNAs from viruses, with *DCL2* as the substitute if *DCL4* has been inactivated. Although *DCL2* has been reported to produce an endogenous siRNA from a convergently transcribed and overlapping gene pair, it is otherwise considered to play a subordinate and redundant role in siRNA biogenesis. Another possibility, however, is that the primary role of *DCL2* has simply not yet been identified. With this idea in mind, we undertook a systematic study of the impact of mutations in *DCL* genes and the effect of viral silencing suppressors on siRNA biogenesis in *Arabidopsis thaliana* carrying different types of silenced transgene. Our experiments focused on three general cases: 1) sense transgene- induced silencing, in which case a construct that was designed to express a reporter gene becomes post-transcriptionally silenced instead, 2) hairpin transgene-induced silencing of a sense transgene, in which case a transgene that produces a self-complementary transcript is used to target an expressing, homologous sense transgene, and 3) processing of the hairpin transcript itself (hairpin transgene self-silencing). We expected that these cases might differ in their *DCL* requirements because sense transgene-induced silencing requires a set of genes that are not required for hairpin-induced silencing of transgenes. Some of these genes are thought to mediate production of dsRNA from the sense transgene transcript and include *RDR6*, *AGO1,* and *SGS3*–which encodes a coiled-coiled domain protein of unknown function, while *HEN1* encodes a methylase that acts downstream of dsRNA and stabilizes siRNAs. The transgenes we used differ in several important respects from those in other studies. First, no other study to date has analyzed the *DCL* requirements for sense transgene-induced silencing. The exact nature of the silencing trigger is still unknown in this type of silencing, but it is thought that the locus gives rise to an aberrant transcript, which might lack normal 5′ or 3′ ends. The aberrant transcript becomes a template for RDR6, which synthesizes the complementary strand and produces dsRNA. Second, with respect to hairpin transgene-induced silencing, the loop portion of the hairpin construct we used is retained in the mature transcript and has sequence identical to that of the targeted sense transgene. Thus, this hairpin construct is unusual in that it has a region that requires *RDR6* to produce dsRNA in addition to one that can directly produce dsRNA, and siRNAs from both regions can target the sense transgene. Furthermore, although hairpin transgene-induced silencing has been the subject of numerous studies, no other study has examined the fate of the hairpin transcript itself and the *DCL* requirements for self-silencing of the hairpin. Lastly, we used transgenes driven by the ubiquitously expressed cauliflower mosaic virus (CaMV) 35S promoter. Consequently, silencing occurs throughout the plant without the need for a silencing signal. In all three cases of silencing we examined, accumulation of secondary siRNAs required *DCL2* but not *DCL4*, providing evidence that DCL2 plays an essential, as opposed to redundant, role in transitive silencing of transgenes. This result strongly suggests that there are natural targets of silencing for which DCL2 plays a primary role. Furthermore, we found that viral suppressors of silencing and a *dcl2* null mutation had similar effects on accumulation of secondary siRNAs, supporting a primary role for *DCL2* in antiviral defense. # Results ## *DCL2* but not *DCL4* is Required for Sense Transgene-induced Silencing To study sense transgene-induced silencing, we used the L1 transgenic line, which carries a direct repeat of the T-DNA insert and is silenced for the *uidA* gene encoding β-glucuronidase (GUS). Silencing of the GUS transgene in line L1 initiates at about the time the plant bolts and is characterized by very low accumulation of GUS mRNA and concomitant accumulation of GUS siRNAs that are *RdR6*-dependent and 21- and 22-nucleotides (nt) in length (lanes 5–6 and 11–12; ). These siRNA size classes have previously been attributed to the activity of DCL4 and DCL2, respectively. About equal amounts of the two size classes are detected using a probe to the 3′-most region of the transcript, whereas only 21-nt siRNAs are detected with a probe to the central region, and no siRNAs of any size class can be detected using a probe to the 5′-most region (lanes 5–6 and 11–12). This drop-off in accumulation of siRNAs at the 5′ end of the L1 GUS transcript was not seen in an earlier study , possibly because the 5′ probe used by that group extended 258-nt further 3′ than ours. To identify which *DCL* genes are required for sense transgene-induced silencing, we crossed the L1 line with *dcl* mutant lines and examined the accumulation of GUS mRNA and siRNAs in F2 progeny that were homozygous for the *dcl* mutation and also carried at least one copy of the L1 GUS transgene. The post-transcriptional silencing induced by a single copy of the L1 GUS locus is comparable to that observed in plants homozygous for the locus (compare lanes 13 and 14). An L1 GUS line that expressed the P1/HC-Pro viral suppressor of silencing was used as a positive control showing expression of the GUS transgene. The *dcl2* and *dcl3* mutant lines were null mutants generated by T-DNA insertion, whereas the *dcl1* mutant lines were partial loss of function mutants. The *dcl4* mutant lines had point mutations in *DCL4* that eliminated or greatly reduced accumulation of the 21 nt species of ta-siR255 (lanes 3–4 and 9–10;). Because 21 nt ta-siRNAs are produced by DCL4, this result confirms that these *dcl4* mutant lines are highly deficient for DCL4 activity. All F2 progeny homozygous for the *dcl3* mutation were transcriptionally silenced for GUS, whether they carried one or two copies of the GUS transgene (Mlotshwa and Vance, manuscript in preparation), preventing examination of the role of *DCL3* in sense transgene-induced silencing in this system. Transcriptional silencing could be avoided, however, in the *dcl2* mutant by using progeny that were hemizygous for the GUS transgene: Run-off transcription assays showed that whereas *dcl2* mutant progeny homozygous for the L1 locus were transcriptionally silenced for GUS, ones hemizygous for the L1 locus were not. The transcriptional silencing induced by T-DNA insertion mutations is likely due to the CaMV 35S promoter in the engineered T-DNA. This propensity of the L1 locus to become transcriptionally silenced in the presence of additional copies of the CaMV promoter highlights the importance of examining siRNA accumulation before concluding that absence of GUS mRNA or activity indicates that the locus is post-transcriptionally silenced. Transcriptional silencing was not a problem in the *dcl1* or *dcl4* mutant lines, which were not T-DNA insertion mutants. Silencing of the L1 GUS transgene was severely impaired in *dcl2* mutant plants, as shown by greatly increased accumulation of GUS mRNA and the absence of GUS siRNAs (compare lanes 14, 16, and 17). Thus, *DCL2* plays a primary role in this *RDR6*-dependent silencing pathway. In contrast, silencing of the L1 GUS transgene was enhanced in both *dcl4-2* and *dcl4-10* mutant plants, as shown by reduced accumulation of GUS mRNA compared to wild type plants (lanes 3–6 and 9–12). Thus, although DCL4 activity produced the major fraction of GUS siRNAs that accumulated in the wild type background, neither *DCL4* nor *DCL4*-dependent 21-nt siRNA is required for sense transgene-induced silencing. The enhancement of silencing seen in the *dcl4* mutants might indicate some inhibitory role of *DCL4* or an earlier onset of silencing due to the accelerated juvenile-to-adult transition seen in these mutants. The *dcl1* mutant plants were slightly impaired for silencing of L1 GUS (lanes 18–20), suggesting that *DCL1* plays a facilitating role in sense transgene-induced silencing like that reported for hairpin transgene-induced silencing of an endogenous gene. Although accumulation of GUS mRNA in the *dcl2* mutant plants is much greater than in wild type, it is considerably less than seen in plants expressing P1/HC- Pro (lanes 14–17). Part of this differential is likely a gene dosage effect due to the fact that the P1/HC-Pro line is homozygous for L1 GUS, whereas the *dcl2* mutant is hemizygous for the transgene. The remaining differential might indicate that absence of DCL2 does not completely eliminate sense transgene- induced silencing, but only reduces it. Finally, the absence of GUS siRNAs of all size classes in the *dcl2* mutant plants suggests that DCL2 or the 22-nt class of siRNAs it generates is necessary for DCL4 and other DCL enzymes to be active on this type of substrate. ## *DCL2* but not *DCL4* is Required for Transitivity in Self-silencing of a Hairpin Transgene Our initial studies of hairpin transgene-induced silencing used northern analysis to examine the fate of the hairpin transcript itself. The self- complementary ΔGUS-SUG transgene in line 306-1 consists of the GUS coding sequence with a 231-nt deletion after nucleotide 558 and an inverted duplication of the 5′ proximal 558-nt at the 3′ end. The 558-nt self-complementary regions form the stem of the hairpin, while the intervening GUS sequence forms the loop. Unlike hairpin transgene constructs that have an intron separating the self- complementary regions, the loop portion of the ΔGUS-SUG transgene is not eliminated from the mature transcript. As a result, it is possible to examine siRNA biogenesis not only from the stem of the hairpin, which can pair to form dsRNA, but also from the loop, which requires the action of an RDR to generate dsRNA. Not surprisingly for a transgene that is a strong inducer of silencing, very little full-length ΔGUS-SUG transcript accumulated in line 306-1 (, lanes 1–4, band labeled hp). Unexpectedly, however, significant quantities of a smaller RNA species accumulated. This species could be detected with a probe specific for the loop (lanes 3–4) but not with one specific for the stem (lanes 1–2), showing that it corresponded to the loop portion of the hairpin transcript. Its size, as determined by migration in agarose gels, corresponds to that of the entire loop (data not shown). The presence of GUS siRNAs confirms that the ΔGUS-SUG transgene in line 306-1 is silenced (lanes 3–4). Preferential accumulation of the loop mRNA, however, indicates that degradation of the hairpin transcript occurs mainly via processing of the stem of the hairpin. Degradation of the loop portion of the transcript, which requires transitive silencing, apparently does not occur very effectively in this system. Analysis of ΔGUS-SUG mRNA and siRNA accumulation in an *rdr6* mutant backgound confirms that elimination of the full-length ΔGUS-SUG transcript and accumulation of siRNAs from the stem of the hairpin are *RDR6*-independent, whereas accumulation of siRNAs from the loop is *RDR6*-dependent: Loop mRNA, but little full-length ΔGUS-SUG transcript, accumulates in *rdr6* mutant plants, as in wild type; however, siRNAs from the loop are eliminated by the *rdr6* mutation, while those from the stem are largely unaffected (lanes 1–2). Thus, as expected, siRNAs from the stem are primary siRNAs, while those from the loop are secondary siRNAs and part of the transitive silencing pathway. The increased accumulation of GUS loop mRNA in the *rdr6* mutant compared to wild type provides additional confirmation that degradation of the loop occurs via transitive silencing. The ΔGUS-SUG primary siRNAs are predominantly 21-nt (lanes 1–4), suggesting that DCL4 is normally responsible for processing the dsRNA stem of the hairpin. In contrast, secondary siRNAs from the hairpin transcript are approximately equal parts 21- and 22-nt (lanes 3–4), as seen for the 3′ end of the L1 GUS sense transgene. To examine the roles of *DCL2* and *DCL4* in self-silencing of the ΔGUS-SUG transcript, we crossed line 306-1 with *dcl2* and *dcl4* mutant lines and examined F2 progeny homozygous for the *dcl* mutation. The hairpin transgene remains silenced in both *dcl* mutants, as indicated by the presence of GUS siRNAs and little accumulation of full-length ΔGUS-SUG mRNA (lanes 11–13). In the *dcl4* mutant, however, accumulation of loop mRNA is nearly eliminated (lanes 12–13), suggesting that impairing DCL4 activity promotes transitive silencing of the hairpin transcript. Enhanced accumulation of loop siRNAs and elimination of the majority of primary siRNAs in the *dcl4* mutant compared to wild type (lanes 12–13) are consistent with a shift to transitive silencing in the mutant. Increased accumulation of full-length ΔGUS-SUG mRNA in the *dcl4* mutant compared to wild type (lanes 12–13) indicates that the reduction in loop mRNA accumulation in the mutant is not due to increased non-specific degradation. Furthermore, it suggests that transitive silencing of the hairpin transcript is not as effective as the *DCL4*-dependent pathway that normally eliminates the stem. In contrast to the shift to transitive silencing produced by the *dcl4* mutation, the *dcl2* mutation impairs transitivity: No loop siRNAs accumulate in the *dcl2* mutant (lanes 5–8). Because loop siRNAs are RDR6-dependent secondary siRNAs (lanes 1–4), this result implies that *DCL2* is required for transitive self-silencing of the hairpin transgene. In addition, accumulation of stem siRNAs, which are predominantly 21-nt, is reduced in the *dcl2* mutant (lanes 5–8), suggesting that DCL2 facilitates DCL4 production of primary siRNAs or enhances their stability. The *dcl2* mutation blocks transitive silencing even in a *dcl4* mutant background: Loop siRNAs are absent and high levels of loop mRNA accumulate in *dcl2 dcl4* double mutant plants (lanes 9–10), indicating that degradation of the hairpin transcript occurs mainly via processing of the dsRNA stem of the hairpin in the absence of both *DCL2* and *DCL4*. Thus, the shift to transitive silencing observed in *dcl4* mutant plants requires *DCL2*. Processing of the dsRNA stem of the hairpin transcript in the *dcl2 dcl4* double mutant presumably involves DCL3, DCL1 or residual low-level DCL4 activity, as 24-nt and a small amount of 21-nt primary (stem) siRNAs accumulate in these plants (lanes 9–10). Altogether, these results show that *DCL2* is required for production of *RDR6*-dependent siRNAs from a hairpin transgene as well as from a sense transgene. In addition, self-silencing of the hairpin transcript provides another example of the hierarchical action of the DCL proteins and reveals that different DCL proteins are preferentially associated with different mechanisms of transcript degradation. In wild type plants, the major pathway for degradation of the ΔGUS-SUG hairpin transcript appears to be *DCL4*-dependent processing of the stem into primary siRNAs. In a *dcl4* mutant, transitive silencing is prominent and involves *DCL2*- and *RDR6*-dependent production of secondary siRNAs. When both *DCL2* and *DCL4* are defective, degradation again occurs via processing of the stem into primary siRNAs, but this time involves *DCL3* and possibly *DCL1*. ## *DCL2* but not *DCL4* is required for transitivity in hairpin transgene-induced silencing of a sense transgene To analyze hairpin transgene-induced silencing of a sense transgene, we used the ΔGUS-SUG locus in line 306-1 to target an expressing GUS locus. Line 6b4 carries a GUS sense transgene that is not silenced, and 6b4 plants accumulate high levels of GUS mRNA but no GUS siRNAs (lane 5). However, the GUS-expressing locus in line 6b4 is silenced in the presence of the ΔGUS-SUG construct, as evidenced by the loss of GUS activity. For our experiments, therefore, we used a 6b4/306 transgenic line obtained by crossing the 6b4 and 306-1 lines. The full- length GUS and ΔGUS-SUG transcripts are similar in size, but the ΔGUS-SUG sequence is 327-nt longer. Northern analysis shows that the 6b4/306 line is silenced for both transgenes because little or no full-length transcript from either one accumulates, but loop mRNA from the hairpin transcript and GUS siRNAs do (lanes 3–4). To detect GUS siRNAs in these experiments, we used the same stem and loop probes (1 and 3, respectively) as in the previous section, but also included probe 2, which corresponds to the region deleted in the ΔGUS-SUG construct. In line 6b4/306, probes 1 (stem) and 3 (loop) can detect siRNAs arising from either the ΔGUS-SUG transgene or the GUS sense transgene, while probe 2 is specific to the GUS sense transgene. Accumulation of the GUS siRNAs detected by probes 2 and 3 in 6b4/306 plants was eliminated in the *rdr6* mutant background (lanes 1–2), confirming that these are *RDR6*-dependent secondary siRNAs and part of the transitive silencing pathway. Accumulation of the GUS siRNAs detected by probe 1, however, was not reduced by the *rdr6* mutation (lanes 1–4), showing that even in the presence of the GUS sense transgene, these are mostly primary siRNAs from the stem of the hairpin transcript. This observation plus the accumulation of loop mRNA suggests that hairpin transgene-induced silencing of the GUS-sense transgene in line 6b4/306 does not have a large transitive component. Indeed, silencing of the GUS-sense transgene in line 6b4/306 is largely unaffected by the *rdr6* mutation, as little or no full-length GUS mRNA accumulates in the *rdr6* mutant plants (lanes 1–2), consistent with earlier work showing that hairpin transgene-induced silencing is *RDR6*-independent. To examine the roles of *DCL2* and *DCL4* in ΔGUS-SUG-induced silencing of the GUS sense transgene, we performed crosses to make homozygous *dcl2*, *dcl4*, and *dcl2 dcl4* mutant progeny of line 6b4/306. The effects of the *dcl* mutations on silencing of the GUS sense transgene in these lines was very similar to their effects on self-silencing of the hairpin transgene shown above for lines carrying the hairpin locus alone. In the 6b4/306 background, the *dcl4* mutation eliminated accumulation of loop mRNA and greatly increased accumulation of secondary siRNAs–including those (probe 2) that could be derived only from the GUS sense transgene transcript–providing additional evidence that impairing DCL4 activity promotes a shift to transitive silencing (, lanes 11–14, probes 2 and 3). The highly abundant secondary siRNAs in the *dcl4* mutant included 24-nt as well as 22-nt siRNAs, showing that both DCL3 and DCL2 produce secondary siRNAs from the targeted sense transgene when DCL4 is defective. The much greater increase in secondary siRNA accumulation caused by the *dcl4* mutation in lines carrying both transgenes than in those having the hairpin locus alone suggests that the GUS sense transgene transcript is a much better substrate for *RDR6*-dependent production of siRNAs than the hairpin transcript. In contrast to *dcl4* enhancement of secondary siRNA accumulation, the *dcl2* mutation eliminated secondary siRNAs–including those (probe 2) that could be derived only from the GUS sense transgene mRNA (lanes 6–9, probes 2 and 3). Accumulation of siRNAs detected by probe 1, which are mostly primary siRNAs, was not greatly affected (lanes 6–9). The *dcl2* mutation also eliminated accumulation of secondary siRNAs in the *dcl4* mutant background (compare lanes 16–19, probes 2 and 3) and restored accumulation of loop mRNA (lanes 16–20). Thus, *DCL2* is required for transitive silencing of the sense transgene target of a hairpin transgene, as well as for transitive silencing of the hairpin itself. The above results provide a basis for understanding the robust nature of hairpin transgene-induced post-transcriptional silencing. Whereas a sense transgene alone activates only an *RDR6*-dependent silencing pathway, a hairpin transgene activates multiple silencing pathways–including one(s) involving only *RDRP*-independent primary siRNAs. Consequently, in the presence of a homologous hairpin transgene, the sense transgene becomes a target for all the silencing pathways activated by the hairpin construct. ## The P1/HC-Pro and P38 viral suppressors block hairpin transgene-induced silencing of a sense transgene, but do not block processing of the stem of the hairpin transcript into primary siRNAs Previous studies in our laboratory showed that the tobacco etch virus (TEV) P1/HC-Pro viral suppressor of silencing altered the accumulation of siRNAs in tobacco, eliminating those derived from sense transgenes and shifting the size distribution of ones derived from inverted repeat and amplicon transgenes. To determine the effect of P1/HC-Pro and other viral suppressors on siRNA biogenesis in the three cases of silencing examined above, we crossed the L1, 306-1, and 6b4/306 lines with *Arabidopsis* lines transgenic for P1/HC-Pro (from turnip mosaic virus) or P38 (from turnip crinkle virus). The P38 viral suppressor carried a C-terminal HA epitope tag, whereas P1/HC-Pro had no tag. In all cases, progeny carrying a viral suppressor transgene exhibited the developmental phenotype associated with expression of the suppressor in the parental line (data not shown). Both of these suppressors restored accumulation of GUS mRNA and eliminated accumulation of GUS siRNAs in progeny of crosses with the sense transgene silenced line L1, similar to the effect of TEV P1/HC-Pro on sense transgene- induced silencing in tobacco. Neither suppressor, however, enabled accumulation of high levels of the full-length ΔGUS-SUG transcript in progeny of crosses with the hairpin-transgenic line 306-1, although accumulation of the full-length transcript was increased (lanes 1–5). In contrast to the full-length transcript, high levels of loop mRNA accumulated in both the P1/HC-Pro and P38 lines (lanes 1–4). Thus, P1/HC-Pro and P38 have little or no inhibitory effect on processing the dsRNA stem of the hairpin, but they suppress the secondary siRNA-dependent pathway responsible for degradation of the loop. Consistent with this interpretation, P1/HC-Pro and P38 reduced accumulation of secondary (loop) siRNAs (lanes 1–5). Interestingly, the suppressors also increased accumulation of primary (stem) siRNAs (lanes 1–5). The relatively minor accumulation of full- length transcript in the P1/HC-Pro and P38 lines suggests that although most ΔGUS-SUG transcripts are degraded via processing of the dsRNA stem into primary siRNAs, some are degraded as a result of being targeted by primary siRNAs and that P1/HC-Pro and P38 block this latter pathway. The increase in primary siRNA accumulation in the viral suppressor lines compared to wild type plants is consistent with the idea that some full-length hairpin transcripts are degraded by transitive silencing in wild type plants, thereby reducing production of primary siRNAs. Alternatively, the suppressors might increase primary siRNA stability, perhaps by binding siRNA duplexes. In progeny of crosses with the 6b4/306 line, both P1/HC-Pro and P38 restored accumulation of full-length mRNA from the GUS sense transgene and eliminated secondary siRNAs (probes 2 and 3) (lanes 3–8). In addition, both suppressors enhanced accumulation of loop mRNA and primary siRNAs. The enhancement of loop mRNA accumulation by P38 was so great that loop mRNA in wild type plants was barely detectable at the exposure appropriate for the samples from P38-expressing plants (lanes 7–9), although it was clearly visible at a 3-fold longer exposure (, image of lanes 7–9 labeled “loop 3x exp”). These results are consistent with the effects of the suppressors on secondary siRNA-dependent silencing seen with the hairpin transgene alone (lanes 1, 2, 9). Hybridizing a duplicate high molecular weight RNA blot with probes specific to the viral suppressors confirmed that both suppressors were being expressed (panels in lanes 1–6 directly below rRNA). The P38 viral suppressor produced several very interesting effects in the above experiments. P38 transgenic plants accumulated much higher levels of loop mRNA than P1/HC-Pro plants (lanes 1–4 and 4B, lanes 3–6), suggesting that P38 is much more effective than P1/HC-Pro at preventing degradation of the loop. P38 also had a more pronounced effect than P1/HC-Pro on the size distribution of primary siRNAs, causing the accumulation of a closely spaced smaller species (lanes 5–6), suggesting that altering the processing of small RNAs is part of the P38 mechanism of action. Some of our viral suppressor results differ from those of another group. P1/HC- Pro inhibition of ta-siRNA accumulation and P38 inhibition of DCL4 activity is not evident in our lines (siR255 probe; primary siRNAs). Conversely, the enhancement of primary siRNA accumulation by P1/HC-Pro and P38 that we observe in hairpin transgene-induced silencing was not detected by the other group. We expect that differences in the plant lines as well as in the inducers and targets of silencing used by the two groups are likely responsible for such discrepancies. Consistent with this expectation and in agreement with our result, an independent group using our plant line observed that P1/HC-Pro does not block ta-siRNA accumulation. It is also interesting to note that P1/HC-Pro enhancement of siRNA accumulation has previously been observed for a hairpin promoter sequence construct that induced transcriptional silencing in tobacco. # Discussion The present work demonstrates that *DCL2* is required for silencing induced by a sense transgene locus and for accumulation of secondary siRNAs in three mechanistically different examples of transgene silencing. Surprisingly, even *DCL4*-dependent 21-nt and *DCL3*-dependent 24-nt secondary siRNAs are eliminated by a *dcl2* null mutation, suggesting that DCL2 is required for DCL4 and DCL3 participation in production of *RdR6*-dependent siRNAs from these transgenes. Our results show that DCL2 plays a primary role in transitive silencing of transgenes and suggest that there are as-yet-unidentified natural substrates for which DCL2 is the primary DCL enzyme. One possible origin of natural substrates of DCL2 is foreign genes that are introduced into the plant genome during infection with pathogens like *Agrobacterium tumefaciens*. Such genes constitute a natural analogue of transgenes. In addition, our observation that a *dcl2* null mutation and two viral suppressors of silencing have the same effect on secondary siRNA accumulation suggests that natural substrates of DCL2 might be produced in viral infection. The similarity in effect of the *dcl2* mutation and the viral suppressors applies only to accumulation of secondary siRNAs, however, and not to suppression of silencing in general: The suppressors eliminated both hairpin and sense transgene-induced silencing, whereas the *dcl2* mutation impaired only the latter. Although DCL2 was initially proposed to play an antiviral role, subsequent studies suggested that the enzyme functions in only a subordinate and redundant capacity in antiviral defense. A key piece of evidence for this conclusion is that 22-nt viral siRNAs accumulated primarily when DCL4 was inactive. Those studies, however, either involved only primary siRNAs or did not distinguish between primary and secondary siRNAs. In contrast, our results point to a requirement for *DCL2* specifically in the production of secondary siRNAs. *RDR6*-dependent production of secondary siRNAs is thought to be particularly important in slowing the systemic spread of viruses by allowing systemically invaded cells to respond before the virus starts replicating. After replication is established, however, viral siRNA production no longer depends on *RDR6*, suggesting that any specific requirement for *DCL2* in antiviral defense might be transient and not detectable in bulk infected tissue. The inhibition of transitivity by P1/HC-Pro and P38 provides additional evidence for the importance of this *RDR6*-dependent branch of silencing in antiviral defense. However, suppression of silencing by these viral proteins must involve more than inhibition of transitivity because they suppress hairpin transgene- induced silencing, which does not require *RDR6*. One possibility is that viral suppressors also inhibit siRNA function by binding to siRNA duplexes. It will be interesting to determine whether inhibition of transitivity results from the dsRNA binding activity of the suppressors or reflects some additional activity. Small RNA pathways have been shown to involve functional modules of specific gene family members, and modules can act alone or sequentially. Thus, *DCL1* and *AGO1* are involved in the biogenesis and function of miRNAs, while *RDR2*, *DCL3*, and *AGO4* are involved in that of hc-siRNAs. *RDR6* and *DCL4* constitute a module that functions in the biogenesis of ta-siRNAs and works downstream of miRNA-directed cleavage. Our results suggest that *DCL2* and *RDR6* also constitute a module in the case of some substrates. These substrates differ from ones previously identified for the *DCL4/RDR6* module in that neither DCL4 nor DCL3 appears to process them in the absence of DCL2, perhaps due to unique structural features of the substrates or their localization in the cell. One possibility is that DCL2 might be required upstream of dsRNA production to recognize the substrates and/or recruit RDR6, after which DCL2 and other DCL enzymes produce siRNAs from the resultant dsRNA. Because the majority of secondary siRNAs in wild type plants are 21-nt for all three cases of silencing we examined, DCL4 appears to be overall the most active one of the enzymes on the *RDR6*-dependent dsRNA substrates produced by transgenes. Alternatively, 21-nt siRNAs might simply be the most stable. For all of three transgenes, however, 21-nt and 22-nt secondary siRNAs are equally abundant at the 3′ end of the GUS sequence, consistent with a specific requirement for *DCL2* early in the *RDR6*-dependent arm of transgene silencing. The processivity of DCL4 appears to be greater than that of DCL2 on these substrates because only 21-nt siRNAs accumulate from the middle region of the GUS sequence. This difference in processivity of the two enzymes might be at least partially responsible for the perception that DCL2 plays a purely subordinate role. The reduced accumulation of loop mRNA and primary siRNA in *dcl2* mutant plants suggests that DCL2 might also stabilize its substrates against non-silencing related nucleolytic degradation. Because transitivity and sense transgene-induced silencing–like ta-siRNA biogenesis-require *RDR6*, DCL4 has been considered the likely DCL enzyme for siRNA production in sense transgene silencing. Moreover, siRNA production from a silenced sense transgene that was engineered to have a miRNA cleavage site was shown to utilize the *DCL4/RDR6* module after miRNA cleavage and not to require *DCL2*, providing support for the expected role of DCL4 in transitivity and sense transgene silencing. The present work, however, shows that *RDR6*-dependent production of siRNAs from transgenes that do not have a miRNA cleavage site differs from ta-siRNA biogenesis. A number of studies have examined the *DCL* requirements of hairpin transgene induced silencing. The studies variously differ from one another and from our work with respect to structure of the hairpin transgene, dependence on a silencing signal, and whether the target of silencing is a transgene or an endogenous gene. Our study is the only one to date that has focused specifically on cell-autonomous silencing of transgenes and systematically distinguished between accumulation of primary and secondary siRNAs. Using this approach, we find that hairpin transgene-induced silencing occurs in *dcl2*, *dcl4*, and *dcl2 dcl4* mutants. The mechanism of target degradation varies, however, depending on which of the DCL enzymes are active. DCL4-dependent processing of the stem of the hairpin into primary siRNAs is the major pathway utilized in wild type and *dcl2* mutant plants. Inactivation of DCL4 promotes a shift to transitive silencing and the production of secondary siRNAs. The very high level of accumulation of secondary siRNAs in *dcl4* mutant plants transgenic for both the ΔGUS-SUG and 6b4 GUS loci suggests that under some conditions, secondary siRNA production itself is a major contributor to degradation of the targeted transcript, consistent with the observation that *AGO1* is not required for silencing of a GUS transgene by the ΔGUS-SUG locus. When both DCL2 and DCL4 are inactive, silencing is again dependent on primary siRNAs, and processing of the stem of the hairpin by DCL3 becomes evident. Our observation that DCL4 completely processes the 558-bp dsRNA stem of a hairpin construct into primary siRNAs, leaving only the loop portion, suggests that the enzyme is highly processive on *RDR6*-independent dsRNA as well as on *RDR6*-dependent dsRNA and is consistent with the observation that *DCL4* is required when silencing depends entirely on production of primary siRNAs from a hairpin transgene. The gene families involved in RNA silencing in plants have evolved to provide a large degree of functional diversity. It is striking that despite this potential for diversity, production of secondary siRNAs relies on *DCL2* in both sense and hairpin transgene silencing, which otherwise differ in their genetic requirements. The DCL enzymes and functional modules undoubtedly evolved to efficiently handle a wide variety of natural substrates–many of which have probably not yet been identified, and it is likely that the enzymes recognize signature structural features of their preferred substrates. The multiplicity of functional pairings is just beginning to be elucidated. For example, although RDR2/DCL3 is a well established module involved in hc-siRNA biogenesis, DCL3 activity without RDR2 and the pairing of RDR2 with DCL4 has recently been proposed for a particular case of hairpin transgene targeting of an endogenous gene. The diversity and great versatility of small RNA pathways in plants is perfectly exemplified by how well prepared plants turned out to be to defend themselves against the recent evolution of genetic engineers. # Materials and Methods ## Transgenic and mutant *Arabidopsis* lines All lines are in the Columbia (Col-0) ecotype. The following transgenic or mutant lines were described previously: L1, 306-1 and 6b4/306, P38 (CP), P1/HC- Pro, *sgs2-1 (rdr6)*, *dcl1-7* and *dcl1-8*, *dcl2-1* (SALK_064627) and *dcl4-2*. The *dcl4-10* mutation is a previously unpublished mutation that arose in an EMS mutagenesis. It is a single mucleotide mutation that produces a glycine to arginine change at amino acid 1403, which is located in the RNAse III domain of DCL4 (data not shown). ## PCR genotyping for mutant and transgene loci The T-DNA primer LBa1 (tgg ttc acg tag tgg gcc atc g) was used with primers DCL2p5 (ttg gat tgc atg cac aca tt) and DCL2p6 (ctc aga aat aaa gat aac agt aag caa at) for *dcl2-1* genotyping. Primer DCL2p5 together with DCL2p6 amplifies a 400-bp product from the wild type locus, while DCL2p5 together with LBa1 amplifies a 600-bp product from the *dcl2-1* locus. Thus, a PCR reaction with all three primers will amplify only the 600-bp product in the case of homozygous *dcl2-1*. Genotyping for homozygous *dcl4-2* was performed as described previously. Homozygous *dcl4-10* plants containing the L1 GUS transgene were identified by their distinctive leaf phenotype, which is similar to that of the *dcl4-2* mutant. Primers L1-306-6b4-F (ttg ggg ttt cta cag gac gga c) and L1-306-6b4-R (cta tcc ttc gca aga ccc ttc c) were used in combination with GUS staining to screen for the presence of different GUS loci. A 250-bp fragment is obtained with the 306-1 locus, a 188 bp fragment with the L1 locus and a 127-bp fragment with the 6b4 locus, due to differences in the 5′ upstream sequences of the GUS constructs. Homozygous *dcl1-7* and *dcl1-8* plants containing the L1 GUS transgene were identified by their distinctive recessive phenotypic defects. Primers 2911F7 (gca ggg ata ctt gaa cat ggc c) and 2911R8 (gtt aac aac cta tgc cac gc) were used for *sgs2-1* genotyping: PCR followed by digestion with BstNI yields three fragments (sizes 250, 200, and 150 bp) from the wild type locus, but only two fragments (200 and 400 bp) from the *sgs2-1* locus. P38-derived primers P38-F (cgc cca atg ggc gat aaa g) and P38-R (cgt ctc ggt cga atg cca gag c) were used to confirm the presence of the P38 transgene in combination with phenotypic and phosphoinothricin selection. RNA isolation, gel blot analysis, and nuclear run-off transcription. RNA was isolated from a mixture of representative aerial tissues of flowering plants. For most experiments, tissues from about ten plants of a given genotype were pooled for RNA isolation. For each genotype, at least two independent RNA preparations were made from separate plants or pools of plants and electrophoresed in neighboring lanes on RNA gels. Total RNA isolation and gel blot analysis of high and low molecular weight RNA were performed as described previously. Nuclear run-off transcription analysis was performed as described previously except that nuclei were isolated from aerial tissues of flowering plants. The minimal sequence (taa tac gac tca cta tag gg) of the T7 promoter was incorporated by PCR into the 3′ or 5′ ends of DNA templates to make RNA probes of antisense polarity to detect mRNAs or of sense polarity to detect antisense siRNAs, respectively. \[α-³²P\]UTP-labeled RNA probes were transcribed in vitro using an Ambion MAXIscript kit with T7 polymerase and hybridized to mRNA blots at 68°C in Ambion ULTRAhyb buffer, or to siRNA blots at 42°C in Ambion ULTRAhyb-oligo buffer. DNA probes were labeled using an Ambion DECAprime II kit and hybridized to mRNA blots in Ambion ULTRAhyb buffer at 42°C. The coding sequence coordinates of the probes for viral suppressors of silencing were: P38, nucleotides 42 to 472; HC-Pro, entire coding sequence. The siR255 probe was prepared by end-labeling the complementary DNA oligonucleotide with \[α-³²P\]ATP using the StarFire™ Oligo Labelling System (Integrated DNA Technologies) as described previously. The probe was hybridized to small RNA blots at 42°C in Ambion ULTRAhyb-oligo buffer. We thank Herve Vaucheret for providing seeds of the L1, 6b4, and 306 lines and Jim Carrington for seeds of the P38 transgenic line. [^1]: Conceived and designed the experiments: VV SM LB. Performed the experiments: SM. Analyzed the data: VV GP SM LB. Contributed reagents/materials/analysis tools: SP XC SM AP ME JL. Wrote the paper: VV GP SM LB. Other: Generated the dcl4-10 mutant: SP AP. [^2]: Current address: Howard Hughes Medical Institute, Skirball Institute of Biomolecular Medicine, New York University School of Medicine, New York, New York, United States of America [^3]: The authors have declared that no competing interests exist.

# Introduction Heterosis is defined as the superior performance of crossbred characteristics as compared with corresponding inbred ones. The utilization of heterosis has become a major strategy to increase the productivity of plants and animals. Despite the successful utilization of heterosis in many crops, there still exists a contradiction between the agricultural practice of heterosis utilization and our understanding of the genetic basis of heterosis and this hampers the effective exploitation of this biological phenomenon. The classical quantitative genetic explanation of heterosis centered on three hypotheses: dominance, over-dominance and epistasis. Evidence of these genetic models remained unavailable until very recent advances in molecular marker technology, high-density linkage maps and genome sequencing. Although much research into the genetic basis of heterosis in crops and plants has been conducted, little consensus has emerged. Research has indicated that heterosis may be attributable to dominance, over-dominance, epistasis or a combination of all of these, depending on the study materials, traits and analytical approach. Typically, little is known about the genetic control of heterosis in the complex polyploid crop rapeseed (*Brasscia napus* L.). Based on the phenotype of the E×R53-DH population and the corresponding BC population, as well as the mid- parent heterosis of the BC population, Radoev et al. (2008) mapped 33 QTL (9 of which showed a significant dominant effect) and a large number of epistatic interactions for seed yield and the three yield-component traits. They concluded that epistasis together with all levels of dominance from partial to over- dominance is responsible for the expression of heterosis in rapeseed. Based on this E×R53-DH population and another E×V8-DH population with the same parent, and using the same experimental design, Basunanda et al. (2010) detected a number of QTL hotspots responsible for seedling biomass and yield-related traits. Given the key role of epistatic interactions in the expression of heterosis in oilseed rape, they supposed that these QTL hotspots might harbour genes involved in regulation of heterosis for different traits throughout the plant life cycle, including a significant overall influence on heterosis for seed yield. However, in both studies, all kinds of genetic effects (A, D and AA, AD/DA, DD) were unable to be estimated in the same population, thus it was difficult to accurately estimate their mode-of-inheritance and relative importance in the expression of heterosis. There were several common patterns described in most of these studies. Firstly, the QTL for yield and yield-correlated traits tended to be clustered in the genome in many crop and model plants, such as rice, maize, wheat, rapeseed and *Arabidopsis*, which suggested the QTL of yield-correlated traits might have pleiotropic effects. However, this kind of pleiotropy has not been well analyzed genetically. Secondly, only a few limited traits were investigated and only a few QTL and epistatic interactions were identified for each trait, so a relatively comprehensive picture of the genetic architecture of heterosis remained unavailable. Thirdly, trials were carried out in only one or two environments and the environmental response of QTL and epistatic interactions for heterosis was not analyzed and thus remains unclear. The main objective of this study was to unravel the genetic architecture of heterosis with QTL mapping in rapeseed, including: (1) determine the level of heterosis for a range of yield-correlated traits; (2) investigate the relationship between molecular marker heterozygosity and heterosis/hybrid performance; (3) identify QTL and epistatic interactions underlying heterosis and estimate their genetic effect, mode-of-inheritance and environmental responses; (4) analyze the relative contribution of all kinds of genetic effects in the expression of heterosis in rapeseed (*Brassica napus* L.). # Results ## Correlation of trait performance and mid-parent heterosis among the 15 investigated traits In the same environment, most pair-wise genetic correlations of performance and mid-parent heterosis were similar. This was understandable since mid-parent heterosis was calculated from trait performance. In different environments, pair-wise genetic correlations differed considerably (mostly in degree, a few in direction), which suggested that genetic correlations depended strongly on the environments. Genetic correlations of performance and mid-parent heterosis among the investigated traits were also calculated across the three environments. In general, significant correlations were observed for 81.9% and 67.6% of the pair- wise combinations of the trait performance and mid-parent heterosis, respectively. Seed yield correlated significantly with the other 14 investigated traits for both trait performance and mid-parent heterosis; negatively for flowering time, maturity time and protein content, and positively for the other 11 ones. Interestingly, the mean *r*² of trait performance was somewhat higher than that of mid-parent heterosis for most traits, ranging from 0.04 and 0.03 (for seed development times) to 0.24 and 0.20 (for seed yield), respectively. ## Traits showing significant heterosis The analysis of variance (in both populations) revealed that genotype, environment and the interaction between them had significant effect on the performance of all the 15 yield-correlated traits, so they were calculated separately for each environment. The broad-sense heritability of these traits ranged from 0.58 (for seed yield) to 0.90 (for flowering time), with a mean of 0.73. The two parents showed significant differences in 38 of the 43 trait- environment combinations. The two populations showed obvious transgressive variation for all of the trait-environment combinations. It should be noted that DH and the reconstructed-F₂ population showed over-F₁ variations for 13 (except seed yield and seed number per plant) and all of the traits respectively in all environments, which indicated that heterozygosity was not always favorable for trait performance. There was significant heterosis on F₁ and F₂ generations compared with the mean of the parents and the DH population, respectively, for the nine (branch number, biomass yield, harvest index, plant height, pod number, pod yield, seed number per pod, seed number per plant and seed yield) and eight (except branch number) traits. Interestingly, for these traits with significant heterosis, the performance of F₁ was significantly higher than the mean of the F₂ population and higher than the mean of the DH population in 19 and all of the 25 trait-environment combinations respectively, which showed an obvious trend of inbreeding depression. According to the significance of heterosis, the 15 yield-correlated traits could be classified into two groups: the nine traits (seed yield, seed number per plant, biomass yield, pod number, harvest index, plant height, pod yield, seed number per pod and branch number) with heterosis and the other six traits (oil content, protein content, maturity time, flowering time, seed weight and seed development time) without heterosis. It should be noted that the correlation coefficients between seed yield and the nine traits with heterosis were all higher than that between the other six traits without heterosis. The analysis of variance revealed that genotype, environment and genotype × environment interaction had significant effect on mid-parent heterosis of the nine traits with heterosis, so they were calculated separately for each environment. For hybrid F₁, seed yield and seed number per plant showed strong mid-parent heterosis, biomass yield and pod number per plant showed moderate mid-parent heterosis, while pod yield, seed number per pod, harvest index, branch number and plant height showed low mid-parent heterosis. For the reconstructed F₂ population, the amount of heterosis varied widely for these traits, from highly negative to highly positive. The average mid-parent heterosis of the reconstructed F₂ population showed similar trend with that of F₁ for the nine traits. It should be noted that in each environment the mid-parent heterosis of some (the proportion is 10.2% for seed yield in S5 environment, data not shown) combinations of reconstructed F₂ population was higher than that of F₁, but the average mid-parent heterosis in the reconstructed F₂ population was in all cases lower than that in F₁. This indicated that heterosis was generally related to the heterozygosity at the population level but poorly correlated with heterozygosity at the individual level. It should be noted that, for these yield-correlated traits, the heritabilities (ranging from 0.40 to 0.60) of mid-parent heterosis were all lower than that (ranging from 0.58 to 0.90) of trait performance. ## Correlation between heterozygosity and hybrid performance/mid-parent heterosis for the nine traits with significant heterosis The correlation between heterozygosity and hybrid performance/mid-parent heterosis was significant for the nine traits with significant heterosis except branch number and seed number per pod, with mean *r*² ranging from 0.001 (branch number) to 0.066 (seed yield) for the different traits, which accorded well with the heterosis level of these traits. Generally, the mean *r*² between heterozygosity and hybrid performance was similar to that between heterozygosity and mid-parent heterosis. Whereas, the mean *r*² (0.026/0.022) between special heterozygosity and hybrid performance/mid-parent heterosis was a little higher than that (0.013/0.014) of general heterozygosity and hybrid performance/mid-parent heterosis in most cases. Interestingly, the mean *r*² between heterozygosity and hybrid performance/mid-parent heterosis was stronger in the S5 environment than in the other two environments, which suggested these correlations were also depended on the environment. Although 47 of the 100 correlations between heterozygosity and hybrid performance/mid-parent heterosis were significant, the *r*² were relatively small (from 1.21% to 18.5%), which suggested that molecular marker heterozygosity could not predict hybrid performance and mid-parent heterosis. ## Genome-wide detection and meta-analysis of QTL for 15 yield-correlated traits A total of 967 QTL (579 significant QTL and 388 suggestive QTL) were identified for the 15 yield-correlated traits in both populations in three environments. Exclusion of 209 non-overlapping suggestive ones, a total of 758 QTL was identified finally. Of which 390 identified QTL were from reconstructed F₂ population (ranging from 11 to 56 for each trait), they were potentially responsible for heterosis and were the objectives of the following analysis. The 390 identified QTL explained 1.4-20.8% (mean = 5.6%) of the phenotypic variance while 92.8% showed only moderate effect, with *R*²\<10% and only one explained \> 20% of phenotypic variance. Furthermore, for the 13 identified QTL with *R*²≥10%, the absolute values of their dominant degree (∣D/A∣) were all \< 1. This suggested that heterosis of these yield-correlated traits was typically controlled by numerous loci with little heterotic effect. To estimate the environmental response of QTL in natural environments, meta- analysis was used to integrate the identified QTL trait-by-trait in different environments. A total of 300 consensus QTL was identified, of which only 77 (25.7%) were repeatedly found in more than two environments and regarded as repeatable QTL, the other 223 (74.3%) were specifically identified in one of the three environments and considered as non-repeatable ones. This indicated that the expression of QTL of yield-correlated traits was strongly dependent on environmental conditions, which is also confirmed by the result that 55.3% (166/300) of consensus QTL showed significant QTL × environment interaction in ANOVA analysis. The proportion of the repeatable QTL was high for flowering time, development time of seeds, pod yield and seed number per pod, and results accorded with the high heritability of these traits. Only 77 consensus QTL were repeatable, whereas 68.8% changed their mode-of-inheritance in different environments. Only 5.2% of the 77 repeatable consensus QTL changed the direction of additive-effect, which suggested that the relative superiority of one allele over the others was stable in different environments. In contrast, 31.2% of the 77 repeatable consensus QTL changed their dominant-effect directions in different environments. In addition, only 20.8% ( = 16/77) of these repeatable consensus QTL showed significant interaction with the environment at P≤0.05, which was lower than that (67.3% = 150/223) of the non-repeatable ones. Therefore, the expression, direction and effect of QTL were all dependent on environmental conditions, which suggested the variability of QTL. The confidence intervals of most consensus QTL determined for each trait overlapped. The 300 consensus QTL for the 15 yield-correlated traits were therefore subjected to a second round of meta-analysis, which resulted in the integration of 220 consensus QTL into 84 pleiotropic unique QTL. ## Genome-wide detection and analysis of epistatic interactions in the reconstructed F₂ population and three environments for 15 yield-correlated traits A total of 522 statistically significant epistatic interactions were identified for the 15 yield-correlated traits in two populations and three environments and most of them were also confirmed by the two-way analysis of variance (data not shown). Of these significant epistatic interactions, 272 were identified from the reconstructed F₂ population (ranging from 11 to 29 for the different traits), potentially responsible for heterosis and were the objectives of the following analysis. Only two epistatic interactions of seed yield, which were detected in different environments and located in similar positions, were considered as repeatable, which suggested epistatic interactions of yield- correlated traits were extremely sensitive to the environmental variation. A total of 136, 103 and 33 epistatic interactions belonged to NN (the two loci involved in epistatic interaction were both with non-significant main-effects), NS (the two loci involved in epistatic interaction was one with significant main-effect and the other one with non-significant main-effect,) and SS (the two loci involved in epistatic interaction were both with significant main-effects) type of epistatic interactions respectively, which indicated most loci of epistatic interactions have no significant effect on trait performance alone but may affect it by epistatic interaction with other loci. The 272 epistatic interactions explained 1.4–18.3% (mean = 5.1%) of the phenotypic variance, while 95.6% showed only moderate effect, with *R*²\<10%. It should be noted that 91.9% of the 272 epistatic interactions occurred between different chromosomes. The proportion of the loci involved in multiple (2–7) epistatic interactions varied from 52.3% (for plant height) to 88.5% (for harvest index) for different traits and with a mean of 68.2% on average, which indicated the prevalence of pleiotropic loci regulating heterosis on an epistatic level. For example, seven epistatic interactions (*eqOIL.13-16/14-26*, *eqPN.13-16/16-28*, *eqSN.11-42/13-16*, *eqSP.13-16/19-12*, *eqSP.11-14/13-16***,** *eqSY.13-16/19-21*, and *eqSY.13-16/19-20*,) shared the common chromosome interval 13-16 indicating existence of a hotspot. ## Mode-of-inheritance of QTL and epistatic interactions Four kinds of QTL mode-of-inheritance (A; PD; D; OD) and three kinds of epistatic interactions mode-of-inheritance (AA; AD/DA; DD) were found for the 15 yield-correlated traits, which accounted for 24.6%, 49.0%, 13.8%,12.6%, and 63.0%, 26.0%, 11.0% respectively. For the same trait, the QTL and epistatic interactions showed an unequal distribution among different mode-of-inheritance categories. For the same mode-of-inheritance category of QTL or epistatic interactions, unequal distribution was also observed among different traits, which suggested that the genetic mechanism underlying the heterosis of different traits might be different. Seed yield and seed number per plant clearly showed the highest proportion of +D/+OD mode-of-inheritance, which accorded well with the highest mid-parent heterosis of both traits. The dominant-effect direction of 41.8% QTL, 54.0% (48 out of 89, 48 from negative and 41 from positive) AD/DA and 48.7% (19 out of 39, 19 from negative and 20 from positive) DD epistatic- effect was negative, which was consistent with the low correlation between marker heterozygosity and mid-parent heterosis/hybrid performance. To test whether the mode-of-inheritance of identified QTL and/or epistatic interactions was associated with the significance of heterosis, a *t* test was used for each mode-of-inheritance category between the nine traits with heterosis and the other six traits without heterosis and no significant differences were found. However, between the 15 yield-correlated traits and 9 seed-quality/metabolic traits (glucosinolates, erucic acid, linolenic acid, linoleic acid, palmitic acid, oleic acid, stearic acid, α-tocopherol and γ-tocopherol contents in seeds, which were not significantly correlated with seed yield and unpublished in the current research), significant and extremely significant differences were found for +D and +OD mode-of-inheritance, respectively. In addition, for the nine traits with significant heterosis, the direction of OD effect was more frequently found to be positive than to be negative. ## Phenotypic effect of QTL and epistatic interactions To test the effect of identified QTL and epistatic interactions on the trait performance of the reconstructed F₂ population for 15 yield- correlated traits, the performance of all kinds of genotypes was calculated (using the marker that was closest to the peak position of the identified QTL and epistatic interactions), compared and sorted. For the single-locus analysis, a homozygote was frequently the best and also the worst genotype, while a heterozygote was the most unlikely best and also worst genotype. For the two- locus analysis, a complementary homozygote (two loci were homologous for Tapidor and Ningyou7 respectively) was frequently the best genotype, followed by a parental homozygote (two loci were homologous for Tapidor or Ningyou7 respectively), a single heterozygote and a double heterozygote for almost all traits. For example in the case of seed yield, it was deduced that, in order to get the best genotype only 39.1% and 8.8% loci of identified QTL and epistatic interactions (21.1% for all loci involved) respectively, should be heterozygous. This accorded well with the previous finding that the seed yield of many lines in the reconstructed F₂ population was higher than that of the F₁ hybrid. # Discussion ## Reconstructed F₂ population is very suitable for heterosis study The reconstructed F₂ population used here holds several unique characteristics for dissecting the genetic architecture of heterosis. Firstly, it is well known that the F₂ population was theoretically the most complete and informative source for most genetic analysis. The genotype of the reconstructed F₂ population was basically the same to that of the F₂ population because the genotype of double haploid lines used in making the reconstructed F₂ population was essentially the same as that of the gamete produced by the F₁ hybrid (except for the possibility that genotypic selections existed in the process of microspore culture). In this sense, the reconstructed F₂ population is more similar to the F₂ population than the immortalized F₂ population produced by the random intercross of recombinant inbred lines. Secondly, each genotype of the reconstructed F₂ population was represented by many individuals and thus permitted replicated experiments in multiple environments, so the reconstructed F₂ population was better than the F₂ and F_2:3 populations. This also increased the power (or decreased experimental error) and reproducibility of QTL detection, and especially facilitated the analysis of environmental response of QTL in natural environments. Thirdly, additive, dominant and all kinds of epistatic effects (including AA, AD/DA and DD) can be well estimated in one population, thus increasing the accuracy of the estimation of dominant degree, mode-of- inheritance and especially the relative importance of all kinds of genetic effects in the expression of heterosis. Therefore, for heterosis study reconstructed F₂ population is also better than BC, TC, NCIII and TTC populations in this sense. However, it should be noted that among all of the available experimental designs, TTC population has the unique potential to identify QTL that is directly linked to heterosis. ## Level of heterosis across traits and species In all environments, seed yield showed the strongest heterosis among the 15 yield-correlated traits, consistent with the findings in other rapeseed research, as well as in other crops and plants, such as rice, , maize, *Arabidopsis*, and tomato. This confirmed the hypothesis that complex traits usually express higher heterosis than component traits. Interestingly, the theoretical mid-parent heterosis of seed yield was calculated as: (1 + 18.4%) × (1 + 10.1%) × (1 + 2%) - 1 = 30.6%, a value which was very clear to the true value (31.4%) of mid-parent heterosis of seed yield (18.4%, 10.1% and 2% was the mid-parent heterosis mean in the reconstructed F₂ population, respectively, for the three yield component traits). In addition, the yield heterosis of the tomato +/*sft* heterozygote could be traced back to component traits, number of flowers per plant and fruit weight. This suggested that the heterosis of complex trait (such as yield) can be well explained by that of the component traits, because the middle and/or weak heterosis of the component traits may result in high heterosis of the complex traits in a multiplicative manner, Generally, the level of mid-parent heterosis for similar traits in the current research as well as other research in rapeseed, rice, wheat, *Arabidopsis*, and tomato were all much lower than that of the corresponding traits in maize. This may be attributable to differences in reproductive biology. Maize is an allogamous species and was supposed to have more deleterious alleles than autogamous species (because in autogamous species, deleterious alleles are possibly eliminated by natural and artificial selection since the individuals are homozygous), so the extent of inbreeding depression in maize was greater than that in rice, wheat, tomato and *Arabidopsis*, the autogamous species, and rapeseed, a partially allogamous crop. ## Mode-of-inheritance of QTL and epistatic interactions No significant difference was found for the proportion of the eight model-of- inheritance categories of QTL between the nine traits with heterosis and the other six traits without heterosis. This suggested that the presence or absence of heterosis was not associated with QTL mode-of-inheritance in the current research, which may be because the dominant effect only accounted for a small proportion of variance when compared with the epistatic effect of these traits. However, between the 15 yield-correlated traits and the 9 seed-quality/metabolic traits, significant and extremely significant differences were found for +D and +OD mode-of-inheritance. This indicated +OD/+D mode-of-inheritance was associated with the traits of yield category, which may be because the occurrence of +OD/+D QTL for yield-correlated traits will increase crop productivity during the processes of domestication. Thus, OD may be an essentially pseudo-OD that involves linked loci with dominant alleles in repulsion. We detected A, PD and D QTL for both yield-correlated and seed- quality/metabolic traits, but OD was basically absent in seed-quality/metabolic traits. This indicated that pseudo-OD due to random linkage is unlikely to be the major genetic basis underlying OD QTL, and thus we favored the true OD model. In fact, +OD/+D QTL was prevalent in almost all research regarding the genetic basis of yield, life-history and reproductive traits in crops. In a tomato introgression line population, +OD QTL was more prevalent for the reproductive traits than nonproductive traits. In a summary research, the dominance effect was found to be larger in life-history traits than in morphological traits. Although only a few studies reported the QTL mapping of metabolic traits, the results all showed that only a few metabolic-QTL showed OD mode-of-inheritance. This suggested that different phenotypic classes may have different dominance relationships among variable alleles, possibly due to differences in the complexity underlying the molecular networks. More importantly, the sign of dominant-effect of OD QTL for the nine traits with heterosis was more frequently found to be positive than to be negative, which suggested that selection also has changed the frequency of the direction of OD effect for these traits of heterosis. This is understandable, since a positive OD effect may undoubtedly increase the heterosis and yield of hybrids. However, no mode-of-inheritance categories and their direction of epistatic interactions showed significant difference in proportion among different phenotypic categories. In fact, this phenomenon seemed to be typical in other crops. In a two-year experiment conducted in an “immortalized F₂” population of an elite rice hybrid known as Shanyou63, the proportions of three kinds of epistatic interactions (AA, AD/DA and DD) were almost the same between reproductive (grain yield, tillers per plant, grains per panicle *etc.*) and non-reproductive (heading date, plant height and panicle length *etc.*) traits. In a two-location experiment conducted in an F_2:3 population in maize, no significant difference was also found in the proportion of three kinds of epistatic interactions between yield traits (such as grain yield, rows number, kernels per row *etc.*) and morphological (ear length, ear diameter and axis diameter *etc.*) traits. This suggested that selection was not effectual at epistatic level during the domestication of rapeseed, as well as other crops. This was understandable: since epistatic interactions were more dependent on the genetic background and environmental variations than QTL, their role was variable, and thus capturing the best gene combination(s) was difficult for breeders. It should be noted that the relative proportion of the four kinds of mode-of- inheritance of QTL showed great differences in different traits and studies. For example, in the same QTL mapping experiment of nine yield traits, the predominant mode-of-inheritance of QTL was over-dominant and additive, in an intraspecific and intrasubspecific rice hybrid. However, in all research in which the three kinds of epistatic effects could be resolved, AA interaction occurred at the highest frequency for all traits, followed by AD/DA and the DD interaction at the middle and lowest frequency, respectively. This confirmed that selection has great but little or no impact on mode-of-inheritance of QTL and epistatic interactions, respectively. In addition, in all cases the practical proportions (usually \>50%, \<40% and \<8%) of AA, AD/DA and DD interactions were all quite different with their theoretical proportions of 25%, 50% and 25%, respectively. This provided the evidence that the identified epistatic interactions were absolutely not the results of chance events. ## Environmental response of QTL and epistatic interactions The meta-analysis of QTL identified in different environments facilitated the exact estimation of the environmental response of QTL. Totally, 74.3% (223) of the consensus QTL and 99.3% (270) of the epistatic interactions for the 15 yield-correlated traits was specifically identified in one of the three environments, which indicated the great impact of natural environments on the genes underlying the heterosis of these yield-correlated traits. These proportions were much higher than the corresponding ones (48.4% and 91.6%) of the other 9 seed-quality/metabolic traits (unpublished data), which accorded well with the broad-sense heritability of these traits. In fact, the high dependency on environment seemed to be a common character of the QTL and epistatic interactions for heterosis in other research. In a two year experiment conducted in an F_2:3 population derived from an elite rice hybrid (Shanyou63), 62.5% QTL and 90.6% digenic interactions for grain yield and the three yield component traits were observed in only one year. In another two year experiment conducted in an “immortalized F₂” population derived from the same rice hybrid, 67.5% QTL and 91.5% digenic interactions for the same four yield traits were detected in only one year. In a two-location experiment conducted in an F_2:3 population derive from an elite maize hybrid, 62.1% QTL and 91.8% of digenic interactions for grain yield and the three yield component traits were detected in only one location. It should be noted that the proportion of environment-specific epistatic interactions was much higher than that of QTL in all cases, which was understandable since the epistatic interactions involved two genetic loci which were also dependent on environmental conditions. It should also be noted that the proportions of environment-specific QTL and epistatic interactions in the current research as well as other rapeseed research were all higher than that in rice, and maize, possibly due to the genome plasticity of polyploids,. This indicated the high variability and plasticity of the genetic architecture of heterosis in rapeseed. Furthermore, of the 77 repeatable consensus QTL for 15 yield-correlated traits, 68.8% changed their mode-of-inheritance in different environments. This proportion was also much higher than that (46.9%) of the 9 seed- quality/metabolic traits (data not shown). This indicated that the relative importance of dominant vs additive effect of QTL of different phenotypic categories may have different sensitivity to environmental variations, possibly due to differences in the intrinsic mechanism of regulation. Interestingly, the additive-effect direction of the repeatable consensus QTL was usually the same in different environments, which was consistent with previous research. This has great significance for genetics and crop breeding: since the relatively favorable alleles identified in one environment were usually relatively favorable in another environment, the actual effect of selection might be well ensured. From an evolutionary point of view, these retained alleles all experienced the processes of far-flung natural or artificial selection, and alleles that were adaptable to changed environments could be successfully retained. Whereas, 24 of the 77 repeatable consensus QTL changed their dominant- effect direction in different environments, this proportion (31.2%) was much higher than that (5.2%) of the additive-effect direction. Furthermore, for the other 53 repeatable consensus QTL with a consistent dominant-effect direction, 54.7% changed their mode-of-inheritance in different environments. For example, the mode-of-inheritance of *qSY.A1-5* was changed from +PD in N6 environment to +OD in an S5 environment. This indicated that the favorable heterozygote identified in one environment was not always favorable in another environment. ## Genetic architecture of heterosis in rapeseed and other species Using a reconstructed F₂ population (that has the maximum similarity to an F₂ population), a multiple-environment experiment and a high- density linkage map, we identified hundreds of QTL and epistatic interactions responsible for the heterosis of 15 yield-correlated traits. Surprisingly, 92.8% identified QTL and 95.6% epistatic interactions explained \<10% of variance. This indicated that heterosis of yield-correlated traits in this cross was mainly controlled by numerous loci with very little effect. In addition, the maximum variances explained by individual QTL and epistatic interactions were 20.8% and 18.3% respectively. Therefore, the development of QTL and epistatic interactions near-isogenic lines, toward fine-mapping and finally cloning the genes responsible for heterosis in this cross would be very challenging. In contrast with the high variability of QTL and epistatic interactions, their most important feature was the high proportion (73.3% and 68.2% respectively) that co-localized at the genomic level. This accorded well with the comprehensive correlation of the mid-parent heterosis/hybrid performance among these yield-correlated traits. These co-localizations indicated the existence of pleiotropic loci regulating heterosis. In fact, most published fine-mapped QTL or genes identified for yield heterosis exhibit pleiotropic effects on at least one or multiple yield-correlated traits. Fifteen of the 21 consensus QTL of seed yield co-localized with other consensus QTL and 7 of them co-localized with more than two consensus QTL. This indicated that, in addition to pleiotropy, the effect of the QTL for seed yield could be a synthetic effect of several tightly- linked QTL of different yield-correlated traits. The multiple co-localized QTL might come from the different environments, which indicated that the environmental conditions contribute to the variability and plasticity of the QTL for seed yield. It should be noted that more than half of the loci of the QTL and epistatic interactions were clustered in several chromosomes. Research from autogamous species, such as *Arabidopsis*, rice, and barley, usually showed that epistasis played a more important role than main-effect as the genetic basis of heterosis. In contrast, data from allogamous crops, such as maize, exhibited the reverse result, demonstrating that main-effect is more important than epistasis. This is not surprising, since co-adapted gene complexes exhibiting favorable epistatic effects can be more easily maintained in autogamous species than in allogamous species. Therefore, it is reasonable for our result to show that epistasis was somewhat more important than main- effect as the genetic basis of heterosis in rapeseed (a partially allogamous crop with an out-crossing rate of 10-30%), and is consistent with other research in rapeseed. According to the theory of classical genetics, only D, AA and DD effect are the genetic components of mid-parent heterosis. Furthermore, the average \|D\| was smaller than the average \|A\|, and their ratios ranged from 0.40 (for branch number) to 0.73 (for seed yield) and with a mean of 0.51. This suggested that dominant effect only accounted for a minor proportion of *R*² of QTL, whereas, AA and DD effects explained a major proportion (67.1%) of *R*² of epistatic interactions. In conclusion, our research showed that epistasis (especially AA epistasis) was the major genetic basis of heterosis in rapeseed (*Brassica napus* L.). ## Implications for evolution and crop breeding The two parents used in this study, Tapidor and Ningyou7, are the representative of two highly diverse gene pools, the European winter-type rapeseed gene pool and the Chinese semi-winter type rapeseed gene pool, both adaptable to their corresponding agro-ecological areas. The proportion of positive (54.9%) and negative (45.1%) additive-effect was basically equal, which indicates that both gene pools harboured alleles adaptable to other agro-ecological areas. One hundred and three epistatic interactions showed significant positive AA interactions, which indicated co-adapted gene complexes retained during the evolution of rapeseed, a phenomenon also found in other species. Oilseed rape (AACC, 2n = 38) originated from the natural hybridization of *Brassica rapa* (AA, 2n = 20) and *Brassica oleracea* (CC, 2n = 18) and the following chromosome doubling, both of which also experienced an evolutionary process of triploidization. Therefore, each gene has an average of 6 copies in rapeseed. If these duplicated genes favorably interacted with each other, this would result in ectopic heterozygosis and the fixed heterosis in inbred lines **.** In fact, many epistatic interactions identified in reconstructed F₂ and DH populations occurred between homologous intervals/blocks (data not shown), which indicated the existence of fixed heterosis loci in rapeseed. Since a high- density linkage map together with detailed chromosome block information was available, it was possible to study the hypothesis of fixed heterosis and demonstrate its advantage in the evolution of polyploids using two-segment near- isogenic lines chosen from the backcross progenies in our laboratory. One hundred and twenty epistatic interactions of the 15 traits showed significant and negative AA interactions, which indicated the complementary homozygote of these epistatic interactions tended to enhance fitness. This also suggested that complementary loci played an important role in the maintenance of genetic variation in the rapeseed population. Therefore, reserving the adapted genes and co-adapted gene complexes (including fixed heterosis loci) in per se gene pool while further pyramiding the favourable genes and gene combinations (including fixed heterosis loci) in another gene pool may be an effective strategy to further improve rapeseed conventional cultivars in both agro-ecological areas. Consistent with the findings in other research in rapeseed as well as other species, a considerable proportion of dominant effect (41.8%) and DD interactive effect (48.7%) was negative, which indicated the general existence of hybrid weakness genes across species. This suggested that heterozygote was not always advantageous for the hybrid performance and mid-parent heterosis in rapeseed. This conclusion was also confirmed by the comparison of phenotypic effects of all kinds of genotypes both at the single and two locus level. Therefore, the knockout or substitution of hybrid weakness genes represents a new avenue to further improve hybrid cultivars. It should also be noted that 58.2% of dominant effect and 51.3% of DD interactive effect was positive, which indicated heterozygosis played an important role in the fitness of natural populations by providing a heterozygous advantage to buffer against recessive alleles and providing genetic plasticity to variable environmental conditions. Although homozygotes of the detected QTL and epistatic interactions were usually the best genotypes in rapeseed as well as in rice and maize, the proportion still needs to be well demonstrated. The most striking finding in this research is that to be the best hybrid, most heterozygous loci (83.2% in this experiment) of all QTL and epistatic interactions in hybrid F₁ should be homozygous, which accorded well with the results that only 19.2% of QTL and 17.4% of epistatic interactions showed positive OD/D and DD/AD(DA) mode-of- inheritance respectively. This suggested that, in most cases, homozygotes were more advantageous for trait performance than heterozygotes. At first view, this conclusion seemed unbelievable, a truth usually neglected, is that, heterosis (usually defined as mid-parent heterosis) and hybrid performance are related but essentially two different concepts, because the latter is more complex and equal to the former plus the parental mean. The cryptic meaning is that a hybrid showing the strongest mid-parent heterosis for a given trait did not always exhibit the best per se manifestation of the same trait. Similarly, a heterozygote may enhance mid-parent heterosis value but decrease per se hybrid performance. Therefore, our conclusion is not intricate, and this has great significance for genetics and crop breeding. Because heterosis usually coincides with the genetic distance between parents, to maximize heterosis, breeders usually adopted parents with greater genetic distance, and as a result, the unadapted germplasm was also adopted in the hybrid breeding scheme. Therefore, the final result is that the breeders get the combinations of max heterosis but not the best hybrids. To avoid the occurrence of this embarrassing situation, we suggest an adapted germplasm with relatively large genetic distance would be a better choice in a hybrid breeding scheme. In addition, our result also suggested the utilization of the residual heterosis of inbred and backcross progenies (such as F₂, F₃ and BC_x *etc*) in rapeseed as well as other partially-allogamous and autogamous crops would be feasible, because the over-F₁ phenomenon for yield and/or biomass was usually found in the subsequent inbred and backcross progenies even for elite hybrids. This research revealed that epistasis played an important role in the genetic architecture of trait performance and heterosis in autogamous and partially- allogamous crops. The research also showed that epistasis is very sensitive to environment, and the epistatic effect varied from one environment to another, thus artificial selection seemed to have little or no effect on it, though it has proved to be effectual at the single-locus level (illustrated by the association between +OD/+D QTL and the traits of yield category, and between positive signs of OD effects and traits with heterosis). This suggested that while challenging, marker-assisted selection to significantly improve the heterosis/hybrid performance of yield traits in the aforementioned crops has great potential. # Materials and Methods ## Design and development of a reconstructed F₂ population A double haploid (DH) population of 202 lines was developed by microspore culture from the F₁ cross between Tapidor (an European winter-type rapeseed cultivar) and Ningyou7 (a Chinese semi-winter type rapeseed cultivar) and named as TNDH. A reconstructed F₂ population was made by making 101 crosses per round between pairs of DH lines randomly chosen from the 202 lines of the TNDH population. In the spring of 2004 and 2005, three and four rounds of crossing were made by hand emasculation and hand pollination, resulting in 303 and 404 crosses respectively. ## Field experiments and trait measurements The two populations (TNDH and reconstructed F₂), two parents (Tapidor and Ningyou7) and F₁ (Tapidor × Ningyou7) were grown in 3 different environments (year-location combinations) in China. The field planting followed a randomized complete block design with three replications. Each plot was 3.0 m² with 30 plants in N6 and S6 environments and 4.0 m² with 40 plants in S5 environments, with a distance of 40 cm between rows and 25 cm between individuals. The seeds were hand sown and the field management followed standard agricultural practice. Twelve representative individuals from the middle of each row in each plot were hand harvested from ground level at maturity. A total of 15 traits were investigated: (1) seed yield (SY, kg/ha), (2) biomass yield (BY, kg/ha), (3) pod number per plant (PN); (4) seed number per pod (SN); (5) seed weight/1000 seeds (SW, g); (6) flowering time (FT, days); (7) maturity time (MT, days); (8) plant height (PH, cm); (9) branch number (BN); (10) development time of seeds (DT, days), calculated from maturity time and flowering time by the formula, DT  =  MT - FT; (11) seed number per plant (SP), calculated from SY and SW by the formula, SP  = 10 × SY (kg/ha)/SW (g/1000); (12) pod yield/100 pods (PY), calculated from SN and SW by the formula, PW  =  SN × SW/10; (13) harvest index (HI), calculated from BY and SY by the formula, HI  =  SY/(SY + BY)); (14) protein content in seeds (PRO), (15) oil content in seeds (OIL). Seed yield per plant was measured as the average dry weight of seeds of the harvested individuals in a plot. Biomass yield per plant was measured as the average total above-ground (except the seeds) dry weight of the harvested individuals in a plot. Pod number was the number of well-filled, normally developed pods on each harvested individual in a plot. Seed number per pod was the average number of well-filled seeds from 100 well-developed pods, sampled from the primary branch in the middle of the harvested individuals in a plot. Seed weight was the average dry weight of 1000 well-filled seeds from three replicate samples, taken from the mixed seeds of the harvested individuals in a plot. Flowering time was measured as the interval between the date of sowing and the date when the first flowers emerged on 50% of the plants in a plot. Maturity time was measured as the interval between the date of sowing and the date when pods on most of the plants in a plot were yellow. Plant height was the height of each harvested individual in a plot, measured from the base of the stem to the tip of the main shoot. Branch number was the number of branches arising from the main shoot of each harvested individual in a plot. The oil and protein content of seeds was measured by Near Infrared Spectroscopy (NIR) using standard methods. ## Statistical analysis Year-location combinations were treated as independent environments. Environment was treated as a fixed effect while genotype (DH or reconstructed F₂ lines) was treated as a random effect. The broad-sense heritability was calculated as: *h*²  =  *σ*²_g/(*σ*²_g + *σ*²_ge/n + *σ*²_e/nr). Where, *σ*²_g is the genetic variance, *σ*²_ge is the interaction variance of genotype with environment, *σ*²_e is the error variance, n is the number of environments and r is the number of replications. The genetic correlation was calculated as: *r*_G  = *cov*_Gxy/(*σ*²_Gx × *σ*²_Gy)^1/2, where, *cov*_Gxy, *σ*²_Gx and *σ*²_Gy were the genetic covariance and variance of the pair-wise traits respectively. The significance of each genetic correlation was determined using a *t* test of the correlation coefficients. The estimation of variance and covariance components were obtained using an SAS GLM procedure. The mean value for three replications in each environment for both populations was used in subsequent QTL analysis for all traits. General heterozygosity was calculated as N_H/(N_T+N_N+N_H). N_T, N_N and N_H were the number of markers with genotypes of Tapidor, Ningyou7 and both parents, respectively. Special heterozygosity was calculated using the same formula but the statistics were restricted to the marker that was significantly associated with phenotype (data not shown). ## Genetic linkage map A total of 786 markers were mapped to the new linkage map generated with the TNDH population using JoinMap 3.0 (<http://www.kyazma.nl/index.php/mc.JoinMap>). This covered 19 chromosomes identified as A1–A10 and C1–C9, with an average distance of 2.7 cM between markers. The threshold for goodness of fit was set to ≤5.0 with logarithm of the odds ratio (LOD) scores 1.0 and a recombination frequency\<0.4. The order of the markers on the linkage map agreed well with our published maps. The genotype of each RC-F2 line was deduced from the corresponding genotype of their parents. ## Genome-wide detection of QTL, meta-analysis and test the result of QTL meta-analysis QTL were detected by composite interval mapping using WinQTL cartographer 2.5 software (<http://statgen.ncsu.edu/qtlcart/WQTLCart.htm>). The number of control markers, window size and walking speed were set to 5, 10 cM and 1 cM respectively. The default genetic distance (5 cM) was used to define a QTL in a specific experiment. The threshold of experiment wise error rate was determined by permutation analysis with 1000 repetitions. LOD values corresponding to P = 0.05 were used for identifying “significant” QTL. To avoid missing QTL with very small effects, a lower LOD value corresponding to P≤0.50 was adopted for the presence of “suggestive” QTL. The overlapping “suggestive” QTL and all the “significant” QTL were admitted and named as “identified-QTL”. The dominant degree of an identified-QTL was defined as d/\|a\|. For mode-of- inheritance of identified-QTL the QTL was defined as additive (\|d/a\|\<0.2), partially-dominant (0.2≤\|d/a\|\<0.8), dominant (0.8≤\|d/a\|\<1.2) and over- dominant (\|d/a\|≥1.2). Since QTL of the same traits or related ones detected in different experiments and mapped to the same region of a chromosome, might in fact be several estimations of the position of one single QTL, algorithms for QTL meta-analysis were used to estimate the number and positions of the meta-QTL underlying the analyzed QTL. This approach, using the *Akaike* information criterion (AIC), provided the basis on which to determine the number of meta-QTL that best fitted the results on a given linkage group. It also grouped the QTL detected in the different experiments into classes that correspond to the same QTL and provided a consensus estimation of QTL positions. Computations were conducted using the *BioMercator2.1* software. At present, the method used in this software cannot distinguish between models with more than four meta-QTL on the same linkage group. If the estimated number of meta-QTL is more than four, *Biomercator2.1* declares the most probable model as one with a number of meta-QTL equal to the number of the analyzed QTL. Then the *Delete* function of the software was used to select specific segments of a linkage group separated by regions with no QTL and separately apply QTL meta-analysis to these segments. The software also provides a method to calculate 95% confidence intervals for the meta-QTL:Where, *S_i²* is the variance of position of the QTL_i and k is the total number of QTL integrated into the meta-QTL. A two-round strategy of QTL meta-analysis was adopted. The QTL identified in different experiments were first integrated into consensus QTL, trait by trait. In the second round of QTL meta-analysis, the consensus QTL for the different traits was integrated into unique QTL. To test the result of QTL meta-analysis, ANOVA implemented in SAS/Stat version 8e was utilized to identify QTL × environment interaction by GLM (generalized linear model) model: P  =  G + E + G × E. Where, P, G, E and G × E represent the phenotype and the effects of genotype, environment and genotype by environment interaction, respectively. The genotype of each consensus QTL was estimated by that of the molecular marker closest to it's peak position. The significant threshold was set as p≤0.05. ## Genome-wide detection of epistatic interactions The maximum-likelihood estimation method in QTLmapper V2.0 software (<http://www.cab.zju.edu.cn/ics/faculty/zhujun.htm>) was employed to detect the epistatic interactions. It was based on mixed linear model and performs composite interval mapping. The walking speed was set to 1 cM. The LR value corresponding to P = 0.005 was used as the threshold for claiming the presence of putative epistatic interactions. The significance of the epistatic effect was further tested by running the submenu of the Bayesian test (using P≤0.005). # Supporting Information The authors thank Mr. Dianrong Li and Mr. Hao Wang (Hybrid Rapeseed Centre of Shaanxi, Dali 715105, China) for the field work and collecting of the phenotypic data. [^1]: Conceived and designed the experiments: JS JM. Performed the experiments: JS. Analyzed the data: JS RL JM. Contributed reagents/materials/analysis tools: JS JM RL YL. Wrote the paper: JS RL JM JZ. [^2]: The authors have declared that no competing interests exist.

# Introduction Brucellosis is considered one of the most important zoonosis. It is responsible for important economic losses due to abortion and culling of infected animals. In man, it causes a systemic febrile illness with a wide spectrum of symptoms, although arthritis is among the most frequent manifestations. In cattle the main causative agent of brucellosis is *Brucella abortus* and during pregnancy it can infect and multiply intensely within trophoblastic cells of the placenta at late gestation causing necrotizing placentitis associated with abundant neutrophilic infiltrate. The capacity of *B*. *abortus* to cause disease is related to its ability to invade host cells, survive intracellularly, and evade antimicrobial defenses of the host. However, the mechanisms related to this ability are not yet fully understood. It is known, however, that once *Brucella* spp. reaches the intracellular environment, the pathogen actively interferes with the host cell metabolism and defense favoring its survival and intracellular multiplication. *Brucella* spp. modulate intracellular trafficking by preventing maturation of phagosomes and blocking endosome-lysosome fusion, which prevents the degradation of bacteria. Although *Brucella* spp. do not have classical virulence factors, these organisms have several known mechanisms associated with pathogenicity. One of these mechanisms is associated with LPS, which differs from other Gram-negative bacteria. *Brucella* spp. LPS is a poor inducer of oxidative burst, reactive nitrogen intermediates, and secretion of lysozyme. Some studies have also shown that *Brucella* spp. LPS reduces TLR4 (Toll-like receptor 4) agonistic activity and that despite being recognized by the receptor, this interaction does not induce cytokine production. The *virB* operon-encoded type IV secretion system (T4SS) is a key virulence factor for *Brucella* spp. This T4SS secretes effector proteins through the envelope of the bacterial cell, and it is required for intracellular survival and *in vivo* persistence of *Brucella* and this expression is induced by acidification of the phagosome after phagocytosis. Effector proteins secreted through the T4SS modulates maturation of the *Brucella*-containing vacuole (BCV), preventing fusion of the BCV with lysosomes at early stages of infection, and driving the interaction between BCV and endoplasmic reticulum (ER) and subsequent maturation of BCV at later stages of infection. Two *Brucella* proteins containing a TIR (Toll intracellular domain/interleukin-1) domain, namely BtpA and BtpB, have been identified in *Brucella* spp. BtpA and BtpB modulate the host inflammatory responses during *Brucella* sp. infection by interfering with TLR signaling. Proteomic analyzes in the context of the interaction between *B*. *abortus* and its target cells are also scarce. These studies are very challenging due to the high complexity of samples and very low concentrations of certain proteins, requiring the use of highly sensitive analytical techniques. All proteomic studies reported to date have used the model of infection of phagocytic cells, whereas the profile of protein expression by trophoblastic cells infected with *B*. *abortus* have not been previously studied. In this study, *ex vivo* culture of trophoblastic cells in CAM explants was associated with proteomic analysis to study the interaction between *B*. *abortus* and trophoblastic cells. # Materials and Methods ## Bacterial strain and growth conditions The inoculum was prepared from cultures of *B*. *abortus* 2308 grown in 20 mL of tryptic soy broth (Difco, USA) for 12–15 h at 37°C under agitation (200 rpm). After incubation, optical density of bacterial suspensions was determined by spectrophotometry (OD₆₀₀) and adjusted to 1.0 x 10⁸ bacteria/mL. Number of bacterial cells was confirmed by serially diluting in PBS (pH 7.4), and plating 100 μL of each dilution on tryptic soy agar (Difco) in duplicate. After 48 h of incubation at 37°C with 5% CO₂, colonies were counted and the number of colony forming units (CFU) was obtained by averaging the duplicates. CFU numbers were determined by the drop count method. Manipulation of *B*. *abortus* was performed under biosafety level 3 conditions. ## Infection of chorioallantoic membrane (CAM) explants with *Brucella abortus* 2308 CAM explants were obtained from three intact bovine uteruses at the final third of gestation collected at a local slaughterhouse (Frigorífico Uberaba Ltda, Sabará, Minas Gerais), as previously described. Gestational age was estimated by measuring the cephalococcygeal length (Crown-rump length). Only placentas from *Brucella*-free fetuses, based on rose Bengal plate agglutination test using amniotic fluid, were included in these experiments. This study was approved by the Committee for Ethical use of Experimental Animals of the Universidade Federal de Minas Gerais, Brazil (CETEA), under protocol 183/2010. Chorioallantoic membranes were aseptically removed from the uterus and immediately placed into RPMI 1640 (Roswell Park Memorial Institute) medium (Invitrogen, USA) with 50 U/mL of penicillin and 50 g/mL of streptomycin (Invitrogen, USA) for 20 min and were then washed two times with RPMI 1640 to remove antibiotic residues. Explants were prepared using snapwell inserts (Snapwell^™ Inserts—Corning, USA) and placed into six-well cell culture plates (Corning) with supplemented medium (RPMI 1640 with 4 mM glutamine, 1 mM pyruvate, 1 mM non-essential amino acids, 2.9 mM sodium bicarbonate, 15% fetal bovine serum) in contact with the trophoblastic and allantoic or amniotic surfaces. The central area of each explant was inoculated with 200 μL of culture medium (supplemented RPMI 1640) containing 2.0 x 10⁷ CFU, which correspond to a multiplicity of infection (MOI) of approximately 1000. The explants were centrifuged for 15 min at 1000 xg and maintained at 37°C in 5% CO₂ for 30 min to allow internalization of bacteria. Extracellular bacteria were eliminated by adding 200 μL of medium with gentamicin (50 mg/mL) (Invitrogen), followed by incubation for 1 h at 37°C under 5% CO₂. After incubation, medium containing gentamicin was removed, the explants were gently washed twice with PBS (phosphate buffered saline—pH 7.4) and then, 200 μL of supplemented RPMI 1640 medium was added in each explant. In this assay, each group was evaluated in triplicate at 0.5, 2, 4, and 8 h after the removal of the gentamicin supplemented RPMI 1640. Explants in the control group were inoculated with sterile supplemented RPMI 1640 medium and submitted to the same conditions. ## Internalization of *Brucella abortus* by trophoblastic cells of CAM explants Three explants from each group (uninfected and infected) obtained from the chorioallantoic membrane of three fetuses were used to determine the number of internalized bacteria. After removal of the RPMI culture medium with gentamycin, the explants were washed twice in PBS pH 7.4 and then lysed with 200 μL of sterile 0.1% Triton X-100 (Roche, Germany). Lysates were serially diluted in PBS pH 7.4 and 100 μL of each dilution were plated on Tryptic Soy Agar (Difco, USA) in duplicate. After 48 h of incubation at 37°C in 5% CO₂, colony counts were performed and the number of CFU was obtained from the average of triplicates. CFU numbers underwent logarithmic transformation followed by analysis of variance (ANOVA). Comparison of means was performed using the Tukey's multiple comparison test with significance level of P≤0.05. ## Extraction of proteins from CAM explants At 0.5, 2, 4, and 8 h after the removal of the medium containing gentamicin, CAM explants were washed three times with RPMI 1640 medium without bicarbonate and without fetal bovine serum. Protein extraction was performed using 200 μL of lysis buffer (8 M urea, 2 M thiourea, 4% w/v CHAPS, 40 mM Tris 1M) containing protease inhibitors (GE Healthcare, UK), which was added to the trophoblastic surface of CAM explants. The explants were maintained under agitation (50 rpm) for 30 min on an orbital shaker (IKA Labortechnik, Germany). Lysates were centrifuged for 30 min at 20.000 xg at room temperature. The supernatant was collected and kept at -80°C. Protein concentration was measured using the 2D Quant Kit (GE Healthcare, UK). ## Two-dimensional gel electrophoresis (2DE) In order to evaluate reproducibility and to determine the time points to be evaluated by DIGE, the proteomic profile was assessed initially using 7 cm gels, pH 4–7. Extracted proteins (50 μg) from CAM explants were added to the rehydration solution (1.25 μL IPG buffer, pH 4–7, 10 μL/L—GE Healthcare, UK) and IEF buffer (8 M urea, 2 M thiourea, 4% CHAPS, 0.0025% bromophenol blue, 10 mg/mL dithiothreitol—GE Healthcare) in a total volume of 125 μL per IPG strip (7 cm, pH 4–7) (GE Healthcare). Samples were incubated with IPG strips on rehydration apparatus (Immobiline DryStrip Reswelling Tray, GE Healthcare, UK) for 12 h and subjected to isoelectric focusing using the Ettan^™ IPGphor^™ 3 Isoelectric Focusing System (GE Healthcare, UK) and the program IPGphor (GE Healthcare, UK). Focused IPG strips were incubated for 15 min in equilibration solution (50 mM Tris-HCl, pH 8.8, 6 M urea, 30% glycerol, 2% SDS, 0.002% bromophenol blue, and 125 mM DTT) and then alkylated for further 15 min in an equilibration solution containing 13.5 mM iodocetamide (GE Healthcare, UK) instead of DTT. Electrophoresis was performed in 12% polyacrylamide gel containing SDS in a vertical electrophoresis apparatus (Ettan DALTsix electrophoresis Unit -GE Healthcare, UK) at 30 mA per gel. Bench Mark Protein^® Lader (Invitrogen, USA) was used as molecular mass marker. Gels were stained with Colloidal Coomassie Brilliant blue G-250 (GE Healthcare, UK) and digitized using the Image Scanner (Amersham Biosciences, England). ## Differential gel electrophoresis (DIGE) Protein samples extracted from CAM explants were labeled using the CyDye DIGE Fluors (minimal dyes) for Ettan DIGE kit (GE Healthcare, UK) according to the manufacturer. To a pool containing 50 μg of an equal mixture of proteins extracted from infected and uninfected CAM explants produced from three fetuses, 400 pmol of dyes were added. Dye swap was performed and Cy2 dye was used as internal standard. Labeled samples and 800 μg of mixture of unlabeled proteins were incubated with the IPG strip (18 cm, pH 4–7—GE Healthcare) in a rehydration apparatus (Immobiline DryStrip Reswelling Tray, GE Healthcare, UK) for approximately 12 h and subjected to isoelectric focus using Ettan^™ IPGphor^™ 3 Isoelectric Focusing System (GE Healthcare, UK) and the program IPGphor (GE Healthcare, UK). Electrophoresis was performed in 12% polyacrylamide gel containing SDS in vertical electrophoresis apparatus (Ettan DALTsix Electrophoresis Unit—GE Healthcare, UK) and scanned using Typhoon Trio (GE Healthcare, UK) with excitation/emission wave lengths specific for Cy2 (488/520 nm), Cy3 (532/580 nm), and Cy5 (633/670 nm). After scanning, gels were stained with Coomassie Brilliant blue G-250 (Thermo Scientific, USA) for marking the spots of interest. ## Image analysis of the gels The 2D Image Master Platinum^™ software (version 6.0, GE Healthcare, UK) was used for image analysis of the two-dimensional gels (2DE) stained with Coomassie Brilliant blue G-250 by a combination of automatic detection and manual detection of spots. To determine the relative amount of each spot was used the method of normalization volume. Mean volumes of each spot were calculated by the software, and spots with at least two-fold increase or decrease were considered for further analysis. Means were compared by the Student t test and considered significant when P≤0.05. For the fluorescent gels, DeCyder^™ 2-D Differential Analysis v7.0 software (GE Healthcare, UK) was used according to the manufacturer's instructions. The spots containing proteins labeled with different fluorescent dyes were co-detected and quantified in the three images obtained from each gel. Based on the ratio of the average volume normalized spots that had at least a two-fold increase or decrease was evaluated for statistical significance. Data was subjected to analysis of variance (ANOVA) and means compared using the Student t test with significance level of P≤0.05. Spots with significant changes were selected for identification by mass spectrometry. ## In gel proteolysis and mass spectrometry (MS) Selected spots were cut in pieces of approximately 1.0 mm³ using sterile scalpel. Discoloration of the gel with 400 μL of a solution containing 25 mM ammonium bicarbonate (NH₄CO₃) pH 8.0 (Synth, Brasil) and 50% acetonitrile (Sigma-Aldrich, USA) was repeated three times for 15 min at room temperature under vigorous agitation. Gel fragments were dehydrated with 200 μL of acetonitrile for 5 min and dried under vacuum centrifugation. Proteins were digested using 20 μL of a solution containing 2.0 mM of NH₄CO₃ (Synth, Brasil) and 20 ng/μL trypsin (Promega, USA), followed by incubation for 16–24 h at 37°C. Peptides were extracted with 5% formic acid (Sigma-Aldrich, USA) and 50% acetonitrile. Protein extracts were concentrated (final volume of 10 μL) in ZipTip^® C¹⁸columns (Millipore, USA) and the final volume was reduced to 5.0 μL in a vacuum centrifuge. For sample analysis in the mass spectrometer, 0.3 μL of sample solution was mixed with the same volume of saturated matrix solution \[10 mg/mL R-cyano-4-hydroxycinnamic acid in 50% acetonitrile/0.1% trifluoroacetic acid\] (Sigma-Aldrich, USA). Raw data for identifying proteins were obtained from ABSCIEX Proteomics Analyzer MALDITOF/TOF^™ System 5800 (Applied Biosystems, USA). The external calibration mode MS was performed using a mixture of four peptides: des-arg1-bradykinin (m/z = 904,468), angiotensin I (m/z = 1296.685); Glu1-fibrinopeptide B (m/z = 1570.677) and adrenocorticotropic hormone (18–39) (m/z = 2465.199) (Applied Biosystems, USA). The MS/MS spectra were calibrated externally using known masses of fragmentations observed in the MS/MS angiotensin I. After data acquisition, a list of peaks was obtained from the raw data of the MS/MS using the Mascot to function Peaks 4000 Series Explore software (Applied Biosystems, USA). ## Identification of proteins in the database All information acquired for each spot, i.e. the mass/charge ratio and intensity of peaks, obtained in the spectrum and the reasons mass/charge and intensities of the peaks relating to each of the five spectra MS/MS has been compiled into one text file. This file was used by MASCOT (Matrix Science, USA) (<http://www.matrixscience.com>) to perform the search in databases. Once complete the search, MASCOT reports the results in a bar graph in which there is a value taken as the limit. Scores below this value indicates random events, i.e. no statistical value. On the other hand, if the score assigned to a given protein exceeds the threshold value identifying the chance of an event being generated randomly is 5% and the larger this value the greater the probability of being correct. The parameters used in the search were: no restriction for protein molecular weight, a trypsin cleavage site lost, variable modifications of methionine (oxidation) and cysteine (carbamidomethylation), formation of pyroglutamate at the N-terminal glutamine without other post-translational modifications. The mass tolerance for peptides in searches was 0.8 Da for MS spectra and 0.6 Da for MS/MS spectra. The data bases of *Bos taurus* and *Brucella abortus* available at NCBI (National Center for Biotechnology Information, USA) were used. # Results ## Internalization of *Brucella abortus* 2308 in bovine trophoblastic cells of CAM explants Considering that placentitis and abortion are the most important manifestation of *B*. *abortus* infection in cattle, and that those are key steps for transmission of the disease, in this study we used a previously developed bovine CAM explant model of infection to evaluate differentially expressed host proteins during the early stages of infection of trophoblastic cells with *B*. *abortus*. In order to ensure that the three independent experiments were in fact comparable, the number of bacteria located within trophoblastic cells, as previously demonstrated, was determined after inoculation of CAM explants with *Brucella abortus* 2308, centrifugation for 15 min at 1000 xg and incubation at 37°C in 5% CO₂ for 30 min, followed by 1 h of incubation with gentamicin. The shows the average number of internalized bacteria (Log CFU/mL) in three explants (triplicates) prepared from three independent experiments (different fetuses). There were no significant differences in the number of *Brucella abortus* internalized in CAM explants from different fetuses under these conditions. ## Protein expression profile of CAM explants infected with *Brucella abortus* The 2D gels (7 cm, pH 4–7) were compared to verify the reproducibility and the occurrence of experimental variation related to the quantity of protein extract added to the gel or/and intensity of the dye. According previous study, 2D gels are considered reproducible when presenting percentage of matching spots higher than 80%, and correlation ratio higher than 0.75. We observed that at each time point after infection, 2D gels had high reproducibility when the triplicates of uninfected CAM explants (% matching spots\> 84%; correlation ration \> 0.77) and triplicate of infected CAM explants (% matching spots\> 88%; correlation ratio \> 0.80) were compared. The proteomic profile of uninfected CAM explants remained stable throughout the course of the experiments (% matching spots = 77%). The same was observed in the profile of infected CAM explants (% matching spots = 80%). No significant changes were observed when the proteomic profile of uninfected CAM explants at various time points were compared. In contrast, four spots with significant changes in volume over time were detected in gels of infected explants: one of the spots was significantly increased at 2 h when compared to 4h, and the other three spots were significantly increased at 4 h when compared to 8 h. To determine the times points to be used for DIGE, an inter-class analysis in which the gels of uninfected and infected samples were compared in the same time post infection, was performed. There were no significant differences in volume of spots when infected and uninfected explants were compared, in each times evaluated. However, spots present in a single experimental group were detected (qualitative differences). Thus, 0.5 and 4 h post infection were chosen to be evaluated by the DIGE since these time points had the highest qualitative differences. ## Differential gel electrophoresis (DIGE) In the analysis by DIGE, there were no statistically significant differences in the volume of the spots when gels from infected or uninfected explants were compared, considering significant in this analysis, values of fold change \> 2 and P≥0.05. Interestingly, although no significant differences were observed in the volume of spots detected in both experimental groups at both time points, several spots were detected in only one experimental group (qualitative differences), as well was observed in 2D gels analysis. These spots were selected and numbered for identification by mass spectrometry (MALDI TOF/TOF). From a total of 103 spots analyzed by mass spectrometry, 74 (72%) were identified, from which 73 (98.64%) corresponded to proteins produced by the host (*Bos taurus*). Only peptides from the spot number 129, identified as adenosylhomocysteinase, matched with significant scores *B*. *taurus* and *B*. *abortus*. Importantly, in this case, search in *Bos taurus* database resulted in identification of seven peptides with high value score (378–17% coverage), while search of the *B*. *abortus* data base identified one peptide with a low score (61–4% coverage), which indicates a greater probability that this protein was in fact expressed by bovine cells. The 74 spots identified corresponded to 51 proteins, i.e. the same protein was identified in different spots. All of them were classified into 11 functional categories by Funcat (Functional Classification of Proteins) (MIPS—Munich Information Center for Protein). In order to better understand the distribution of proteins according to their function and experimental group from which they were identified, a Venn diagram showing the identified proteins in uninfected and infected CAM explants at 0.5 and 4 h was constructed. In uninfected CAM explants group (0.5 and 4 h) were mainly identified proteins related to the biogenesis of cellular components (cytokeratin 8, dynein light chain roadblock-type 1, F-actin-capping protein subunit beta, transgelin), proteins with binding or cofactor function (3-hydroxyisobutyrate dehydrogenase, mitochondrial precursor, albumin protein, alpha-actinin-4, alpha-fetoprotein precursor, beta actin, gelsolin isoform a, gelsolin isoform b, hemoglobin subunit beta, prostaglandin reductase 2, transitional endoplasmic reticulum ATPase) but also related to the cell structure (gelsolin isoform a, gelsolin isoform b, alpha-actinin-4), proteins related to the metabolism (3'(2'),5'-nucleotidase bisphosphate 1 3-hydroxyisobutirate dehydrogenase, creatine kinase B-type, inositol-3-phosphate synthase 1, precursor transthyretin) and related to protein fate (activator of 90 kDa heat shock protein ATPase homolog 1, cathepsin D, endoplasmin precursor, heat shock cognate 71 kDa protein). It is also important to emphasize that most of the spots in which these proteins were identified, were selected based on the comparison between the two groups of uninfected CAM explants (77,5%, 38/49). Together these findings suggest that the differentially expressed proteins observed in uninfected CAM explants are related to the maintenance and stabilization of cellular organization and adaptation to cell culture after a stress condition related to the process of CAM explants preparation. In addition, in *B*. *abortus*-infected CAM explants group the highest number of proteins identified was classified as related to metabolism or cell rescue defense and inflammation. Proteins with cell rescue, defense and inflammation functions (n = 15) were identified only in explants infected with *B*. *abortus* when compared to uninfected explants: 8 (53,3%) proteins identified 0,5 h post infection (aldose 1-epimerase, aldose reductase, biliverdin reductase A, heat shock protein beta-1, high-mobility group box 1, protein disulfide-isomerase A3 precursor, thioredoxin-dependent peroxide reductase and toll-interacting protein) and 7 (46,7%) proteins identified 4 h post infection (aldose reductase, complement component 1 Q subcomponent-binding protein, heat shock protein beta-1, high-mobility group box 1, NADH dehydrogenase \[ubiquinone\] iron-sulfur protein 8, protein disulfide-isomerase A3 precursor and thioredoxin-dependent peroxide reductase). Moreover, the proteins aldose 1-epimerase, heat shock protein beta-1, high-mobility group box 1-like protein disulfide-isomerase A3 precursor thioredoxin-dependent reductase peroxide, mitochondrial precursor, has been identified in both groups of infected explants. The proteins albumin, alpha-fetoprotein precursor and gelsolin isoform b were identified in more than one experimental group but with different values of pI or molecular mass indicating possible post-translational modifications. # Discussion Despite being bovine brucellosis a zoonotic disease that causes significant economic losses worldwide due to abortion and culling of infected animals, the pathogenic mechanisms are not yet fully understood. Considering our results with what has been reported in the literature, we suggest the hypothesis that at earlier stages of infection with virulent *B*. *abortus* 2308, bovine CAM explants exhibited increase abundance of proteins directed to recognition of bacteria, mostly activation of the innate immune response, involving proteins related to TLR signaling and ROS production, as well as expression of proteins associated with intracellular trafficking and inflammation. Accordingly, we will direct our discussion to the simultaneous involvement of the proteins identified here and that have been previously linked to *Brucella* and their interactions. One of the most interesting proteins identified in CAM explants during the initial stages of infection with *B*. *abortus* was the Toll-interacting protein (TOLLIP), a protein that modulates TLR signaling and also control membrane trafficking processes by its interaction with proteins and phosphoinositides. As TOLLIP is a protein that negatively modulates the inflammatory response, expression of this protein in CAM explants infected by *B*. *abortus* at 0.5 h post inoculation may be related to the state of immunosuppression observed at early stages of infection, condition that can contribute to bacteria evasion of initial immune response, important step in the establishment of infection. The suppression of proinflammatory response by trophoblastic cells at early stages of *B*. *abortus* infection has been previously shown since transcriptomic analysis demonstrated a reduction of transcription of genes associated with TNF superfamily, e.g. lymphotoxin beta, tumor necrosis factor, and ligand chemokine—CXC motif at 4 h post infection. Recent published data from our group indicated that this active suppression of proinflammatory responses induced by *B*. *abortus* in trophoblastic cells requires a functional T4SS and the BtpB TIR-containing protein. Increased expression of TOLLIP maybe a host mechanism for controlling inflammation or it may be induced by the pathogen in order to evade an effective immune response. It is known that *Brucella* spp. is able to subvert immune response by the producing BtpA and BtpB that bind directly to MyD88 thereby preventing TLR signaling. Another protein that participates in TLR signaling, and that was identified in infected trophoblastic cells in this study was the high-mobility group box 1 (HMGB1), a DAMP (damage associated molecular pattern), which participates in signaling danger to other cells, activates innate and adaptive immune responses and promotes tissue regeneration. HMGB1 is an endogenous ligand of TLR, which may explain its ability to induce cellular activation and inflammatory responses similar to those initiated by LPS. Considering that HMGB1 is passively released by necrotic cells or actively secreted by activated cells of the immune system, its ligation to several receptors that induce inflammation including TLR2, TLR4, TLR9, and RAGE, activating NF-kB and inducing secretion of proinflammatory cytokines, expression of HMGB1 by cell of CAM explants infected with *B*. *abortus* can have a significant immunomodulatory effect and can potentially impact the outcome of pregnancy, and along with other mechanisms may play a relevant role in the development of necrotizing placentitis that is associated with *B*. *abortus*-induced abortion in cattle. The Ras-related protein Rab-11A (RAB11A) was upregulated in explants infected with *B*. *abortus*. This protein is related to transport of TLR to phagocytic vesicles containing Gram-negative bacteria and consequently with mechanisms of pathogen recognition and immune response. Another protein related to intracellular trafficking that was identified in infected CAM explants was SCAMP2 (Secretory carrier membrane proteins 2), which is part of a group of transmembrane proteins expressed in most eukaryotic cells that participate in traffic vesicles between Golgi apparatus and plasma membrane and also plays a role in exocytosis and endocytosis. The involvement of SCAMPs in formation of *Salmonella* sp. containing vacuole (SCV) has been described. Several intracellular pathogens are able to manipulate the secretory pathways of the host cell, including *Brucella* spp. which interacts with the endoplasmic reticulum to establish a niche for multiplication and to evade the host immune response. The proteomic screening in this study also identified proteins related to generation of reactive oxygen species (ROS), and inflammatory responses induced by oxidative stress (Aldose reductase—AKR1B1, NADH dehydrogenase \[ubiquinone\] iron-sulfur protein 8—NDUFS8). Induction of ROS causes loss of intracellular redox homeostasis, with altered cell signaling and development of pathological processes. Moreover, *Brucella* spp. is capable to produce superoxide dismutase (SOD), which protects against endogenous superoxide produced by aerobic metabolism and respiratory burst from the host cell. Production of SOD prevents bacterial death and enables the establishment and maintenance of intracellular bacteria, and thus it is important for the survival of *Brucella* spp. in its intracellular niche. Conversely, biliverdin reductase (BVR), thioredoxin- dependent peroxide reductase (PRDX3—peroxiredoxin 3) and heat shock protein beta-1 (HSPB1) or Hsp27 are proteins related to cytoprotection against oxidative stress and were also identified in *B*. *abortus*-infected CAM explants. Detection of these proteins in infected explants suggests an attempt of host cells to reduce the damaging effects of ROS, although these proteins may also favor intracellular survival of *Brucella* spp. *B*. *abortus* is capable of inducing a strong neutrophilic and necrotizing placentitis that is associated with transcription of CXCL8 and CXCL6 in CAM explants and *in vivo* in the placenta of experimentally infected cows. In present proteomic analysis of infected CAM explants resulted in the identification of several proteins potentially related to recognition of pathogens and tissue inflammation that had not been reported in previous studies. Among these proteins, there was upregulation of the complement subcomponent1 Q-binding protein (C1QBP), an intracellular protein found predominantly in association with mitochondria and the nucleus whose expression can be activated by the action of proinflammatory cytokines such as IFN-γ, TN F-α, or LPS; protein disulfide isomerase A3 (PDIA3), a protein that belongs to the superfamily of enzymes thioloxiredutases that is specifically associated with peptide presentation via class I MHC and whose overexpression has been recently related to immunological processes; aldose 1-epimerase or galactosemutarotase (GALM), an enzyme that catalyzes interconversion of β-D- galactose to α-D-galactose, whose increase in expression has been described in inflammatory processes. Interestingly, in *B*. *abortus* CAM explants there was low abundance of adenosylhomocysteinase (AHCY), a highly conserved enzyme that catalyzes the reversible hydrolysis of S-adenosyl-L-homocysteine to adenosine and homocysteine so it is considered a key enzyme in immune response. The use of AHCY inhibitors for therapeutic purposes in inflammatory and immune diseases is being studied, since inhibition of AHCY is associated with immunosuppression. # Conclusions This study clearly demonstrated changes in the protein expression profile of bovine trophoblastic cells in CAM explants in early stages of *B*. *abortus* infection and that infection induces increase or production of proteins that may be associated with the necrotic placentitis seen in infection of cattle placenta. Several of the proteins that were upregulated during infection are associated with modulation of the innate host immune response to infection with *Brucella abortus*. Therefore, this study contributes to improving our understanding of the mechanisms related to abortion caused by infection with *B*. *abortus* in cattle. # Supporting Information This study was funded by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), and the Fundação de Amparo a Pesquisa do Estado de Minas Gerais (FAPEMIG—PRONEX). JPSM, APL and RLS have fellowships from CNPq. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JPSM HMA APL. Performed the experiments: JPSM SFP. Analyzed the data: JPSM SFP ADC JP. Contributed reagents/materials/analysis tools: JP RLS HMA APL. Wrote the paper: JPSM RLS HMA APL.

# Introduction One of the striking patterns in geographic distribution of terrestrial biodiversity is the increase in species richness towards lower latitudes in several groups of organisms, including birds. The possible causes for this pattern is one of the highly debated topics in ecology and evolution, even though no definitive conclusion was yet been achieved,. The Neotropical area alone holds a third of the recognized extant bird species (about 3,300 out of 10,000), with a biodiversity hotspot in the tropical forests. Moreover, recent phylogenies suggest the number of species in the area is underestimated because reproductively isolated lineages are frequently described in these studies. In stark contrast to bird taxonomy in temperate zones, genetic evidence for species limits in the Neotropics is often discordant with traditional taxonomy due to the high incidence of species complexes. These complexes commonly feature gradual variation in morphological and behavioural characters, masking the occurrence of similar species that can be uncovered with genetic analyses. DNA barcodes based on the 5′ portion of the cytochrome oxidase I gene (*COI*) linked with specimens vouchers and locality information provides a rapid and inexpensive method to identify species and detect ‘provisional new species’. Pilot DNA barcode surveys in birds of North America, sister-species pairs, and birds of Korea were successful in either identifying recognized species of birds, and detecting some potential new species, except for a minor proportion of cases where species are very recently diverged or hybridize. Critics questioned if the success observed in North American birds could be extrapolated to the tropics, where species clearly exhibit a higher level of phylogeographic subdivision. However, DNA barcoding has subsequently proved to be highly successful in identifying Neotropical species of birds; all 16 species (100%) of antbirds (Thamnophilidae) that were barcoded and 494 of 500 (95.8%) species of birds of Argentina had distinguishable *COI* signatures. The screening of Argentinean birds also detected 21 species with deep intraspecific structure, and revealed more complex patterns of regional divergence in the Neotropical than in the North American avifauna. Even though more species will doubtlessly be shown to share barcodes when complete coverage of species and genera is available, it is clear that large-scale sequencing of *COI* associated with vouchered specimens and locality information is a valuable tool in understanding genetic differentiation within and among species of birds. In this study we increased the coverage of Neotropical bird species that have been barcoded by adding 637 samples from 431 species, with higher representation in tropical forest areas of Brazil and Guyana, but also including samples from localities ranging from Mexico to Argentina and Chile. We compared these sequences with previously published sequences of congeneric species of Neotropical birds, totaling 1,431 samples from 561 different species of birds, 296 of which were represented by multiple individuals. We showed that a high success rate in species identification (93%) with DNA barcodes can be achieved in this large sample of avian biodiversity from the mega-diverse Neotropical region similar to that obtained in broad geographic surveys in the Nearctic and Palearctic regions of the world. Additionally, a higher percentage (12%) of species had multiple deep phylogeographic splits than in previous surveys, some of which are likely reproductively isolated lineages. # Results ## Species identification in Neotropical birds About 93% of the species in our sample (520 out of 561) did not share sequences with any other species included in the analysis, and when multiple individuals were sampled (296 species), mean genetic distances among individuals were lower than to the closest species from the same genus. Kimura 2-Parameter genetic distances (K2P) within-species had a wide range (0 up to 13.7%), with more than 75% of the observations below 1% K2P. Conversely, 10% of the pairwise comparisons were higher than 3% K2P (range 3.1–13.7%), overlapping considerably with among-species variation. Pair-wise comparisons among-species of the same genus were distributed from 0.08 to 20.3% K2P with most of the comparisons observed between 5–15% K2P. Extremely high genetic distances suggestive of higher rates of evolution or ancient divergences were observed among species within *Trogon* (Trogoniformes) and *Crypturellus* (Tinamiformes), with maximum distances of 19, and 20.3%, respectively. One specimen identified as *Nothoprocta ornata* differed from other species in the genus by 24.7%, but was only 0.34% divergent from specimens of **Tinamotis pentlandii.** Hence it was either incorrectly identified in the field or possibly is a hybrid between the two genera, as both species occur near to the collecting locality in Chile. A total of 41 species did not have unique barcodes, of which 21 share sequences with other species. Eight of those species despite being reciprocally monophyletic, or represented by one sample, only differed from their sister species by 1 to 6 diagnostic characters, or 0.14 to 0.86% K2P distance. Aggregation of closely related haplotypes in phylogenetic trees can either represent distinct taxonomic units, or random branches of lineages within the same taxonomic group. To distinguish between the two scenarios, we applied a statistical test of taxonomic distinctiveness proposed by Rosenberg for sister species differing by less than 1% K2P. With the current limited sampling of individuals, chance occurrence of reciprocal monophyly between species could not be rejected (p\>0.05), so they were considered not distinguishable by *COI* barcodes. Species pairs differed by few nucleotide substitutions, with marginal values for the test of chance occurrence of reciprocal monophyly (0.01\<p\<0.05). Thus the following species groups were considered to be distinguishable by *COI* barcodes: the ducks *Anas puna*/*versicolor* (p = 0.01), the greenfinches *Carduelis atrata/barbata/versicolor* (p = 0.01) and the orioles *Icterus cayanensis*/*chrysocephalus* (p = 0.03). Sixteen species had multiple divergent clusters (K2P genetic distances between 1.54 up to 13.7%), not recovered monophyletic with COI, that often corresponded to samples from different areas of endemism or ecoregions (- cat. IV). A few exceptions were observed, where paraphyletic divergent specimens were found in the same geographic locality. For instance, specimens from the long-tailed hermit (*Phaethornis superciliosus*) from Aripuanã and Juruena, both within the Rondonian area of endemism, were 8% divergent, and the specimen from Juruena differed from a scale-throated hermit (*P. eurynome*) from Southern Atlantic forest by 7.4%. Even more strikingly, two samples of the yellow-margined flycatcher (*Tolmomyias assimilis*) from Napo were 8.3% divergent. The species pair of thrushes *Turdus albicollis/leucomelas* were paraphyletic in their *COI* sequences, sharing barcodes in their Amazonian distribution. ## Deep genetic structure within Neotropical bird species Deep intraspecific divergences in 48 species overlapped widely with among- species distances (K2P 1.6 to 7.8%). These genetically structured species belong to 21 bird families from nine different bird orders, most frequently represented by antwrens (Thamnophilidae, Passeriformes). Most of the species with deep genetic structure were broadly distributed in the Neotropics, and several are subdivided into multiple subspecies. Often samples from different areas of endemism or different ecoregions were the most divergent within species. Some species showed genetic discontinuities in some pairs of geographic areas, but not in others, such as the ochre-bellied flycatcher (*Mionectes oleagineus*). Samples from the Napo, Imeri and Guyanian areas of endemism were not very distinct genetically, but specimens from Belém were 2.76% divergent from the others. All samples of the white-shouldered antshrike (*Thamnophilus aethiops*) from different areas of endemism (Belém, Rondonian, Imeri, and Napo) had deep instraspecific genetic variation. The deepest split was between Belém and the other areas, and then the next split was between Rondonian, Imeri, and Napo. ## Phylogeographic patterns Deep intraspecific divergences in different species were often located between the same pairs of areas of endemism or ecoregions. The most common pattern observed was between the Napo and Rondonian areas of endemism, followed by Belém and Rondonian. However, phylogeographic splits between areas varied in depth in different species. For example, distances between Napo and Rondonian were 8% in *Tolmomyias assimilis*, *Myrmotherula longipennis*, and *Conopophaga aurita*, 5% in *Thamnomanes caesius*, *Cyphorhinus arada*, *Hylophylax naevius*, *Schiffornis turdina*, and *Jacamerops aureus*, and 2% in *Myrmoborus myotherinus*, *Hylophylax poecilinotus*, *Deconychura longicauda*, *Deconychura stictolaema*, *Hemithraupis flavicollis*, *Thamnophilus aethiops*, and *Monasa morphoeus*. Between Guyanian and Napo areas of endemism instraspecific divergences of 5 and 2% were observed in *Dendrocincla fuliginosa* and *Jacamerops aureus*, respectively. Conversely, K2P genetic distances were close to zero within the species *Dixiphia pipra* and *Nyctidromus albicollis*. Genetic divergences between Belém and Napo were from 2 to 6%, with a wide range of intermediate levels: 2.4 (*Phaethornis ruber*), 2.5 (*Formicarius colma*), 2.8 (*Mionectes oleaginous*), 3.3 (*Xiphorhynchus guttatus*), 3.6 (*Thamnophilus aethiops*), 4.2 (*Jacamerops aureus*), 5.6 (*Dendrocincla fuliginosa*) and 5.7% (*Turdus leucomelas*). # Discussion ## Identification of Neotropical species with DNA barcodes Despite the high success we obtained in Neotropical bird species identification with DNA barcodes (93%), comparable to previous barcode surveys in birds, most of the genera and species were not sampled across their entire distribution, which overestimates its potential to differentiate species. This was observed for at least two species previously distinct with DNA barcodes, *Anas sibilatrix*, and *Celeus lugubris*, who were shown to be sharing sequences with *Anas flavirostris* and *Celeus elegans*, respectively, when samples from other areas of their geographic range were included in this study. When comprehensive genus and species coverage becomes available in Neotropical birds, more species are likely to not have unique DNA barcodes. Nonetheless, more certainty will be achieved overall in the identification of species with *COI* barcodes because we will be able to better address monophyly of lineages and to verify the frequency with which individuals from different populations within species complexes are exchanging genes. In most of the genera for which we had better species coverage for *COI*, such as *Paroaria*, *Coryphospingus*, *Hemithraupis*, *Cyanerpes*, *Cyanocompsa*, *Mimus*, *Phacellodomus*, and *Dendrocincla*, species did not share barcodes. Even though we obtained only single sequences for many species, they will contribute to future systematic efforts as part of the public standardized DNA barcode library. They also will aid in faster identification of specimens that are difficult to identify morphologically, such as embryos and eggs, which will positively impact the conservation of avian wildlife in the Neotropical region. ## Species not identified by DNA barcodes Among the species we considered not identifiable with COI barcodes, some were very closely related with very similar barcode sequences (category III). Our sampling was not comprehensive enough to reject their monophyly by chance, but once more individuals from different areas of their range are included stronger support might be adduced for their reciprocal monophyly. On the other hand some species might be recovered as not monophyletic with increased sample sizes, due to unsorted ancestral polymorphism or hybridization. In that case they would not be identified by DNA barcodes at the species level, suggesting that future studies should employ multilocus phylogenetic inference with faster evolving nuclear sequences in a coalescent framework to try to resolve species lineages. Once larger sample sizes are available for closely related species, character- based approaches implemented automatically, such as in CAOS, are preferable to genetic distance levels to determine their distinctiveness, as distance levels within and among species can overlap considerably even when substitutions among species are fully sorted. The species recovered as non-monophyletic (category IV) are strong candidates for taxonomic revision, and some of their divergent lineages might correspond to different species. For instance, the divergent lineages within the bearded flycatcher (*Myiobius barbatus*) belong to different recognized subspecies: *amazonicus*, *insignis*, and *mastacalis*. They currently are allopatric, have morphological differences and differ in their K2P genetic distances by 12.6–13.7%. The three subspecies clades were not recovered as monophyletic with COI barcodes because the ruddy-tailed flycatcher (*Terenotriccus_erythrurus*) and the black-tailed myiobius (*Myiobius_atricaudus*) were included in the species clade. Similarly, specimens from North and South Atlantic forest of the rufous gnateater (*Conopophaga lineata*) differ by 9.6%. However, the lineages from the two localities are not monophyletic because the chestnut-belted gnateater (*Conopophaga aurita*) and the hooded gnateater (*Conopophaga roberti*) are embedded in this group, as shown previously with more comprehensive sample sizes and mitochondrial markers. The morphological characters used to define these lineages as members of a single species could be under strong stabilizing selection, and thus not mirroring the accumulation of mutations through time in neutral genes like *COI*. Most cases of paraphyly in birds are caused by incorrect taxonomy. Alternatively, paraphyletic species can arise when geographically isolated lineages merge in part of their distribution before complete reproductive isolation has evolved. Phylogeographic studies including samples from their entire geographic range and from the closely related species are needed to properly understand their diversification patterns, and establish their taxonomic status. The 17 species that shared barcodes with closely related species in sympatry likely experienced hybridization, or recent speciation and incomplete lineage sorting, or could simply be examples of incorrect taxonomy or sample misidentification. For instance, the flightless steamer duck (*Tachyeres pteneres*) shares barcodes with the flying steamer duck (*Tachyeres patachonicus*) in Argentina, even though these species are very distinct morphologically. In this example, misidentification of the sample is less likely. A multigene phylogeny of four duck genera also reported difficulty in resolving the relationships among species of *Tachyeres*, and attributed this to a rapid diversification of the group, with possible incomplete lineage sorting, founder effects, and introgression. The tawny-crowned greenlet (*Hylophilus ochraceiceps*) had intraspecific clusters differing by almost 7% sequence divergence between Napo/Imerí and Rondonian endemic areas, and shared barcodes with the grey-chested greenlet (*H. semecinereus*) in their Rondonian distribution. Both species are comprised of multiple subspecies, and some of their variants are morphologically alike. The current taxonomy of the genus might not be an accurate reflection of lineage relationships, but misidentification of samples cannot be ruled out. Two species pairs occurring in allopatry were not reciprocally monophyletic: the bicolored (*Gymnopithys leucaspis*) and rufous-throated (*Gymnopithys rufigula*) antbirds, and the ochre-collared piculet (*Picumnus temminckii*) and spotted piculet (*Picumnus pygmaeus*). In both cases they are morphologically distinct and do not share identical barcodes with the other species; genetic distances among samples were around 0.5% and 1.0%, respectively. In these examples the lack of reciprocal monophyly could be result of recent speciation and shared ancestral polymorphism, and hybridization. A faster evolving marker such as the control region or larger mitochondrial sequences might recover their reciprocal monophyly. ## Complex patterns of population structure detected with DNA barcodes Our results agree with previous hypotheses that complex patterns of speciation were responsible for the high diversity in Neotropical bird species, and strongly supports the view that most avian species in the region are narrowly endemic rather than widely distributed. Several hypotheses were proposed to explain the patterns of taxon distribution in the Amazonian lowland region. The forest *refugia* hypothesis, suggested that cycles of expansion and retraction of dry patches within forest areas were associated with interglacial and glacial periods, and this could create multiple events of isolation among widely distributed groups, promoting speciation. The *riverine* hypothesis suggested that the formation of the rivers in the Amazon region could have acted as important geographic barriers to promote speciation, as they delimit most areas of endemism. This would have started at least by the end of the Miocene with the uplift of the Northern portion of the Andes. Another proposal is the *marine incursions* hypothesis, in which sea-level rises of about 100 m in the Quaternary and late Tertiary are suggested to have fragmented the Amazonian lowland into a large number of true islands and archipelagos, favoring active allopatric speciation. The wide range of divergence levels we observed within the 61 non-monophyletic and monophyletic species with deep intraspecific variation (1–13% K2P distances), together with the high incidence of recently evolved species, is consistent with speciation events starting well before the Pliocene and Pleistocene, and extending to more recent geologic periods. Although several groups of species have similar patterns of genetic and geographic breaks among the same areas of endemism, different levels of genetic distances between the same areas were also recovered in other species. The wide range of intraspecific genetic distances observed between a pair of geographical localities might reflect multiple vicariant events that have occurred at different geological times, or they could reflect multiple dispersal events that followed a major isolation process, or variation in rates of evolution across different species, whose populations were isolated by a single vicariant event. Additionally, a significant relationship was observed in previous studies between interspecific levels of cross-barrier genetic differentiation within the forest stratum at which a species forages in Neotropical rain forest. More comprehensive taxon sampling and estimates of times of diversification that take into account variation in rates of evolution across lineages are needed to properly associate the diversification of a particular taxon with geographical events. We have chosen not to flag divergent lineages as provisional new species, because our sampling was not comprehensive enough to properly quantify genetic variation in each locality in different species, such as the red-eyed vireo (*Vireo olivaceus*) and the ultramarine grosbeak (*Cyanocompsa brissonii*). Specimens of red-eyed vireo from Puna+Napo and Atlantic Forest were genetically divergent (2–3%), but haplotypes from the Atlantic Forest and Puna were observed in the Chaco. Similarly, specimens of ultramarine grosbeak from Caatinga and Puna were also divergent (2.7%), and both haplotypes are also found in Chaco. Both species may have reinvaded the Chaco after being isolated on the borders of this area. To check if these lineages deserve species recognition it is important to investigate if the highly divergent specimens in sympatric zones are reproductively isolated. Some of the deep intraspecific lineages we described in this study were reported previously, such as the difference among thrush-like Schiffornis (*Schiffornis turdina*) from Rondonian and Napo areas of endemism. Others, such as the whiskered myiobius (*Myiobius barbatus*) from Belém, Para2 and Atlantic forest will likely prove to be different species. DNA barcodes of several new species of Neotropical birds will contribute to a deeper understanding of the systematics and diversification of these taxa in the area. Assuming the current species taxonomy, studies of historical patterns of diversification of species in the area can be obscured since many species were revealed not to be monophyletic. Moreover, a high number of species in the Neotropical realm are comprised of multiple divergent lineages, thus the sample sizes of barcoded individuals and other markers within and among species in the area need to be higher than in other biogeographic areas that are not as taxon- diverse. This can be achieved by complementary efforts of independent research groups. Common and divergent patterns of genetic distances observed within and among closely related species suggest that multiple geographic processes have shaped the distribution of avian taxa in the Neotropics, and DNA barcodes surveys will continue to reveal many more interesting geographic patterns in the region. # Materials and Methods ## Taxon sampling We analyzed 637 individuals from 431 species of Neotropical bird species from two tissue collections: Laboratório de Genética e Evolução Molecular de Aves (LGEMA) in the Universidade de São Paulo, São Paulo, and The Royal Ontario Museum in Toronto (ROM), with high representation in the Amazon lowlands and Atlantic Forest. Whenever available, individuals from different localities of their distribution range were sampled (GenBank numbers JN801479 - JN802115, project “Neotropical-BRAS” in the completed projects section of the Barcode of Life Data System- BOLD). To increase intraspecific sampling and to compare more closely related congeneres, we added sequences of individuals from the same species and same genera of Neotropical birds from the study of birds from Argentina (project “Birds of Argentina-Phase I-BARG” in the completed projects section of BOLD), thus extending our survey to 1,431 samples from 561 different species. ## DNA extraction and amplification DNA was extracted by a membrane purification procedure in glass fiber-filtration plates (Acroprep 96 Filter Plate- 1.0 µm Glass, PALL Corporation), and collected in PCR plates. Sequences of about 700 base pairs (bp) were obtained from the 5′end of the mitochondrial gene Cytochrome oxidase I (*COI*). Polymerase Chain Reaction (PCR) amplifications were performed in 12.5 µL reactions in a buffer solution containing 10 mM Tris-HCl (pH8.3), 50 mM KCl, 2.5 mM MgCl₂, 0.01% gelatin, 0.4 mM dNTPs, 0.2 µM of each primer, 1 U *Taq* Polymerase (Invitrogen) and 20–25 ng of DNA. Cycle conditions were: an initial denaturation at 94°C for 5 min, 36 cycles of 94°C for 40 sec, 50°C for 40 sec and 72°C for 1 min, and a final extension at 72°C for 7 min. Bird universal primers used in *COI* amplifications were LTyr (forward – TGTAAAAAGGWCTACAGCCTAACGC) and COI907aH2 (GTRGCNGAYGTRAARTATGCTCG) resulting in a long but very stable amplified product of about 910 bp. This primer set successfully amplified the 5′ end of *COI* across a wide range of bird species. The amplified segments were purified by excising bands from agarose gels and centrifuging each through a filter tip. Sequences were obtained on an ABI3730 (*Applied Biosystems*) according to the manufacturers' suggested protocols using the same primer LTyr to sequence the 5′end, and the internal primer COI748Ht (reverse- TGGGARATAATTCCRAAGCCTGG) to sequence the reverse 3′end, resulting in a sequenced product of about 750 bp. Sequences were checked for ambiguities in CodonCode Aligner (*CodonCode Corporation*), and Geneious 5.3. ## Data analyses Sequences were aligned in Geneious 5.3 using the Geneious alignment algorithm, with gap penalty set as 12.8, and gap extension penalty set as 3. Species and genera counts were performed in the software environment R 2.12. Genetic distances were calculated under the Kimura 2 –Parameter model (K2P) for all pair-wise comparisons in the matrix using PAUP4b10. Two datasets of genetic distances were built in R: the first, including all within-species comparisons; and the second, including among-congener comparisons (excluding within-species ones). We wrote R scripts to summarize the mean, variance, maximum, and minimum genetic distances per species and among congeners, respectively, using the first two datasets. Frequency plots of pairwise genetic distances for congeners of different species, and with only within species comparisons were built in R. The maximum likelihood tree topology for the complete dataset was calculated in Geneious 5.3 using PHYML. The best fit-model (General Time Reversible with proportion of invariable sites and gamma, GTR+I+Γ, I = 0.5, Γ = 0.42) was selected with jModelTest with a sample of the original dataset including one or two representative samples of each bird family. Species were considered not distinguishable by DNA barcode if: a) they were not monophyletic; b) they shared barcodes with other species; or c) their intraspecific variation overlapped with the lowest 5% of among-species variation, and reciprocal monophyly of sampled individuals could not be distinguished from random branching at p = 0.05 with the test for chance occurrence of reciprocal monophyly. Within-species clusters with minimum pairwise distances higher than 1.5% K2P were considered for analyses, because this level of genetic distances overlapped with more than 5% of among congeners comparisons , but information on clades differing by less than 1.5% K2P distance is also available. Species without unique barcodes were sorted into the following non-exclusive categories: I) they share barcodes with species occurring in sympatry or II) they share barcodes with species occurring in allopatry, or III) were monophyletic differing from their sister species by few mutations, or IV) paraphyletic species with lineages more than 1.5% divergent. For all the paraphyletic and monophyletic species with deep intraspecific divergences, we compared the genetic discontinuities with the geographic locality of the samples. Because areas of endemism are known to harbor unique biota, and many subspecies of birds are delimited also by these zones, we classified the sample localities of individuals according to the areas of endemism in the Amazon and in the Atlantic forest where they occur. We adopted the revised areas of endemism in Amazon and Atlantic forest from Bates *et al.* and Borges. Samples collected in other localities were classified according to their respective ecoregion according to the simplified map from Haffer. # Supporting Information We thank A. C. Mendez, Alexandre Aleixo, Alexande Martensen, Alexandre Uezu, Camila Ribas, E. Machado, Fernando d'Horta, Fernando Nodari, Guilherme Brito, Gustavo Gabanne, Luis Fabio Silveira, Pedro Develey, Renato Gaban Lima, and Rodrigo Pessoa for collecting the biological samples deposited at LGEMA and used in this study. For permission to barcode loaned samples we thank the American Museum of Natural History and Field Museum of Natural History; Nicole Leung and Nicolle Domnick for their help with sample processing, Bernd Schierwater and three anonymous reviewers for valuable suggestions on the manuscript, and Cedrik Juillet for advice in developing the R scripts. [^1]: Conceived and designed the experiments: EST PG. Performed the experiments: EST PG CYM. Analyzed the data: EST PG. Contributed reagents/materials/analysis tools: CYM AJB. Wrote the paper: EST PG CYM AJB. [^2]: The authors have declared that no competing interests exist.

# Introduction Quantum dots (QDs) are semiconductor nanocrystals that are gaining in popularity over organic fluorophores in applications ranging from bioimaging and analytical assays to electronic displays, solid-state lighting and photovoltaics. QDs commonly consist of a CdSe, CdTe, ZnSe, or PbSe core coated with a ZnS shell to enhance stability and optical properties. In some cases, the shell is further functionalized with thiols or amphiphilic polymers to make the nanocrystals soluble in aqueous solvents and to allow for biomolecule conjugation.\[–\] As QDs become more prevalent in consumer products that will be used, reused, recycled, and landfilled, concerns have been rising about their impact on humans and the environment. Studies conducted with cultured eukaryotic cells have revealed that QDs can exert cytotoxic effects through a variety of mechanisms. These include leaching of toxic heavy metals from the inorganic core, (photo)generation of reactive oxygen species that induce oxidative stress, and direct or indirect damage to genomic DNA and biological membranes.\[,–\] Parameters such as size, shape, composition, and surface coating(s) can all impact cytotoxicity outcomes and do so in a mechanism-specific (and cell-specific) manner. For instance, while ZnS shells or polymer coatings can reduce the cellular toxicity of CdSe nanocrystals, they do little to prevent photo-induced DNA damage. Similar observations have been made in animals, where the situation is further complicated by the route of exposure and where long-term retention in the liver, spleen, kidney, and lymph nodes is of primary concern. There is considerably less information on how QDs interact with prokaryotes although these abundant microorganisms will be first to come into contact with engineered nanomaterials that find their way into the environment. Furthermore, most studies conducted to date have focused on cadmium-based QDs (CdS, CdSe and CdTe cores) produced with different synthesis schemes and coatings, and used at different doses with a variety of strains and culture conditions.\[–\] To answer a growing demand for the production of functional nanomaterials through environmentally friendly processes, we previously described a set of “designer” proteins that support the low-temperature and aqueous fabrication of undoped and transition metal-doped ZnS QDs to which antibodies can be conjugated by simple mixing.\[–\] Because zinc is not as toxic as cadmium, these particles should exhibit low cellular toxicity upon core dissolution, and because they are capped by proteins as part of the manufacturing process, their shell is already biologically-relevant. Here, we used *E*. *coli* as a model organism to investigate the penetration requirements, subcellular localization, induction of stress responses, and long- term fate of luminescent ZnS:Mn nanocrystals fabricated with a minimized designer protein. Our results suggest that such protein-coated fluorophores are environmentally benign because their uptake requires membrane destabilization, they only induce oxidative stress at high doses, and they are rapidly diluted by cell division. # Results and Discussion ## Uptake by *Escherichia coli* requires membrane destabilization We recently reported that BB-CT43, a minimized designer protein consisting of a linear ZnS binding peptide (CT43) fused to an antibody-binding domain derived from *S*. *aureus* Protein A (BB) is suitable for the one-pot synthesis of ZnS:Mn QDs. These luminescent nanocrystals are produced when the designer protein caps the growth of the inorganic core at about 4 nm, a process that is schematically illustrated in. With their protein shell, the particles have an overall hydrodynamic diameter of 9.5 ± 2 nm and a zeta potential of -16.5 ± 6 mV. They exhibit a strong emission peak at 590 nm under UV illumination, can be decorated with antibodies by simple mixing and are stable for months without aggregation or degradation of optical properties. Unlike traditional QDs, these fluorophores are manufactured using mild aqueous conditions, do not contain highly toxic heavy metals such as cadmium, and sport a built-in protein shell coat. Thus, they should have a minimal impact on microbial life and the environment. As a first test of this hypothesis, we studied the uptake of BB-CT43-stabilized QDs by *Escherichia coli*, a well-studied gram-negative organism whose envelope consists of a 5 nm-thick phospholipid bilayer inner membrane, a 12 nm-thick interstitial space known as the periplasm, a 1 to 2 nm-thick peptidoglycan layer, and a 13 nm-thick, negatively charged outer membrane composed of a lipopolysaccharide outer leaflet and a phospholipid inner leaflet. Like all prokaryotes, *E*. *coli* lacks the endocytosis pathways responsible for nanoparticle uptake by eukaryotes. In addition, bacterial porins, which allow free diffusion of small molecules across prokaryotic membranes through 1–2 nm pores, should be too small to allow even the smallest QDs to enter the cell. Nevertheless, Hirschey and coworkers reported that CdSe/CdS QDs stabilized by citrate, isocitrate, succinate, or malate readily penetrate *E*. *coli* when their inorganic core is smaller than 6 nm. By contrast, Wenhua *et al*. found that the uptake of mercaptoacetic acid-stabilized QDs with 3 to 4 nm CdSe/CdS cores require chemical destabilization of the outer membrane, while Nadeau and coworkers reported that internalization of adenine-coated CdSe QDs strictly depends on photo-induced membrane damage and purine metabolism. To determine whether BB-CT43-stabilized nanocrystals would be uptaken by unmodified *E*. *coli*, we incubated 0.5 μg/mL of nanoparticles with mid- exponential phase *E*. *coli* cells for 2h at room temperature. Cells were washed to remove nonspecifically bound particles, pelleted by centrifugation and exposed to UV light. Under these conditions, there was no detectable nanocrystal uptake. However, in agreement with the results of Wenhua and coworkers, fluorescent material colocalized with sedimented cells if they were first made chemically competent by incubation with 100 mM CaCl₂ at low temperature. This treatment transiently affects the integrity of the outer membrane and is routinely used for introducing naked DNA into cells, although the precise mechanisms at play remain unknown. Not unexpectedly, uptake was dose-dependent and we observed a linear increase in cell fluorescence at 590 nm (the QD emission maximum) when the particle concentration was increased from 0.5 to 2.5 μg/mL. We conclude that chemical disruption of the outer membrane is required for the uptake of QDs coated with protein shells and whose inorganic cores are less photo reactive than CdSe/CdS. ## Internalized QDs localize to the cytoplasm To confirm that the nanocrystals were not simply adsorbed to the surface of the outer membrane, competent cells incubated with 0.5 μg/mL of BB-CT43-stabilized QDs as above were subjected to spheroplasting. This procedure strips *E*. *coli* of its outer membrane and peptidoglycan layer, cause release of periplasmic contents in the surrounding medium, and leads to loss of rod shape and the formation of spherical vesicles bounded by the inner membrane. shows that spheroplasts remained fluorescent, indicating that the QDs were either associated with the inner membrane or had translocated to the cytoplasm. Because confocal microscopy does not allow one to unambiguously distinguish between these two possibilities, we took advantage of the fact that the emission spectrum of UV-excited ZnS:Mn nanocrystals overlaps the absorption spectrum of the fluorescent protein mCherry. Thus, Forster Resonance Energy Transfer (FRET) should occur between the two fluorophores if they co-localize to the same cellular compartment and are separated by distances smaller than 10 nm. To test this idea, we first recorded the fluorescence emission spectra of competent cells that had taken up QDs or had been exposed to buffer alone following excitation at 280 nm. Subtraction of the two spectra eliminated the contribution of background fluorescence and revealed a weak but clear peak centered at 590 nm and corresponding to ZnS:Mn emission (orange). Next, we introduced a plasmid expressing mCherry at high level in the cytoplasm of *E*. *coli* and confirmed that the emission spectrum of these cells exhibited the characteristic shape and 610 nm emission maximum of mCherry upon excitation at 590nm (, inset). Finally, we made competent mCherry-producing cells, exposed them to QDs or buffer, and collected fluorescence emission data following excitation at 280 nm. The subtracted spectrum (red) shows that the QD-associated peak at 590 nm peak completely disappeared to the profit of a 610 nm peak corresponding to mCherry emission. We conclude that nonradiative energy transfer occurs between QDs (donor) and mCherry (acceptor) and therefore that both species are located within a few nanometers of each another in the cytoplasm. How QDs (or plasmid DNA for the matter) translocate across the peptidoglycan layer, periplasm and inner membrane to reach the cytoplasm remains unclear. One possible explanation is that they rely on the transient and CaCl₂-induced opening of sites where the outer and inner membranes come into intimate contact. Such adhesion zones, known as Bayer’s patches, were identified microscopically over 40 years ago, but their existence has remained controversial in spite of supporting biochemical evidence. ## High doses of ZnS:Mn nanocrystals are required to induce an oxidative stress responses We next investigated how the presence of BB-CT43-stabilized nanocrystals in the cytoplasm would impact cell physiology. Prokaryotes have evolved complex and redundant mechanisms to survive exposure to environmental stresses. Many of these processes rely on increasing the synthesis of protective proteins (e.g., molecular chaperones, proteases, DNA repair enzymes and reductases) through upregulation events that are often controlled at the transcriptional level. Previously, we described *E*. *coli* cells harboring single-copy gene fusions between the stress-inducible *ibp* or *sulA* promoters and the *lacZ* gene (which encodes β-galactosidase). These strains report on the amount of stress experienced by the cell as a result of cytoplasmic protein misfolding (*ibp*::*lacZ* fusion) or DNA damage (*sulA*::*lacZ* fusion) by producing the easily assayed enzyme, β-galactosidase. Because QD cytotoxicity has repeatedly been correlated with oxidative damage,\[,–\] we constructed an additional isogenic strain bearing a single-copy gene fusion between the oxidative stress responsive promoter of the major *E*. *coli* catalase (the OxyR-regulated *katG* gene product) and *lacZ*. The functionality of the reporter panel was first confirmed using hydrogen peroxide, nalidixic acid and ethanol at concentrations known to cause extensive oxidative stress, DNA damage, or protein misfolding, respectively. These chemicals caused an about 3-fold induction of the corresponding promoters (positive controls). Next, the three strains were made chemically competent, exposed to QDs, and cultures were assayed for β-galactosidase activity after 3h. While there was no detectable induction of any of the stress promoters when QDs were supplied at the concentration used in all above experiments (0.5 μg/mL), addition of 2.5 μg/mL nanocrystals was as effective as the use of 10 μM H₂O₂ in inducing the *katG* promoter. Of note, however, there was no statistically significant activation of either the *ibp* or *sulA* promoter under the same conditions. While the dependency of toxicity on QD dose is not particularly surprising, our results indicate that it takes highly concentrated solutions of nanocrystals to fully induce the bacterial oxidative stress response and that BB-CT43-stabilized QDs do not cause appreciable protein misfolding or DNA damage under the same conditions. ## QD fluorescence is rapidly lost in growing cells The persistence of toxicants in the environment can lead to their long-range transport and bioaccumulation at toxic doses in animal and human tissues. To gain information on the *in vivo* stability of BB-CT43-stabilized QDs, we first incubated cells that had internalized nanocrystals in phosphate buffered saline (PBS) for 24h at temperatures ranging from 4 to 42°C. There was no significant change in the fluorescence of cell pellets indicating that protein-capped nanocrystals are stable for extended periods of time in quiescent cells exposed to a physiologically relevant range of temperatures. To determine if growth or metabolic activity would influence this outcome, QD- loaded cells were taken in LB medium or PBS and incubated at 37°C, the optimum growth temperature for *E*. *coli*. While non-growing cells held in PBS did not lose their initial fluorescence, we observed a linear decrease in fluorescence over time and nearly complete disappearance of the signal after 3h of cultivation in LB medium (, closed symbols). Because cells experienced a 1h lag phase and exponential growth only started about 2h after transfer to LB (open symbols), the nearly 50% loss of fluorescence that occurs over the first 1.5h of cultivation cannot be attributed to QD dilution by cell division. Indeed, when the experiment was repeated in the presence of the translational inhibitor kanamycin, we observed a similar fluorescence loss over the first 1.5h but, remarkably, no further decrease thereafter. Thus, although *de novo* protein synthesis and/or cell growth are not implicated in initial signal loss, they are necessary for complete elimination of QD fluorescence. There are several possible explanations for the initial fluorescence loss: dissolution or extrusion of about 50% of the internalized nanocrystals or substitution of the BB-CT43 shell by host species that change the QD optical properties. We do not believe that chemical dissolution of the nanocrystals is a likely mechanism since it would be unlikely to abruptly stop in kanamycin- treated or quiescent cultures ( and Figs). To directly test the possibility that active extrusion was involved, we repeated the experiment of in isogenic cells containing or lacking TolC, a trans- periplasmic protein that functions as an exit duct for the expulsion of a wide variety of small drugs and proteins from the cytoplasm to the growth medium. The lack of significant difference in the kinetics and extent of fluorescence loss in *tolC*⁺ and *tolC* ruled out the involvement of TolC- dependent QD export. While we cannot rule out extrusion through other systems, we favor a mechanism in which endogenous species replace at least some of the bound BB-CT43 at the ZnS:Mn surface and cause a decrease in emission intensity through fluorescence quenching. Such quenching phenomena have previously been described and exploited for ZnS:Mn QDs. This explanation is consistent with our previous finding that ZnS:Mn nanocrystals fabricated with BB-TrxA::CT43 have about 30% lower emission intensity than those synthesized with BB-CT43 owing to fluorescence quenching by the TrxA domain. It is also in agreement with the fact that the fluorescence of kanamycin-treated cultures reaches a plateau after 1.5h, a time that is presumably needed to modify the surface of all internalized nanocrystals. Why samples taken in PBS do not experience a similar initial decrease in fluorescence remains unclear but the process appears to require metabolic activity. Irrespective of the precise mechanism of initial fluorescence loss, the data of shows that the QD signal is rapidly lost in actively growing cultures due to dilution by cell division. # Conclusions We have shown here that protein-coated ZnS:Mn nanocrystals can translocate in a dose-dependent manner to the cytoplasm of *E*. *coli*. The process requires transient destabilization of the cell outer membrane and is reminiscent of bacterial transformation. Once in the cytoplasm, biofabricated QDs do not cause a significant induction of the unfolded protein or SOS responses. However, they lead to oxidative stress when supplied at very high concentrations (2.5 μg/mL). Although internalized QDs are stable over a broad range of temperature in quiescent cells, they are rapidly diluted in dividing cells. Taken together, our results suggest that biomimetic fluorophores designed with low toxicity cores and biologically-relevant shells are unlikely to cause significant damage to the microbial ecosystem. These design principles may prove useful for the production of other environmentally benign nanomaterials. # Materials and Methods ## QD uptake by competent cells AB734, an *E*. *coli* K-12 strain containing a mutation in the *lacZ* gene but otherwise wild type was obtained from the *E*. *coli* Genetic Stock Center. To prepare competent cells, 500 mL cultures were grown in LB medium at 37°C to *A*₆₀₀ ≈ 0.4, and cells were sedimented by centrifugation at 8,000*g* for 8 min and resuspended in 100 mL of 100 mM CaCl₂ or phosphate buffered saline (PBS;150 mM NaCl, 10 mM Na₂HPO₄, 2 mM KH₂PO₄). After 30 min incubation on ice and centrifugation at 8,000*g* for 8 min, cells were taken into 12.5 mL of 100 mM CaCl₂ (or PBS for a non-competent control) and held on ice overnight. Glycerol was added to a 10% (v/v) final concentration and aliquots (200 μL) were stored at -80°C for future use. For uptake experiments, competent or control cells were thawed at room temperature, washed twice with PBS with intervening cycle of centrifugation at 4,000 rpm for 5 min in a microfuge, and resuspended in 900 μL of the same buffer. QDs (approximately 100 μL for a dose of 0.5 μg/mL) were added and the mixture was incubated at room temperature for 2h without shaking. Cells were washed twice with PBS to remove unincorporated QDs. ## QD subcellular localization Cells that had uptaken QDs were stripped of their outer membrane and peptidoglycan layer by spheroplasting. Briefly, samples prepared as above were resuspended in 200 μL of buffer A (100 mM Tris-HCl, pH 8.0, 0.5 M sucrose, 0.5 mM EDTA), and 10 μL of a 2 mg/mL solution of lysozyme was added, followed by 400 μL buffer A, and 400 μL of ddH₂O. After 20 min at room temperature, spheroplasts were recovered by centrifugation at 12,800*g* for 30 s and resuspended in 100 mM Tris-HCl, pH 8.0, 0.3 M sucrose, 10 mM MgCl₂. Samples were visualized on an optical microscope at 50x magnification. ## FRET experiments AB734 (pmCherry-mut2) cultures were grown to *A*₆₀₀ = 0.4 at 37°C in LB medium supplemented with 50 μg/mL kanamycin. Production of mCherry was induced by addition of 0.2% L-arabinose and cultures were collected after 3 h of growth at 37°C. Cells were made competent by CaCl₂ and ice treatment as above and stored in 200 μL aliquots. After two wash cycles and resuspension in 900 μL PBS, one sample was incubated for 2 h with 100 μL of BB- CT43-stabilized QDs or the same volume of PBS to serve a control. After 2 wash cycles with PBS, samples were diluted 20-fold in PBS and fluorescence emission spectra were recorded with excitation at 280 nm or 590 nm. Control samples of AB734 cells lacking the pmCherry-mut2 plasmid and incubated or not with QDs were prepared as above and fluorescence spectra were recorded following excitation at 280 nm. The spectra of show QD-free AB734 emission subtracted from QD-treated AB734 emission with excitation at 280 nm (orange), and QD-free mCherry-producing AB734 subtracted from QD-treated mCherry-producing AB734 with excitation at 280 nm (red). ## QD fate Aliquots (200 μL) of AB734 cells that had uptaken QDs as above were used to inoculate 2 mL of LB media in multiple 15 mL culture tubes supplemented or not with 50 μg/mL of the translational inhibitor kanamycin. Cultures were transferred to 37°C water bath. At the indicated time points, culture absorbance was recorded at 600 nm and samples (2 mL) were subjected to centrifugation at 5,000 rpm for 5 min in a microfuge. Cells were resuspended in 50 μL of PBS, deposited on quartz microscope slide and photographed on a UV table with excitation at 303 nm. Median fluorescence in square areas encompassing about 60% of the droplets and excluding their edges was quantified in the red channel using the histogram function of Adobe Photoshop. Fluorescence loss was quantified by subtracting the fluorescence of control samples from that of QD- loaded cells at the indicated time points. ## Stress responses Strains ADA110 (AB734 λϕ*ibp*::*lacZ*)) and ADA510 (AB734 λϕ*sulA*::*lacZ*)) have been described previously. ADA710 (AB734 λϕ*katG*::*lacZ*)) was constructed by lysogenizing AB734 with a bacteriophage λ derivative bearing the oxidative stress-responsive *katG*::*lacZ* translational fusion and isolated from BGF931 (a kind gift from Dr. Gonzalez-Flecha) through standard protocols. The three strains were made chemically competent by CaCl₂ treatment and incubated or not with QDs as above. After resuspension in buffer, 200 μL of culture was used to inoculate 5 mL of LB medium. Samples were either exposed to buffer (negative control), known stress response inducers (4% ethanol, 15 μg/mL nalidixic acid, or 10 μM H₂O₂) or 0.5 μg/mL or 2.5 μg/mL of BB-CT43-stabilized QDs. After 3h incubation at 37°C, cells were lysed and β-galactosidase activities determined as described. ## Analytical techniques UV-visible absorption spectra were recorded on a Beckman DU640 spectrophotometer. Fluorescence and phosphorescence emission spectra were recorded using 1 mL of sample on a Hitachi F4500 fluorescence spectrophotometer with excitation at 280 nm and excitation and emission slit widths set at 2.5 nm (fluorescence) or excitation at 316 nm and excitation and emission slit width at 2.5 nm and 10 nm, respectively (phosphorescence). The wavelength region corresponding to the second order diffraction peak of the excitation light was omitted. # Supporting Information We are grateful to Elyse Shapiro for constructing ADA710. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FB BJFS. Performed the experiments: BJFS. Analyzed the data: FB BJFS. Wrote the paper: FB BJFS.

# Introduction Recent investigations into the pathogenesis of colorectal cancer (CRC) in the general population have indicated that 15% to 35% of CRCs arise through the serrated neoplasia pathway, which differs from the adenoma–carcinoma sequence. Among serrated polyps, sessile serrated lesions (SSLs) and traditional serrated adenomas (TSAs) are considered premalignant lesions. In the general population, the prevalences of SSLs and TSAs are 5%–15% and \<1%, respectively. *BRAF* mutations, CpG island hypermethylation phenotype (CIMP)-positive status, and microsatellite instability are associated with SSL development, whereas *KRAS* mutations are involved in TSA development. Ulcerative colitis (UC) is associated with an increased risk of developing CRC. CRC in patients with UC develops through a carcinogenesis pathway distinct from the dominant pathway in sporadic CRC. Risk factors for CRC include a long disease duration, extensive colitis, and more severe or persistent inflammation. Serrated polyps in patients with UC have been reported in several recent articles. In these studies, serrated polyps were found in 1.2%–1.7% of patients with UC and accounted for 11%–23% of the neoplasias in patients with UC. Serrated epithelial change has been described as a potential risk factor and precursor of colorectal dysplasia and cancer in patients with UC. Serrated polyps in patients with UC have been recognized as histologically and biologically distinct from serrated epithelial change, which shows a high frequency of *TP53* mutations and low frequency of *KRAS/BRAF* mutations. Ko et al. examined serrated polyps in 78 patients with inflammatory bowel disease (IBD), including 56 with UC. The authors reported that SSLs occurred mainly in the proximal colon and contained the *BRAF* mutation, whereas TSAs occurred mainly in the distal colon of men and contained *KRAS* mutations. Miller et al. examined TSA-like lesions in 30 patients with IBD, including 22 with UC, and reported that *KRAS* and *BRAF* mutations were detected in 59% and 16% of the patients, respectively. With the increasing number of reports of serrated polyps in patients with UC, these polyps have become recognized as dysplastic variants that are histologically distinct from conventional dysplasia. The most recent histological studies classified serrated polyps in patients with UC into three subtypes: SSL-like dysplasia, TSA-like dysplasia, and serrated dysplasia not otherwise specified (SD NOS). However, the histological terminology differed among the previous studies, and few studies have examined the clinical characteristics (including risk factors for development) and biological characteristics (including the *KRAS*, *BRAF*, and CIMP status) of serrated polyps in patients with UC based on such accurate histological classification. Therefore, the present study was performed to clarify the yet unknown prevalence, risk factors, and clinical and biological characteristics of serrated polyps in patients with UC. # Materials and methods ## Data collection, patients, and neoplasia Consecutive patients with UC who underwent total colonoscopy at Yokohama City University Medical Center, a tertiary IBD center in Japan, from 2000 to 2020 were identified in this retrospective single-center study. Colonoscopies were performed by experienced endoscopists with knowledge of IBD. In particular, for patients with a \>7-year disease duration, surveillance colonoscopies with panchromoendoscopy were performed by one of three experienced endoscopists (MN, TO, or RK). The inclusion criteria for endoscopic resection in UC patients were as follows: neoplasias that were well-circumscribed endoscopically, neoplasias that had no evidence of invisible dysplasia in the surrounding mucosa on the basis of confirmational biopsies, and neoplasias that had no evidence of submucosal invasion. When the lesions were diagnosed as neoplasia on the basis of endoscopic observation or biopsy, endoscopic resection was performed regardless of size. When the neoplasia could not be resected endoscopically, surgical resection was selected. Established UC databases and endoscopy and pathology reports were reviewed to ensure complete case capture. Patients in whom the diagnosis of UC was uncertain, or who had an IBD unclassified status or underwent only flexible sigmoidoscopy, were excluded. Among the patients with UC who underwent total colonoscopy, data for those with colonic neoplasia were extracted. Indefinite for dysplasia and hyperplastic polyps were not considered neoplastic. Data on the demographic and clinical parameters (sex, age at diagnosis of neoplasia and UC, disease duration, disease extent, course of disease, location and endoscopic features of the neoplasia, and treatment of the neoplasia) were obtained from the medical charts. To compare the clinical and endoscopic findings between serrated polyps in patients with UC and without IBD, we reviewed 249 consecutive serrated polyps in patients without IBD (216 SSLs and 30 TSAs) that were resected endoscopically or surgically at Yokohama City Medical Center during the study period. ## Classification and histological evaluation of neoplasia All pathological slides of neoplasia were re-reviewed and classified in accordance with the most recent World Health Organization classification by two expert gastrointestinal pathologists (YA and SC); one of whom (YA) was highly specialized in UC-associated dysplasia or cancer. We classified neoplasia into five categories on the basis of the histological findings and whether the neoplasia was located in colitis-affected segments, as follows: serrated polyps in colitis-affected segments, conventional dysplasia (intestinal-type dysplasia in colitis-affected segments), serrated polyps in colitis-unaffected segments (usually proximal to the extent of the colitis), sporadic adenomas in colitis-unaffected segments, and invasive carcinoma. Additionally, serrated polyps in colitis-affected segments were histologically classified into three subtypes in accordance with previous studies, as follows: SSL-like dysplasia, TSA-like dysplasia, and serrated SD NOS. Briefly, SSL-like dysplasia is characterized by distorted serrated crypts with prominent basal crypt dilatation (i.e., dilated L- or inverted T-shaped crypts) at the interface with the muscularis mucosa. TSA-like dysplasia is characterized by a villiform growth pattern with columnar cells with intensely eosinophilic cytoplasm and ectopic crypts, creating a prominent serrated profile. SD NOS was defined as serrated dysplasia without definite features of SSL-like or TSA-like dysplasia, with a complex serrated architecture and evidence of dysplasia. The representative endoscopic and histological features of SSL-like dysplasia, TSA- like dysplasia, and SD NOS are shown in. Serrated polyps in colitis-unaffected segments were classified as hyperplastic polyps, SSLs, TSAs, or unclassified serrated adenomas. ## Definitions Persistent active colitis was defined as endoscopically active colitis lasting \>6 months. Whether neoplasia was located in colitis-affected segments was determined on the basis of clinical and endoscopic data for the most active inflammation during the course of UC treatment. The proximal colon comprised the cecum, ascending colon, and transverse colon, whereas the distal colon comprised the descending colon, sigmoid colon, and rectum. ## DNA extraction and genetic and epigenetic analyses DNA samples were purified from archived formalin-fixed, paraffin-embedded blocks of serrated polyps and invasive cancer that had been endoscopically or surgically resected from patients with UC. The pathologists selected the appropriate tissue blocks for DNA extraction. We used a laser microdissection system for selective isolation of the neoplastic sections, avoiding foci of inflammation, and then extracted the DNA. Neoplasia specimens containing adequate DNA quantity and quality were provided for genetic analysis. *KRAS* and *BRAF* mutations were detected using a droplet digital polymerase chain reaction system. The CIMP status was evaluated in accordance with previous reports \[, \]. The details of the laser microdissection system, DNA extraction, and genetic and epigenetic analyses are shown in. ## Statistical analysis Data were analyzed using JMP Pro 12 (SAS Institute Inc., Cary, NC). The prevalence and clinical characteristics were compared among neoplasia groups using Fisher’s exact test or the Wilcoxon rank sum test. Statistical significance was set at *P* \< 0.05. ## Ethical considerations This study was approved by the Ethics Committee of Yokohama City University Medical Center (Protocol number: A130926011). All patients with UC whose neoplastic DNA was extracted and genetically analyzed provided written informed consent in accordance with the tenets of the Declaration of Helsinki. An opt-out for the present study was published on the Web. # Results ## Prevalence of serrated polyps in patients with UC shows an overview of the neoplasias among the consecutive patients with UC included in the present study. During the study period, 2035 patients with UC underwent total colonoscopy, and 252 neoplasms from 187 patients were identified. Of 219 neoplasms (after excluding 33 invasive cancers), 26 serrated polyps and 132 cases of conventional dysplasia were observed in colitis-affected segments, whereas 10 serrated polyps and 51 sporadic adenomas were found in colitis-unaffected segments. ## Clinical characteristics of patients with serrated polyps and other neoplasms shows the endoscopic characteristic of serrated polyps in patients with UC and without IBD. Serrated polyps in patients with UC were more frequent in the distal colon and significantly smaller in size compared with those in patients without IBD (42% vs. 24%, respectively, *P* = 0.04, and 10 mm vs. 16 mm, respectively, *P* \< 0.01). However, there was no significant deference in neoplasia morphology and border description. Regarding the subtype of serrated polyps (SSL/SSL-like dysplasia and TSA/TSA-like dysplasia), there were no significant differences in neoplasia location, neoplasia morphology, and border description ( and Tables). shows the patient and neoplasia characteristics of serrated polyps in colitis- affected and -unaffected segments. shows representative results of the comparisons among the categories of neoplasia (serrated polyps in colitis- affected and colitis-unaffected segments, conventional dysplasia in colitis- affected segments, and sporadic adenomas). The detailed characteristics of each category are shown in. The age at diagnosis of neoplasia, age at diagnosis of UC, and duration of UC did not differ between patients with serrated polyps and those with conventional dysplasia found in colitis-affected segments (50 vs. 57 years, *P* = 0.11; 37 vs. 40 years, *P* = 0.055; and 12.1 vs. 10.0 years, *P* = 0.21, respectively). However, the age at diagnosis of neoplasia and age at diagnosis of UC in patients with sporadic adenomas (67 and 52 years, respectively) were significantly older than those in patients with each neoplasia in colitis-affected segments (*P* values are shown). Serrated polyps in colitis-affected segments were more frequent in men than in women, while those in colitis-unaffected segments were more frequent in women than in men (percentages in men: 88% vs. 20%, respectively; *P* \< 0.001). Patients with either type of neoplasia in colitis-affected segments were more likely to have extensive colitis (serrated polyps: 88%, conventional dysplasia: 74%) and a history of persistent active colitis (serrated polyps: 58%, conventional dysplasia: 52%) compared with patients with either type of neoplasia in colitis-unaffected segments (serrated polyps: 40% and 0%, respectively; sporadic adenomas: 10% and 22%, respectively) (*P* values are shown). Neoplasia size was similar (serrated polyps in colitis-affected segments: 9 \[interquartile range: –\] mm, conventional dysplasia: 9 mm, serrated polyps in colitis-unaffected segments: 10 mm), with the exception of sporadic adenomas (4 mm) (*P* values are shown). Regarding previous treatment, immunomodulators were used significantly more frequently in patients with serrated polyps in colitis-affected segments than in those with polyps in colitis-unaffected segments. ## Clinical characteristics of serrated polyps in colitis-affected segments Of 26 serrated polyps in colitis-affected segments, 15 (58%), 6 (23%), and 5 (19%) were categorized as SSL-like dysplasia, TSA-like dysplasia, and SD NOS, respectively. The characteristics of these polyps are shown in. All subtypes of serrated polyps in colitis-affected segments were common in men. SSL-like dysplasia was common in the proximal colon, whereas TSA-like dysplasia and SD NOS were common in the distal colon. Although statistical analysis was not performed owing to the small number of samples, SD NOS was more likely to be associated with large neoplasia size (23 mm) than with SSL-like dysplasia (12 mm) or TSA-like dysplasia (8 mm). TSA-like dysplasia was more likely to have a polypoid morphology (67%) compared with SSL-like dysplasia (7%) and SD NOS (20%). ## Locational and biological characteristics of serrated polyps in patients with UC shows the locational distribution of serrated polyps in patients with UC, including 26 polyps in colitis-affected segments and 10 polyps (9 SSLs and 1 TSA) in colitis-unaffected segments. SSL-like dysplasia was distributed in all segments of the colorectum, and 10 lesions (67%) were located in the proximal colon. All cases of TSA-like dysplasia were located in the distal colon (2 \[33%\] in the rectum and 4 \[67%\] in the sigmoid colon). Four (80%) SD NOS lesions were located in the distal colon (one in the descending colon and three in the rectum), and one was located in the proximal colon. In contrast, all serrated polyps in colitis-unaffected segments were located in the proximal colon. Of 36 serrated polyps in both colitis-affected and -unaffected segments in patients with UC, genetic and epigenetic evaluations were performed in 18 specimens (13 affected and 5 unaffected) from which DNA of adequate quality and quantity was obtained. *BRAF* mutations were observed in 75% (3/4) of cases of SSL-like dysplasia in colitis-affected segments and in all SSLs in colitis- unaffected segments, whereas *KRAS* mutations were observed in all cases of TSA- like dysplasia in colitis-affected segments and in all TSAs in colitis- unaffected segments. In SD NOS, 75% (3/4) of the specimens showed *KRAS* mutations and 25% (1/4) showed *BRAF* mutations. CIMP-positive status for serrated polyps in colitis-affected segments was observed in 60% (3/5) of cases of TSA-like dysplasia but in only 25% of cases of SSL-like dysplasia and SD NOS. The genetic and epigenetic status of 16 invasive cancers in colitis-affected segments were also evaluated. *KRAS* and *BRAF* mutations and CIMP-positive status were observed in 13% (2/16), 0% (0/16), and 19% (3/16) of invasive cancers in colitis-affected segments. # Discussion This study showed the prevalence of serrated polyps in patients with UC and is the first, to our knowledge, to suggest the increasing prevalence of these polyps, which have been observed more frequently recently. The clinical and biological characteristics of SSL-like dysplasia and TSA-like dysplasia in patients with UC are basically similar to those of each counterpart in individuals without IBD. However, SSL-like dysplasia in patients with UC was more frequently observed in men than in women, in the present study, a finding that contrasts the known female dominance of SSLs in individuals without IBD. Although many other issues remain unknown, our findings suggest a new pathway in the development of UC-associated neoplasia. The present study showed that the prevalence of serrated polyps was 1.8% (36/2035) in patients with UC, and these polyps accounted for 14% (36/252) of all neoplasias in patients with UC. This proportion is similar to that in a previous study that reported a prevalence of serrated polyps of 1.2%–1.7% in patients with IBD. Additionally, serrated polyps accounted for 11%–23% of all neoplasias in the study. Regarding ethnicity, the present study is the first, to our knowledge, to investigate the prevalence of serrated polyps in the Asian population. Meta- analyses of the non-IBD population have reported that the prevalence of serrated polyps in Western countries was higher than those in Eastern countries. The possible reason for the similar proportions of serrated polyps in patients with UC between the present study and previous studies is differences in the study periods. Two previous studies included serrated polyps detected between 2005 and 2007 and 2000 and 2013. Because serrated polyps (SSLs and TSAs) were considered hyperplastic polyps before 2010, SSLs and TSAs may have been overlooked without biopsy or endoscopic resection. To clarify the differences in the prevalence between Eastern and Western countries, an accumulation of recent cases in both ethnic groups is desirable. Previous studies have shown that serrated polyps in patients with UC have clinical and biological characteristics similar to those in individuals without IBD. For instance, SSLs are usually found in women and in the proximal colon, whereas TSAs are usually found in men and in the distal colon. Our results also showed similar neoplasia location, neoplasia morphology, and border description for both SSLs/SSL-like dysplasias and TSAs/TSA-like-dysplasias in patients with UC and without IBD. The difference in the neoplasia size between the groups may be associated with different indications for endoscopic resection between patients with UC and without IBD. Notably, our results demonstrated a unique clinical characteristic in SSL-like dysplasia in colitis-affected segments. SSLs in colitis-unaffected segments showed clinical and biological similarity to those in patients without IBD (all SSLs in colitis-unaffected segments were found in women and in the proximal colon). In contrast, most cases of SSL-like dysplasia in colitis-affected segments were found in men, although their locational and biological characteristics were similar to those of SSLs in colitis-unaffected segments (predominantly in the proximal colon and with *BRAF* mutation). These sex-related differences may arise from differences in the biological mechanisms of development between serrated pathways with and without background inflammation. In this context, although previous studies indicated that SSLs in patients with UC were dominant in women and in the proximal colon, the distribution of SSLs was not always based on the presence or absence of background inflammation. Conversely, TSA-like dysplasia in colitis-affected segments showed clinical and biological characteristics similar to those of TSAs in colitis-unaffected segments (predominantly in men, in the distal colon, and with *KRAS* mutations). These results suggest that the specific development of SSL-like dysplasia in men with UC is derived from a unique tumorigenic pathway. Serrated polyps in colitis-affected segments were more common in patients with a long disease duration and a history of persistent active colitis than in patients without these characteristics, and this was also true in the development of UC-related dysplasia. Therefore, chronic inflammation must be involved in the development of both dysplasia and serrated polyps in patients with UC. Notably, both patients who have IBD with dysplasia and patients without IBD who have serrated polyps are likely to develop synchronous and metachronous CRCs. Considering this similarity, serrated polyps in patients without IBD may also develop on a background of chronic inflammation. In fact, previous studies have shown that the concentrations of some key inflammatory factors, such as tumor necrosis factor α, cyclooxygenase-2, interleukin-4, and interleukin-1β, are increased in patients with serrated polyps without IBD. These studies suggest that chronic inflammation is correlated with the development of serrated polyps in both patients with UC and individuals without IBD. Long-term follow-up of our patients and other observational studies are needed to further examine the multiple developmental pathways of serrated polyps in patients with UC. Recently, SD NOS was newly described as serrated polyps without definite features of SSL-like and TSA-like dysplasia. However, only one study reported a small number of cases of SD NOS, and the clinical and biological characteristics were not assessed. The present study is the first, to our knowledge, to show the clinical and biological characteristics of SD NOS. Our study showed a similar distribution (predominantly located in the distal colon) and similar biological characteristics between TSA-like dysplasia and SD NOS, suggesting that SD NOS is a subtype of TSA-like dysplasia. Additionally, considering the longer duration of UC and larger size of SD NOS compared with the duration and size of TSA-like dysplasia, SD NOS may be an enlarged lesion of TSA-like dysplasia induced by long-term inflammation. Similarly, Ko et al. reported that 32% of the serrated polyps in patients with IBD were TSA-like, and this rate is consistent with our rate of the combination of TSA-like dysplasia and SD NOS (42%). Although the reason for the higher proportion of TSA-like dysplasia (with SD NOS) than the proportion of TSAs in patients without IBD is unclear, chronic inflammation may also be involved in the development of TSA-like dysplasia. No previous studies have assessed the CIMP status of serrated polyps in patients with UC. Although CIMP-positive status is important in the serrated neoplasia pathway, recent studies showed that CIMP-positive status was exclusively associated with SSLs in the proximal colon and advanced age. The low rate of a CIMP-positive status of SSL-like dysplasia in our study was considered to be associated with the neoplasia location (33% were in the distal colon) and relatively young age of our study population. In patients without IBD, TSAs in the proximal colon frequently show *BRAF* mutation and CIMP positivity, whereas TSAs in the distal colon show *KRAS* mutations and CIMP negativity. In our study, although all cases of TSA-like dysplasia were located in the distal colon and showed *KRAS* mutations, 60% (3/5) were also CIMP-positive. The reason for this discrepancy is unclear, but TSA-like dysplasia arising from chronic inflammatory mucosa may develop through a pathway distinct from TSAs in non- inflamed mucosa. The proportion of *KRAS* and *BRAF* mutations and CIMP-positive status in serrated polyps in colitis-affected segments tended to be higher in invasive cancer in colitis-affected segments in this study. These results suggest that most serrated polyps may not be precursors of colitis-associated invasive cancers. Although the optimal management of serrated polyps in patients with UC is still unclear, our results suggest that colectomy is excessive for serrated polyps in colitis-affected segments. However, considering that serrated polyps are known precursors of CRCs in patients without IBD, serrated polyps in patients with UC should be treated endoscopically as for those in patients without IBD. Although there was no UC-associated dysplasia within SSLs or TSAs in the present study, high-grade dysplasia with a hyperplastic polyp was detected in a 19-year-old man in our hospital (this case was not included in the present study because the lesion did not meet the diagnostic criteria for SSL, TSA, or SD NOS). This case suggests that some serrated polyps in patients with UC, although rare, have malignant potential. To determine the appropriate management of serrated polyps in patients with UC, further accumulation of cases is desirable. Our study has some limitations. The setting was a single center, and the study had a small sample size, which included only Japanese patients. Undoubtedly, the number of serrated polyps in patients with UC was smaller than those of conventional dysplasia and sporadic adenomas. Another limitation is the lack of data regarding a family histology of CRC, obesity, smoking, and alcohol intake, which are risk factors for SSLs. # Conclusions The present study showed the detailed clinicopathological and biological characteristics of serrated polyps in patients with UC. Serrated polyps in colitis-affected segments were common in men with extensive colitis and a long duration of UC, suggesting that chronic inflammation might be involved in the development of serrated polyps in patients with UC. # Supporting information 10.1371/journal.pone.0282204.r001 Decision Letter 0 Suzuki Hiromu Academic Editor 2023 Hiromu Suzuki This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 3 Jan 2023 PONE-D-22-31131Serrated Polyps in Patients with Ulcerative Colitis: Prevalence and Unique Clinicopathological and Biological CharacteristicsPLOS ONE Dear Dr. Kunisaki, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. This manuscript was carefully reviewed by 2 experts. Although both reviewers highly evaluated this study, they suggested several points which need to be addressed before acceptance. For instance, it is important to see whether endoscopic findings of serrated lesions in UC-patients differ from those in non- UC patients. Reviewer 1 also suggested additional mutation and methylation analyses, while the editor considers it is not mandatory because KRAS, BRAF mutations and CIMP status are already determined in the samples. Reviewer 2 indicated possible difficulties in comparing samples collected in different time periods because of the inconsistent clinical definition of serrated lesions. Reviewer 2 also suggested additional molecular analysis in in invasive cancer. Analysis of KRAS/BRAF mutations and CIMP status in the invasive cancers may help us to understand whether those cancers develop through the serrated pathway. Please respond to each of the reviewer comments. Please submit your revised manuscript by Feb 17 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Hiromu Suzuki, M.D., Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf>. 2\. You indicated that you had ethical approval for your study. Please clarify whether minors (participants under the age of 18 years) were included in this study. If yes, in your Methods section, please ensure you have also stated whether you obtained consent from parents or guardians of the minors included in the study or whether the research ethics committee or IRB specifically waived the need for their consent. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The paper is potentially interesting and seems to be convincing, however, I have several concerns with this paper as shown below. 1\. Comparison should be made between the endoscopic findings of serrated lesions in UC patients compared to non-UC patients. 2\. The authors should determine mutations of driver genes of colorectal serrated lesions such as APC, GNAS, and RNF43. 3\. The authors should analyze all 36 serrated lesions for genetic alterations and methylation statuses. 4\. The difference between Asian patients and non-Asian patients is of interest of readers. Are there ethnic differences? The authors are encouraged to add more discussion in details on this point. 5\. Grammatical and typographical errors in the text need to be corrected by a native English speaker. Reviewer \#2: In this manuscript, the authors study clinical and molecular features of serrated polyps in patients with ulcerative colitis (UC), retrospectively. Clinicopathological and biological characteristics of serrated polyps in colitis-affected segments is potentially interesting, but several concerns need to be addressed. 1, Neoplasias enrolled in this study was classified into five categories based on histological findings. Were all neoplasias endoscopically or surgically resected? In fig3G, median size of sporadic adenoma was less than 5mm. Criteria of endoscopic resection should be shown. 2, It is difficult to compare the prevalence of serrated polyps in patients with UC from 2000 to 2009 and from 2010 to 2020. Endoscopic findings of serrated polyp were not established, and almost serreted polyps were classified into hyperplastic polyp before 2010, as the authors are discussing. If the authors want to show the increasing incidence of serrated polyps, the data after 2010 should be used. 3, Molecular features of serrated polyp in colitis affected and unaffected segments were interesting. Molecular features of invasive cancer should be examined. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0282204.r002 Author response to Decision Letter 0 26 Jan 2023 Responses to the Reviewers’ Comments Reviewer \#1: The paper is potentially interesting and seems to be convincing, however, I have several concerns with this paper as shown below. 1\. Comparison should be made between the endoscopic findings of serrated lesions in UC patients compared to non-UC patients. Response: We thank Reviewer \#1 for the insightful comment, which has helped us improve the Results and Discussion sections of our manuscript. As Reviewer \#1 pointed out, a comparison of the clinical and endoscopic characteristics between serrated polyps in patients with UC and without IBD is important. Accordingly, we reviewed 249 consecutive serrated polyps in patients without IBD (216 SSLs and 30 TSAs) that were resected endoscopically or surgically at our hospital during the study period. We compared the endoscopic characteristics of serrated polyps in patients with UC and without IBD. Serrated polyps in patients with UC were more frequent in the distal colon and significantly smaller in size than those in patients without IBD (42% vs. 24%, respectively, P = 0.04, and 10 mm vs. 16 mm, respectively, P \< 0.01). However, there was no significant deference in neoplasia morphology and border description. Next, we compared the endoscopic characteristics on the basis of the subtype of serrated polyps (SSL/SSL-like dysplasia and TSA/TSA-like dysplasia). There was no significant deference in neoplasia location, neoplasia morphology, and border description between patients with UC and without IBD for both SSL/SSL-like dysplasia and TSA/TSA-like dysplasia. These results support the finding in previous studies that serrated polyps in patients with UC were similar to those in patients without IBD. However, our results showed a unique characteristic of SSL-like dysplasia in colitis-affected segments: these were common in men. We added this information in the methods and discussion of the revised manuscript. We also created new Table 1, S1 Table, and S2 Table. Previous Tables 1, 2, and 3, and S1 Table have been renumbered accordingly. 2\. The authors should determine mutations of driver genes of colorectal serrated lesions such as APC, GNAS, and RNF43. Response: We thank Reviewer \#1 for the insightful comment. As Reviewer \#1 pointed out, APC, RNF43, and GNAS mutations have been reported in addition to KRAS or BRAF mutations in TSAs in patients without IBD. Accordingly, we examined somatic mutations in serrated polyps in colitis- affected segments using next-generation sequencing (NGS). However, sufficient quality and quantity of DNA were obtained in only six samples. After excluding samples that did not meet the inclusion criterion of successful sequencing, only three samples remained (two TSA-like dysplasias and one SD NOS). Among the three samples that could be evaluated, KRAS mutations were detected in all cases, whereas APC and GNAS mutations were not detected. In our study, RNF43 mutations were not evaluated because the Cancer Hotspot Panel v2 (Thermo Fisher Scientific) that we used did not include RNF43. Because only three samples could be examined, the above data were insufficient to discuss somatic mutations other than KRAS and BRAF mutations of serrated polyps in patients with UC. Therefore, we presented only the data for NGS in this response. 3\. The authors should analyze all 36 serrated lesions for genetic alterations and methylation statuses. Response: We thank Reviewer \#1 for the insightful comment. As Reviewer \#1 pointed out, analysis of all 36 lesions is desirable for more accurate genetic and epigenetic characterization. However, in our study, only 18 samples had sufficient quality or quantity of DNA for analysis. Therefore, it is difficult to perform additional examinations. The possible reasons for not obtaining sufficient quality or quantity of DNA are as follows: First, a long time has passed since the samples were collected. Second, before 2017, the specimens were fixed in 20% formalin buffer, which caused tissue damage. We would like to examine the DNA of more cases under improved preservation conditions, in the future. 4\. The difference between Asian patients and non-Asian patients is of interest of readers. Are there ethnic differences? The authors are encouraged to add more discussion in details on this point. Response: We thank Reviewer \#1 for the insightful comments, which have helped us improve the Discussion section of our manuscript. As Reviewer \#1 have pointed out, the prevalence of serrated lesions in the non- IBD population differs between ethnic groups, and it is very important to consider ethnic differences in serrated polyps in patients with UC. To the best of our knowledge, no study has compared the prevalence of serrated polyps in patients with UC between Eastern and Western countries. All previous studies that reported the prevalence of serrated lesions in patients with UC were from Western countries. Therefore, to our knowledge, the present study is the first to report the prevalence in the Asian population. Two meta-analyses of the non-IBD population reported that the prevalence of serrated polyps in Western countries was higher than those in Eastern countries. In comparison, the prevalence of serrated polyps in patients with UC in our study (1.8%) was similar to those in previous studies from Western countries (1.2%–1.7%). The possible reason for this similarity is differences in the study periods. The two previous studies included serrated polyps detected between 2005 and 2007 and 2000 and 2013, respectively. Because serrated polyps (SSLs and TSAs) were considered hyperplastic polyps before 2010, SSLs and TSAs may have been overlooked without biopsy or endoscopic resection. To clarify the differences in the prevalence between Eastern and Western countries, the accumulation of more recent cases in both ethnic groups is desirable. We added the above information in the revised discussion. 5\. Grammatical and typographical errors in the text need to be corrected by a native English speaker. Response: We thank Reviewer \#1 for the insightful comment, which has helped us improve our manuscript. The revised manuscript has been edited by a native English-speaking Medical Editor. Reviewer \#2: In this manuscript, the authors study clinical and molecular features of serrated polyps in patients with ulcerative colitis (UC), retrospectively. Clinicopathological and biological characteristics of serrated polyps in colitis-affected segments is potentially interesting, but several concerns need to be addressed. 1, Neoplasias enrolled in this study was classified into five categories based on histological findings. Were all neoplasias endoscopically or surgically resected? In fig3G, median size of sporadic adenoma was less than 5mm. Criteria of endoscopic resection should be shown. Response: We thank Reviewer \#2 for the insightful comments. All neoplasias evaluated in the present study were resected endoscopically or surgically. As Reviewer \#2 pointed out, the Japanese guidelines recommend endoscopic resection for adenomas ≥ 6 mm in size or for superficial depressed-type lesions even when the lesion measures ≤ 5 mm in non-IBD patients. However, no indication for endoscopic resection on the basis of size has been established for colitis- associated dysplasia. In patients with UC, the distinction between sporadic adenoma and colitis-associated dysplasia is sometimes difficult even after observation with magnifying endoscopy and biopsy. Therefore, in our hospital, when the lesions are diagnosed as neoplasia on the basis of endoscopic observation or biopsy, the lesions are resected regardless of size. In this study, the inclusion criteria for endoscopic resection in UC patients were as follows: neoplasias that were well-circumscribed endoscopically, neoplasias that had no evidence of invisible dysplasia in the surrounding mucosa with confirmational biopsies, and neoplasias that had no evidence of submucosal invasion. When the neoplasia could not be resected endoscopically, surgical resection was performed. We added the criteria for endoscopic resection in patients with UC in the Methods section. 2, It is difficult to compare the prevalence of serrated polyps in patients with UC from 2000 to 2009 and from 2010 to 2020. Endoscopic findings of serrated polyp were not established, and almost serreted polyps were classified into hyperplastic polyp before 2010, as the authors are discussing. If the authors want to show the increasing incidence of serrated polyps, the data after 2010 should be used. Response: We thank Reviewer \#2 for the insightful comments. As Reviewer \#2 pointed out, data after 2010 should be used to investigate the increasing incidence of serrated polyps. However, we think that it is inappropriate to discuss the increasing incidence of serrated polyps in post-2010 data on the basis of the present data because of the short study period. Additionally, considering that the definition of serrated polyps was revised in 2019, further accumulation of serrated polys detected after 2019 is needed to investigate the prevalence of serrated polyps. Therefore, we deleted text that mentioned the increasing incidence of serrated polyps in the abstract, results, and discussion. We also deleted original Fig 2, and current Figs 3 and 4 have been renumbered accordingly. 3, Molecular features of serrated polyp in colitis affected and unaffected segments were interesting. Molecular features of invasive cancer should be examined. Response: Reviewer \#2 for the insightful comments, which have helped us improve the results and discussion in our manuscript. In accordance with your recommendation, we examined the KRAS and BRAF mutations and CIMP-positive status of invasive caner in colitis-affected segments. Sixteen invasive carcinoma samples with adequate DNA quality and quantity were examined. KRAS and BRAF mutations and CIMP-positive status were observed in 13% (2/16), 0% (0/16), and 19% (3/16) of invasive carcinomas, respectively, in colitis-affected segments. Considering that the proportions of KRAS and BRAF mutations and CIMP-positive status of invasive cancers tended to be lower than those of serrated polyps in colitis-affected segments, most serrated polyps may not be precursors of colitis-associated invasive carcinoma. These results suggest that colectomy is excessive for serrated polyps in colitis-affected segments. However, considering that serrated polyps are known precursors of CRCs in patients without IBD, serrated polyps in patients with UC should be treated endoscopically, as for those in patients without IBD. In our hospital, high-grade dysplasia with a hyperplastic polyp was detected in a 19-year-old man (this case was not included in the present study because the lesion did not meet the diagnostic criteria for SSL, TSA, and SD NOS). This case suggests that some serrated polyps in patients with UC, although rare, have malignant potential. To determine the appropriate management of serrated polyps in patients with UC, further accumulation of cases is desirable. We added the details of these additional examinations in the methods, results, and discussion. We also created a new supplementary table (S4 Table) explaining the results the of genetic and epigenetic examinations of invasive carcinoma. Additionally, we revised the total number and percentage of CIMP-positive serrated polyps in colitis-affected segments in Table 4 (original Table 3) because of a text error in the original manuscript (wrong: 6 (50%) → correct: 5 (38%)). 10.1371/journal.pone.0282204.r003 Decision Letter 1 Suzuki Hiromu Academic Editor 2023 Hiromu Suzuki This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 10 Feb 2023 Serrated polyps in patients with ulcerative colitis: Unique clinicopathological and biologial characteristics PONE-D-22-31131R1 Dear Dr. Kunisaki, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Hiromu Suzuki, M.D., Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have responded appropriately to my concerns, providing additional data. This reviewer thinks this paper is improved. Reviewer \#2: Clinicopathological and biologial characteristics of serrated polyps in colitis-affected segments is interesting. The revised manuscript is improved and acceptable for publication. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Yasushi Sasaki Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0282204.r004 Acceptance letter Suzuki Hiromu Academic Editor 2023 Hiromu Suzuki This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Feb 2023 PONE-D-22-31131R1 Serrated Polyps in Patients with Ulcerative Colitis: Unique Clinicopathological and Biological Characteristics Dear Dr. Kunisaki: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Hiromu Suzuki Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: ‡ KH, AT, YI, WU, KO, HO, TO, HK, YS and HK also contributed equally to this work.

# Introduction *Plasmodium falciparum* is the species responsible for the vast majority of malaria-related morbidity and mortality. Serious clinical complications frequently arise due to dramatic modification of the structural and functional properties of *P. falciparum*-infected erythrocytes (IEs). During the intraerythrocytic stage, malaria parasites interact with and affect both the plasma membrane and the membrane skeleton of the IEs. Alterations in biochemical, structural and adhesive properties of the host membrane occur as the parasite develops within the erythrocyte. Recently, a large study showed that many parasite proteins contribute at different degrees to remodel the host erythrocyte. Among parasite-secreted proteins, several types of enzymes (like kinases, several still uncharacterized and phosphatases) are trafficked to the erythrocyte membrane indicating that post-translational modifications may contribute to establish successful intracellular parasite proliferation. Phosphorylation of membrane skeleton proteins of both parasite and host origin have been described during *P. falciparum* infections. Selective phosphorylation of host membrane skeleton proteins include protein 4.1, β-spectrin, ankyrin and band 3,. It has also been established that phosphorylation of some of these proteins modulate their interactions with other membrane proteins , and, consequently, the membrane mechanical functions and membrane stability. In addition, erythrocyte membrane skeleton phosphorylation was suggested to be involved in the regulation of malaria parasite invasion and development. However, the molecular events involved in the phosphorylation of membrane skeleton proteins have not yet been identified. Recently we showed that some members of the *apicomplexa*-specific FIKK kinase family are transported to the erythrocyte membrane via Maurer's clefts. *Fikk* is a single copy gene in most *Plasmodium* species but has expanded in *P. falciparum* to 20 related members dispersed mostly on subtelomeric regions of 11 of the 14 parasites chromosomes. Nineteen *Pf fikk* genes possess the *Plasmodium* exported element/host targeting motif downstream of a signal or anchor sequence required for transport across the parasitophorous vacuole. Despite the fact that these proteins share a common structure, the N-terminal regions are highly variable, suggesting that individual members of this family may have access to distinct substrate pools since their variable N- terminal can probably target them to different locations. Due to the restricted homology with well-characterized kinase domains, the FIKK proteins did not cluster within any of the kinase groups described in higher eukaryotes. In this work, we have analyzed the biological role of two members of the *P. falciparum* FIKK kinase family (FIKK7.1 and FIKK12) in IEs. We show that both FIKK kinases are non-essential for parasite growth *in vitro*. However, the absence of functional copies of either *Pf fikk7.1*(*MAL7P1.144*) or *Pf fikk12* (*PFL0040c*) resulted in altered rigidity of the IEs using single cell micromanipulation and both mutant parasites showed changes in the phosphorylation pattern of two distinct proteins of the erythrocyte membrane skeleton. # Results ## Single targeted disruption of two *fikk* genes in *P. falciparum* leads to viable blood stage development We have previously reported that FIKK proteins are exported to different locations in the IE. Using both GFP-tagging and specific antibodies against FIKK12 we noticed that this protein was transported to the erythrocyte membrane. In this initial analysis we also noticed that *Pf fikk7.1* was more than 3-fold up-regulated in ring stage FCR3-CSA-selected (chondroitin sulfate A) parasites when compared to CD36-selected parasites. To investigate the biological role of these two members of the *Pf fikk* gene family, we established two parasite lines with single gene disruption by double crossover recombination. The pHTK- FIKK7.1 and the pHTK-FIKK12 vectors contain the human dihydrofolate reductase (*hdhfr*) gene flanked by 5′ and 3′ segments of the *Pf fikk7.1* and *Pf fikk12* genes, respectively. FCR3 parasites were transfected with the pHTK-FIKK constructs and selected on WR99210 and ganciclovir to generate two insertional disruptant mutants, the FIKK7.1-KO and the FIKK12-KO. After drug selection, the mutants were cloned by limiting dilution and genetically characterized. Clones were screened by polymerase chain reaction (PCR) analysis for the disruption of the *Pf fikk7.1* or *Pf fikk12* gene as well as for the absence of contaminating wild type gene (data not shown). To confirm that the pHTK-FIKK vector had integrated into the respective *Pf fikk* gene, Southern blots were performed using genomic DNA derived from parental FCR3 or recombinant parasites previously digested with AluI to test FIKK12-KO or HindIII to test FIKK7.1-KO. Radiolabelled probes from *Pf fikk7.1* 5′ and 3′ segment or *Pf fikk12* 5′ and 3′ segment were used for hybridization. These hybridizations showed bands of the expected size, indicating that the integration occurred at the predicted site within the *Pf fikk* genes. Transcription of the specific genes was analysed by real-time PCR. No transcript was detected in the respective KO-line; i.e. no *Pf fikk7.1* transcript was detected in the FIKK7.1-KO line and no *Pf fikk12* transcript was detected in the FIKK12-KO line. The KO parasites were viable and there was no obvious difference in the parasite growth and multiplication rates between the KO parasites and wild type FCR3 strain (data not shown), suggesting the FIKK7.1 and FIKK12 proteins are non- essential for the parasite growth of laboratory isolates. ## Analysis of the adhesive properties of IEs and trafficking process in FIKK-KO parasites Given the transit of several FIKK kinases via Maurer's clefts and the presence of a variable N-terminal domain in each protein we hypothesized that the kinase activity of these proteins might be involved in the trafficking of parasite- encoded variant surface molecules, such as the adhesion molecule *Plasmodium falciparum* erythrocyte membrane protein 1 (PfEMP1). To test this hypothesis we analyzed the adhesive properties of the transgenic lines to endothelial cell- expressed receptors by static *in vitro* adhesion assays. The FIKK7.1-KO and FIKK12-KO parasite lines were selected to adhere to human placenta choriocarcinoma Bewo cells expressing CSA receptor or Chinese hamster ovary 745 cell expressing CD36 receptor. The levels of parasite cytoadherence were quantified after the third round of selection. Both KO parasite lines cytoadhered to CSA and CD36 in similar numbers to the wild type parasites, suggesting that the disruption of these *Pf fikk* genes neither prevented the trafficking of PfEMP1 to the surface of the infected erythrocyte membrane nor their ability to bind to different receptors. To examine if the PfEMP1 switching rates in the KO parasites were altered, we performed binding assays with receptors immobilized on plastic over a 2-month period after which the numbers of bound IEs per mm² were quantified. The binding phenotype of the CSA-selected KO parasites was analyzed with two different receptors (purified bovine CSA and human recombinant CD36) at three different intervals: 1 week, 5 weeks and 7 weeks post-selection (Supplementary ). Both KOs and wild type parasites maintained similar CSA-binding phenotypes during the long *in vitro* 7 weeks cultivation and they did not bind to CD36, though as expected the binding efficiency to CSA decreased gradually in the three CSA-selected parasite lines. CD36-selected parasites conserved a strong affinity for CD36 during the all experiment (data not shown). These data suggest that FIKK12 and FIKK7.1 do not interfere with parasite adhesion to CSA and CD36 receptors. Furthermore, the assay we used did not reveal any apparent alteration in the switching rate to other adhesion phenotypes. We also tested the possibility that the FIKK7.1 and/or FIKK12 might be involved in the trafficking of other parasite-encoded variant multi-gene families. To address this question, the localization of Rifin (repetitive interspersed family), Stevor (subtelomeric variable open reading frame) and Surfin (surface- associated interspersed family) proteins was analyzed in the transgenic parasites and compared to wild type parasites by immunofluorescence assay using specific antibodies. The genes of these three families are located close or within the subtelomeric chromosomic region. And different protein members have been localized in the IEs Maurer's clefts. Similar immunofluorescence patterns were detected in IEs with mature stages of FIKK7.1-KO, FIKK12-KO and FCR3 parental parasites. We were unable to detect differences in the Maurer's cleft localizations of A-type Rifin, Surfin 4.2 and Stevor between the KO and wild type lines, suggesting that FIKK7.1 and FIKK12 do not control trafficking of these multigene families into the erythrocyte membrane. ## Erythrocytes infected with mutant FIKK7.1-KO and FIKK12-KO parasites have altered membrane rigidity To evaluate whether FIKK proteins play a role in alteration of the mechanical properties of the IE membrane skeleton and, therefore overall erythrocyte deformability, we measured and compared the shear elastic modulus of uninfected erythrocytes and highly synchronized erythrocytes infected with pigmented trophozoites of either parental FCR3 parasites or mutant parasite lines. Using single cell micromanipulation, we observed that infection of erythrocytes with *P. falciparum* FCR3 caused an approximately 4-fold increase in the rigidity of the erythrocyte membrane. In contrast, the overall level of rigidification of erythrocytes infected with either of the FIKK-KO parasites was reduced when compared to erythrocytes infected with parental FCR3 parasites. When compared to the level of rigidification of the erythrocyte membrane induced by FCR3 parental parasites, the rigidity of erythrocytes infected with FIKK7.1-KO parasites was significantly lower (P\<0.0005 by Mann Whitney U test). Similarly, the median level of rigidity of erythrocytes infected with FIKK12-KO parasites was also lower than those infected with FCR3 but the difference did not quite reach statistical significance at the 95% confidence level (P = 0.06) These suggest that both FIKK7.1 and FIKK12 play a role in modulating the overall level of membrane rigidification induced by *P. falciparum* infection. ## Phosphorylation levels of two different proteins in the erythrocyte membrane skeleton are altered in FIKK12-KO and FIKK7.1-KO parasites We took a phospho-proteomics approach in an attempt to identify proteins which are regulated by the FIKK kinases. Since most of the FIKK members are transported into the erythrocyte membrane via Maurer's clefts, we investigated the level of phosphorylation in ghost fractions prepared from IEs. We first evaluated the purity of the ghost samples, confirming that only erythrocyte membranes and Maurer's cleft were present without contamination by other parasite components. The punctate pattern of Maurer's cleft proteins revealed by antibodies against the Pf332 antigen was observed in the ghost preparations as previously described and the intraparasitic maker Pfhsp70 was detected in the total IEs extracts but not in the ghost fraction, indicating no contamination of parasite cytosolic proteins in the ghosts (Supplementary). The purified ghost fractions from ring (10–14 hours after infection), trophozoite (18–22 hours after infection) and schizont (32–36 hours after infection) stages of the FIKK12-KO and the parental FCR3 parasites were analyzed by gel-based 1 dimensional electrophoresis (1-D). Given that phosphorylation in *P. falciparum* blood stages occurs in a stage-specific manner, tightly synchronized parasites, within a window of +/− 2 hours, were used in this analysis. FIKK12-KO and FCR3 ghosts from the three intraerythrocytic stages were lysed in the presence of protease and phosphatase inhibitors, after which the same protein amount was subjected to 1-D SDS-PAGE. The gel was sequentially stained with the phosphoprotein-specific Pro-Q Diamond fluorescence dye, followed by SYPRO Ruby staining to detect total protein. The experiment was performed independently three or four times in each stage. Comparison of the phosphorylation pattern between the KO and wild type parasites preparations showed a clear distinct profile. A protein with an apparent molecular weight of 80 kDa was significantly less stained in FIKK12-KO trophozoites than in wild type parasites. This differential phosphorylation level was reproducibly observed at the trophozoite stage but neither in ring nor schizont stages. To investigate if the detected change in phosphorylation could result from non- specific protein staining by Pro-Q Diamond, the FCR3 ghost sample was treated with λ–protein phosphatase (λ-PPase), a Mn²⁺-dependent enzyme with activity towards phosphorylated serine, threonine, tyrosine, and histidine. Treated and untreated ghost samples were resolved by 1-D and compared by Pro-Q Diamond and Coomassie blue staining. With the exception of one protein with high molecular weight (over 250 kDa), all proteins were sensitive to λ-PPase treatment and showed a strong reduction, or no reactivity to the phospho- specific Pro-Q Diamond staining, validating the specificity of the Pro-Q Diamond staining. FIKK7.1-KO ghost extracts from highly synchronized parasites were also analyzed for changes in the phosphorylation pattern. We detected a band with an estimated molecular weight of 300 kDa in schizont extracts from wild type ghosts, which was absent in FIKK7.1-KO extracts. No difference was seen in the 80 kDa band, indicating that FIKK7.1 and FIKK12 target distinct proteins in the erythrocyte membrane skeleton. This differential phosphorylation level was reproducibly observed at the schizont stage but neither in ring nor trophozoite stages. # Discussion In this work, we demonstrate that two members of the FIKK kinase family are involved in the remodeling of erythrocyte membrane skeleton proteins. Importantly, each analyzed FIKK member apparently targets a distinct protein at a different time point of the asexual blood cycle. Our experimental evidence suggests that FIKK12 targets a protein of 80 kDa at trophozoite stage and FIKK7.1 targets another protein of approx. 300 kDa at schizont stage. A recent study reported changes in the phosphoproteome of IEs and identified numerous *P. falciparum* phosphorylated proteins and 77 human proteins as phosphorylated in IEs whereas only 48 were detected in uninfected erythrocytes. This indicates an elevated level of post-translational modifications of the host cell by parasite kinases. Given that most of the *Pf fikk* genes were shown to be transcribed in blood stages and N-terminal regions are unique to each paralog, this raises the possibility that each FIKK protein might have different functions in the IEs and that other FIKK proteins could have other biological roles including trafficking, adhesion and antigenic variation. A reverse genetic screen identified different parasite proteins that are related with changes in the deformability of the erythrocyte membrane. Inactivation of several genes (none of the *fikk* genes was analyzed) showed moderate decreases in the deformability of the IEs, validating that several proteins contribute to the overall IEs rigidity. We may expect that these proteins interact with the erythrocyte membrane skeleton or may facilitate the trafficking of skeleton- related proteins. We previously showed that FIKK12 is exported into the cytoplasm of the erythrocyte via Maurer's cleft at ring stage and disassociates from the Maurer's cleft at trophozoite stage. Taken together, this raises the possibility that the exported FIKK12 protein might interact with the membrane skeleton of IEs at the trophozoite stage and be involved in the phosphorylation of proteins of the erythrocyte membrane skeleton. Potential candidates for FIKK12 substrates include the erythrocyte protein protein 4.1 with a molecular weight of 80 kDa. Protein 4.1 plays different roles in protein attachment to the membrane and is involved in regulating the membrane mechanical stability, which is decreased in the protein 4.1-deficient erythrocytes. Interestingly phosphorylation of protein 4.1 has been suggested to affect its role in protein- protein interaction at the erythrocyte membrane skeleton and erythrocyte membrane deformability. Our results show that disruption of *Pf fikk12* resulted in less phosphorylation of the band at 80 kDa, however this was not completely blocked. One possibility is that this substrate might be a substrate for a variety of kinases in the erythrocyte and that FIKK12 might phosphorylate a particular site. For FIKK7.1 we identified a potential substrate with a high molecular weight, although the parasite protein PfEMP1 has a molecular weight in this range and is only detectable after radioactive labelling, at this stage we cannot conclude that a parasite protein in the erythrocyte membrane is the target of FIKK7.1. Furthermore, we do not know if these FIKK kinases target directly erythrocyte membrane proteins or are involved in a phosphorylation cascade. At this stage, we can not rule out the possibility that differences in the phosphoproteomes of the KO lines are derived from the lack of activity of non FIKK kinase regulated by FIKK12 or 7.1. Deletion of both *Pf fikk7.1* and *Pf fikk12* in the FCR3 strain demonstrates that they are not crucial for parasite replication *in vitro*, and since the transgenic parasite lines show normal multiplication rates they are probably not involved in the invasion process. However, our erythrocyte deformability data strongly suggest that these proteins are involved in changes of the erythrocyte membrane skeleton to meet the needs of the growing intracellular parasite. The resulting modification of the rheological properties of IEs and may have a role in splenic clearance by IEs *in vivo*. When we commenced this work, we hypothesized that the FIKK proteins could be involved in the process of IE adhesion and antigenic variation of *var* genes, since this process is unique to *P. falciparum*. It was recently reported that the phosphorylation of PfEMP1 cytoplasmic domain by casein kinase II alters the association of this domain with knob associated histidine-rich protein and interestingly that inhibition of phosphorylation reduced the cytoadherence of IEs to two endothelial receptors. We, however, could not see any differences in the adhesion properties, switching rate and trafficking of multigene families between the two KO parasite lines and the wild type FCR3 strain. In conclusion, we present for the first time that FIKK kinases are likely involved in remodelling the membrane of IEs and our data point to changes in the cellular mechanical properties in a stage-specific and target specific manner. Work is ongoing to identify the specific proteins that are phosphorylated by these FIKK kinases. In the future it will be interesting to investigate if the *Pf fikk* genes perform similar functional roles in gametocytes and infected hepatocytes. # Materials and Methods ## *Plasmodium falciparum* cultures and adhesion assays *P. falciparum* blood stage parasites from the FCR3 strains were cultured using modifications to the method described by Trager and Jensen. IEs were selected on the human placental derived BeWo cell line (European Collection of Cell Cultures) to obtain the FCR3-CSA parasite line or on Chinese Hamster Ovary cells-745 (CHO-745; American Type Culture Collection) expressing CD36 to obtain the FCR3-CD36 parasite line. Pannings were repeated three times, and parasites were tested for their ability to bind purified CSA (Sigma) or recombinant human CD36 (R&D Systems). Cytoadhesion assays on receptors immobilized on plastic petri dishes were carried out as described. Briefly, plastic Petri dishes were coated overnight at 4°C with phosphate-buffered saline (PBS) containing 1 mg/ml CSA sodium salt from bovine trachea (Sigma), 1 mg/ml chondroitin sulfate C sodium salt from shark cartilage (Sigma), 10 µg/ml recombinant human CD36 (R&D Systems). All spots were blocked with 1% bovine serum albumin (BSA) for 1 h at room temperature (RT) before trophozoite-IE (5×10⁷ IEs/ml) were allowed to adhere. The average number of adherent IE (±SEM) for four different fields in duplicate spots was determined in two to three independent experiments after fixing with 2% glutaraldehyde in PBS for 2 h at RT and staining the plates with Giemsa. Pictures were taken with a Nikon camera. Lucia software was used to count the number of bound IEs per mm². ## Plasmids and transfection *Pf fikk7.1* and *Pf fikk12* were targeted using fragments amplified by PCR from FCR3 strain genomic DNA with the following oligonucleotides: *fikk7.1* 5' segment 5′-GAG*ccgcgg*TAGTACATTGTATAATAAAATATAACGC-3′ and 5′-CGC*agatct*CAAGAGATTATCATTTTTATTTTGC-3′ and 3′ segment 5′-ACGC*ccatgg*CTGTGGATATGTTGTAATGGTATC-3′and 5′-CGA*cctagg*CTATAAATATAATATTATGTATGCAC-3′; *fikk12* 5′ segment 5′-GAG*ccgcgg*ATGTATATTTTGAGAAATATGTTCTG-3′ and 5′-CGC*actagt*TCGTCCTCTTTTAAATTAGACATAC-3′ and 3′ segment 5′-ACGC*ccatgg*CAGATAAATTAAGACATATAGATAAAAAG-3′ and 5′-CGA*cctagg*TTATGTTTCGTTAAACCATGGGTGTG-3′ (enzyme restriction sites in italics). These PCR fragments were sequentially cloned into pHTK using the SacII/BglII and NocI/AvrII sites for *fikk7.1* and SacII/SepI and NocI/AvrII sites for *fikk12*, to derive pHTK-*FIKK7.1* and pHTK-*FIKK12*, respectively. Ring-stage FCR3-CSA *P. falciparum* parasites were transfected with 100 µg plasmid DNA and cultured with WR99210 (10 nM) (Jacobus Pharmaceutical Co. Inc.) after cultures were established parasites where double crossover homologous recombination events had occurred were selected with 4 µM ganciclovir (Roche). The resistant parasites were cloned by limiting dilution. ## Southern blot analysis of the two KO lines Genomic DNA was digested using the following enzymes: HindIII (FIKK7.1 transfectants) and AluI (FIKK12 transfectants) and size fractionated on 0.8% agarose gels. Southern blot were performed, as described previously. Specific probes were amplified with the same set primers used for the initial cloning. Membranes were hybridized at high-stringency conditions at 60°C overnight and washed twice with 0.2x saline-sodium citrate (SSC) and 0.1% SDS at 60°C for 30 min. ## Immunofluorescence microscopy Synchronized IEs were washed in PBS, cell pellets were resuspended in 10 vols of PBS and a monolayer was set onto microscope slides. Parasites were air-dried and fixed for 30 min at RT in 4% paraformaldehyde and 0.0075% glutalaldehyde. Slides were washed with PBS and incubated with the primary antibodies diluted in 0.1% BSA: rabbit anti-Rifin-A 565 antibody 1∶100 (serum raised against a peptide in the highly conserved C-terminus of RIFIN-A, recognizes this type of RIFIN specifically), rabbit anti-Stevor antibody 1∶400 (obtained from rabbits immunized with a peptide designed on the basis of conserved STEVOR regions and recognizes the subset of *stevor* genes transcribed in parasite population) and rabbit anti-Surfin 4.2 antibody 1∶800 (antibody against PFD1160w that detects specifically Surfin 4.2) for 45 min at RT. After washing cells were incubated for 30 min with mouse anti-rabbit secondary antibody conjugated with fluorochrome (Invitrogen) in a 1∶500 dilution. Slides were washed thoroughly in PBS and mounted in Vectashield anti-fading with DAPI (Vector Laboratories). Images were captures using a Nikon Elipse 80i optical microscope. ## Ghost preparation Erythrocyte ghosts were essentially prepared as previously described with some modifications. Synchronized ring (10–14 hours after infection), trophozoite (18–22 hours after infection) or schizont (32–36 hours after infection)-infected erythrocytes were extensively washed in RPMI 1640 medium and lysed for 15 min at 4°C in 40 vols of hypotonic buffer \[10 mM sodium-phosphate buffer, pH7.4 containing protease inhibitor cocktail tablet (Roche) and phosphatase inhibitors, 400 nM Okadaic acid (Sigma) and phosphatase inhibitor cocktail 2 (Sigma)\]. The lysates were then separated by centrifugation at 15,000×g for 30 min at 4°C into a cytosolic fraction and a pellet fraction containing ghosts and free parasites. The erythrocyte ghosts were collected at the top of the free parasite pellet, washed extensively five times in the hypotonic buffer. Ghosts were also prepared from uninfected erythrocyte for a mocked control. Ghost samples from each parasite stage were prepared in three or four independent experiments and analyzed by phospho-proteomics. Twenty µg of the purified ghosts were analyzed by SDS-PAGE using 4–12% polyacrylamide gel followed by immunoblot assay using anti-Pf332 and anti-Pfhsp70 antibodies to confirm the purity of the ghost preparation. In some experiments, 20 µg of ghost proteins were subjected to dephosphorylation prior SDS-PAGE analysis. For this samples were incubated for 30 min at 25°C with 4,000 U of λ-PPase (New England BioLabs) in reaction buffer (50 mM Tris-HCl pH7.5, 0.1 mM EDTA, 5 mM dithiothreitol, 0.01% Brij 35, and 2 mM MnCl₂). ## Staining procedures Polyacrylamide gels were fixed in 50% methanol, 10% acetic acid, and stained using Pro-Q Diamond phosphoprotein gel stain (Molecular Probes) in accordance with the manufacturer's instructions. The stained gel was visualized on a Typhoon 9400 scanner (Amersham Biosciences) using excitation at 532 nm and 560 nm longpass emission filter, normal sensitivity (600 V), 3 mm focal plane and 50 mm resolution. Total protein was restained with SYPRO Ruby total protein gel stain (Molecular Probes), and visualized on the scanner using excitation at 457 nm and 610 nm bandpass emission filter and the same settings as above. Images were acquired as gel file format, imported into ImageJ software and stored as tiff file format for subsequent analysis. ## Measurement of membrane shear elastic modulus by micropipette aspiration Single cell micropipette aspiration was used to determine the shear elastic modulus of erythrocyte membranes as previously described. Briefly, the membrane of individual erythrocytes was aspirated progressively into glass micropipettes (internal diameter 1.3 µm) over a range of increasing negative hydrostatic pressures. The shear elastic modulus of the membrane skeleton was determined by measuring the length of a membrane tongue (L) aspirated from the erythrocyte into the pipette for a range of aspiration pressures (P) and calculated from the linear regression of dL/dP. The range of aspiration pressures was 1.0 - 4.5 mmH₂O for uninfected erythrocytes or 1.0–12.0 mmH₂O for infected erythrocytes. All measurements were performed at RT (approximately 20–25°C). # Supporting Information We would like to thank Denise Mattei and Ross Coppel for helpful discussions. We are grateful to the generous gift of the Rifin and Surfin antibodies by Mats Wahlgren and Stevor antibodies by Peter Preiser. We also thank Fiona Glenister for assistance with the micromechanical measurements. [^1]: Conceived and designed the experiments: MCN MO BMC AS. Performed the experiments: MCN MO CSB. Analyzed the data: MCN MO BMC. Wrote the paper: MCN MO BMC AS. [^2]: Current address: Respiratory and Meningeal Pathogens Research Unit, University of the Witwatersrand, Johannesburg, South Africa [^3]: Current address: National Center for Global Health and Medicine, International Medical Center of Japan, Tokyo, Japan [^4]: The authors have declared that no competing interests exist.

# Introduction and background Proton cancer therapy (PCT) is an approved type of radiotherapy that utilizes high-energy H⁺ projectiles to fight cancer. The ultimate effect of this radiation is to damage the DNA of cancerous cells. If left unrepaired, this damage produces various anomalies in cancerous cells that eventually lead to their death (apoptosis). While PCT damages both healthy and cancerous cells, the latter have a high rate of division and a reduced ability to repair damaged DNA. Thus, cancer cells are particular vulnerable to radiation attacks. PCT radiation is applied as collimated beams of H⁺ projectiles at an initial kinetic energy of 70–250 MeV. As they penetrate the patient’s body, these projectiles lose their energy through molecular interactions until they reach thermal energy in deep tissues. The radiation energy deposited in the tissues is a measure of its potential for DNA damage. Typically, a plot of the radiation energy loss *vs*. the radiation travelled distance exhibits a maximum of energy deposition. Conventional X-ray therapy exhibits a broad deposition maximum not far after the photons’ penetration into the body, followed by a gradual energy loss at deeper penetrations. In contrast, PCT (and other ion therapies) exhibits a sharp maximum peak known as the Bragg peak; that peak occurs just before the H⁺ projectiles are stopped in deep tissues. Thus, by focusing Bragg peaks on a deep tumor, PCT inflicts a maximum DNA damage on that region and a minimum DNA damage on the surrounding healthy tissues. In all radiotherapies, the radiation predominantly interacts with cellular H₂O because the latter constitutes about 70% of the human cell mass. This interaction triggers water radiolysis reactions—i.e. a series of cascade hydrolytic reactions producing DNA-damaging species. In PCT, water radiolysis generates various secondary species: (a) free radicals (e.g. H⁺ + H₂O → H⁺ + H·+ OH·)\[–, \], (b) secondary ions (e.g. H⁺ + H₂O → 2H⁺ *+* OH^-)\[–, \], (c) reactive molecules (e.g. H⁺ + 2H₂O → H⁺ + H₂ + H₂O₂), and (d) localized heat in the medium. These species can react with H₂O and generate similar tertiary species and so forth. All these reactive products eventually reach cellular DNA and cause various types of damage: DNA bases’ fragmentations and deletions, and single-, double- and clustered-strand DNA breaks. PCT comprises various processes spanning different space (*l* = 10⁻¹⁰–10⁰ m) and time (*t* = 10−21–10⁵ s) scales. For instance, water radiolysis reactions, DNA damage at the genome level and tumor remissions lie at the microscopic (10⁻¹⁰ ≤*l*≤10⁻⁸ m), mesoscopic (10⁻⁸ ≤*l*≤10⁻³ m) and macroscopic (10⁻³ ≤*l*≤10⁰ m) scales, respectively. The scale of a process determines the appropriate methods for its study. Thus, in theoretical/computational studies, microscopic water radiolysis reactions can be feasibly simulated with *ab initio* quantum-mechanics methods. In contrast, mesoscopic energy-loss and Bragg peak processes are only tractable with Monte Carlo (MC) models. Quantum-mechanics and MC methods act in synergy: the former predict properties (e.g. reaction cross sections) required as input data for MC simulations \[, –\], and the latter calculate proper radiation doses for treatments. Although PCT is clinically approved, various PCT details at the microscopic scale remain uncertain. Knowledge of those details is essential for a rational design of PCT seeking to maximize its therapeutic power and minimize its side effects. While the predominant paradigm for cancer research is experimental/clinical, theoretical/computational methods can reveal microscopic details of PCT more exhaustively than experimental/clinical techniques and without putting human subjects at risk. Therefore, time-independent scattering and time-dependent dynamics methods have been applied to computationally feasible prototypes of PCT reactions (e.g. H⁺ + (H₂O)_1–4 to model water radiolysis reactions and H⁺ + DNA/RNA bases to model DNA proton damage). Among quantum-mechanics methods for PCT, the electron nuclear dynamics (END) theory offers distinctive capabilities to study PCT reactions. END is a (1) time-dependent, (2) direct and (3) non-adiabatic method to simulate chemical reactions. These attributes are valuable for PCT simulations because they afford: (1) time-dependent detail, (2) independence from predetermined potential energy surfaces, and (3) capability of describing high-energy non-adiabatic processes \[e.g. electron transfers (ETs)\]. Among different END versions \[, \], the simplest-level (SL) END (SLEND) describes the nuclei and electrons in terms of classical mechanics and a Thouless single-determinantal wavefunction, respectively. Thus, SLEND possesses a suitable balance between accuracy and computational feasibility to simulate large PCT prototypes (cf. previous SLEND studies of H⁺ + (H₂O)_*n*, *n* = 1, 2, and 3–4 and of H⁺ + DNA/RNA bases and DNA base pairs). Following the aforesaid precedents, we present herein an exploratory SLEND study of PCT water radiolysis reactions with the H⁺ + (H₂O)_1-6 prototypes at *E*_*Lab* = 100 keV. This energy is selected because it corresponds to the average Bragg peak energy in bulk water, the medium where water radiolysis occurs. However, no *ab initio* quantum-mechanics methods can simulate bulk water due to prohibitive computational cost and, therefore, those methods treat the above-mentioned water-clusters prototypes. Surprisingly, most quantum-mechanics studies of water radiolysis have utilized the smallest prototype: H⁺ + H₂O, whose “cluster” is the farthest from being bulk water. While these H⁺ + H₂O studies were indeed useful for investigating radiolysis processes, they could not completely capture the processes occurring in bulk water. For that reason, previous SLEND studies explored H⁺ + (H₂O)_*n* prototypes with n = 2 and 3–4. However, these studies still concentrated on the smallest possible clusters and involved a limited number of proton-cluster orientations and simulations. Therefore, to overcome all the discussed limitations, we study herein the H⁺ + (H₂O)_1-6 prototypes that include ten isomers in a larger series of clusters (H₂O)_1-6 (cf.) and involve a larger number of proton-cluster orientations (60) and simulations (25,020). The selected (H₂O)_1-6 series contains the initial terms of the long progression from molecular H₂O to bulk water. Specifically, the first terms in this progression, (H₂O)_1-5, have mono- and di-cyclic quasi-planar/multiplanar structures exhibiting no “bulky” shapes (cf.), but two isomers in the last term—the prism and cage isomers of (H₂O)₆—have multi-cyclic, three-dimensional structures exhibiting drop-like shapes (cf.). In fact, these two (H₂O)₆ isomers, particularly the prism, are considered the smallest possible drops of water—i.e. the minimum water structures that manifest the three-dimensional hydrogen-bond structure and solubility properties of bulk water. Based on those facts, we expected that this series would reveal the earliest manifestations of bulk water effects on PCT properties; however, the present results do not display such manifestations and suggest that even larger water clusters should be considered (cf. Results and Discussion Section). Despite that outcome, all the predicted properties and reactions of H⁺ + (H₂O)_2-6 have never been measured or calculated before; therefore, the present results are truly predictive and fill a gap in the medical physics literature. Furthermore, these results are important to understand PCT more thoroughly and to model water radiolysis processes and radiation dosages with MC methods. Finally, it should be noticed that our simulated phenomena are the first processes occurring upon a short-time direct collision of a proton with moderate size clusters. Other post-collision phenomena contributing to PCT such as local temperature increases and hypothesized shock waves in water require for their modelling longer simulation times, larger clusters and even different theoretical methods; therefore, those phenomena are not reproduced by the current simulations. # Methodology The END theory and its SLEND version have been reviewed in detail in Refs.; therefore, we provide a brief outline of them. END is a variational, time- dependent, direct, and non-adiabatic dynamical method. END prescribes a total trial wavefunction $\left| \Psi_{Total}^{END} \right\rangle = \left| \Psi_{N}^{END} \right\rangle\ \left| \Psi_{e}^{END} \right\rangle$, which consists of nuclear $\left| \Psi_{N}^{END} \right\rangle$ and electronic $\left| \Psi_{e}^{END} \right\rangle$ wavefunctions, and treats $\left| \Psi_{Total}^{END} \right\rangle$ under the time-dependent variational principle (TDVP). The various versions of END differ in the kind of descriptions for the nuclei and electrons (e.g., density functional theory for electrons). In SLEND, the nuclear wavefunction $\left| \Psi_{N}^{SLEND} \right\rangle$ for a system having *N*_*N* nuclei is the product of 3*N*_*N* frozen, narrow Gaussian wave packets: $$\left\langle \mathbf{X} \right.\left| {\Psi_{N}^{SLEND}(t)} \right\rangle = \left\langle \mathbf{X} \right.\left| {\mathbf{R}(t),\ \mathbf{P}(t)} \right\rangle = {\prod\limits_{A = 1}^{3N_{N}}{\exp\left\{ {- \left\lbrack \frac{X_{A} - R_{A}(t)}{2\Delta R_{A}} \right\rbrack^{2} + iP_{A}(t)\left\lbrack {X_{A} - R_{A}(t)} \right\rbrack} \right\}}}$$ with average positions **R**_*A*(*t*), average momenta **P**_*A*(*t*) and widths Δ*R*_*A*. To lower computational cost, the zero-width limit, Δ*R*_*A* → 0, is applied to the nuclear wave packets after constructing the SLEND quantum Lagrangian (cf. next paragraph). That procedure generates a classical nuclear dynamics but with full retention of the nucleus-electron non-adiabatic coupling terms (cf.). As proven previously, classical nuclear dynamics does not impair the accuracy of PCT simulations because they happen at high collision energies. The SLEND electronic wavefunction $\left| \Psi_{e}^{SLEND} \right\rangle$ is a spin-unrestricted, single-determinantal wavefunction in the Thouless representation. Specifically, taking *N*_*e* occupied {*ψ*_*h*}, and *K* – *N*_*e* virtual {*ψ*_*p*} molecular spin-orbitals (MSOs), $\left| \Psi_{e}^{SLEND} \right\rangle$ is: $$\left| \Psi_{e}^{SLEND} \right\rangle = \left| {\mathbf{z}(t);\mathbf{R}(t),\mathbf{P}(t)} \right\rangle = \det\left\{ {\chi_{h}\left\lbrack {\mathbf{x}_{h};\ \mathbf{z}(t),\mathbf{R}(t),\mathbf{P}(t)} \right\rbrack} \right\} = \text{exp}\left\lbrack {\sum\limits_{h = 1,\ p = N_{e} + 1}^{N_{e},\ K}{z_{ph}(t)b_{p}^{\dagger}b_{h}}} \right\rbrack\left| 0 \right\rangle;$$ where $\left| 0 \right\rangle = \left| \psi_{N_{e}} \right.\ldots\left. \psi_{1} \right\rangle$ is an unrestricted Hartree-Fock (UHF) reference state and {*χ*_*h*} are dynamical spin-orbitals (DSOs) $$\chi_{h}\left\lbrack {\mathbf{x};\ \mathbf{z}(t),\mathbf{R}(t),\mathbf{P}(t)} \right\rbrack = \psi_{h}\left\lbrack {\mathbf{x};\ \mathbf{R}(t),\mathbf{P}(t)} \right\rbrack + {\sum\limits_{p = N_{e} + 1}^{K}{z_{ph}(t)\psi_{p}}}\left\lbrack {\mathbf{x};\ \mathbf{R}(t),\mathbf{P}(t)} \right\rbrack;\ \left( {1 \leq h \leq N_{e}} \right)$$ with complex-valued molecular coefficients {*z*_*ph*(*t*)}. The MSOs are obtained at initial time at the UHF level. The MSOs are constructed with travelling atomic basis set functions $\phi_{A_{i}}\left( {\mathbf{r}_{i};\mathbf{R}_{A},\mathbf{P}_{\mathbf{A}}} \right)$—i.e., Slater- type orbitals in terms of contracted Gaussian-type orbitals on the moving nuclear centers **R**_*A*(*t*) and augmented with electron translation factors (ETFs) to include explicit nuclear momenta **P**_*A*(*t*) effects. MSOs and DSOs are spin-unrestricted; therefore, the unrestricted determinant \|**z**(*t*);**R**(*t*),**P**(*t*)⟩ can reasonably describe bond- breaking/-forming processes. The Thouless representation provides a non- redundant and singularity-free parameterization of an evolving single- determinantal state. The SLEND dynamical equations are obtained according to the TDVP. First, the quantum Lagrangian $L_{SLEND} = \left\langle \Psi_{Total}^{SLEND}\ \right|\ \left( {i/2} \right)\ \left( {\overset{\rightharpoonup}{\partial}/\partial t} \right. - \left. {\overset{\leftarrow}{\partial}/\partial t} \right) - {\hat{H}}_{Total}\ \left| \ \Psi_{Total}^{SLEND} \right\rangle/\left\langle \Psi_{Total}^{SLEND}\ \right|\ \left. \Psi_{Total}^{SLEND} \right\rangle^{- 1}$ is constructed and then the zero-width limit is applied to the nuclear wave packets. Subsequently, the stationary condition is imposed to the quantum action *A*_*SLEND*, $\delta A_{SLEND} = \delta{\int_{t_{1}}^{t_{2}}{L_{SLEND}(t)dt}} = 0$. The resulting SLEND dynamical equations are: $$\begin{bmatrix} {i\mathbf{C}} & 0 & {i\mathbf{C}_{\mathbf{R}}} & {i\mathbf{C}_{\mathbf{P}}} \\ & & & \\ 0 & {- i\mathbf{C}^{*}} & {- i\mathbf{C}_{\mathbf{R}}^{*}} & {- i\mathbf{C}_{\mathbf{P}}^{*}} \\ & & & \\ {i\mathbf{C}_{\mathbf{R}}^{\dagger}} & {- i\mathbf{C}_{\mathbf{R}}^{T}} & \mathbf{C}_{\mathbf{R}\mathbf{R}} & {- \mathbf{I} + \mathbf{C}_{\mathbf{R}\mathbf{P}}} \\ & & & \\ {i\mathbf{C}_{\mathbf{P}}^{\dagger}} & {- i\mathbf{C}_{\mathbf{P}}^{T}} & {\mathbf{I} + \mathbf{C}_{\mathbf{P}\mathbf{R}}} & \mathbf{C}_{\mathbf{R}\mathbf{P}} \\ \end{bmatrix}\begin{bmatrix} \frac{d\mathbf{z}}{dt} \\ \frac{d\mathbf{z}^{*}}{dt} \\ \frac{d\mathbf{R}}{dt} \\ \frac{d\mathbf{P}}{dt} \\ \end{bmatrix} = \begin{bmatrix} \frac{\partial E_{Total}}{\partial\mathbf{z}^{*}} \\ \frac{\partial E_{Total}}{\partial\mathbf{z}^{}} \\ \frac{\partial E_{Total}}{\partial\mathbf{R}} \\ \frac{\partial E_{Total}}{\partial\mathbf{P}} \\ \end{bmatrix}$$ where *E*_*total* is the total (nuclear and electronic) energy and the dynamic metric matrices are $$\left( \mathbf{C}_{XY} \right)_{ik,\ jl} = \left. {- 2{Im}\frac{\partial^{2}\ln S}{\partial X_{ik}^{'}\partial Y_{jl}}} \right|_{\begin{array}{l} {\mathbf{R}^{\prime} = \mathbf{R}} \\ {\mathbf{P}' = \mathbf{P}} \\ \end{array}};\mspace{9mu}\left( \mathbf{C}_{X_{ik}} \right)_{ph} = \left. \frac{\partial^{2}\ln S}{\partial z_{ph}^{*}\partial X_{ik}} \right|_{\begin{array}{l} {\mathbf{R}^{\prime} = \mathbf{R}} \\ {\mathbf{P}' = \mathbf{P}} \\ \end{array}};\mspace{9mu}\mathbf{C}_{ph,\ qg} = \left. \frac{\partial^{2}\ln S}{\partial z_{ph}^{*}\partial z_{qg}} \right|_{\begin{array}{l} {\mathbf{R}^{\prime} = \mathbf{R}} \\ {\mathbf{P}' = \mathbf{P}} \\ \end{array}}$$ where *X*_*ik* and *Y*_*jl* denote either **R**_{*A* = *i*,*k*} or **P**_{*A* = *j*,*l*} and *S* = ⟨**z**(*t*), **R**′(*t*), **P**′(*t*)\|**z**(*t*), **R**(*t*), **P**(*t*)⟩. **C**_**R** and **C**_**RR** are equivalent to the ordinary non-adiabatic coupling terms. Neglect of these terms seriously impairs the accuracy of the SLEND non-adiabatic dynamics; therefore, those terms are kept in the present calculations. However, to lower computational cost, ETFs are not included in the present basis set so that **C**_**P** = **C**_**PR** = 0. The accelerated SLEND equations thus obtained have been successfully applied to various reactive and non-adiabatic processes, from a few eV to the keV regime\[–\] and up to 900 keV (cf. also the seminal END Ref., especially page 948, where this approximation is applied exhaustively to high-energy non-adiabatic processes). At initial time, the reactants are prepared with positions $\left\{ \mathbf{R}_{A}^{i} \right\}$, momenta $\left\{ \mathbf{P}_{A}^{i} \right\}$ and Thouless electronic state \|**z**^(*i*) = **0**, **R**^(*i*)⟩ = \|0⟩, i.e. the UHF ground state of the reactants’ super-molecule. When the reaction accesses the non-adiabatic regime, **z**(*t*) ≠ **0** and \|**z**(*t*),**R**(*t*)⟩ becomes a superposition of ground \|0⟩ and excited \|… *h* → *p* …⟩ UHF states (cf.) $$\left| {\Psi_{e}^{SLEND}(t)} \right\rangle = \left| {\mathbf{z}(t);\mathbf{R}(t),\mathbf{P}(t)} \right\rangle = {\sum_{h = 1,\ p = N_{e} + 1}^{K,\ N_{e} + 1}{z_{ph}(t)\left| \ldots\ h\rightarrow p\ \ldots \right\rangle}} + \ \ldots$$ # Computational details ## Software All the present SLEND simulations were computed with our END program <span class="smallcaps">PACE</span> (Python-Accelerated Coherent states Electron- nuclear dynamics, T. V. Grimes, E. S. Teixeira and J. A. Morales, Texas Tech University, 2010–2016; cf. Ref., Sect. 4). PACE combines several advanced computer science techniques such as a mixed programming language (Python for logic flow and Fortran and C++ for calculations), intra- and internode parallelization, and the fast OED/ERD atomic integral package from the ACES III/IV program. In addition, the water clusters’ geometries at initial time were computed with the NWChem <span class="smallcaps">\[</span><span class="smallcaps">\]</span> and <span class="smallcaps">GAMESS \[</span><span class="smallcaps">\]</span> programs. ## Water cluster structures at the initial states Present SLEND simulations start with the super-molecular systems H⁺ + (H₂O)_1-6 optimized at the UHF level and having projectile- to-target \[H⁺-to-(H₂O)_1-6\] separations ≥ 30.00 a.u. (cf.). Integration of the SLEND requires the evaluation of their various terms at numerous adaptive time steps on a total of 25,020 trajectories. Thus, for feasibility’s sake, the medium size basis sets 6-31G\* \[for H⁺ + (H₂O)_1-6\] and 6-31G\*\* \[for H⁺ + (H₂O)_1-5\] are adopted. These basis sets provide good water clusters’ geometries and energies (cf. Refs.; for these basis sets’ dynamical performance, cf. Electron Transfer Properties and Further Analysis Sub-Sections). Numerous theoretical and experimental studies have been devoted to determine the geometries and energies of water clusters because these systems are prototypes to study the structural and solubility properties of bulk water. In fact, the scientific literature about water clusters is vast and growing. Therefore, for brevity’s sake, we limit ourselves to cite herein only the water clusters studies closely related to this investigation. It is well-known that the (H₂O)_2-3 clusters present one isomer each; however, the (H₂O)_*n*, *n* ≥ 4, clusters present a variable number of isomers that rapidly increase with the cluster size *n*. (H₂O)_1-5 present mono- and di-cyclic quasi- planar/multiplanar structures, whereas (H₂O)_*n*, *n* ≥ 6, present multi-cyclic, three-dimensional structures in addition to quasi- planar/multiplanar ones (cf.). The three-dimensional (H₂O)₆ isomers (e.g. the prism and cage isomers named hexamer a and b) are considered the smallest drops of water. In general, theory and experiments agree in regard to the structures and relative energies of the (H₂O)_1-5 isomers, but discrepancies arise in the energy orders of the (H₂O)_*n*, *n* ≥ 6, isomers. Recent spectroscopy experiments have identified the cage isomer as the lowest-energy (H₂O)₆ structure. However, *ab initio* calculations at various levels of accuracy have disagreed on whether the prism, the cage, both of them, or the chair isomer(s) is(are) the lowest-energy (H₂O)₆ structure(s). Ultimately, the most accurate calculations with the coupled-cluster with singles, doubles and perturbative triples \[CCSD(T)\] method have identified the prism isomer as the lowest-energy (H₂O)₆ structure. For this investigation, we selected ten representative isomers in the (H₂O)_1-6 series: each single isomer of (H₂O)_1-3, the two isomers of (H₂O)_4, two isomers of (H₂O)₅ out of a total of at least four _, and three isomers of (H₂O)₆ out of a total of at least twelve (cf.). The isomers were calculated at the UHF/6-31G\* \[(H₂O)_1-6\] and /6-31G\*\* \[(H₂O)_1-5\] levels. When two or more isomers of a (H₂O)_n cluster are considered, they are depicted (cf.) and listed (cf.) in their increasing order of energies \[e.g. hexamer a (prism), hexamer b (cage), and hexamer c since $E_{\text{Hexamer\ b}}^{\text{Cage}} - E_{\text{Hexamer\ a}}^{\text{Prism}}$ = 4.2 kJ/mol and $E_{\text{Hexamer\ c}}^{} - E_{\text{Hexamer\ a}}^{\text{Prism}}$ = 15.4 kJ/mol with UHF/6-31G\*). Notice that the UHF prism isomer is the lowest-energy (H₂O)₆ structure as predicted by CCSD(T). The first depicted/listed isomer is also the lowest-energy structure in its whole set of isomers. The present UHF calculations of (H₂O)_1-6 are not intended to contribute to the resolution of the discrepancies about the (H₂O)₆ energies because these calculations do not attain the accuracy of CCSD(T). Instead, these optimizations provide a good description of the water clusters for subsequent SLEND/6-31G\* and /6-31G\*\* simulations. ## Initial states preparation and simulation times Once the water clusters are optimized at the UHF/6-31G\* and /6-31G\*\* levels, the super-molecular systems H⁺ + (H₂O)_1-6 are assembled for the initial conditions (cf.). The (H₂O)_1-6 targets are prepared at rest in their equilibrium geometries with their centers of mass placed at the origin of the laboratory-frame coordinate axes; the (H₂O)_1-6 major (pseudo-)planes of symmetry are placed with maximum coincidence with the x-y plane. The H⁺ projectile is first prepared with position $\mathbf{R}_{\text{H}^{+}}^{0} = \left( {b \geq 0} \right.,\ 0,\mspace{9mu}\left. {+ 30\ \text{a}.\text{u}.} \right)$ and momentum $\mathbf{P}_{\text{H}^{+}}^{0} = \left( 0,\ 0, - \left. p_{\text{H}^{+}}^{z} \right) \right.$, where *b* ≥ 0 is the projectile impact parameter measured from the (H₂O)_1-6 centers of mass, and $p_{\text{H}^{+}}^{z}$ corresponds to *E*_*Lab* = 100 keV (cf. , panel I). Having set $\mathbf{R}_{\text{H}^{+}}^{0}$ and $\mathbf{P}_{\text{H}^{+}}^{0}$, various projectile-target relative orientations can be generated by rotating a (H₂O)_1-6 target according to ordinary Euler angles prescriptions. However, such a procedure involves the electronic re-optimization of the (H₂O)_1-6 targets at each new orientation. Therefore, since the H⁺ bare ion requires no electronic optimization, we adopted the easier and equivalent procedure of keeping a (H₂O)_1-6 target fixed while rotating the H⁺ projectile around. The definite initial conditions of the H⁺ projectile $\mathbf{R}_{\text{H}^{+}}^{i}$ and $\mathbf{P}_{\text{H}^{+}}^{i}$ are therefore obtained by rotating $\mathbf{R}_{\text{H}^{+}}^{0}$ and $\mathbf{P}_{\text{H}^{+}}^{0}$ through the *extrinsic* Euler angles in the order: 1^st, 0⁰ ≤ *γ* \< 360⁰, 2^nd, 0⁰ ≤ *β* ≤ 180⁰, and 3^rd, 0⁰ ≤ *α* \< 360⁰, around the *space-fixed* z, y, and z axes, respectively (cf., panels I, II and III for each angle rotation). *α* and *γ* are measured from the +x axis employing the $\mathbf{R}_{\text{H}^{+}}^{0}$ and $\mathbf{P}_{\text{H}^{+}}^{0}$ projections on the x-y plane and *β* is measured directly from the +z axis (cf.). These rotations define projectile-target relative orientations Ω_*i* = (*α*_*i*, *β*_*i*, *γ*_*i*). The definite initial conditions of the H⁺ projectile for the simulations, $\mathbf{R}_{\text{H}^{+}}^{i}$ and $\mathbf{P}_{\text{H}^{+}}^{i}$, are shown in, panel IV. *α* and *β* determine the direction of an axis of incidence whereby an incoming H⁺ trajectory runs parallel to that axis with a lateral separation *b* ≥ 0; *γ* is the polar angle of that trajectory around the axis of incidence. In all simulations, the selected values of the Euler angles correspond to a 60-point grid: Ω_*i*, 1≤*i*≤60, developed in Ref.. This grid displays a uniform sampling of the orientation space and provides a numerical quadrature that ensures the invariance of Euler-angle integrals under several rotation operations (e.g. Wigner D-matrices satisfy ${\int\limits_{0}^{2\pi}{\int\limits_ {0}^{\pi}{\int\limits_{0}^{2\pi}D_{M^{\prime}M}^{J}}}}\ \left( {\alpha,\ \beta,\ \gamma} \right)\ \sin\beta d\alpha\ d\beta d\gamma = 0$ for 2 ≤ *J* ≤ 5). For a given orientation Ω_*i*, *b* is varied according to the grids defined in. Simulations for the calculation of 1-ET ICSs utilize the grids denoted as \[*b*\]₁ in and run for a total time of 30.00 a.u. (0.7257 fs); this simulation time ensures that the final projectile-target separation is at least equal to the initial one (30.00 a.u.). Simulations for the prediction of fragmentations utilize the grids denoted as \[*b*\]₂ in and run for a total time of 1,000 a.u. (24.19 fs); this much longer simulation time permits the manifestation of post-collision fragmentations that have longer time scales than those of the 1-ET processes. The described initial conditions generate a total of 25,020 trajectories to complete the H⁺ + (H₂O)_1-6 study. ## Final states analysis and properties calculations By the end of a simulation, various auxiliary codes in the PACE package identify and analyze the final products and calculate dynamical properties. The most important properties calculated herein are the cluster-to-proton 1-ET total ICSs, *σ*_1−*ET*, corresponding to H⁺ + (H₂O)_1–6 → H + different cluster products $$\sigma_{1 - ET} = \frac{1}{4\pi}{\int\limits_{0}^{\infty}{\int\limits_{0}^{2\pi}{\int\limi ts_{0}^{\pi}{\int\limits_{0}^{2\pi}{b\mspace{9mu} P_{1\text{-ET}}\ \left( {\alpha,\ \beta,\ \gamma,\ b} \right)db\mspace{9mu}\sin\beta}}}}}\ d\alpha\ d\beta\ d\gamma$$ where *P*_1−*ET*(*α*, *β*, *γ*, *b*) is the probability of a 1-ET process from a bound electronic state of the cluster to a bound electronic state of the projectile, henceforth named bound-to-bound 1-ET for brevity’s sake. Notice that for atom-atom collisions involving spherically symmetric potentials, *P*_1−ET (*α*, *β*, *γ*, *b*) → *P*_1−ET (*b*), and Eq simplifies to the familiar expression: $\sigma_{1 - \text{ET}} = 2\pi{\int\limits_{0}^{\infty}b}\mspace{9mu} P_{1 - \text{ET}}(b)\ db$. In the present systems, an outgoing projectile can in practice capture up to two electrons: H⁺ + A → H^1−*n* + A^−*n*, 0 ≤ *n* ≤ 2, because the probability of forming unstable H^1−*n* with *n* ≥ 3 is negligible. Under these conditions, *P*_*n*−ET (*α*, *β*, *γ*, *b*), 0 ≤ *n* ≤ 2, are $$\begin{array}{l} {P_{0 - \text{ET}}\left( {\alpha,\ \beta,\ \gamma,\ b} \right) = \left( {1 - N_{\alpha}} \right)\left( {1 - N_{\beta}} \right);} \\ {P_{1 - \text{ET}}\left( {\alpha,\ \beta,\ \gamma,\ b} \right) = N_{\alpha}\left( {1 - N_{\beta}} \right) + N_{\beta}\left( {1 - N_{\alpha}} \right);} \\ {P_{2 - \text{ET}}\left( {\alpha,\ \beta,\ \gamma,\ b} \right) = N_{\alpha}N_{\beta}} \\ \end{array}$$ where *N*_*α* and *N*_*β* are the number of *α*- and *β*-spin electrons in the outgoing projectile calculated from their respective electron densities $\rho_{\alpha}^{out.\ proj.}\left( \mathbf{r} \right)$ and $\rho_{\beta}^{out.\ proj.}\left( \mathbf{r} \right)$: $$N_{\alpha} = {\int{\rho_{\alpha}^{out.\ proj.}\left( \mathbf{r} \right)}}\ d\mathbf{r};\mspace{9mu} N_{\beta} = {\int{\rho_{\beta}^{out.\ proj.}\left( \mathbf{r} \right)}}\ d\mathbf{r}$$ Eqs and are evaluated at the final simulation time, when the outgoing projectile and the clusters are well separated by at least 30.00 a.u. of length; therefore, *N*_*α* and *N*_*β* are the number of electrons unequivocally assigned to the distant outgoing projectile. Moreover, at those separations, *N*_*α* and *N*_*β* from and those from any ordinary electron population analyses (Mulliken, Löwdin, etc.) become identical; this assures that *N*_*α* and *N*_*β* are free of any arbitrary criterion for electrons’ distributions over the projectile and clusters. Other calculated properties are the orientation-averaged 1-ET probabilities ${\overline{P}}_{1 - \text{ET}}(b)$, ${\overline{P}}_{1 - \text{ET}}(b) = \left( {8\pi^{2}} \right)^{- 1}{\int_{0}^{2\pi}{\int_{0}^{\pi}{\int_{0}^{2\pi}P_{1 - \text{ET}}}}}\ \left( {\alpha,\ \beta,\ \gamma,\ b} \right)\ \sin\beta$, and their *b* – weighted counterparts, $b{\overline{P}}_{1 - \text{ET}}(b)$, where $\sigma_{1\text{-ET}} = 2\pi{\int_{0}^{\infty}b}\mspace{9mu}{\overline{P}}_{1 - \text{ET}}\ (b)\ db$. ${\overline{P}}_{1 - \text{ET}}(b)$ and $b{\overline{P}}_{1 - \text{ET}}(b)$ reveal more mechanistic details than *σ*_1-*ET*. # Results and discussion ## Electron-transfer properties The first property calculated in this investigation is the cluster-to-proton total 1-ET ICS, *σ*_1−*ET*, for the H⁺ + (H₂O)_n systems, *n* = 1–6, at *E*_*Lab* = 100 keV at the SLEND/6-31G\* (*n* = 1–6) and SLEND/6-31G\*\* (*n* = 1–5) levels. Both SLEND *σ*_1−*ET* for the monomeric system (*n* = 1) are listed in along with their available counterparts from four experiments denoted as Exp. A to D, respectively. In addition, includes results from two alternative theories, namely, Theory A: the basis generator method (BGM), and Theory B: the continuum distorted wave-eikonal initial state (CDW-EIS) approximation. From, one finds that the average experimental ICS, ${\overline{\sigma}}_{1 - ET}^{Exp.}$, and its average relative error, *e*_*Exp*., are 1.27 Ų and ± 10.62%, respectively. The theoretical $\sigma_{1 - ET}^{Theo.}$ ‘s and their average relative deviations ${\overline{\Delta}}_{Theo.}$ from the experimental values are: 1.54 Ų and +21.8% for SLEND/6-31G\*, 1.00 Ų and +21.0% for BGM, and 0.589 Ų and -53.4% for CDW-EIS. Only the BGM result is within the error bars of one experiment, Exp. D, with Δ_*Theo*. = 11.5%, but it lies on the lowest fringe part of the error bar range. The SLEND/6-31G\* result is very close to the result from Exp. C with Δ_*Theo*. = 11.6% and not far from entering the upper part of the error bar range. In absolute quantitative terms, the BGM and SLEND/6-31G\* results are at the same level of accuracy and their agreement with the experimental data should be deemed satisfactory given the difficulty to both measure and predict the present 1-ET processes. Deviations of the obtained magnitude are not uncommon in measurements and predictions of similar complex processes (cf. Ref. for the case of one experiment and four different theories including SLEND, where even higher deviations are observed). The CDW- EIS result compares less favorably with the experimental ones, being roughly a half of its experimental counterparts (${\overline{\Delta}}_{Theo.}$ = -53.4%). The SLEND/6-31\*\* result also compares less favorably with the experimental ones, but, opposite to CDW-EIS, its value is roughly twice as much as the experimental one (${\overline{\Delta}}_{Theo.}$ = 63.0%). The reason and remediation of the SLEND/6-31\*\* *σ*_1−*ET* overestimation will be discussed in detail in the Further Analysis Sub-section. It suffices to say here that this overestimation results from the *σ*_1−*ET* contamination with electron transfers to the continuum of unbound states. The SLEND *σ*_1−*ET* for the polymeric systems (*n* = 2–6) are listed in. Unfortunately, to the best of our knowledge, there are no alternative experimental or theoretical ICSs for H⁺ + (H₂O)_2-6; therefore, current SLEND *σ*_1−*ET* for these systems are predictive. To facilitate the comparison among all the considered *σ*_1−*ET*, we plot them as a function of the cluster size *n* in. There, each set of SLEND/6-31G\* and /6-31G\*\* *σ*_1−*ET* is fit to the scaling formulae *σ*_1−*ET* (*n*) = *cn*^2/3, where *c* are fitting coefficients reported in. These formulae are by no means arbitrary because they can be justified on physical grounds as follows. The volume *V*(*n*) of the (H₂O)_*n* clusters should be approximately proportional to their size *n*, *V*(*n*) ∝ *n*. If the clusters are represented by the minimal spheres enclosing all their atoms, then their volume *V*(*n*) and external area *A*(*n*) are $V\ (n) = \left( {4/3} \right)\ \pi\mspace{9mu} R_{n}^{3}$ and $A(n) = 4\pi\mspace{9mu} R_{n}^{2}$, respectively, where *R*_*n* is the radius of the (H₂O)_*n* sphere; therefore, *A*(*n*) ∝ *V* (*n*)^2/3 ∝ *n*^2/3. In turn, the *σ*_1−*ET* are proportional to the effective external area *A*(*n*) of the (H₂O)_*n* exposed to the incident H⁺ for ET processes; therefore, *σ*_1−*ET* ∝ *A*(*n*) ∝ *n*^2/3 ⇒ *σ*_1−*ET* (*n*) = *cn*^2/3. In relative quantitative terms, the SLEND/6-31G\* and /6-31G\*\* *σ*_1−*ET* fit remarkably well into the physically justified formulae *σ*_1−*ET* (*n*) = *cn*^2/3 with correlation factors *R*² = 0.983 in both cases (cf.). This indicates that regardless of their absolute quantitative performance, the SLEND *σ*_1−*ET* scale correctly with the number of water molecules in the clusters. Therefore, with these fitting formulae, one can estimate the *σ*_1−*ET* of the immediately larger clusters: e.g., $\sigma_{1 - ET}^{6 - 31G^{*}}$ = 5.96 and 6.51 Å for *n* = 7 and 8, respectively; these estimated values should be interpreted as average *σ*_1−*ET* over the various (H₂O)₇ and (H₂O)₈ isomers, respectively. Inspection of Tables and and reveals that the SLEND/6-31G\*\* *σ*_1−*ET* are always higher in value than the SLEND/6-31G\* ones for each cluster as was the case with the H₂O monomer. Inspection of and reveals that the SLEND *σ*_1−*ET* for the various isomers appearing at a given *n* ≥ 4 do not significantly differ in their values; this implies that these *σ*_1−*ET* are rather insensitive to the varying isomers’ structures as targets. A similar finding was observed in the *σ*_1−*ET* of H⁺ + DNA/RNA bases at *E*_*Lab* = 80 keV, where similar bases differing in their structure even more than isomers exhibited close values of *σ*_1−*ET*. We expected that various *σ*_1−*ET* (*n*) = *cn*^2/3 formulae differing in their coefficient *c* would exclusively connect values from different sets of clusters—e.g. a single *σ*_1−*ET* (*n*) = *cn*^2/3 formula might have only fit well with results from the quasi-planar clusters (monomer, dimer, trimer, tetramer a, pentamer a and hexamer c), another single formula with other type of clusters, etc., but that is not case. In fact, the uniformity among the SLEND *σ*_1−*ET* (*n*) values for isomers at a given *n* led us to fit all of them with a single formula per basis set. Furthermore, we expected that the *σ*_1−*ET* of the drop-like prism and cage (H₂O)₆ isomers would differ sharply from the *σ*_1−*ET* of the quasi-planar/multiplanar isomers so that it would be impossible to fit drop-like and non-drop-like *σ*_1−*ET* with a single formulae. Such a hypothetical fitting failure would manifest as a “phase transition” discontinuity from non-drop-like to drop-like *σ*_1−*ET*. However, no such “phase transition” is observed in the selected series of (H₂O)_1-6 isomers. Thus, unlike the case of structural and solubility properties, the “magic number” of six waters in the prism and cage (H₂O)₆ isomers do not bring about any hint of water radiolysis processes in bulk water. Likely, an extension of the current (H₂O)_1-6 series with the (H₂O)_7-20 isomers may bring about some type of bulk-water manifestations. Total 1-ET ICSs *σ*_1−*ET* are not very detailed properties and cannot reveal some dynamical details of 1-ET processes. Therefore, Figs and show the orientation-averaged 1-ET probabilities ${\overline{P}}_{1 - ET}^{}(b)$ and their *b* – weighted counterparts $b{\overline{P}}_{1 - ET}^{}(b)$ vs. *b* for the considered (H₂O)_1-6 isomers. With both basis sets, the ${\overline{P}}_{1 - ET}^{}(b)$ are high in value at small impact parameters (roughly, *b*≤6 a.u.) corresponding to close projectile-cluster encounters but they decrease rapidly at larger impact parameters. The ${\overline{P}}_{1 - ET}^{}(b)$ vs. *b* plots show a variable number of maximum peaks (from one to three) depending on the considered water cluster. The $b{\overline{P}}_{1 - ET}^{}(b)$ vs. *b* plots exhibit similar patterns to those of the ${\overline{P}}_{1 - ET}^{}(b)$ but modulated by the *b* value. As the integrands of the *σ*_1−*ET*, the $b{\overline{P}}_{1 - ET}^{}(b)$ plots indicate that the most important contributions to the *σ*_1−*ET* come from 1-ET processes arising from intermediate impact parameters (roughly, 2 a.u. ≤ *b* ≤ 9 a.u.). ## Fragmentation reactions Unlike time-independent scattering methods applicable to PCT, SLEND simulations can reveal the reactants-to-products time evolution of the chemical reactions underlying the 1-ET processes. To study those reactions, the SLEND/6-31G\* simulations to calculate *σ*_1−*ET* of H⁺ + (H₂O)_1-4 (only tetramer a for *n* = 4) were prolonged from their simulation times of 30.00 a.u. (0.7257 fs) to 1,000 a.u. (24.19 fs) using the impact parameter grids \[*b*\]₂ in. This much longer simulation time allows for the manifestation of fragmentation reactions that occur at longer time scales. In fact, no fragmentation was observed within the original time of 30.00 a.u. As shows, the final SLEND electronic wavefunction \|**z**(*t*),**R**(*t*)⟩ is a superposition of various UHF states corresponding to various products’ channels, e.g. $\left. \text{H}_{proj.}^{+} + \text{H}_{2}\text{O}\rightarrow\text{H}_{proj.}^{q_{1} = + 1} + \text{H}_{2}^{q_{2} = 0} + \text{O}_{}^{q_{3} = 0} \right.$ or $\left. \rightarrow\text{H}_{proj.}^{q_{1} = 0} + \text{H}_{2}^{q_{2} = + 1} + \text{O}_{}^{q_{3} = 0} \right.$ or $\left. \rightarrow\text{H}_{proj.}^{q_{1} = - 1} + \text{H}_{2}^{q_{2} = + 1} + \text{O}_{}^{q_{3} = + 1} \right.$, etc., where $\text{H}_{proj.}^{q_{1} = 0, + 1, - 1}$ is the incoming/outgoing projectile; these channel states occur with different probabilities. Therefore, when is applied to all the well-separated fragments at final time, *N*_*α* and *N*_*β* and their corresponding charges *q*_*i* (e.g. *q*₁ = 1−*N*_*α*−*N*_*β* for the final $\text{H}_{proj.}^{q_{1}}$) are not necessarily integer numbers corresponding to canonical chemical species (e.g. $\text{H}_{proj}^{q_{1} = + 1}$, $\text{H}_{proj}^{q_{1} = 0}$ and $\text{H}_{proj}^{q_{1} = - 1}$) but fractional numbers as the averages of the number of electrons and charges over the channels’ probabilities. This was always the case in all the present simulations not leading to clusters fragmentations (i.e. $\left. \text{H}_{proj.}^{+ 1} + \left( {\text{H}_{2}\text{O}} \right)_{n}\rightarrow\text{H}_{proj.}^{q_{1}} + \left( {\text{H}_{2}\text{O}} \right)_{n}^{1 - q_{1}} \right.$, with *q*_*i* continuously varying in the range −1≤*q*_*i*≤+1). However, as in previous SLEND studies of H⁺ + (H₂O)_1–4 at *E*_*Lab* = 1 keV, the present simulations leading to clusters fragmentations always bring about outgoing projectiles $\text{H}_{proj.}^{q_{1}}$ and clusters fragments $\text{A}^{q_{2}}$ with *q*₁ = +1 and *q*₂ = 0, respectively (cf. caption). Thus, the present SLEND simulations predict that the fragmentation channel leading to outgoing H⁺ and neutral fragments predominates over the others. However, it is known experimentally that proton- water collisions lead to fragmentations into ions. SLEND can properly describe fragmentations into ions as shown in previous studies \[e.g. cf. Refs. \]. Therefore, to allow the manifestation of those types of fragmentations here, it will be necessary to prolong even further the simulation time of the present calculations or, more likely, increase the number of total simulations by using a finer grid. Such improvements entail further computational cost and will be attempted later. The predicted fragmentations are: **H**^**+** **+ H**_**2****O simulations:** 9 out of 252 simulations exhibited the H₂O target fragmenting into: H + OH (2 simulations), H + H + O (6 simulations) and H₂ + O (1 simulation, cf. ). **H**^**+** **+ (H**_**2****O)**_**2** **simulations:** 9 out of 540 simulations exhibited the (H₂O)₂ target fragmenting into: H₂O + HO + H (2 simulations), H₃O + O + H (1 simulation), H₂O + 2H + O (5 simulations) and H₃O + OH (1 simulation, cf.). **H**^**+** **+ (H**_**2****O)**_**3** **simulations:** 3 out of 540 simulations exhibited the (H₂O)₃ target fragmenting into: H₃O + OH + H₂O (2 simulations) and H + OH + 2H₂O (1 simulation). **H**^**+** **+ (H**_**2****O)**_**4** **(tetramer a) simulations:** 10 out of 748 simulations exhibited the (H₂O)₄ target fragmenting into: (H₂O)₂ + 2H₂O (4 simulations), H₃O + OH + 2H₂O (4 simulations), an H + OH + 3H₂O (2 simulations) In conclusion, present SLEND/6-31G\* simulations predict the DNA-damaging radicals H, OH, O and H₃O, and the innocuous species H₂O and (H₂O)₂ as water radiolysis products. To illustrate the predicted fragmentations, we present a few animation stills from some representative simulations. shows five sequential stills of the animation of $\left. \text{H}_{proj.}^{+} + \text{H}_{2}\text{O}\rightarrow\text{H}_{proj.}^{+} + \text{H}_{2} + \text{O} \right.$ and a sixth panel plotting the Mulliken populations of the H_*proj*., H₂ and O moieties vs. time. This last panel permits the observation of the time evolution of the electrons. Mulliken populations are basis-set dependent and, more importantly, somewhat arbitrary in the way they distribute electrons over neighboring fragments. However, in previous SLEND studies, Mulliken populations were good predictors for the time- evolution of the electrons over atoms and fragments. Furthermore, when all the fragments are well-separated at final time, the Mulliken populations converge to the unequivocal *N*_*α* and *N*_*β* in (cf. caption). In , the colored spheres represent the classical nuclei (white = H and red = O), and the colored clouds represent selected electron density iso-surfaces (from red = lowest density to blue = highest density). The incoming projectile $\text{H}_{proj.}^{+}\ $ passes in between the H atoms of H₂O, hits the O atom and bounces back. As a result of this collision, the H₂ and O moieties of H₂O break apart; the ejected H₂ moiety undergoes a series of strong oscillations ranging from the near dissociation of H₂ into H atoms to these atoms’ recombination back to H₂. Finally, shows the nine predicted fragments in H⁺ + (H₂O)₂. ## Further analysis and improvements SLEND/6-31G\*\* *σ*_1−*ET* compares unsatisfactorily with experiments in contrast to SLEND/6-31G\* *σ*_1−*ET*. Indeed, it is surprising that SLEND/6-31G\*\* performs worse than SLEND/6-31G\* in these calculations since common knowledge dictates that the 6-31G\*\* basis set is better than the 6-31G\* one. 6-31G\*\* is constructed from 6-31G\* by augmenting the latter with p-type basis functions on the hydrogen atoms; thanks to this augmentation, 6-31G\*\* provides better time-independent molecular properties than 6-31G\*. For instance, the UHF/6-31G\*\* energies and geometries of (H₂O)_1–6 are more accurate than the UHF/6-31G\* ones. However, this comparative time-independent performance does *not* necessarily extend to time-dependent calculations since these basis sets were designed to calculate static properties. To explain the SLEND/6-31G\*\* *σ*_1−*ET* overestimation, one should remember that the main component in the bound-to- bound SLEND *σ*_1−*ET* is the 1-ET probability *P*_1−ET (cf. Eqs). However, as derived in Ref. and supposed in previous SLEND studies, the *P*_*n*−ET, 0 ≤ *n* ≤ 2, in assume that the probabilities of ETs from the target A to the projectile H⁺ are dominated by transitions with electrons transferring into the localized, discrete, bound states of H: H⁺ + A: → H⋅ + A⋅⁺ \[pure charge-transfer (CT) processes\]; instead, transitions with electrons scattering into the delocalized, continuous, unbound states of H are considered negligible: H⁺ + A: → (H⁺ + *e*⁻) + A⋅⁺ \[direct ionization (DI) processes\]. Typical quantum chemistry basis sets, such as 6-31G\* and 6-31G\*\*, are ultimately based on localized primitive Gaussian functions so that occupied spin-orbitals {*ψ*_*h*} below the Fermi level represent localized, bound states in the discrete part of the spectrum. However, as a by-product of the UHF procedure, diffuse virtual spin-orbitals {*ψ*_*p*} above the Fermi level approximately represent some of the delocalized, unbound states in the continuous part of the spectrum. Therefore, the virtual space constitutes the so-called quasi-continuum that may accommodate DI processes. However, if the basis set is not large, the DI contributions of a small quasi-continuum to the ET processes become negligible in comparison to the CT contributions of the occupied space. Under those conditions, the ET probabilities *P*_*n*−ET in basically correspond to bound-to-bound (occupied-space-to-occupied-space) CT processes. For that reason, with relatively small basis sets, those *P*_*n*−ET consistently rendered correct bound-to-bound CT *σ*_1−*ET* in various systems. However, if the basis set is large, the DI contributions of an enlarged quasi-continuum may become substantial. If so, the *P*_*n*−ET in and resulting *σ*_*n*−*ET* no longer correspond to pure bound-to-bound CT processes since they get contaminated with bound-to-quasi-continuum DI contributions. Therefore, one can hypothesize that SLEND with the smaller 6-31G\* basis set can predict genuine CT *σ*_*n*−*ET* via but not with the larger 6-31G\*\* one. To verify the above hypothesis, we performed a series of SLEND simulations on the simple model system: H⁺ + H at 40 keV ≤ *E*_*Lab* ≤ 100 keV with the 6–31⁺⁺G\*\* basis set. The latter produces the best DI results for H⁺ + H as shown shortly. However, instead of using *P*_1−*ET* in for *σ*_1−*ET*, the final-time electronic wavefunction $\left| {\Psi_{e}^{SLEND}\left( t_{f} \right)} \right\rangle$, was projected onto the ground \|0⟩ and excited states \|… *h* → *p* …⟩ of the target and the projectile. In this way, the evaluation of ET probabilities could distinguish the cases with the electron transferring into bound or unbound states of H. In the 40 ≤ *E*_*Lab* ≤ 80 keV range, where experimental results are available, the DI ICSs, *σ*_*DI*, deviate less than 10% from experimental data, with the best agreement at *E*_*Lab* = 60 keV: $\sigma_{DI}^{SLEND}$ = 13.9Å and $\sigma_{DI}^{EXPT.}$ = 13.8 Å ⇒ deviation Δ_*Theo*. = 0.7%. Notably, these calculations produced accurate results even though ETFs were neglected as in ; this gives extra support to the ETFs’ neglect in the H⁺ + (H₂O)_*n* simulations and ruled it out as a source of the SLEND/6-31G\*\**σ*_1−*ET* overestimation. For H⁺ + H at *E*_*Lab* = 100 keV, SLEND *σ*_*DI* is 0.92 Ų. If *P*_1−*ET* in is used to calculate CT ICSs, *σ*_*CT*, a *σ*_*DI* contribution of 0.92 Å will be spuriously added to the *σ*_*CT* making it overestimated. If this DI contribution is assumed to be similar to that in H⁺ + H₂O, the overestimated SLEND/6-31G\*\* $\sigma_{CT}^{SLEND}$ via can be corrected by subtracting the $\sigma_{DI}^{SLEND}$ part from it: $\sigma_{CT}^{SLEND}$ = 2.06 Å2–0.92 Ų = 1.14 Ų; this places SLEND/6-31G\*\* $\sigma_{CT}^{SLEND}$ within the range of the four experimental values and closest to that from Exp. D: $\sigma_{CT}^{Expt.D}$ = 1.13 Å ⇒ Δ_*Theo*. = 0.8%. A similar correction might occur with the SLEND/6-31G\* $\sigma_{CT}^{SLEND}$ but it will be far smaller due to a smaller virtual space as suggested by previous calculations with comparable basis sets. The calculation of the CT *σ*_*CT* in H⁺ + H₂O is more complicated than that of H⁺ + H because the former has more than one electron. For H⁺ + H₂O, numerous excited states \|… *h* → *p* …⟩ from \|0⟩ forming a full CI expansion should be generated so that $\left| {\Psi_{e}^{SLEND}\left( t_{f} \right)} \right\rangle$ in can be projected on all those states. This more demanding capability is not currently available in PACE but is under development. # Conclusions To model microscopic processes in PCT, the SLEND method was applied to the H⁺ + (H₂O)_*n* systems at *E*_*Lab* = 100 keV with the 6-31G\* (*n* = 1–6) and 6-31G\*\* (*n* = 1–5) basis sets. Ten (H₂O)_1–6 clusters were selected for this study: eight exhibit mono- and di-cyclic quasi-planar/multiplanar structures and two others, the prism and cage (H₂O)₆ isomers, exhibit multi-cyclic, three-dimensional, drop-like structures. These “smallest-drop” clusters were purposely included in this study in an attempt to reproduce early manifestations of bulk-water properties in PCT. Short-time SLEND/6-31G\* (*n* = 1–6) and /6-31G\*\* (*n* = 1–5) simulations render cluster-to-projectile total 1-ET ICS, *σ*_1−*ET*, and 1-ET probabilities. In absolute quantitative terms, SLEND/6-31G\**σ*_1−*ET* compares satisfactorily with alternative experimental and theoretical results only available for *n* = 1, and exhibits almost the same accuracy of the best alternative theoretical result from BGM calculations. SLEND/6-31G\*\* overestimates *σ*_1−*ET* and a detail account about the cause and remediation of this effect was presented. In relative quantitative terms, both SLEND/6-31\* and /6-31G\*\* *σ*_1−*ET* precisely fit into physically justified scaling formulae *σ*_1−*ET* (*n*) = *cn*^2/3 as a function of the cluster size *n*. Long-time SLEND/6-31G\* (*n* = 1–4) simulations predict the formation of the DNA-damaging radicals H, OH, O and H₃O. While “smallest-drop” isomers were included, no incipient manifestations of bulk-water PCT properties are observed. Therefore, to capture bulk-water effects, simulations with larger water clusters are currently underway. The authors thank Prof. Takao Oi (Sophia University, Japan) for providing the optimized geometry of the prism isomer of (H₂O)₆. Present calculations were performed at the Texas Tech University High Performance Computer Center and the Texas Advanced Computing Center at the University of Texas at Austin. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** JAM. **Data curation:** AJP EST JAM. **Formal analysis:** AJP CS EST JAM. **Funding acquisition:** JAM. **Investigation:** AJP CS EST JAM. **Methodology:** AJP JAM. **Project administration:** JAM. **Resources:** JAM. **Software:** AJP EST JAM. **Supervision:** JAM. **Validation:** AJP CS EST. **Visualization:** AJP EST. **Writing – original draft:** AJP JAM. **Writing – review & editing:** JAM. [^3]: Current address: Department of Chemistry and Biochemistry, Lipscomb University, Nashville, Tennessee [^4]: Current address: Department of Chemistry and Industrial Hygiene, University of North Alabama, Florence, Alabama, United States of America

# Introduction Intuitively, we all grasp the concept of cognitive effort. We all know what it is like to feel that a cognitive task is demanding and effortful. However, defining cognitive effort is not straightforward. Cognitive effort refers to the degree of engagement in demanding cognitive tasks, as opposed to tasks that can be completed using routine or habitual behavior, which require little effort. An influential neuroeconomics approach to effort postulates that deciding whether to invest cognitive effort comes down to investigating the relevant costs and benefits (see for example). More specifically, several theories have proposed that effort may be primarily implicated in the decision to engage cognitive control resources (\[–\]; for a review see). The broad concept of cognitive control comprises cognitive operations such as planning a new strategy, evaluating it, controlling its execution and correcting possible errors. It kicks in when routine activation of behavior is no longer sufficient for optimal performance. Cognitive control thus allows us to perform intelligent, purposive behavior. When performing basically any task or when we aim to achieve a goal, we need to stay focused and inhibit dominant, yet irrelevant information to prevent being distracted from our task. Whenever an irrelevant source of information interferes with our task performance, we will exert additional cognitive control in order to preserve task performance and resolve conflict. Conflict can be defined as the simultaneous activation of incompatible representations. In an experimental context, conflict tasks such as the Eriksen flanker task are often used to measure cognitive control. This task consists of a central target arrow flanked by distractor arrows. Congruent (i.e., flankers and target point in the same direction) and incongruent (i.e., flankers and target point in opposite directions) stimuli are presented and participants must indicate the direction of the central target arrow as fast and as accurately as possible. Responses to incongruent trials are slower and are more error prone than to congruent trials. This difference in performance between incongruent and congruent trials constitutes the flanker effect. Here, dominant yet task irrelevant stimuli (i.e., flanker arrows) need to be suppressed in order to perform this task well. When people encounter interference, such as conflict as defined above, cognitive control allows them to adjust their behavior to overcome it. However, exerting cognitive control comes at a cost and this is where cognitive effort comes in. The neuroeconomic theories treat cognitive effort as the opportunity cost of exerting cognitive control (i.e., the decision to expend cognitive control is accompanied by the missed benefit of avoiding effort; see for a review). Cognitive control usually drives cognitive effort and exerting more cognitive control would lead to a higher effort cost. While we are performing a demanding cognitive task, not only do we have a sense of cognitive effort, we are also subjectively aware that we are initiating, executing and controlling our thoughts and actions. The feeling of being in charge when we perform voluntary actions is called the sense of agency. The brain appears to actively construct the sense of agency using motor actions, sensory feedback, previous experiences, cause-and-effect inferences, and so on. The core of sense of agency is the association between a voluntary action (e.g., pressing a light switch) and an outcome (e.g., the light in the room goes on). Experimentally, sense of agency can be explicitly measured by simply asking participants to judge whether their action caused an outcome. Alternatively, implicit measures can be used, such as the compression of perceived time between action and outcome. In the intentional binding paradigm, participants judge the perceived time of a voluntary action or a subsequent outcome. It has been shown that voluntary actions (as compared to involuntary movements) are perceived as shifted in time towards their subsequent outcomes and that the outcomes are perceived as shifted towards the voluntary actions that caused them. Based on this implicit measure, sense of agency is defined as the compression (or underestimation) of the perceived time interval between action and outcome. Crucially, effort and sense of agency have been linked to each other, dating back even to the 19^th century (for an overview see). It has been argued that without the subjective experience of effort there could not be any feeling of agency or causality, but only mere facts of behavior. According to this view, it is the conscious experience of effort that makes self-knowledge possible. If effort is a crucial determinant of sense of agency, then experimental manipulations of cognitive effort should also influence the accompanying sense of agency. A few studies have shown that sense of agency can indeed be influenced by increased cognitive effort. Howard et al. instructed participants to memorize two (low effort) or eight (high effort) letters. After encoding, a maintenance period started by a self- initiated action which ended after a variable duration with an outcome (i.e., a tone). Participants’ estimation of the duration of this maintenance period was used to measure intentional binding. The results showed that a high cognitive effort context decreased intentional binding compared to a low cognitive effort context. Similar results were obtained by Hon et al. who used an explicit instead of an implicit measure of sense of agency. Contrarily, Demanet et al. observed increased intentional binding in a high physical effort context. In the study of Sidarus and Haggard, participants responded to flanker trials (i.e., the action) which triggered the appearance of a colored circle after a variable delay (i.e., the outcome). They judged how much control they felt over the colored circles that were triggered by their actions. Results indicated that incongruent flanker trials led to lower sense of agency compared to congruent and neutral flanker trials. Vastano et al. reached the same conclusion using an implicit intentional binding measure of sense of agency. However, although Wang et al. also found that sense of agency ratings were lower for incongruent trials, this appeared to be the case only when the previous trial was a congruent trial and thus did not contain conflict. Similarly, Di Costa et al. observed that sense of agency measured with intentional binding is increased *after* making an error. Thus, previous studies have shown that cognitive effort can be both detrimental and facilitative for the experienced sense of agency. We hypothesize that the reason for the nuances in these findings might lie in the use of differential time windows in which cognitive effort operates. Indeed, the findings seem to depend on whether the effort that was taken into account was exerted on the current trial, the previous trial or across a block of trials. When the *current* trial contained a high level of effort (e.g., incongruent or error trial), sense of agency on that trial decreased. However, sense of agency seemed to be increased on trials *following* high effort. In *contexts* with high effort, the results are mixed: some studies found increased sense of agency while others observed decreased sense of agency. Sidarus and Haggard argued that when an increase of effort can be anticipated, the required cognitive control and expected self-engagement in the task can become part of action planning and prediction. In contrast, effort investment that is sudden and unexpected, for example triggered by unexpected or unpredictable conflict, cannot be predicted and included in action planning. Based on this, we speculate that when high effort is required unpredictably or unexpectedly (e.g., on the current trial), sense of agency will decrease; contrarily, when high effort is anticipated (e.g., after a conflict or in a high conflict context), it should be accompanied by increased sense of agency. We formulated three specific hypotheses. First, in a block of trials where effort is required unpredictably, encountering a high- effort trial will lead to a decreased sense of agency *on that same trial* compared to a low-effort trial. Second, in a block of trials where effort is required unpredictably, encountering a high-effort trial will lead to an increased sense of agency *on the next trial* compared to a low-effort trial. Third, in a block of trials where effort can be anticipated (i.e., frequent high-effort trials), sense of agency will be increased compared to a block of trials where effort cannot be anticipated (i.e., scarce high-effort trials). In order to assess these hypotheses we will use a flanker task to manipulate conflict and hence the required cognitive control and, in turn, the required effort (i.e., on congruent trials, no conflict is present and thus the required cognitive control and the effort cost is minimal; contrarily, on incongruent trials, conflict is present and thus the required cognitive control and effort cost is larger). In Experiment 1 sense of agency will be measured implicitly using the intentional binding paradigm. # Experiment 1 ## Method ### Participants Sixty participants were recruited through the Experiment Management System of the Katholieke Universiteit Leuven. They received course credit for participation. All participants provided written informed consent. All of them had normal or corrected-to-normal eyesight, were not colorblind and were able to operate a keyboard and mouse. We used the following exclusion criteria for participants: response times and/or intentional binding exceeding 2.5 *SD*, and/or error rates above 20%. However, none of the participants met these exclusion criteria. Thus, all 60 participants were included for the analyses (6 males, mean age = 18.52, *SD* = 1.13, range 18–26). This study was approved by the Social and Societal Ethics Committee (SMEC) from KU Leuven (G-2019 01 1493). The study was also preregistered on the Open Science Framework (OSF, osf.io; doi:[10.17605/OSF.IO/EM3GQ](https://doi.org/10.17605/OSF.IO/EM3GQ)) and the raw data can be retrieved from <https://osf.io/em3gq/> (doi:[10.17605/OSF.IO/EM3GQ](https://doi.org/10.17605/OSF.IO/EM3GQ)). ### Apparatus The experiment was administered in a computer room in small groups. Stimuli were presented on a 22” monitor with a refresh rate of 60 Hz located approximately 60 cm from the participant. Stimulus presentation and response registration were controlled by PsychoPy v.3.1.0 (Psychology software for Python;). ### Design and procedure The experiment was composed of active and passive trials. During *active trials* an Eriksen flanker task was used in which a string of numbers was presented in white against a grey background in the center of the screen on each trial (font = Consolas, height = 1.2 deg). A central target number was flanked by two distractor numbers on both sides. Participants were instructed to respond as fast and as accurately as possible to the central target arrow and to ignore the flankers. The flanker stimuli could either trigger the same response as the target (i.e., congruent trials; e.g., “22222”), or trigger a different response as the target (i.e., incongruent trials; e.g., “33233”). The stimuli used were 1, 2, 3 and 4. Participants had to respond by pressing the corresponding key (1, 2, 3 or 4) on an Azerty keyboard. Specifically, the keys of interest were indicated by stickers: sticker of the number “1” was placed on the “d” key, a “2” on the “f” key, a “3” on the “j” key and a “4” on the “k” key. To speed up response time, participants were instructed to keep their left middle finger on the “1”, their left index finger on the “2”, their right index finger on the “3” and right middle index finger on the “4”. Active trials were announced by a white fixation cross (1000ms) that was followed by the flanker stimulus until a response was provided. After a variable delay of between 500 and 1250ms (in steps of 250ms;), a colored circle was presented on the screen for 200ms (size = (2, 2) deg). The color of the circle depended on the response made on the current trial. Specifically, different colors were linked to passive trials (see below) and to button presses on “1”, “2”, “3” and “4”, leading to five colors (i.e., RGB color space values of pink \[1.000, -1.000, 1.000\], yellow \[1.000, 1.000, -1.000\], cyan \[-1.000, 1.000, 1.000\], orange \[1.000, 0.000, -1.000\] and blue \[-1.000, -1.000, 1.000\]). Thus, on active trials an action (i.e., a button press) triggered the appearance of a colored circle, whereas on passive trials no action triggered the appearance of a colored circle. Note that we avoided the use of green and red to prevent any associations with correct/incorrect connotations. Which color was related to which response was randomized across participants. After the disappearance of the circle, participants performed an Interval Reproduction Task in which they reported the estimated length of the delay between the disappearance of the flanker stimulus and the appearance of the circle. They responded by continuously pressing the space bar for the estimated duration. For a schematic overview of an active trial, see. During *passive trials* a string of neutral stimuli (i.e., “00000”) was presented in black against a grey background in the center of the screen. Participants were instructed that these trials did not require a response. As participants did not respond to these stimuli, they were therefore non-agentic. If participants did respond to a passive trial, this was registered as an error. Passive trials were announced by a black fixation cross (1000ms) followed by a neutral stimulus (i.e., “00000”) that was presented for a duration jittered between 500 and 1500ms. After a variable delay of between 500 and 1250ms a colored circle was again presented. Next, participants again had to perform the Interval Reproduction Task in which they estimated and reproduced the length of the delay between the neutral stimulus disappearance and the circle appearance. For a schematic overview of a passive trial, see. Note that on passive trials both the fixation cross and the neutral stimulus were presented in a different color (i.e., black) compared to the active trials (i.e., white) to make the non- agentic nature of these trials very clear to the participants. The experiment started with several practice phases. First, participants received the instructions about the active trials and practiced only the flanker task for 10 trials (five congruent and five incongruent trials). Next, instructions were provided about the presence of the passive trials and participants practiced this on 10 trials where both active and passive trials were presented (six passive, two congruent and two incongruent trials). During these two practice phases, participants received feedback about their responses (i.e., “correct” or “wrong” and their response time in seconds). After this, participants were informed that a circle would appear after the disappearance of the flanker or neutral stimulus and that the color of this circle would depend on their response. They received instructions about the Interval Reproduction Task. Finally, they were also told that after each block they would receive a few additional questions. The first question assessed the level of experienced effort after each block (i.e., “how much effort did you have to exert during the previous block?”). Three additional questions assessed the subjective feeling of agency over congruent, incongruent and passive trials, respectively, during the previous block. For each type of trial (e.g., congruent trials), the trials were listed on the screen (e.g., 11111, 22222, 33333 and 44444) and participants were asked: “To what extent did you have the feeling that you caused the colored circle to appear on the following trials?”. For all of these post-block questions, participants indicated their response using a continuous rating scale ranging from “very little” to “a lot” by a mouse button click on the selected location on the rating scale. In a final practice block, they practiced the full procedure for 10 trials (4 passive, 3 congruent and 3 incongruent trials). After these practice phases, the experimental phase began where feedback was no longer provided. Active and passive trials were presented intermixed and the inter-trial interval was 1000ms. Experimental trials were divided into three blocks, each containing 216 trials. The blocks varied depending on the ratio of incongruent, congruent, and passive trials: one block with ratio 4:1:1 (i.e., MI block), one with ratio 1:4:1 (i.e., MC block), and one 5:5:2 (i.e., EQ block) ratio of incongruent to congruent to passive trials. The block that had mostly incongruent trials (MI block) served as the high effort context and the bock that had mostly congruent trials (MC block) served as the low effort context. The block with an equal ratio of congruent and incongruent trials (EQ block) served as the block where effort was required unpredictably, to assess the impact of current and previous trial effort on sense of agency. The order of trials in each block was randomized. The order of the blocks was counterbalanced across participants. Finally, after each block, participants received the additional questions to assess exerted effort and experienced agency over each trial type, as described above. In between blocks, participants could take a self-paced break. The whole experiment did not take more than one hour. The experiment was followed by a short debriefing explaining the goal of the study. ### Statistical analysis Active trials exceeding 2.5 *SD* of the overall mean RT were excluded from all analyses (1.7% of trials). For RT and intentional binding analyses, erroneous flanker trials were removed (5.9% of trials). For trial-by-trial analyses assessing hypotheses 1 and 2, the first trial (0.46% of trials) and trials following an erroneous flanker trial (5.8% of trials) were also excluded in the EQ block. No trials met our exclusion criterion with regards to intentional binding (i.e., trials exceeding 2.5 *SD* of the overall mean intentional binding). First, we calculated the measure of intentional binding for active and passive trials based on the Interval Reproduction Task. Specifically, for active trials, the participant’s estimated delay (in milliseconds) was subtracted from the actual delay between their response on the flanker stimulus (i.e., response) and the appearance of the colored circle (i.e., outcome) for each trial (i.e., actual delay—estimated delay). For the passive trials, we subtracted the participant’s estimated delay from the actual delay between the disappearance of the neutral stimulus and the appearance of the colored circle for each trial. Note that a value of zero indicates a perfect estimate, and hence no intentional binding. Positive values indicate an underestimation of the time period, and thus intentional binding. As a control check, we assessed whether our manipulation of intentional binding was successful by looking at the difference in average attentional binding between active (i.e., congruent and incongruent) and passive trials using a paired-samples *t*-test. If our task was indeed successful in eliciting intentional binding, we should observe that intentional binding in active trials was larger than in passive trials (where no action is present). Note that comparing active (i.e., action) and passive (i.e., no- action) trials is an established paradigm in this field (e.g.,). Second, to test hypotheses 1 and 2, we focused on the EQ block where effort is required unpredictably (i.e., equal proportion of incongruent and congruent trials). For hypothesis 1, we examined whether encountering a high-effort trial leads to a decreased sense of agency *on that same trial* compared to a low- effort trial. For hypothesis 2, we examined whether encountering a high-effort trial leads to an increased sense of agency *on the next trial* compared to a low-effort trial. We conducted a 2x2 repeated measures ANOVA analysis with Current and Previous trial congruency as within-subjects factors (both with 2 levels: congruent and incongruent) on intentional binding (in ms), RTs (in ms) and error rates (in %) as (separate) dependent variables. Third, to test hypothesis 3, we examined whether sense of agency was increased in a block of trials where effort can be anticipated (i.e., MI block), compared to a block of trials where effort cannot be anticipated (i.e., MC block). For this purpose, a block analyses was conducted. We performed a 2x2 repeated measures analysis with Block (2 levels: MC and MI) and Current trial congruency (2 levels: congruent and incongruent) as within-subjects factors on intentional binding, RTs and error rates as (separate) dependent variables. Note that in the analyses described above, we did not pool all trials across blocks to assess the general effect of current trial congruency on intentional binding. As the different contexts that were created by manipulating the proportion of congruent trials (i.e., EQ, MC and MI) trigger different cognitive control mechanisms, and hence differential congruency effects (see for example), we a priori decided not to pool across blocks. However, an additional exploratory repeated measures analysis across all blocks with current trial congruency as within-subjects factor revealed that the main effect of congruency on intentional binding was not significant (*F*(1,59) = 0.58, *p* =.46, *η*²_p = 0.010). Finally, we looked at the subjective ratings of experienced sense of agency and cognitive effort reported after each block for exploratory purposes. We conducted a repeated measures analysis with block (3 levels: EQ, MC or MI) as within-subjects factor on the reported experienced effort after each block and a repeated measures analysis with trial type (3 levels: congruent, incongruent or passive) as within-subjects factor on the reported explicit sense of agency after each block. ## Results ### Intentional binding check A paired-samples *t*-test confirmed that intentional binding was larger on active compared to passive trials (407 versus 293ms, *t*(59) = -7.41, *p* \<.001). This ensures that our implicit measurement of sense of agency was successful. presents the participants’ average estimated delays as a function of block and actual delay (i.e., 500, 750, 1000 or 1250ms) for the correct active trials (excluding trials exceeding 2.5 *SD* of the overall mean RT) and the passive trials. ### Trial-by-trial analyses These analyses were conducted on the EQ block. We conducted a 2x2 repeated measures analysis with Current and Previous trial congruency as within-subjects factors (both with 2 levels: congruent and incongruent) on intentional binding (in ms), RTs (in ms) and error rates (in %) as (separate) dependent variables. Means and *SD*s for each of these dependent variables in relation to current and previous congruency are reported in. The results are also depicted on. With regard to *intentional binding*, we found no significant main effects of current or previous congruency (resp. *F*(1,59) = 0.15, *p* =.70, *η*²_p = 0.003 and *F*(1,59) = 0.011, *p* =.92, *η*²_p \< 0.001), nor a significant interaction (*F*(1,59) = 0.16, *p* =.69, *η*²_p = 0.003). This indicates that we observed no difference in intentional binding depending on current or previous congruency. With regard to *RT*, we found a significant main effect of current congruency (*F*(1,59) = 90.65, *p* \<.001, *η*²_p = 0.61) indicating that participants were slower on incongruent compared to congruent trials (on average 719.5ms versus 664.5ms). Additionally, we observed a significant interaction between current and previous congruency (*F*(1,59) = 8.85, *p* =.004, *η*²_p = 0.13). This interaction reflects a typical Gratton effect: a decreased congruency effect after an incongruent compared to a congruent trial (40ms versus 70ms). The observation of this Gratton effect ensures that our manipulation of current and previous congruency was successful in order to trigger trial-by-trial adaptations. The main effect of previous congruency did not reach significance (*F*(1,59) = 0.95, *p* =.33, *η*²_p = 0.016). With regard to *error rates*, we only observed a significant main effect of current congruency (*F*(1,59) = 21.56, *p* \<.001, *η*²_p = 0.27) indicating that participants made more errors on incongruent compared to congruent trials (on average 7.38% versus 4.39%). The main effect of previous congruency and the interaction between current and previous congruency did not reach significance (resp. *F*(1,59) = 1.32, *p* =.26, *η*²_p = 0.022 and *F*(1,59) = 0.77, *p* =.38, *η*²_p = 0.26). As an exploratory analysis that was not included in our preregistration, we additionally conducted a 4x2x2 repeated measures analysis with Actual delay (4 levels: 500, 750, 1000 or 1250ms), Current and Previous trial congruency (both with 2 levels: congruent and incongruent) as within-subjects factors on intentional binding (in ms). We only observed a main effect of Actual delay (*F*(3,56) = 285.67, *p* \<.001, *η*²_p = 0.94), with more intentional binding for longer delays (specifically, 132.5, 313, 501 and 670ms for the increasing delays). None of the other main effects or interactions reached significance. ### Block analyses These analyses were conducted on the MC and MI blocks. We conducted a 2x2 repeated measures analysis with Block (2 levels: MC and MI) and Current trial congruency (2 levels: congruent and incongruent) as within-subjects factors on intentional binding (in ms), RTs (in ms) and error rates (in %) as (separate) dependent variables. Means and *SD*s for each of these dependent variables in relation to current and previous congruency are reported in. The results are also depicted on. With regard to *intentional binding*, we found no significant main effect of congruency (*F*(1,59) = 1.56, *p* =.22, *η*²_p = 0.026), nor a significant interaction between block and congruency (*F*(1,59) = 0.048, *p* =.83, *η*²_p = 0.001). The main effect of block also did not reach significance, but showed a slight trend (*F*(1,59) = 3.00, *p* =.089, *η*²_p = 0.048). Indeed, we observed slightly more binding in the MI compared to the MC block, as hypothesized (419.5 versus 398.5ms). With regard to *RT*, we found a significant main effect of congruency (*F*(1,59) = 286.29, *p* \<.001, *η*²_p = 0.83) indicating that participants were slower on incongruent compared to congruent trials (on average 738ms versus 671.5ms). Additionally, we observed a significant interaction between block and congruency (*F*(1,59) = 45.80, *p* \<.001, *η*²_p = 0.44). This interaction reflects a typical proportion congruency effect: a decreased congruency effect in an MI block compared to an MC block (44ms versus 89ms). The observation of this proportion congruency effect ensures that our manipulation of block was successful in order to trigger blockwise adaptations. The main effect of block did not reach significance (*F*(1,59) = 0.056, *p* =.81, *η*²_p = 0.001). With regard to *error rates*, we observed a significant main effect of congruency (*F*(1,59) = 31.93, *p* \<.001, *η*²_p = 0.35) indicating that participants made more errors on incongruent compared to congruent trials (on average 7.80% versus 4.77%). A main effect of block was also observed (*F*(1,59) = 10.91, *p* =.002, *η*²_p = 0.16), indicating that participants made more errors on the MC block compared to the MI block (7.17% versus 5.40%). Finally, the interaction between block and congruency also reached significance (resp. *F*(1,59) = 1.32, *p* =.26, *η*²_p = 0.022), again indicating a proportion congruency effect: a decreased congruency effect in an MI block compared to an MC block (1.38% versus 4.68%). As an exploratory analysis that was not included in our preregistration, we additionally conducted a 4x2x2 repeated measures analysis with Actual delay (4 levels: 500, 750, 1000 or 1250ms), Block (2 levels: MC and MI) and Congruency (2 levels: congruent and incongruent) as within-subjects factors on intentional binding (in ms). We only observed a main effect of Actual delay (*F*(3,56) = 372.89, *p* \<.001, *η*²_p = 0.95), with more intentional binding for longer delays (specifically, 128, 328, 518.5 and 682ms for the increasing delays). None of the other main effects or interactions reached significance. ### Post-block subjective effort and agency analyses With regard to *subjective effort* experienced during each block (EQ, MC or MI), we conducted a repeated measures analysis with block (3 levels: EQ, MC or MI) as within-subjects factor on the reported experienced effort after each block. We observed no significant differences between the three blocks (*F*(2,58) = 0.14, *p* =.87, *η*²_p = 0.005). Indeed, the reported experienced effort was very similar in the EQ, MC and MI blocks (resp. 6.44, 6.53 and 6.36). With regard to the *subjective sense of agency* experienced for congruent, incongruent and passive trials, we conducted a repeated measures analysis with trial type (3 levels: congruent, incongruent or passive) as within-subjects factor on the reported explicit sense of agency after each block. We observed significant differences between the experienced agency over congruent, incongruent and passive trials (*F*(2,58) = 16.83, *p* \<.001, *η*²_p = 0.37). Specifically, participants reported a higher sense of agency over incongruent trials (4.81) compared to congruent (4.2, *F*(1,59) = 24.61, *p* \<.001, *η*²_p = 0.29) or passive trials (4.1, *F*(1,59) = 7.98, *p* =.006, *η*²_p = 0.12). The explicit sense of agency did not differ between congruent and passive trials (4.1, *F*(1,59) = 0.12, *p* =.73, *η*²_p = 0.02). Note that this effect did not interact with block: a repeated measures analysis with trial type (3 levels: congruent, incongruent or passive) and block (3 levels: EQ, MC or MI) as within-subjects factors again only revealed a main effect of trial type (*F*(2,58) = 16.83, *p* \<.001, *η*²_p = 0.37). ## Discussion Experiment 1 confirmed that our manipulation of cognitive control was successful: incongruent trials led to a slower and more erroneous response than congruent trials (i.e., flanker effect). After an incongruent trial, participants also increased their exertion of cognitive control compared to after a congruent trial (i.e., Gratton effect). Finally, when conflict could be anticipated (i.e., MI block), the flanker effect was reduced (i.e., proportion congruency effect). Our implicit intentional binding measure also seemed to be successful: participants reported shorter time intervals on active trials compared to passive trials. However, we were unable to confirm the results of previous studies: we observed no decrease in binding on incongruent trials, nor an increase in binding after incongruent trials. We did observe a slight trend towards increased binding in a context where conflict could be anticipated (i.e., MI block). Interestingly, our explicit measure of sense of agency, administered after each block, did indicate that participants experienced more agency over incongruent trials than over congruent trials. According to the two-step account of agency, the implicit and explicit approaches to measure sense of agency, capture separate aspects of the sense of agency, namely the implicit feeling of agency and the explicit judgement of agency. The implicit feeling of agency stems from a low-level comparator involving motor action planning and prediction, and sensory feedback processes. The explicit judgement of agency relies on higher- order causality judgements based on contextual factors and beliefs. Furthermore, although correlations between intentional binding and subjective sense of agency have been reported, whether binding truly reflects sense of agency is debated. Based on this, we decided to conduct the experiment again, but now using an explicit measure of sense of agency on each trial (see also). # Experiment 2 ## Method ### Participants Sixty participants were recruited from the Experiment Management System of the Katholieke Universiteit Leuven. They received course credit or a monetary reward for participation. All participants provided written informed consent. None of the them had participated in Experiment 1. All of them had normal or corrected- to-normal eyesight, were not colorblind and were able to operate a keyboard and mouse. We used the following exclusion criteria for participants: response times and/or error rates above 20%. Two participants did not meet these criteria based on their response time. Thus, 58 participants were included for the analyses (9 males, mean age = 20.67, *SD* = 4.57, range 17–40). This study was approved by the Social and Societal Ethics Committee (SMEC) from KU Leuven (G-2019 05 1652). The study was also preregistered on OSF (osf.io; doi:[10.17605/OSF.IO/EM3GQ](https://doi.org/10.17605/OSF.IO/EM3GQ)) and the raw data can be retrieved from <https://osf.io/em3gq/> (doi: [10.17605/OSF.IO/EM3GQ](https://doi.org/10.17605/OSF.IO/EM3GQ)). ### Apparatus, design, procedure and statistical analyses The apparatus, design and procedure were identical to Experiment 1, except for the following changes. Instead of using an implicit measure of sense of agency, as we did in Experiment 1 (i.e., intentional binding based on the Interval Reproduction Task), we now used an explicit measure of sense of agency. Specifically, after the disappearance of the colored circle, participants now had to indicate their explicit sense of agency (i.e., “I felt like I caused the circle to appear”) using a 9-point scale ranging from “strongly disagree” to “strongly agree”. This was the case for both active and passive trials. For a schematic overview of both trial types, see. Statistical analyses were also the same as in Experiment 1, except that this explicit sense of agency was now used as dependent variable instead of intentional binding. Note that, as explained in Experiment 1, we did not pool all trials across blocks to assess the general effect of current trial congruency on sense of agency ratings. However, an additional exploratory repeated measures analysis across all blocks with current trial congruency as within-subjects factor revealed that the main effect of congruency on sense of agency ratings was not significant (*F*(1,57) = 1.81, *p* =.18, *η*²_p = 0.031). Active trials exceeding 2.5 *SD* of the overall mean RT were excluded from all analyses (i.e., 1.9%). For RT and intentional binding analyses, erroneous active flanker trials were removed (i.e., 4.5%). For trial-by-trial analyses assessing hypotheses 1 and 2, the first trial of the block (0.46%) and trials following an erroneous flanker trial (i.e., 3.8%) were also excluded in the EQ block. For one participant, the sense of agency and effort ratings after each block were not recorded. Therefore, this participant was excluded from the post-block analyses. ## Results ### Sense of agency rating check A paired-samples *t*-test confirmed that the reported sense of agency was larger on active compared to passive trials (5.51 versus 2.56ms, *t*(57) = -9.79, *p* \<.001). This ensures that our explicit measurement of sense of agency was successful. presents the participants’ sense of agency ratings as a function of block and delay (i.e., 500, 750, 1000 or 1250ms) for the correct active trials (excluding trials exceeding 2.5 *SD* of the overall mean RT) and the passive trials. ### Trial-by-trial analyses These analyses were conducted on the EQ block. We conducted a 2x2 repeated measures analysis with Current and Previous trial congruency as within-subjects factors (both with 2 levels: congruent and incongruent) on sense of agency (score on a scale from 1 to 9), RTs (in ms) and error rates (in %) as (separate) dependent variables. Means and *SD*s for each of these dependent variables in relation to current and previous congruency are reported in. The results are also depicted on. With regard to *sense of agency*, we found no significant main effect of previous congruency (*F*(1,57) = 0.045, *p* =.83, *η*²_p = 0.001), nor a significant interaction (*F*(1,57) = 0.16, *p* =.69, *η*²_p = 0.003). The main effect of current congruency was not significant, but showed a trend (*F*(1,57) = 3.29, *p* =.075, *η*²_p = 0.055): incongruent trials received a slightly higher agency rating than congruent trials (5.51 versus 5.39). With regard to *RT*, we found a significant main effect of current congruency (*F*(1,57) = 41.57, *p* \<.001, *η*²_p = 0.42) indicating that participants were slower on incongruent compared to congruent trials (on average 800ms versus 728.5ms). The main effect of previous congruency did not reach significance (*F*(1,57) = 0.75, *p* =.39, *η*²_p = 0.013). We also did not observe a significant interaction between current and previous congruency (*F*(1,57) = 1.82, *p* =.18, *η*²_p = 0.031). This implies that we did not observe a typical Gratton effect: the congruency effect was not significantly decreased after an incongruent compared to a congruent trial (60ms versus 83ms). With regard to *error rates*, we observed no significant effects (all *p* \>.55). As an exploratory analysis that was not included in our preregistration, we additionally conducted a 4x2x2 repeated measures analysis with Delay (4 levels: 500, 750, 1000 or 1250ms), Current and Previous trial congruency (both with 2 levels: congruent and incongruent) as within-subjects factors on sense of agency ratings (score between 1–9). We observed a main effect of Actual delay (*F*(3,55) = 3.41, *p* =.024, *η*²_p = 0.16), with slightly decreasing sense of agency for longer delays (specifically, 5.63, 5.51, 5.38 and 5.28 for the increasing delays). We also observed a main effect of Current trial congruency (*F*(1,57) = 4.28, *p* =.043, *η*²_p = 0.070) with a slightly higher sense of agency for incongruent trials compared to congruent trials (5.51 and 5.38, respectively). None of the other main effects or interactions reached significance. ### Block analyses These analyses were conducted on the MC and MI blocks. We conducted a 2x2 repeated measures analysis with Block (2 levels: MC and MI) and Current trial congruency (2 levels: congruent and incongruent) as within-subjects factors sense of agency (score on a scale from 1 to 9), RTs (in ms) and error rates (in %) as (separate) dependent variables. Means and *SD*s for each of these dependent variables in relation to current and previous congruency are reported in. The results are also depicted on. With regard to *sense of agency*, we found a significant main effect of congruency (*F*(1,57) = 6.94, *p* =.011, *η*²_p = 0.11), indicating that participants reported slightly more agency on incongruent compared to congruent trials (5.64 versus 5.46). The main effect of block did not reach significance (*F*(1,57) = 0.42, *p* =.52, *η*²_p = 0.007), nor did the interaction between block and congruency (*F*(1,57) = 0.024, *p* =.88, *η*²_p \< 0.001). With regard to *RT*, we found a significant main effect of congruency (*F*(1,57) = 113.85, *p* \<.001, *η*²_p = 0.67) indicating that participants were slower on incongruent compared to congruent trials (on average 811ms versus 734ms). Additionally, we observed a significant interaction between block and congruency (*F*(1,57) = 5.52, *p* =.022, *η*²_p = 0.088). This interaction reflects a typical proportion congruency effect: a decreased congruency effect in an MI block compared to an MC block (60ms versus 94ms). The observation of this proportion congruency effect ensures that our manipulation of block was successful in order to trigger blockwise adaptations. The main effect of block did not reach significance (*F*(1,57) = 0.56, *p* =.46, *η*²_p = 0.010). With regard to *error rates*, we observed a significant main effect of congruency (*F*(1,57) = 6.65, *p* =.013, *η*²_p = 0.10) indicating that participants made more errors on incongruent compared to congruent trials (on average 4.86% versus 3.78%). The main effect of block (*F*(1,57) = 2.19, *p* =.14, *η*²_p = 0.037) and the interaction between block and congruency (*F*(1,57) = 0.50, *p* =.48, *η*²_p = 0.009) did not reach significance. As an exploratory analysis that was not included In our preregistration, we additionally conducted a 4x2x2 repeated measures analysis with Actual delay (4 levels: 500, 750, 1000 or 1250ms), Block (2 levels: MC and MI) and Congruency (2 levels: congruent and incongruent) as within-subjects factors on sense of agency ratings (score between 1–9). We observed a main effect of Actual delay (*F*(3,55) = 6.32, *p* =.001, *η*²_p = 0.269), with slightly decreasing agency ratings for longer delays (specifically, 5.77, 5.61, 5.48 and 5.35 for the increasing delays). We also observed a main effect of Congruency (*F*(1,57) = 6.61, *p* =.013, *η*²_p = 0.10), with a slightly higher sense of agency for incongruent trials compared to congruent trials (5.65 and 5.46, respectively). None of the other main effects or interactions reached significance. ### Post-block subjective effort and agency analyses With regard to *subjective effort* experienced during each block (EQ, MC or MI), we conducted a repeated measures analysis with block (3 levels: EQ, MC or MI) as within-subjects factor on the reported experienced effort after each block. We observed no significant differences between the three blocks (*F*(2,55) = 0.82, *p* =.44, *η*²_p = 0.029). Indeed, the reported experienced effort was very similar in the EQ, MC and MI blocks (resp. 4.86, 4.87 and 5.20). With regard to the *subjective sense of agency* experienced for congruent, incongruent and passive trials, we conducted a repeated measures analysis with trial type (3 levels: congruent, incongruent or passive) as within-subjects factor on the reported explicit sense of agency after each block. We observed significant differences between the experienced agency over congruent, incongruent and passive trials (*F*(2,55) = 48.90, *p* \<.001, *η*²_p = 0.64). Specifically, participants reported a higher sense of agency over incongruent trials (5.71) compared to congruent (5.03, *F*(1,56) = 18.78, *p* \<.001, *η*²_p = 0.25) or passive trials (2.63, *F*(1,56) = 96.65, *p* \<.001, *η*²_p = 0.63). The explicit sense of agency also differed between congruent and passive trials (*F*(1,56) = 60.03, *p* \<.001, *η*²_p = 0.52). Note that this effect did not interact with block: a repeated measures analysis with trial type (3 levels: congruent, incongruent or passive) and block (3 levels: EQ, MC or MI) as within-subjects factors again only revealed a main effect of trial type (*F*(2,55) = 45.32, *p* \<.001, *η*²_p = 0.62). ## Discussion Next to confirming again that our manipulation of cognitive control and our measurement of sense of agency were successful, we only obtained a small effect of congruency on the sense of agency, both in the trial-by-trials and the block analyses: incongruent trials received a slightly higher agency rating than congruent trials. This is in line with the post-block reported explicit sense of agency in Experiment 1, but contrary to our hypothesis. In the general discussion we will elaborate this further. # General discussion The aim of this study was to examine the effect of cognitive effort on sense of agency. We were interested in shedding light on the contradictory literature about the facilitative or/and detrimental nature of cognitive effort on sense of agency. Based on previous studies we suggested that temporal aspects might determine when effort facilitates or impedes sense of agency. We formulated concrete hypotheses. First, during a block of trials where effort is required unpredictably, encountering a high-effort trial would lead to a decreased sense of agency *on that same trial*, but an increased sense of agency *on the next trial* compared to a low-effort trial. Second, in a block of trials where effort can be anticipated (i.e., frequent high-effort trials), sense of agency would be increased compared to a block of trials where effort cannot be anticipated (i.e., scarce high-effort trials). Furthermore, we aimed to explore the effect of cognitive effort on both implicit and explicit measures of sense of agency. We used a flanker task while also varying the proportion of conflict present in a block of trials, creating three conditions (MC, MI and EQ block) in which the required cognitive control (and hence effort) varied. Our results showed that our manipulations of cognitive control, intentional binding and sense of agency were successful. With regards to the trial-by-trial analyses, and contrarily to our expectations, we observed no decreased intentional binding or sense of agency *on* incongruent trials, nor an increased intentional binding or sense of agency *after* incongruent trials. Remarkably, in the trial-by-trial analyses of Experiment 2, we observed a trend towards an increase of sense of agency on current incongruent trials compared to congruent trials (which became significant when actual delay was taken into account). This increased sense of agency for incongruent trials was also observed in the block analyses of Experiment 2 and in the post-block analyses of both experiments. With regards to the block analyses of Experiment 1, we observed a slight trend towards increased intentional binding in a context where conflict could be anticipated (i.e., MI block) compared to a context where conflict was scarce (i.e., MC block). This trend was not present for sense of agency in Experiment 2. Our results are not in line with previous studies observing lower sense of agency for incongruent compared to congruent trials. A striking difference between our study and these previous studies, is the interval between action and outcome and hence when participants are asked to introspect about their sense of agency. Previous studies used short intervals (ranging from 100 to 500ms), whereas our intervals ranged from 500 to 1250ms. We chose these longer intervals (see also) to be able to use the spacebar Interval Reproduction Task, as a measure of intentional binding, in Experiment 1. But the moment when sense of agency is tapped into, might be crucial. After shorter action-outcome intervals, perhaps the experience of conflict is still predominant, leading to smaller agency ratings on incongruent trials. After longer intervals, perhaps more metacognitive processes might come into play, leading participants to reappraise their action as an accomplishment of successfully solving an incongruent trials, leading to higher agency ratings on incongruent trials. It has already been shown that conflict experience precedes metacognitive experience, which might only occur at a later time. In order to further elucidate this, we should include a broader range of action-outcome intervals in the experimental design. Interestingly, our results are in line with Damen et al., who, using quite a different free choice priming task, found that clearly visible incongruent primes increased agency compared to congruent primes. Based on this observation, another possible explanation for the discrepancy in results might be the exact way in which sense of agency is probed. Both in our study and the study of Damen et al. we particularly asked participants whether they felt they had *caused* the outcome. In contrast, Sidarus and Haggard and Wang et al. asked participants how much *control* they felt over the outcome that was triggered by their actions. This subtle difference in instructions, might have shifted participants’ focus from how much control they felt (which arguably might be larger for congruent trials) to how much causality they experienced (which might be higher for incongruent trials which require the overruling of habitual behavior). A similar argument can be made for the post-block assessment of sense of agency. In previous studies, participants had to rank the colored circles and rate their sense of control for each of them after each block of trials. In our study, in contrast, participants reflected on the extent to which they felt they had caused the colored circles to appear on congruent, incongruent and passive trial specifically. Next to the difference between the focus on control versus causality, additionally, we specifically highlighted the aspect of congruency here. This in turn might have led to a reappraisal of their actions in response to the clearly more difficult, incongruent trials. A follow-up study manipulating instructions might be very informative with regards to the role of even subtle nuances in sense of agency instructions. In this study, we departed from the view that sense of agency and effort are tightly related and that a subjective experience of effort might be a prerequisite for any feeling of agency or causality. From this, we speculated that experimental manipulations of cognitive effort should also influence the accompanying sense of agency. However, whether our manipulation of effort was successful, is doubtful. Although only assessed after each block and only per block type and not per trial-type, participants reported equal experiences of effort for MC, MI and EQ blocks in both experiments. Thus, although our manipulation of cognitive control was successful (e.g., reduced congruency effects in MI compared to MC blocks, indicative of increased cognitive control in MI blocks), this might not have been accompanied by differences in experienced effort. Indeed, the constructs of cognitive effort and cognitive control are not identical, and the possibility of effortless exertion of cognitive control (i.e., “flow”) has been suggested. Alternatively, even though it is expected that in an MI block the overall cognitive control, and hence effort experience, is larger, perhaps the intense effort experienced on the rare incongruent trials in an MC block cancelled out the effort differences at the block level. Finally, it could also be the case that the experienced differences in effort in our experiments were too subtle to be crudely captured only after a block of trials. Indeed, despite the absence of crude differences in effort between blocks, we still observed a trend of a difference in intentional binding between blocks. On the other hand, the lack of clear effort differences between the blocks, might also explain why our block-based results are minimal. Contrary to what we expected, we did not find an increased explicit sense of agency in a context where conflict could be anticipated (i.e., MI block) compared to a context where conflict was rare (i.e., MC block), but we did observe a slight trend towards increased intentional binding in the MI block compared to the MC block. These results could be linked to the fact that the MI block probably was not effortful enough. In any case, replicating the experiments using a task that induces stronger differences in experienced effort between the different conditions (e.g., Stroop task), and regularly assessing the experienced effort for different trial types and blocks seems advisable. Our findings seem to imply a dissociation between our intentional binding and sense of agency measures. First, despite no apparent crude difference in experienced effort, the explicit sense of agency measure was still able to pick up some (trends towards) differences between experienced sense of agency for congruent and incongruent trials, whereas the intentional binding measure was not. It could be that intentional binding is a less sensitive measure (especially when there are no strong differences in experienced effort). Alternatively, the lack of clear effort differences between congruent and incongruent trials, might also inhibit differences at the assumed implicit, pre- reflective level of sense of agency, but not the at the explicit reflective, inferential, belief-like level. Second, across both experiments, the delay between the action and the outcome had an effect on our measures of sense of agency: participants showed more intentional binding, but lower sense of agency ratings when the delay increased (i.e., 500, 750, 1000 or 1250ms). This is in line with previous studies. Temporal binding measured using interval estimation is typically larger for longer intervals ( but see for the reversed result), whereas higher agency ratings are often found for shorter intervals. We are not the first to highlight differences between different measures of sense of agency. Recently, Imaizumi and Tanno have suggested that intentional binding measured using time interval might originate from a different mechanism. Suzuki et al. even showed that intentional binding might not necessarily reflect sense of agency, but might be accounted for by multisensory causal binding, without necessarily being related to intention or agency. We sought to combine methods typically used in intentional binding tasks, with methods employed to study the effect of effort on explicit sense of agency. More specifically, in line with most binding studies , we compared an active condition to a passive condition to assess the baseline level of intentional binding. This differs from previous studies that have assessed the effect of action selection fluency (effort) on explicit sense of agency, which typically do not include passive trials. While comparing active and passive trials remains the most common way of assessing intentional binding, it should be noted that this has been criticised for not adequately controlling for other processes that differ between these two types of trials. In addition, the inclusion of passive trials in our studies, leading to a clearly felt difference in agency between active and passive trials, might have obscured more subtle differences between the different active trial types (e.g., congruent and incongruent trials). Therefore, an experiment presenting only active trials might be useful to expose perhaps elusive differences in agency between active trial types. The current study examined the effect of cognitive effort on sense of agency using explicit and implicit measures of sense of agency. We showed that trials requiring the exertion of more cognitive control lead to a higher sense of agency. This result was contrasted to previous studies to establish potential reasons for this contradictory finding. Future studies should ensure that conditions are sufficiently effortful, use a broad range of action-outcome intervals and contrast (even subtle) different ways of probing sense of agency. The authors wish to thank Katleen Vandist and Elke Van Lierde for their help with the collection of the data. [^1]: The authors have declared that no competing interests exist.

# Introduction The geometric dilution of precision (GDOP) describes the accuracy of positioning from the geometry of the satellites visible to the receiver. Errors in a GPS receiver’s determination of its position are typically the GDOP times the measurement error; in other words, the GDOP is an error amplification factor, which means that a smaller GDOP generally results in a more accurate position. Therefore, a smaller GDOP is better. It has been shown that the number of satellites cause the GDOP to move in the opposite direction, i.e., a higher number of satellites leads to a lower GDOP value, and a lower number of satellites generally produces a bad GDOP and results in a corresponding reduction in the positioning error. The simplest method is to choose a combination of all of the satellites, all of which must be in view to achieve optimal position accuracy with a minimal GDOP; however, some GPS receivers can receive and parse navigation signals from only a limited number of satellites in real time because of restrictions on hardware resources. Therefore, at least four satellites should be selected for three-dimensional positioning to balance hardware resources and positioning performance, however, a minimum of 5 satellites is required to evaluate the consistency and reliability for the GPS receiver autonomous integrity monitoring(RAIM), meanwhile, more than five pseudo-ranges are need to be used to identify the contaminating measurement in the procedure of fault detection and exclusion (FDE). Calculating the GDOP typically involves multiple mathematical matrix operations, such as inversions and transformations, particularly for the optimal satellite selection method, which involves selecting the subset with the lowest GDOP using all of the satellites. However, the approach used for GDOP calculation creates a heavy burden for the receiver because the calculation’s complexity increases exponentially with the number of visible satellites. Satellite selection is a classical combinatorial optimization problem, the solution methods of which have been the subject of intense study in recent years. One solution method includes geometric methods, such as the fast satellite selection approach, quasi-optimal satellite selection, and optimal selection. Miaoyan Zhang proposed a novel algorithm based on selecting the subset of in-view satellites that has a closer distance to optimal solution from the aspect of geometry when there are more than four satellites for multiple constellations. The other solution methods include statistical and machine learning approaches based on neural networks, support vector machines and fuzzy logic. Simon initially proposed the neural network-based approach as a way to predict the GDOP and complete the classification, and back-propagation neural networks and optimal interpolative nets have been employed to achieve the two objectives, respectively. Jwo proposed several types of neural network mapping relationships for different classes of relationships between inputs and outputs and compared the performance of different network architectures. Mosavi studied the relationships of eigenvalues between the visibility matrix and its inverse matrix for GDOP. Azami developed several improved neural network training algorithms, including the Levenberg Marquardt (LM), modified LM and resilient BP (RBP) methods. ChihHung addressed the approximation of the GPS GDOP using support vector machines (SVMs). In this paper, we develop a novel modified genetic algorithm to select the optimal subsets of satellites for GPS receivers. Because of the global optimization capability and fast convergence of the modified genetic algorithm, this approach effectively searches the entire solution spaces and determines the optimal satellite subset after evolving through a number of generations without the number limitation of required visible satellites, which improve the adaptability and flexibility of the algorithm under a unified framework. # Materials ## The Geometric Dilution of Precision The mathematical model behind GPS is based on pseudo-range measurements between the receiver and visible satellites. Without loss of generality, supposing that (*x*_*u*, *y*_*u*, *z*_*u*) represents the three- dimensional coordinates of the receiver’s position in the ECEF system, (*x*_*j*, *y*_*j*, *z*_*j*) denotes the coordinates of the *j*th satellite’s position in the ECEF system, and *t*_*b* indicates the equivalent distance of the time difference between the receiver arrival time and the satellite signal sent time. The receiver acquires information from *n*(*n* \> 4) satellites and determines the pseudo-range value. $$\rho_{j} = \left( {\left( x_{j} - x_{u} \right)^{2} + \left( y_{j} - y_{u} \right)^{2} + \left( z_{j} - z_{u} \right)^{2}} \right)^{1/2} + ct_{b}(j = 1,2,\cdots,n)$$ It is shown that is a nonlinear function because of the square root mathematical operation. In general, this class of problem can be linearized using the mathematical principle of Taylor series and be written in a concise matrix form as $$\begin{array}{r} {{\Delta\rho} \approx {G\Delta x} = {G\Delta x} + \epsilon} \\ \end{array}$$ where the order of measurement matrix *G* is *n* × 4 (*n* ≥ 4), which can be represented as $$\begin{array}{r} {G = \begin{pmatrix} a_{x_{1}} & a_{y_{1}} & a_{z_{1}} & 1 \\ a_{x_{2}} & a_{y_{2}} & a_{z_{2}} & 1 \\ \cdots & \cdots & \cdots & \cdots \\ a_{x_{n}} & a_{y_{n}} & a_{z_{n}} & 1 \\ \end{pmatrix}} \\ \end{array}$$ To simplify this problem, we assume that the measurement error has a Gaussian distribution independently. The least squares solution for the super definite matrix is typically given by $$\begin{array}{r} {{\Delta x} = \left( G^{T}G \right)^{- 1}G^{T}{\Delta\rho}} \\ \end{array}$$ To illustrate the position accuracy, the covariance of Δ*x* is $$\begin{array}{ll} {cov\left( {\Delta x} \right)} & {= E\left\lbrack {\left( {\Delta x} \right)\left( {\Delta x} \right)^{T}} \right\rbrack} \\ & {= E\left\lbrack {\left( {G^{T}G} \right)^{- 1}G^{T}\left( {\Delta\rho\Delta\rho^{T}} \right)G\left( {G^{T}G} \right)^{- 1}} \right\rbrack} \\ & {= \left( {G^{T}G} \right)^{- 1}cov\left( {\Delta\rho} \right)} \\ \end{array}$$ It is reasonable to assume that all of the errors in the pseudo-range measurements are stationary random processes for short time intervals, which are independent and identically distributed with a variance of $\sigma_{uere}^{2}$. Let $$\begin{array}{r} {cov\left( {\Delta x} \right) = \begin{pmatrix} \sigma_{x_{u}}^{2} & \cdot & \cdot & \cdot \\ \cdot & \sigma_{y_{u}}^{2} & \cdot & \cdot \\ \cdot & \cdot & \sigma_{z_{u}}^{2} & \cdot \\ \cdot & \cdot & \cdot & \sigma_{ct_{b}}^{2} \\ \end{pmatrix}} \\ \end{array}$$ The GDOP, which represents the amplification of the equivalent ranging errors in the measurement into the receiver’s position solution, is $$\begin{array}{r} {GDOP = \sqrt{\sigma_{x_{u}}^{2} + \sigma_{y_{u}}^{2} + \sigma_{z_{u}}^{2} + \sigma_{ct_{b}}^{2}}/\sigma_{uere}^{2}} \\ \end{array}$$ According to linear algebra and matrix theory, when *λ*_*i* are the eigenvalues of an invertible matrix *A*, $\lambda_{i}^{- 1}$ are assumed to be the eigenvalues of the inverse of the matrix *A*⁻¹. The measurement matrix $H = \left( G^{T}G \right)^{- 1}$ is always revertible, and the eigenvalues $\lambda_{i}^{- 1}$ of $H = \left( G^{T}G \right)^{- 1}$ can be found using the eigenvalues *λ*_*i* of (*G* *^T* *G*). This leads to a significant reduction in the number of calculations required to find the matrix inverse; the eigenvalues *λ*_*i* of (*G* *^T* *G*) can be quickly calculated using QR decomposition. Let $$\begin{aligned} {GDOP =} & \sqrt{tr\left( \left( G^{T}G \right)^{- 1} \right)} \\ = & \sqrt{tr\left( (H) \right.} \\ = & \sqrt{\lambda_{1}^{- 1} + \lambda_{2}^{- 1} + \lambda_{3}^{- 1} + \lambda_{4}^{- 1}} \\ \end{aligned}$$ where *tr*(⋅) represents the trace function of the matrix and *c* is the constant for the speed of light. The satellite selection process is to select a subset of satellites visible in the current view for the purpose of the best positioning accuracy, which corresponds to selecting the rows of the visibility matrix *G* that produce the minimum GDOP as follows: $$\begin{array}{r} {S = {\left\{ s_{i} \right\},1 \leq i \leq k,k \in (4,5,6,7)}} \\ \end{array}$$ where *s*_*i* is the identification number of the satellite currently in view such that $$\min\limits_{\forall{({G_{i},G_{j},G_{l},G_{m}})} \in G}\,\left( {GDOP} \right) = \sqrt{tr\left( \left( {G_{k}^{T}G_{k}} \right)^{- 1} \right)} = \sqrt{\lambda_{1}^{- 1} + \lambda_{2}^{- 1} + \cdots + \lambda_{k}^{- 1}}$$ ## Genetic Algorithm Preliminaries A genetic algorithm (GA) is a robust global optimization approach that operates according to the principles of evolution, such as inheritance, selection, crossover and mutation. In a GA, chromosomes are used to encode a candidate solution, and the fitness function is the indication factor that represents the quality of the individuals. As the population evolves, its average fitness gradually increases based on the principle of the ‘survival of the fittest’; excellent genes that are more fit are usually selected stochastically. The selected individuals are operated through the process of the genetic operator, such as selection, crossover and mutation, until the algorithm reaches termination. In previous studies, GA was primarily used for micro-strip antenna optimization, resolving ambiguity and multi-path mitigation in satellite navigation. This paper proposes a novel modified genetic algorithm (MGA) for GPS satellite selection. # Methods Because individuals evolve from generation to generation, the optimal solution is found in a short period of time. Several strategies were used to improve the performance of the algorithm. First, the elitist strategy reserves the most fit gene of the previous generation to improve the average fitness of the next generation. Second, adaptive mechanisms are used to adjust the mutation and crossover rates to maintain each generation’s diversity. Finally, a hybrid method accelerates the convergence. Without loss of generality, we establish the following mathematical definitions: 1. *MaxGenNumber* represents the maximum number of evolution cycles. 2. *PopSize* represents the size of the population. 3. *Pop* represents the population. 4. *Ind*_*i* represents the *i*th chromosome. 5. *Fitness*(*Ind*_*i*) represents the fitness of the *i*th chromosome. ## Representation The integer code includes a clear description of the satellite selection problem with population diversity followed by binary formation. It is assumed that *N* GPS satellites are visible in the current view, which are coded as (*k*₁, *k*₂, ⋯, *k*_*N*). The objective is to select the four signals *Ind*_*i* = (*k*_*i*, *k*_*j*, *k*_*m*, *k*_*n*) that provide the minimum GDOP value and that satisfy the following constraint: $$\left\{ \begin{array}{l} {k_{i} \leq k_{j} \leq k_{m} \leq k_{n}} \\ {1 \leq i \leq j \leq m \leq n \leq N} \\ \end{array} \right.$$ where the size of the solution space is $C_{N}^{4}$, *Ind*_*i* represents an individual, and *k*_*i* is a gene on the chromosome. Without loss of generality, we assume that the list of currently visible satellites is \[2, 5, 7, 14, 15, 20, 21, 25\] and possible solutions are \[2, 5, 7, 14\] or \[2, 5, 7, 15\]. A total of 5 or 6 satellites must be selected when considering RAIM and fail detection and exclusion (FDE), which means that the code’s dimension must increase to, for example, \[2, 5, 7, 14, 15\] or \[2, 5, 7, 14, 15, 20\]. ## The Fitness Function The fitness function is a special type of objective function that indicates the current number of individuals; a larger number is better in most cases. We minimize the GDOP as follows to obtain a more accurate position: $$\begin{array}{r} {Fit(t) = \frac{1}{\sqrt{tr\left( \left( G_{4}^{T}G_{4} \right)^{- 1} \right)}}} \\ \end{array}$$ Note that according to, the fitness function is monotonic and non-negative, and the number of calculations can be reduced by omitting the square root, which yields $$\begin{array}{r} {Fit^{\prime}(t) = \frac{1}{tr\left( \left( G_{4}^{T}G_{4} \right)^{- 1} \right)}} \\ \end{array}$$ ## Selection During each generation, individuals are selected by the *Ps* probability to produce the offspring, in which the individuals with higher fitness have considerably higher probabilities of being selected. We define the fitness selection operator as *T*_*s* : *S*^*PopSize* → *S*, where *S* represents the solution space and *PopSize* is the population size. Then, $$\begin{array}{r} {P\left\{ {T_{s}\left( {Pop} \right) = {Ind}_{i}} \right\} = {P\left\{ Ind_{i} \right\}} = \frac{Fitness\left( {Ind}_{i} \right)}{\sum\limits_{k = 1}^{PopSize}{Fitness\left( {Ind}_{k} \right)}}} \\ \end{array}$$ where *Fitness*(*Ind*_*i*) is an individual’s fitness and the probability distribution function *P*{*Ind*_*i*} satisfies the following condition: $$\left\{ \begin{array}{l} {P\left\{ {Ind}_{i} \right\} \geq 0} \\ {\sum\limits_{i = 1}^{PopSize}{P\left\{ {Ind}_{i} \right\} = 1}} \\ \end{array} \right.$$ The role of the selection operator is to select the individuals that reproduce stochastically; the roulette wheel selection method based on a proportional- selection mechanism is the most widely used strategy in which a greater fitness implies a higher selection probability. To reduce the time required to find the optimal solution, the elitist reservation strategy is used. Using a ranking based on the population’s fitness, the best chromosomes are reserved for the next generation. This process accelerates the algorithm’s convergence. ## Adaptive Crossover Operators Crossover is a critical genetic operator that is also called recombination. In this process, a couple of solutions are selected to create offspring that inherit the characteristics from the “parents”. The process continues until a new population is created. Without loss of generality, *T*_*c* : *S*² → *S* is a stochastic map with cross probability *p*_*c* for the single-point crossover operator. ∀(*Ind*₁, *Ind*₂) ∈ *S*², *Y* ∈ *S*, there is $$P\left\{ {T_{c}\left( Ind_{1},Ind_{2} \right) = Y} \right\} = \left\{ \begin{array}{lc} \frac{kp_{c}}{Len} & {Y \neq Ind_{1}} \\ {\left( 1 - p_{c} \right) + \frac{kp_{c}}{L}} & {Y = ind_{1}} \\ \end{array} \right.$$ where *Len* is the length of the chromosome. In general, the cross probability *P*_*c* strongly influences the performance of the MGA. During the initial stage of the MGA’s evolution, if the best chromosome is considerably more fit than the other individuals, the optimum chromosome has a substantially higher probability of being selected, which causes the algorithm to converge to a local minimum. Therefore, a gene with a high fitness should be restricted from over-reproducing to maintain the gene diversity at the initial stage of population evolution. In turn, the average fitness of all individuals is approximately equal to the maximum value of the population. Consequently, the possibility of selecting an individual with an average fitness is equal to that of selecting the chromosome with the highest fitness, which eliminates competition between excellent individuals. Then, the selection operation becomes random, which results in the worst performance when searching for the best chromosome during the post-evolutionary stage. is a demonstration of single- point crossover operation for MGA-based satellite selection. The following novel adaptive strategy for generating the cross probability for different evolutionary stages that controls the gene diversity to a certain, but not large, degree is proposed: $$P_{c} = \left\{ \begin{array}{lr} {P_{cmax} - \frac{\left( {p_{cmax} - p_{cmin}} \right)\left( {Fitness^{\prime} - Fitness_{avg}} \right)}{Fitness_{\max} - Fitness_{avg}}} & {Fitness^{\prime} \geq {Fitness_{avg}}} \\ P_{cmax} & {Fitness^{\prime} \leq {Fitness_{avg}}} \\ \end{array} \right.$$ where *Fitness*_*max* and *Fitness*_*avg* denote the maximum and minimum fitness of the population, respectively. *Fitness*′ is the larger one between the pair of parents. *p*_*cmax* denotes the maximum probability of crossover. *p*_*cmin* denotes the minimum probability of crossover. ## Adaptive Mutation Operators The mutation operator brings a new gene into the population, keeping the population’s diversity from converging prematurely. is an example of single- point mutation operation for MGA-based satellite selection. The mutation operator is used to alter an individual gene for the purpose of population diversity; however, the MGA degenerates into a random search when the mutation probability *P*_*m* is high. In contrast, the probability of producing certain useful genes is zero when *P*_*m* is low. To improve the algorithm’s ability to find a global optimum and to avoid converging prematurely or becoming stuck at a local minimum, the adaptive mutation is $$P_{m} = \left\{ \begin{array}{lr} {P_{mmax} - \frac{\left( {p_{mmax} - p_{mmin}} \right)\left( {Fitness^{\prime} - {Fitness_{avg}}} \right)}{{Fitness_{\max}} - {Fitness_{avg}}}} & {Fitness^{\prime} \geq {Fitness_{avg}}} \\ P_{mmax} & {Fitness^{\prime} \leq {Fitness_{avg}}} \\ \end{array} \right.$$ where *Fitness*_*max* and *Fitness*_*avg* denote the maximum and minimum fitness of the population, respectively. *Fitness*′ is the larger one between the pair of parents. *p*_*mmax* is the maximum probability mutation. *p*_*mmin* is the minimum probability mutation. ## The Hybrid Genetic Algorithm To utilize the benefits of the two classes of approaches, a hybrid genetic algorithm that combines the advantages of an MGA and fast selection is proposed. To allow the algorithm to quickly find the best solution, individuals may be initially created in the areas with the higher probability for optimal solutions. It is easy to obtain an equivalent formulation of when the receiver position offset is described in the ECEF coordinate system. $$\widetilde{G}\mspace{180mu}\begin{bmatrix} {\bigtriangleup e} \\ {\bigtriangleup n} \\ {\bigtriangleup u} \\ {\bigtriangleup\delta t_{u}} \\ \end{bmatrix} = b$$ where \[△*e* △*n* △*u* △*δt*_*u*\]^*T* is the state variable to be determined; then, the geometry matrix $\widetilde{G}$ becomes $$\widetilde{G} = \begin{bmatrix} {- cos\theta^{(1)}sin\alpha^{(1)}} & {- cos\theta^{(1)}cos\alpha^{(1)}} & {- sin\theta^{(1)}} & 1 \\ {- cos\theta^{(2)}sin\alpha^{(2)}} & {- cos\theta^{(2)}cos\alpha^{(2)}} & {- sin\theta^{(2)}} & 1 \\ \vdots & \vdots & \vdots & \vdots \\ {- cos\theta^{(N)}sin\alpha^{(N)}} & {- cos\theta^{(N)}cos\alpha^{(N)}} & {- sin\theta^{(N)}} & 1 \\ \end{bmatrix}$$ where *α*^(*n*) and *θ*^(*n*) are the elevation and azimuth angles of the *n*th satellite, respectively. The main idea behind this step is to randomly generate all of the chromosome subsets with geometries similar to that of the optimal subset of the population by grouping all of the satellites and randomly selecting one satellite from each group to form a chromosome subset according to their azimuths and elevations. ## Termination This evolutionary process for MGA satellite selection arrives at an endpoint when one of the following conditions has been reached: 1. A solution is found that satisfies the minimum criterion, *GDOP* \< 3. 2. The number of generations reaches 20. # Results To demonstrate the performance of the proposed MGA satellite selection algorithm, the OEMStar, which is one of NovAtel’s OEM global navigation satellite system receiver platforms, was used to collect broadcast ephemeris data and to calculate all of the satellites’ positions in the ECEF coordinate system every second for 12 hours. The OEMStar receiver was placed at the author’s institute, which has ECEF coordinates of \[−2258692.95, 4405376.94, 4007987.94\]. At the beginning of the evolutionary process, four different satellite signals were randomly selected to compute the GDOP from the geometry matrix G, using. The following experiments were conducted using Matlab 2010 on a personal computer with an Intel Core(TM)2 Duo CPU and 2 GB of memory. ## Simulation Parameters The primary parameters of the proposed MGA satellite selection algorithm include the population size, the crossover probability, the mutation probability and the number of evolution iterations. These parameters influence the convergence speed and accuracy of the algorithm. A high crossover probability accelerates the creation of individuals and increases the possibility of an excellent gene being destroyed, which negatively affects the process of evolution. In contrast, when the crossover probability is low, the speed at which new individuals are created decreases, and the search process stagnates. With regard to the mutation probability, a high probability results in the MGA being closer to a random search algorithm, and a low probability reduces the likelihood of producing new individuals and of premature convergence. lists the values of all of the parameters used in the simulations. # Discussion GPS satellites orbit the earth once approximately every 12 hours, and as shown in, there are approximately 7–12 satellites visible from any point on the earth at any given time. shows the cumulative probability distribution function for the number of visible satellites. shows that the GDOP calculated using the all- in-view (AIV) method is minimized, and the proposed MGA’s performance is equivalent to that of the optimal selection algorithm. The GDOP of the optimal selection algorithm is slightly better than that of the AIV method, whose performance, however, is better than that of the neural network and the fast selection algorithm, particularly in terms of accuracy and efficiency. A comparison is shown in. A comparison of the residuals of the GDOPs of the various algorithms is provided in, where it can be observed that the proposed MGA is the most accurate and has the smallest residuals. shows the average fitness function and the best individual fitness versus the number of iterations. The average fitness function gradually decreases over time as the process of evolution continues, which shows that the algorithm converges and that the best chromosome is eventually obtained. The proposed novel MGA satellite selection algorithm is not only suitable for the common problem of selecting 4 satellites but can also be used to select 5 or 6 satellites under RAIM or FDE conditions without modification. presents the GDOP performance curve computed by the MGA, which shows that the GDOP gradually decreases as the number of satellites selected, namely, 4, 5 and 6, increases. # Conclusion A novel method for selecting satellites for GPS use was proposed and shown to provide both parallel and global convergence. The MGA can select a subset of satellites of any size that satisfies the position requirements using an elite conservation strategy, adaptive mechanism, hybrid genetic algorithm and reasonable designs of the fitness function, and the selection, crossover and mutation operators. Comprehensive simulations were conducted and demonstrated that the MGA-based satellite selection method can effectively select an optimal subset of the available satellites in both conventional and RAIM modes. The latter is more feasible and adaptable to the GPS receivers that are used in handset equipment and mobile phones. An unexpected discovery was that this method can be applied to not only single constellation systems such as GPS but also to multi-constellation systems because of its ability to select more than 4 satellites. # Supporting Information The authors would like to thank the Associate Editor and anonymous reviewers for their valuable comments, which improved the presentation of this paper. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JS. Performed the experiments: GX YK. Analyzed the data: JS GX YK. Contributed reagents/materials/analysis tools: GX. Wrote the paper: JS. Contributed to the use of GPS receiver: GX.

# Introduction Improvement in lifestyle and changes in consumption habits mean that livestock production aims to provide sufficient meat of improved quality. Meat quality and production are influenced by intramuscular fat (IMF) content and skeletal muscle development. For example, the content of IMF and the myofiber type can affect meat quality traits such as flavor, juiciness, water holding capacity and tenderness\[–\]. In chickens, IMF is not visible and not anatomically separable, which makes it difficult to investigate the mechanism of its deposition. Protein profiles of whole muscle, therefore, are important in understanding the mechanisms for both muscle development and IMF deposition. The short lifespan of chickens makes them an excellent model for studying various aspects of development. Some of the molecular markers of muscle structure/metabolism in livestock have also been identified by genome scans, but no studies using proteomics technology have linked muscle growth and IMF content in chickens. It is desirable to analyze the expression profile of proteins in chicken skeletal muscle at different ages. Many studies have characterized proteins from 2-DE gels in pigs, cattle, and layers at different stages of embryonic development to early growth after hatching. Doherty et al have characterized the proteome of layer chicken breast muscle using two-dimensional gel electrophoresis (2-DE) from 1 to 27 days after hatching. Fifty-one proteins had mass spectra that matched existing chicken proteins in on-line databases. For many of these proteins, there were dramatic changes in relative expression levels during the 27 days of growth. Proteomic profiling of the breast muscle of Thai indigenous chickens during the growth period were also analyzed using the 2-DE method. A total of 259, 161, 120 and 107 protein spots were found to be expressed in the chicken breast muscles at 0, 3, 6 and 18 weeks of age, respectively. From these proteins, five distinct spots were significantly associated with chicken age. These were characterized and showed homology with phosphoglycerate mutase 1 (PGAM1), apolipoprotein A1 (APOA1), triosephosphate isomerase 1 (TPI1), heat shock protein 25 kDa (HSP25) and fatty acid binding protein 3 (FABP3). In addition, by application of an isobaric tagging for relative and absolute quantification (iTRAQ)-based approach, the proteomes of bovine embryos at the zygote and 2-cell and 4-cell stage with MII oocytes as a reference were quantitatively analyzed. Bioinformatic analysis of 87 proteins that differed significantly in abundance between the four stages revealed proteins involved in the p53 pathway, lipid metabolism, and mitosis, indicating that these processes may play pivotal roles in embryonic development. All of these studies showed the utility of proteomics as a tool for uncovering the molecular basis of physiological differences in muscle during these growth periods. Compared with the methods previously used (1D and 2D gels), the isobaric tags for relative and absolute quantitation (iTRAQ) analysis in the present study is more accurate and has been widely applied to investigate the proteome of different organisms at different developmental stages. The present study used advanced proteomics methodology (iTRAQ) to identify differentially expressed proteins in breast muscles of slow-growing chickens at different post-hatching ages. # Materials and Methods ## Ethics Statement All of the animal experiments were conducted in accordance with the Guidelines for Experimental Animals established by the Ministry of Science and Technology (Beijing, China). Animal experiments were approved by the Science Research Department (in charge of animal welfare issue) of the Institute of Animal Sciences, CAAS (Beijing, China). ## Animals Forty female Beijing-You chickens were obtained at day 1 from the Institute of Animal Science, Chinese Academy of Agricultural Sciences (Beijing, China), and were randomly assigned to four groups of 10. Individuals were reared in stairstep caging under continuous lighting using standard conditions of temperature, humidity and ventilation. The same diet was fed to all chickens and a three-phase feeding system was used: the starter ration (d 1 to d 28) with 21.0% crude protein and 12.12 MJ/kg, the second phase (d 28 to d 56) with 19.0% crude protein and 12.54 MJ/kg, and the last phase (after d 56) with 16.0% crude protein and 12.96 MJ/kg. Feed and water were provided ad libitum during the experiment. All birds were fasted for 12 h, and weighed before being killed by stunning and exsanguination. The left breast muscles were collected from 10 chickens at day 1 (hatching), 56 (fast growth age), 98 (marketing age) and 140 (first egg age). All samples (200–300 mg) were snap-frozen and stored at −80°C before analysis. The entire right breast and livers were collected, weighed and stored at −20°C for phenotypic measurement. ## IMF measurement IMF content of breast and fat content of the liver were determined by Soxhlet extraction, as described previously, and expressed as percentages of the dry weight of the breast muscle. ## Protein extraction Frozen breast muscle tissues (\~200 mg) was homogenized in 1 mL lysis buffer containing 7 M urea (Sigma, St Louis, MO, USA), 2 M thiourea (Sigma), 4% (w/v) 3–3 (cholamidopropyl) dimethylammonio-1-propanesulfonate (CHAPS; Sigma), 65 mM dithiothreitol (DTT; Sigma), and 0.05% (w/v) protease inhibitor (Sigma). The homogenates were held on ice for 30 min and centrifuged for 30 min at 12,000 *g*, to remove insoluble components. The total protein concentration of each sample was determined with a 2-D Quant kit (GE Healthcare, Pittsburgh, PA, USA). The ten samples of each age group were pooled using equal amounts of protein then the four pools were diluted to the same concentration with Tris-buffered saline (TBS) before iTRAQ labeling. The samples were stored at −80°C until analysis. Each pool was tested twice. ## iTRAQ labeling After precipitation with acetone, the protein (200 μg) of each pool was dissolved with 1 M DTT for 1 h at 37°C and kept in the dark with 1 M indole-3-acetic acid (IAA) for 1 h at room temperature. Samples were dissolved and centrifuged twice with 120 μl UA (8 M urea in 0.1 M Tris.HCl, pH 8.5), and then re-dissolved and centrifuged three times with 100 M lautyltrethylammonium bromide (LTEAB) (1 M). The proteins (2–4 μg) were digested with trypsin (trypsin: protein = 1:50; Sigma) and incubated at 37°C overnight. Each peptide pool was then passed through a 0.2-μm centrifugal filter for 20 min at 10,000 *g* at 20°C. Labeling of each pooled sample was 2-plex, where two reporter tags were used; hatchling samples were labeled with reporter tags 113 and 117; those from day 56 with reporter tags 114 and 118; the pool at day 98 with reporter tags 115 and 119; and the last pool from day 140 was labeled with reporter tags 116 and 121. The four 2-plex labeled samples were then combined into a single 8-plex sample mixture and dried by centrifugal evaporation. ## Strong cation exchange (SCX) separation and reverse phase liquid chromatography tandem mass spectrometry (RPLC-MS/MS) The combined peptide mixture was analyzed by RPLC-MS/MS for simultaneous identification and quantification. The sequence of a peptide is determined from the products that are generated from proteolytic cleavage of the protein and the relative quantity of a given peptide among the treated samples is determined from the intensities of reporter ion signals also present in the MS/MS scan. iTRAQ-8 plex labeling reagents (Applied Biosystems, Foster City, CA, USA) were added to the peptide samples, which were incubated at room temperature for 2 h. The digested protein samples were separated using multidimensional liquid chromatography (LC). In the first dimension, the peptide mixtures were fractionated using an Ultimate LC system (Shimadzu 20AD, Kyoto, Japan) connected to an SCX column (Polysulfoethyl column, 2.1 mm × 100 mm, 5 u, 200 A; Nest Group, Southborough, MA, USA). A linear binary gradient from solvent A (10 mM KH₂PO₄ (Sinopharm Chemical Reagent Co. Ltd, Shanghai, China) and 25% acetonitrile (ACN, pH 2.6; Fisher Scientific, Fair lawn, NJ, USA), to solvent B (10 mM KH₂PO₄, 0.35 M KCl (Sinopharm Chemical Reagent Co. Ltd.), 25% ACN, pH 2.6 was applied: 0%–5% solvent B over 5 min, 5%–25% solvent B over 35 min, then 35%–100% solvent B over 10 min, with a flow rate of 200μl/min and detection at 214/280 nm. The entire run lasted 1 h, and 20 SCX fractions were collected. These fractions were vacuum dried (rotation vacuum concentrators, Christ RVC 2–25; Christ, Germany) and re-dissolved in 0.1% formic acid (Tedia, Fairfield, OH, USA) and 5% ACN. Based on the SCX chromatograms, the 20 SCX fractions were combined into eight pools then desalted by ZORBAX 300SB-C18 column (5 μm, 300 Å, 0.1 × 150 mm; Microm, Miami, FL, USA). The pooled SCX fractions were automatically injected by a Famos autosampler and separated by an UltiMate capillary LC system (Dionex/LC Packings) and fractionated on a C18 PepMap main column (5μm, 300 Å, 0.1 × 150 mm; Microm, Miami, FL, USA) using a linear binary gradient (solvent A: 0.1% formic acid, 5% ACN; solvent B: 0.1% formic acid, 95% ACN). High Performance Liquid Chromatography (HPLC) linear gradients were from 0% solvent B (5 min) to 35% (70 min) and from 35% to 100% (120 min) at a flow rate of 0.3μl/min. The peptides were eluted from the LC column and automatically deposited using a Probot spotting device. Mass spectrometry (MS) was conducted with a QSTAR XL instrument (Applied Biosystems). ## Data analysis Peptide identification from the QSTAR XL data was carried out using the Paragon algorithm in the ProteinPilot 4.2 software package (Applied Biosystems). MS/MS was performed on the four most abundant ions and the proteins identified by searching the SWISSPROT-vertebrate and National Center for Biotechnology Information (NCBI) databases. The following parameters were used for searching: trypsin as enzyme, fixed modification of methyl methanethiosulfate labeled cysteine, iTRAQ as sample type, no special factors, biological modification, thorough identification search, and fragmentation mass accuracy, which were built-in functions of ProteinPilot software, and the Paragon method was adopted. Then the name, function, International Protein Index (IPI), and similar characteristics were obtained from the Uniprot database. For protein-abundance ratios measured using iTRAQ, 1.5-fold up-regulation and 0.75-fold down- regulation change and the p-value \< 0.05 (the p-value is generated from the peptide ratios used for quantitation) were set as the threshold for significant changes. ## Western blotting The pools of proteins from each age group were mixed (4:1) with 5× sample buffer \[0.5 ml 0.5 mM Tris.HCl (pH 6.8), 0.1 g SDS, 0.005 g bromophenol blue, 0.5 ml glycerol, 0.078 g DTT\]. Proteins (40 μg) were boiled for 5 min and separated on SDS-PAGE in running buffer (25 mM Tris.HCl, pH 8.3, 1.4% glycine, 1 g SDS; Mini- PROTEAN Tetra Electrophoresis Cell, 0.75 mm thickness; Bio-Rad, Hercules, CA, USA) in two stages (30 min at 80 V, and 60 min at 120 V). The gels were transferred to Polyvinylidene Fluoride (PVDF) membranes (Millipore, Billerica, MA, USA) in ice-cold transfer buffer (25 mM Tris.HCl, pH 8.3, 1.4% glycine, 20% methanol; Mini Trans-Blot Electrophoretic Transfer Cell; Bio-Rad) at 200 mA for 1 h. Membranes were blocked with 5% non-fat milk (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) in TBST (10 mM Tris.HCl, pH 8.0, 150 mM NaCl, and 0.1% Tween 20) for 90 min. Primary antibodies for Heat shock protein beta-1 (HSPB1) (diluted 1:400; Abcam, Cambridge, MA, USA), Apolipoprotein (APO)A1 (diluted 1:300; Biorbyt, Cambridge, UK), aldehyde dehydrogenase (ALDH)1A1 (diluted 1:300; Biorbyt), malate dehydrogenase (MDH)1 (diluted 1:300; Biorbyt), annexin (ANX)A6 (diluted 1:800; Sigma), and fast skeletal muscle troponin T isoform (TNNT3) (diluted 1:800; Sigma) were incubated overnight at 4°C. Membranes were washed twice for 10 min in TBST and once for 10 min in TBS. Horseradish peroxidase (HRP)-labeled anti-goat and anti-mouse secondary antibodies (Thermo Scientific Pierce, Waltham, MA, USA) were diluted 1:15,000 in TBST and incubated with the membranes for 90 min. After washing twice for 10 min in TBST and once for 10 min in TBS, immunoreactive proteins were visualized using a chemiluminescent HRP substrate (Millipore) in a dark room. The exposed films were analyzed for their gray-scale value using Image J. # Results and Discussion ## Proteomic analysis of breast muscle Traits related to breast muscle weight and IMF content were measured. The breast muscle absolute weight was obviously increased, but the breast muscle weight, relative to body weight, increased more slowly with age. Another study, using Beijing-You and western-type broilers, also showed that the breast muscle weights significantly increased with growth of the chickens. Saneyasu, et al.also investigated the change of body and breast muscle weights at 7, 14, 28, and 49 days of age, and showed significant increases in both with age. The relative weight of the breast muscle increased slowly, indicating a slight favoring of its growth over that of the whole body. Chartrin et al. investigated lipid deposition in breast muscle of mule ducks at days 1 to 98 and found that there are two periods of IMF deposition. The first, from day 1 to 42, is when lipids (mainly phospholipids and cholesterol provided by the egg yolk) stored in the adipocytes during embryonic life were transferred to the muscle fibers and used for growth and energy requirements and the second, after day 42, is when muscle again stores lipids. The present result is consistent with that finding as the content of IMF was highest at day 1, decreased dramatically by day 56, then increased again from day 56 to 140. The pooling strategy was adopted in this study, as it minimize the differences due to subject-to-subject variation and better identifies characteristics of the population. Clustering showed that the protein expression profile was consistent for each of the repetitions. Over 5000 proteins were identified and those accurately identified in breast muscle at days 1, 56, 98 and 140 were 494. Details of all accurately identified proteins as well as those at each sampled age are shown in. Cluster analysis of all proteins expressed at different ages showed that the proteins in breast muscle at post-hatching ages (days 56, 98 and 140) were more similar than those at hatching, and the proteins at day 56 and day 98 were similar. Two hundred and four differentially expressed proteins were defined and analyzed (P \< 0.05, indicating that the quantity in one pool was \> 1.5 or \< 0.7, compared to the other pool, for each pairwise comparison). To gain insight into the changes between each stage, four groups of proteins by Gene Ontology analysis were compared: day 56 vs. day 1, day 98 vs. day 56, and day 140 vs. day 98. Proteins related to glucose and intermediary metabolism were abundant from day 1 to 56; proteins involved in muscle development were abundant from day 56 to 98; and from day 98 to 140, translation and protein folding processes were abundant. The protein expression profiles were similar at the three post- hatching ages (days 56, 98 and 140) but differed from those at day 1 by clustering analysis. The molecular mechanisms of muscle development and IMF deposition are different at hatching and post-hatching stages. ## Highly upregulated or downregulated proteins in hatchlings compared to post-hatching chickens The proteins were defined as age-specific, highly upregulated or downregulated when the content differed \>10 fold compared to that at other ages. As shown in, many proteins were highly upregulated at day 1, such as ADP-ribosylhydrolase like 1 (ADPRHL1), alpha-2-HS-glycoprotein (AHSG), apolipoprotein A (APOA1, APOAIV), histone family (H1, H2B-VII, HIST2H2AC, H4), thymocyte nuclear protein 1 (THYN1), myosin light polypeptide 6 (MYL6), isocitrate dehydrogenase \[NADP\](IDH), peptidyl-prolyl cis-trans isomerase (PPIA), sarcalumenin (SRL), tubulin beta-7 chain (TUB7) *inter alia*. Some proteins were highly downregulated at day 1, which were mostly involved in energy metabolism and muscle development; for example, adenylate kinase isoenzyme 1 (AK1), fructose- bisphosphate (ALDOA.ALDOB.ALDOC), creatine kinase S-type, mitochondrial (CKMT2), desmin (DES), fructose-1, 6-bisphosphatase 2 (FBP2), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), glycogen phosphorylase (GPH1), glucose-6-phosphate isomerase (GPI), L-lactate dehydrogenase A chain (LDHA), phosphoglycerate kinase 1 (PGK1), malate dehydrogenase 1and 2 (MDH1, MDH2), malic enzyme 1 (ME1), phosphoglycerate mutase 1(PGAM1), phosphorylase B and L chain (PYGB,PYGL), acylphosphatase (ACYP), alpha-actinin-2 (ACTN2), troponin C, skeletal muscle (TNNC2), troponin I, fast skeletal muscle (TNNI2), fast skeletal muscle troponin T isoform (TNNT3), triosephosphate isomerase 1 (TPI1), tropomyosin alpha-1 chain (TPM1) etc. ## Over-represented pathways in breast muscle from day 1 to 56 Between days 1 and 56, breast muscle weight increased about 100-fold and IMF decreased about 100-fold. Thus, this is a critical stage for muscle growth and depletion of IMF. Comparing the protein profiles at days 1 and 56, 191 differing proteins were identified (the fold difference was \> 1.5 or \< 0.7). The 19 metabolic pathways were enriched during this fast growing stage based on a KEGG pathway analysis. The protein–protein interaction network of the differentially expressed proteins identified for this interval (days 56 vs. 1) was also analyzed by web-tool STRING 10.0 (<http://string-db.org>); there were two functional modules. The first related to metabolic pathways, including glycolysis/gluconeogenesis pathway, insulin signaling pathway, and lipid metabolic pathway. This module might relate to the significant changes in breast muscle, such as fast muscle growth and IMF deposition. The second module involved the ribosomes. ## Active pathways and key differentially expressed proteins at day 1 Within the 19 metabolic pathways, some differentially expressed proteins that were highly expressed at day 1 compared to day 56 were involved in the lipid metabolic pathway, such as valine, leucine and isoleucine degradation, and fatty acid degradation and elongation in mitochondria. Of eight differentially expressed proteins involved in the fatty acid degradation pathway, five (CPT-2, ACADL, HADHA, HADHB and ACAA2) were more abundant at day 1 than at day 56, and ALDH1A1 and ALDH2 were significantly lower at day 1 compared to day 56 (Abbreviations used in this paper are listed in Tables). It was deduced that lipid oxidation was more active and fatty acid synthesis was less active at day 1 compared to day 56. The differentially expressed proteins, APOA1 and APOAIV, were identified , which was consistent with previous studies. The reason may be that a large number of lipoproteins take up cholesterol from the yolk sac membrane at hatching, and the production of the APOs was stimulated by lipoproteins. Synthesis of APOAI in the skeletal muscle of hatchling chicks acts as a local lipid transporter in early post-hatching development. This is consistent with the phenotype where the content of IMF at day 1 is about 8 times higher than that at day 56. Expression of histone family and ribosomal family proteins differed greatly between hatching and post-hatching (56, 98 and 140 days). Histones play a pivotal role in regulating gene expression by controlling the access of key regulatory factors and complexes to chromatin, which is essential for transcription, DNA replication, DNA repair and DNA recombination. The organization of chromatin is considered to be regulated by post-translational modification of histones, such as methylation, acetylation, phosphorylation and ubiquitination. Many studies have shown that myogenesis is controlled through sequential chromatin regulation by the selection of the histone variant and the appropriate histone modification. For example, in mouse embryos, a bivalent modification of H3K4me3 and H3K27me3 was formed on H3.3-incorporated skeletal muscle genes before embryonic skeletal muscle differentiation. Ribosomal proteins, in conjunction with rRNA, make up the ribosomal subunits involved in the cellular process of translation and protein biosynthesis. It has been demonstrated that differential mRNA translation controls protein expression of specific subsets of genes during myogenesis, and one of a subset of transcripts that is enriched for mRNAs encoding ribosomal proteins is regulated at the translational level. Ribosomal proteins were also highly expressed during mature adipogenesis. A large group of ribosomal proteins was identified in chickens, which may partly explain why differentiation of myocytes and preadipocytes within muscle occurs mainly before hatching. ## Active pathways and key differentially expressed proteins at day 56 Many metabolism-related proteins were more abundant at day 56 than at day 1. There were 24 upregulated proteins, related to cytoskeleton and actin binding, including actin, cofilin, desmin, actinin, myosin, calpain, calmodulin, troponin, myomesin and myozenin. These proteins were involved in muscle-development-related pathways, such as glycolysis/gluconeogenesis, hypertrophic cardiomyopathy (HCM), insulin signaling, cardiac muscle contraction and dilated cardiomyopathy (DCM). Sixteen upregulated proteins related to the glycolysis/gluconeogenesis pathway were more abundant at day 56, and eight proteins in the HCM pathway and six (LMNA, DES, TTN, TPM1, TPM2, MYH6) were highly abundant at day 56. The period from day 1 to 56 is a fast growing stage for skeletal muscle. The identified proteins play a critical role in all skeletal and cardiac muscle in the early stages of development. Specifically, the energy from glycolysis/gluconeogenesis metabolism is needed for developing skeletal muscle. Insulin signaling pathway proteins are involved in the proliferation and differentiation of preadipocytes and myocytes\[–\], and are prominent for coordinating myofiber growth, muscle hypertrophy and muscle regeneration\[–\]. calmodulin (CALM), connectin/titin (TTN) and troponin C (TnC) are activated by the second messenger Ca²⁺ and stimulate expression of troponin I (TnI), troponin T (TnT), actin, myosin and muscle development\[–\]. In the present study, the amount of TTN, TnC (TNNC2), TnI (TNNI2), TnT (TNNT3), actin (ACTA1), myosin (MYBPC1, MYBPC2, MYBPH, MYLPF, MYH6), M line (MYOM2) and Z line (MYOZ1, MYOZ3) was enhanced, related to muscle development at day 56. Muscle contraction and hypertrophy are the main mechanisms influencing muscle development, along with the insulin signaling and HCM pathways at day 56. ## Over-represented pathways in breast muscle from day 56 to 98 (market age) From day 56 to 98, breast muscle weight and IMF about doubled and thus fast growth of muscle and IMF deposition continues. For the local slow-growing chickens used here, market age is around day 98. There were 44 differentially expressed proteins identified when comparing days 56 and 98. Seven significant enrichment pathways were identified by KEGG analysis, of which six are involved in muscle development. The only significant functional module in the network analysis of protein interaction, however, related to the ribosomes. There were 12 differentially expressed proteins related to cytoskeleton and actin binding, which was consistent with the high muscle growth rate at this stage. Four more abundant proteins at day 98 than at day 56 related to HCM and three related to focal adhesion. This suggests that the muscle development at this stage was closely connected with muscle growth, and these proteins may play important roles in focal adhesion and HCM during skeletal muscle development. The proteins included ACTN4, extracellular matrix proteins (LAMA2 and LAMB1), MYH6, TTN, DES and TPM2. Previous studies have made inroads into understanding the mechanism underlying muscle development. The enzymes related to cytoskeletal protein binding, including MYBPC2, MYBPC1, PDLIM3, ANXA2, SPTAN1, KBTBD10 and TUB7 were highly abundant in breast muscle at day 98 compared to day 56. ANXA2 is a Ca²⁺-binding protein implicated in several biochemical processes, including cell proliferation, ion-channel activation, cytoskeleton rearrangement, cell–cell interactions and the bridging of membranes. ANXA2 forms junctions between lipid bilayer structures through molecular bridging of their external leaflets. From day 56 to 98, focal adhesion, tight junctions and HCM play an important role in muscle development. Extracellular matrix proteins had a key role in muscle growth at day 98, which was different from day 56. ANXA2 may play an important role in lipid metabolism. ## Over-represented pathways in breast muscle from day 98 to 140 From day 98 to 140, breast muscle weight and IMF increased about 1.3-fold, so muscle growth and IMF deposition are slowing from that occurring in the previous phase. For the local slow-growing chickens used here, day 140 is near sexual maturity. Comparing the proteins at days 98 and 140, 58 were identified as being differentially expressed with cytoskeletal and ribosomal proteins being less abundant at day 140 than at day 98. The cytoskeleton is present in all cells, and plays important roles in cellular processes such as differentiation and apoptosis. As key regulators of cellular architecture, cytoskeletal components contribute to physical processes such as adhesion and migration\[–\]. Ribosomal proteins relate to the cellular processes of translation and protein biosynthesis. This decrease in abundance suggests that the rate of growth and the metabolic processes were slower at 140 than at 98 days. The only functional module clustering in the protein–protein interaction network involved the ribosomes. Three pathways were identified by KEGG analysis, with 58 differentially expressed proteins, including those of fatty acid metabolism and the proteins related to lipid metabolism, ACADL, APOA1 and HADH, were more abundant at day 140 compared to day 98. So, the capacity for lipid production was increased and that of oxidation decreased, resulting in lipid accumulation, perhaps explaining the higher IMF content in mature birds compared to younger birds. ## Verification of content of key proteins by Western blotting To validate the results of the iTRAQ testing, Western blotting was used to examine the relative contents of six key functional proteins at the different ages. Two proteins (MDH1 and TNNT3) related to muscle development, and four proteins (ALDH1A1, ANXA6, APOA1, HSPB1) related to lipid metabolism. There was acceptable consistency between the results of Western blotting and the fold- change of differentially expressed proteins from iTRAQ analysis. No internal reference was used here in the Western blotting because the abundance of β-actin, β-tubulin, histone and GAPDH all differed significantly at hatching and the post-hatching ages. The control sample used for Western blotting was a composite of breast muscle proteins made by pooling the different ages. MDH1 plays an important role in transporting NADH equivalents across the mitochondrial membrane, controlling tricarboxylic acid (TCA) cycle pool size and providing contractile function, so the increased expression of MDHs is required for the high demands of energy metabolism in developing tissues, especially in those with high metabolic rate such as heart, skeletal muscle, and brain. Troponin-mediated Ca²⁺-regulation governs the actin-activated myosin motor function which plays a key role in the regulation of striated muscle contraction in vertebrates. Point mutations in the *cTnT* gene have been found in human familial hypertrophic cardiomyopathy, and the expression of TnT isoform is regulated during heart and muscle development and adaptation, suggesting that TnT plays an important role in muscle growth and function. ALDHs are known to participate in oxidizing a plethora of endogenous and exogenous aldehydes. ALDH1A1 was up-regulated in omental and intramuscular preadipocytes during differentiation, and the increased levels of ALDH1A1 in the obese omental fat might be involved in fat accumulation. Annexin A6 (AnxA6) is a Ca²⁺ and phospholipid binding protein that acts as a scaffolding protein and regulates cholesterol transport along endo- and exocytic pathways. Loss of AnxA6 alters both lipid and glucose homeostasis, resulting in increased lipolysis and high density lipoprotein increased in *AnxA6* KO mice. Apolipoprotein A1 (apoA1) is the major apolipoprotein constituent of the high-density lipoprotein (HDL) and is involved in reverse cholesterol transport. Variants in the apolipoprotein A1 (*APOA1*) gene play an important role in the regulation of lipid transport\[–\]. Synthesis of APOAI in the skeletal muscle of hatchling chicks acts as a local lipid transporter for early post-hatching development. Heat shock protein beta 1 (HSPB1), a member of the heat-shock family of proteins, is a relatively small (27 kDa) molecular chaperone protein associated with cellular development, differentiation, and signal transduction. HspB1 and its regulator genes (*FAS*, and *AGT*) were shown to be good candidate genes associated with intramuscular fat content in the longissimus muscle of Korean cattle. # Conclusion In summary, the present study provides a useful resource for further investigating the roles of proteins expressed differentially in skeletal muscle at different developmental stages. Such efforts will enable better understanding of the molecular mechanisms of muscle development in chickens. The changes in protein abundance with age have not been documented previously, and the extent of the changes found here was unexpected. This study is the first step in understanding post-hatching development on a proteome-wide scale, and indicates the complexity of such an analysis. In addition, the present results suggest that APOA1 and HSPB1 may be useful as molecular markers of IMF deposition in chickens. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceived and designed the experiments:** RRL GPZ JWe. **Performed the experiments:** RQF JL. **Analyzed the data:** JWa MQZ. **Contributed reagents/materials/analysis tools:** HXC QHL JS. **Wrote the paper:** JL RQF.

# Introduction Rheumatoid arthritis (RA) causes significant costs for society due to the increased use of healthcare resources, sick leaves and early retirements. Consequently, effective anti-rheumatic therapies have the potential to reduce societal costs while also improving the patients’ quality of life. According to current Finnish Care Guideline, treatment of RA should be initially treated with a combination of methotrexate (MTX), hydroxychloroquine (HCQ), sulfasalazine (SSZ) and a low-dose glucocorticoid. In case of an insufficient response or intolerance, biological disease modifying anti-rheumatic drugs (bDMARDs), i.e. abatacept (ABA), tocilizumab (TCZ), rituximab (RTX), sarilumab (SAR), and tumour necrosis factor (TNF) inhibitors including etanercept (ETN), adalimumab (ADA), infliximab (IFX), certolizumab pegol (CTZ), and golimumab (GOL) are prescribed. The bDMARDs have comparable efficacy and not significantly differing safety profiles. bDMARDs are recommended to be used in combination with methotrexate rather than as monotherapy due to better efficacy and reduced immunogenicity. However, ABA and TCZ as monotherapy have been shown to have similar efficacy as in combination with MTX. A previously published systematic review indicated RTX as the most cost- effective bDMARD among patients with an insufficient response to bDMARD treatment. Previous cost-effectiveness analyses were mostly based on the efficacy measured in randomized controlled trials (RCT), and might therefore, have limited generalizability to routine healthcare owing to stringent inclusion criteria and brief follow-up. Nevertheless, a previous study based on observational data showed RTX to provide small savings and quality-adjusted life year (QALY) gains as a second line treatment as compared with TNF inhibitors. Whether these findings are generalizable to Finnish healthcare is unknown. Also, introduction of biosimilars has lowered the treatment costs of some TNF- inhibitors as well as rituximab. The treatment of RA, especially in the field of biological drugs, has changed a lot during previous years, and therefore, there is a need for cost-effectiveness analyses based on real-world data reflecting current treatment practice and providing valuable information for health-care decision making. The objective of this study was to evaluate the cost- effectiveness of ABA, TCZ, and TNF inhibitors as compared with RTX in RA patients, who have previously been treated with TNF inhibitor using Finnish patient-level registry data. # Materials and methods ## Model structure We developed a patient-level simulation model using R statistical programming language 3.2.2 to estimate costs and outcomes associated with different bDMARDs in the treatment of RA. The population consisted of RA patients who had previously used a TNF inhibitor as their first bDMARD and were about to begin their second bDMARD. The model simulated four alternative treatment regimens: ABA, RTX, TCZ and a second TNF inhibitor. In this simulation TNF inhibitors were considered together as a single group rather than as individual drugs due to the same mechanism of action and similar effectiveness. Different routes of administration for ABA and TCZ were also pooled as single groups. In the base case analysis, the choice of admin route of ABA and TCZ was based on the National Register for Biologic Treatment in Finland (ROB-FIN). Every treatment regimen was simulated by identical cohorts of 1,000 patients. In case the simulated bDMARD treatment was discontinued either due to lack of efficacy or adverse events, patients were switched to another bDMARD in the beginning of the next cycle until the patient had exhausted all four treatment options. After that, patients were assumed to be treated with a 6^th line treatment until death. The outline of the model is presented in. Patients were assumed to remain on any given treatment for at least 6 months. At the end of each six-month period, the model individually evaluated for each person whether he or she would continue treatment, discontinue it or die. The length of the time period was set to six months as this was the average time interval for routine care visits to rheumatologists in the (ROB-FIN). The simulation used regression models to predict outcomes and costs for each patient individually in each period. Each patients’ characteristics, history of drug use and past treatment responses were recorded in the model and utilized as predictors for future outcomes and costs. The analysis was conducted from a societal perspective as the study included both direct and indirect costs. Half- cycle correction was applied to both outcomes and costs. We analyzed direct costs and both direct and indirect costs separately. Health outcomes also were expressed as QALYs. Based on the Finnish recommendations for health economic evaluations, all costs and benefits were discounted at 3.0% annually. Primary outcome of the simulation was incremental cost-effectiveness ratio (ICER) per QALY including only direct costs. ## Data sources Primary data source for the model was the ROB-FIN and all assumptions were based on observed data unless otherwise mentioned. ROB-FIN has been described in previous publications. In brief, ROB-FIN is a longitudinal observational cohort study established in 1999 to monitor effectiveness and safety of biologic drugs in treatment of RA and was originally based on structured data collection forms submitted by rheumatologists on patients´ routine care visits to outpatient specialized healthcare. Starting in 2007, most of the data have been retrieved from electronic patient monitoring systems. Additional data on the patients’ hospitalization and outpatient visits as well as sick leaves and disability pensions were acquired from national healthcare registers, which were linked to ROB-FIN using social security numbers. Observed ROB-FIN data from baseline visit as well as subsequent follow-up visits were used to construct the regression models later used to predict patients’ treatment response, utility and costs in the simulation. Independent variable selection for the regression models was based on Akaike Information Criteria although the choice of bDMARD therapy and its interactions with other variables were always included where appropriate. Missing data among disease activity parameters were imputed by multiple imputation whereas information on treatments and use of healthcare resources were considered complete. Patients lost to follow-up were treated as uninformative censoring while information on treatment discontinuations were utilized in the modeling. ## Model inputs ### Baseline characteristics The model population were sampled with replacement among the patients included in ROB-FIN about to start their second bDMARD therapy thus preserving any potential correlation between the variables. The baseline variables included in the model comprised age, sex, weight, Body Mass Index (BMI), Health Assessment Questionnaire (HAQ), DAS28, time from diagnosis of rheumatic disease, Rheumatoid Factor (RF) status, and concomitant use of MTX, SSZ and HCQ along with the patients’ healthcare costs during the past 12 months. The patients´ weight, BMI and RF status were fixed at the baseline values. ### Clinical effectiveness Treatment effectiveness was defined as an achievement of at least American College of Rheumatology (ACR) 20% improvement, a moderate European League Against Rheumatism (EULAR) response, or a Disease Activity Score 28 (DAS28) value of less than 3.2. The actual observed data were used to construct regression models, which in turn were used to predict the responses in the simulation. Based on expert opinion, the treatment response to the 6^th line treatment options was assumed to be the same as averaged treatment response to all biological treatments in the model. The patients’ underlying HAQ score was assumed to remain fixed at the baseline value. The change in HAQ score in comparison to baseline was modeled and subsequently predicted using a linear regression model. Any effect the treatment could have on these disease activity parameters was assumed to be temporary and re-evaluated in the next period. ### Clinical safety In addition to lack of efficacy, treatment might be discontinued due to adverse events and other reasons. The risk of discontinuation due to adverse events and other reasons besides lack of efficacy was observed to be 0.068 across all bDMARDs in average based on the available data in ROB-FIN. After every six-month time period 0.017, 0.010 and 0.009 per cent of MTX, SSZ and HCQ users discontinued the use of the said co-treatment, respectively. ### Mortality Mortality rate adjusted by age and sex for patient-level data was based on the life table in 2017 published by Statistics Finland. In this model, RA was associated with mortality of general population. Despite using lifetime horizon in the model, we assumed that patients would die at the latest at the age of 100 due to lack of mortality data beyond that age. ### Utility Quality-adjusted life years (QALY) were calculated corresponding to EuroQol five dimensions questionnaire (EQ-5D-3L) utilities predicted from the HAQ scores utilizing a multinomial logistic regression model. The regression model was based on data from a survey including both HAQ and EQ-5D-3L conducted in Finland in 2009. Obtained EQ-5D-3L health stages were valued with the Finnish tariff. ### Costs Direct costs comprised drug costs, administration costs of infusions, costs of switching, outpatient and inpatient care, while indirect costs included early retirement due to RA and sick leave. Costs of drugs were based on the Finnish price list including the retail price without value added tax of drugs and the dose in the label. The costs for infusion drugs were the wholesale prices of Helsinki and Uusimaa Hospital District (Aaltonen T, personal communication, April 20, 2016). The prices of biosimilar IFX, ETN, and ADA were used in the base case analysis, whereas the price of biosimilar of RTX was taken into account in sensitivity analysis. Administration costs of infusions including intravenous treatment cost prices of bDMARDs for RA at Finnish hospitals were derived from the Finnish study by Soini et al. The price for TNF inhibitor group was an average for individual TNF inhibitors weighed by their actual usage. Similarly, the costs for ABA and TCZ were weighed averages based on the prices of subcutaneously and intravenously administered products. In the base case analysis, the bDMARD was subcutaneously administered for 15 and 7 per cent of ABA and TCZ users, respectively. Drug costs, administration costs of infusions and dosages are presented in. Cost of switching was assumed to be equal to the cost of one healthcare visit to internal medicine specialist in specialized outpatient healthcare. Based on the expert opinion and in line with their effectiveness, the costs of the 6^th line treatment option were considered to be same as the averaged costs of currently available biological treatments. Patients’ biannual healthcare and indirect costs were modeled using a linear regression model based on actual observed data and later predicted in the simulation using this model. Drugs in outpatient care were in 2019 euros and all other costs were converted to 2017 euros using the price indices of Statistics Finland. ## Sensitivity analyses The model was stochastic in nature and the 1000 unique model runs with different seeds for random number generator quantified the variability in the results. Several subgroup and deterministic sensitivity analyses were carried out based on 300 model runs to explore uncertainty and heterogeneity of the model results. As ABA and TCZ can be administered either subcutaneously or intravenously, we evaluated the influence of the choice of the administration route on the results in the deterministic sensitivity analyses. Even though HAQ progression was associated with age and the number of biologic treatment in the base case analysis, we used annual HAQ progression rates of 0.03 and 0.06 in the sensitivity analyses. Also, we included a scenario where we employed a British tariff to value EQ-5D health stages. Although the patent of RTX has expired, RTX biosimilar approved for the treatment of RA is not on the market in Finland. Therefore, we considered RTX biosimilar with a price discount of 30% in comparison to the reference medicinal product in sensitivity analysis. The time horizon of 10 years and the discounting rate of 0% and 6% were explored in the sensitivity analyses. In subgroup analyses we evaluated the heterogeneity of the results related to the use of the concomitant csDMARD therapy, age, body mass index (BMI), gender, presence of RF, prior use of MTX, and primary response to the first TNF inhibitor. ## Ethical approval Ethical approval for this study was granted by the Helsinki University Central Hospital ethical committee (73/13/03/00/2014). The study permit to use the patient records and cost data was granted by the Finnish National Institute for Health and Welfare (THL/1497/.5.05.00/2013), Finnish Population Registry (262/410/16) and the Social Insurance Institution of Finland (Kela 7/522/2016). Written informed patient consents were acquired from patients who had been included in ROB-FIN prior to the introduction of the electronic patient monitoring systems. # Results ## Baseline characteristics At baseline, median age of patients included in the study was 56 years. Most of the patients were female. Median HAQ score was 1.1, whereas median DAS28 was 4.6. More than half of the patients had a treatment response to the first TNF inhibitor. The characteristics of the patients at the baseline are presented in. ## Base-case analysis Lifetime drug costs without administration costs and costs of switching were the lowest for RTX, but when administration costs and costs of switching were included, drug costs were the lowest for TNF inhibitors. ABA had the highest drug costs. However, ABA had the lowest healthcare costs, while RTX had the highest healthcare costs. In total, the lowest and highest direct costs were associated with the TNF inhibitors and RTX, respectively. Indirect costs ranged from 148,718 € for TNF inhibitors to 165,300 € for RTX. Drug costs including administration costs and costs of switching represented over half of the total costs. QALYs ranged from 9.405 of RTX to 9.661 of TNF inhibitors. In our model, patients died in average at the age of 85.59 (standard deviation 9.96). TNF inhibitors, ABA, and TCZ had lower costs and higher QALYs than RTX, and therefore, they were dominant in comparison to RTX. TNF inhibitors are the most cost-effective treatment option, as they have the lowest costs and the highest lifetime QALYs. ## Sensitivity analysis The results regarding costs and QALYs from the subgroup and deterministic sensitivity analysis are presented in and Tables. Instead of intravenous administration, the use of self-administered ABA and TCZ decreased the administration costs of drugs, but the choice of route of administration had no effect on other costs. Similarly to base case results, RTX had the highest costs and lowest QALYs when they were discounted at 0%, but when the discounting rate of 6% was used, TCZ had the lowest QALYs. When the time horizon of 10 years was used, RTX had the lowest drug costs including the costs of switching and the administration costs, while TCZ was associated with the lowest QALYs. As compared with RTX, TNF inhibitors were dominant, whereas the incremental cost- effectiveness ratio for ABA was 144,213 €/QALY when the time horizon of 10 years was used. RTX as a second-line treatment option had the highest direct costs even when the price discount of 30% for RTX biosimilar was used, owing to the highest outpatient and inpatient costs for RTX. Indirect costs increased and QALYs decreased in case the bi-annual HAQ progression was fixed at 0.03 or 0.06. Removing half-cycle correction had little effect on the results. QALYs decreased when the health stages were valued with the British tariff. The costs were higher and QALYs were lower for women than for men. As compared with patients with no response to the first TNF inhibitor, primary responders had slightly higher costs and QALYs. The QALYs ranged from 11.960 to 12.370 for patients with negative RF status, while QALYs ranged from 9.159 to 9.415 for patients with positive RF status. Also the costs were higher among patients with negative RF status as compared to RF-positive patients. Furthermore, concomitant use of MTX or other non-biologic therapies lead to increased QALYs in comparison to non-use of MTX or biologic monotherapy, respectively. # Discussion Based on our patient-level simulation model using real-world data from Finland, TNF inhibitors, ABA, and TCZ as a second-line biologic treatments for RA were dominant as compared with RTX. Even though a commonly referred threshold for cost-effectiveness has not been published in Finland, we have used the willingness to pay threshold of 40,600€, representing the Finland´s gross domestic product per capita in 2017 per QALY gained in this study. According to our model, RTX was associated with the lowest drug costs, but when administration costs and costs of switching were included, TNF inhibitors had the lowest drug costs. RTX had the highest healthcare costs, and in total, RTX was associated with the highest drug costs. TCZ was associated with the lowest effectiveness. TNF inhibitors had the lowest costs and the highest QALYs, and therefore, they were the most cost-effective treatment option. Consistent with our results, previous studies showed similar effectiveness between bDMARDs in patients with RA failing on TNF inhibitor, whereas other studies suggest that changing to RTX is more effective than switching to an alternative TNF inhibitor. Unlike our results, previously published results by Lindgren et al. found that RTX treatment was associated with the lowest overall costs and was the most effective option. Similarly, in a head-to-head RCT between RTX, TNF-inhibitors, ABA and TCZ, RTX was associated with both the lowest costs and the highest QALY gain. The Finnish study reported also RTX to be most cost-effective treatment alternative for patients with RA who have failed TNF inhibitor treatment. According to this study, life years were highly similar between all bDMARDs, but they were the highest for RTX. The main advantage of this study was that real-world data on costs and effectiveness were based on registry data from the same population. Baseline characteristics of the model and population and the assumptions used in the model were mainly based on ROB-FIN data representing routine clinical practice. Compared to the previous studies mostly based on efficacy derived from RCTs, our simulation was based on observational data and therefore results are likely to be more generalizable to the Finnish healthcare setting. In addition, cost data derived from comprehensive Finnish national registers were employed whenever possible. To eliminate the bias caused by confounders in the observational data, we used several regression models to predict the patients’ treatment response, utility and costs in the simulation. Our model memorized changes in patient characteristics over time and enabled simulation based on individual patients’ history. The history of what had happened was an important aspect of the model because it affected the occurrence of future events, their consequences and valuations, and many other aspects of the simulation. The simulation followed up patients for their complete lifetime, because RA is a chronic disorder that progresses over time. As such, long-term consequences of any differences in disease progression, effect on life expectancy, or drug discontinuation rates were assessed. The first TNF inhibitor was not included in this model as it was assumed to be identical between the comparators and would not affect the results. Reasons for discontinuation of RTX treatment were severely underreported in ROB-FIN data and therefore, the risk of adverse events was assumed to be equal across all treatment regimens. Also, the discontinuation probabilities for non-biologic co-therapies were assumed to be similar between the users of different biologics. Because of high costs of original bDMARDs, interest has grown in biosimilars that are comparable to the reference medicinal product in terms of efficacy and safety. Many original biologic drugs have reached, or are approaching, patent expiry. This will lead to increasing development and use of biosimilar drugs in the future, offering considerable savings in comparison with the reference medicinal product. Price competition after patent expiry may also reduce the price of reference medicinal product, leading to remarkable cost savings. Biosimilars of ETN, IFX, and ADA were considered in this analysis. The price of RTX biosimilar used for the treatment of RA was not available, but we used price discount of 30% for RTX biosimilar in the sensitivity analysis based on the Finnish legislation concerning the confirming a reasonable wholesale price for medicinal products. However, the price discount may be even bigger in the future. A potential limitation of this study is related to the dosing interval of RTX. As compared with other biologics, RTX has a unique mode of action and long dosing interval. In maintenance treatment, RTX infusions were often administered on demand in Finland. Consequently, the patients treated with RTX came to visit the rheumatologist only after the effect of the previous infusion began to wear off, which is very likely to lead to an underestimation of the effectiveness of RTX in our data, and therefore, leading to overestimation of healthcare costs. Based on the registry data used in this study, RTX was administered every 7.98 months in maintenance treatment. Similar dosing interval of RTX was reported by Keystone et al, whereas mean dosing interval of RTX varied between 11 and 13 months in daily clinical practice in Finland according to the study by Valleala et al. Therefore, a relatively short dosing interval used in this study might also be one reason to cause high costs of RTX treatment. In this model RA was associated with mortality of general population, which is another limitation of this study. According to the study by Kroot et al, mortality of RA patients was comparable with the expected mortality of the general population of the Netherland up to 10 years of RA. This finding was in line with the study by Lindgqvist and Eberhardt et al.. Furthermore, Lacaille et al found that mortality gap between RA and the general population in the first five years was not observed in people with RA onset after year 2000. However, it is notable that the patient populations in these studies comprised of patients with recent onset of RA. Although mortality has decreased among RA patients over the past decades, the general belief is that patients with RA, especially the more severe cases, have a shortened life expectancy compared with the general population. We assume that differences in QALYs between bDMARDS would have been bigger, but the order of these results would have been the same, if the association between disease activity and mortality in RA patients had been considered in the model. We did not however, have sufficient data to create a prediction model for mortality, which can be considered a limitation. Furthermore, patients’ erosive progression could not be taken into account as data on this subject was not available in ROB-FIN. The effect could be mitigated by inclusion of HAQ scores, which have been shown to be correlated with the presence of joint erosions. Some bias could also be caused by the lack of a generic health related quality of life instrument. When we employed another valuation method in sensitivity analysis, the QALYs decreased when health stages were valued with the British tariff. # Conclusion Our patient-level simulation based on observational data showed that TNF inhibitors were associated with the lowest costs and highest QALYs, whereas RTX had the highest costs and lowest QALYs. TNF inhibitors, ABA, and TCZ were dominant in comparison to RTX. As TNF inhibitors had the lowest costs and highest QALYs, they were the most cost-effective treatment option. # Supporting information We acknowledge the support given to the National Register for Biologic Treatment in Finland (ROB-FIN) by the Finnish Society for Rheumatology and all the rheumatologists who have contributed to data collection. [^1]: HR has received a fee for speaking from Pfizer. KT has received fees for consultancy from Novartis and Pfizer. KP has received fees for speaking and consultancy from Pfizer, MSD, Abbvie, BMS, Roche, Lilly, Novartis, Sandoz, and Sanofi. DN has received fees for consultancy from AbbVie, BMS, Lilly, MSD, Novartis, Pfizer, Roche, and UCB. SH works at ESiOR Oy, which carries out studies, consultancy, education, reporting and health economic evaluations for several pharmaceutical (including companies producing biologic RA drugs), food industry, diagnostics and device companies, hospitals, and academic institutions. ESiOR Oy did not play any role in study design, data collection and analysis, decision to publish and did not provide any financial support to work or author. All other authors have declared no conflicts of interest. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# 1 Introduction Over the past few decades, functional MRI has widened our understanding of the functional organization of intrinsic brain networks and their role in cognition and behavior. Classical univariate (i.e., voxel-wise) analyses of fMRI signal (i.e., blood-oxygenation level-dependent, or BOLD) have been instrumental in probing the specialized function of brain regions. More recent approaches using functional connectivity and network neuroscience portray a complex and multi- scale set of interactions between brain structures. Following this view, a wide array of graph theoretical and complex systems tools have been used to describe BOLD dynamics. Despite these efforts, we still lack a unified mechanistic framework that overcomes three key limitations. First, the features of the BOLD signal that are important for neural activity are unclear. Several prior studies demonstrate a relation between BOLD and slow amplitude features of cortical activity, and between BOLD and the hemodynamic response function (HRF). These studies imply that the low frequency component of the BOLD signal contains information relevant to underlying neural dynamics, although it is also clear that the signal contains artifact. Due to the mixture of signal and artifact in the BOLD time series, it is possible that the common practice of band-pass filtering the BOLD signal at low frequencies may exclude functionally relevant signal. Second, many graph theoretic and network analyses are inherently descriptive in nature, and lack the power to give a generative understanding of the relationship between model inputs and outputs (for extensions of these approaches that move beyond description into explanation and prediction, see). Finally, model-based approaches often treat the brain as an isolated system by ignoring external input, or assuming an artificial profile of internal and external noise. To address these three limitations, we develop a generative framework that explicitly includes exogenous input (e.g., external sensory or subcortical structures’ inputs), and provide evidence that the brain’s activity can be fruitfully understood in the context of its natural drivers. Specifically, we use a multivariate autoregressive model with unknown inputs to capture the spatiotemporal evolution of the BOLD signal driven by extra-cortical inputs. These models have been used to characterize and predict the evolution of several synthetic and biological systems. For instance, Chang and colleagues (2012) leveraged a multivariate linear dynamical system’s framework and the patients’ intracranial EEG to model the cortical impulse response to the direct electrical stimulation. Many prior studies use this and similar methods such as Granger causality and dynamic causal modeling (DCM) for understanding the directed functional connectivity of BOLD \[, –\]. While some prior studies account for the effect of exogenous input, they typically assume a simple known and abstract form of the input function. Moreover, the inability of models such as DCM to capture signal variations beyond those caused by the external inputs makes the connectivity estimation highly dependent on the assumed number and form of the inputs. In this work, we treat the exogenous inputs to the cortex as *unknown* parameters of a linear time-invariant (LTI) system, which we estimate following recent developments in linear systems theory. We use these developments to provide new insights into how the brain responds to ongoing task requirements, and to shine a light on factors that contribute to the dynamics of cortical functional connectivity. To demonstrate our approach’s utility, we begin with a proof-of-concept where we consider synthetic examples for which we retrieve the external inputs’ spatiotemporal profiles of a known LTI system. We demonstrate that unknown external inputs result in apparent changes in internal system parameters, and consequently, in estimated external inputs’ error. Also, we show that using internal system parameters estimated from time windows without external stimulation significantly improves our ability to extract external inputs’ profile from periods with external stimulation, expect for simulations with relatively low external inputs and signal-to-noise. Next, we test the hypothesis that variations in cortical dynamics during different tasks or cognitive states can be accurately modeled as external excitations on fairly stable interactions between cortical regions. Specifically, we recover the unknown external cortical inputs during resting- state and task scans for 96 subjects with the lowest motion artifact from the Human Connectome Project (HCP). Our results demonstrate that using system parameters estimated from resting-state scans enables uncovering the expected spatiotemporal profiles of external sensory (i.e., visual cues) and task-related extra-cortical inputs, while system parameters estimated from task scans result in highly inaccurate input estimations. In addition, an in-depth examination of estimated inputs during task scans reveals the spatiotemporal patterns of other task-related inputs that were not captured by the abstract task regressors. Lastly, we measure the non-stationarity of estimated external inputs over resting-state scans to examine the assumption of the system’s time-invariance and to identify exogenous determinants of the BOLD signal’s non-stationarity. Recently, the nature of non-stationarity of BOLD signal and dynamic functional connectivity has been a topic of scientific debate, as several recent publications paint seemingly contrasting portraits of the processes’ stationarity underlying the brain’s functional dynamics. However, to the best of authors’ knowledge, no study examines the BOLD signal’s stationarity in the context of time-varying external inputs and their effects. Our results show that the inputs to several brain regions, most notably over default mode network, estimated from the resting-state scans display significantly high non- stationarity compared to other brain regions. Together, we demonstrate that our framework allows us to uncover spatiotemporal patterns and dimensionality of unknown cortical drivers. These findings offer insight into how a relatively static relation between brain regions and exogenous drivers can give rise to complex cortical dynamics and contribute to their non-stationarity. # 2 Materials and methods ## 2.1 Linear time-invariant (LTI) dynamical systems with external inputs Each region *i* of interest (ROI) from which the BOLD signal is collected provided us with a time series described by *x*_*i*\[*k*\] at sampling point *k* = 0, …, *T*. A total of *n* = 100 regions are considered and the collection of these signals is captured by the vector $x\lbrack k\rbrack = \left\lbrack x_{1}\lbrack k\rbrack\mspace{720mu}\ldots\mspace{720mu} x_{n}\lbrack k\rbrack \right\rbrack^{\intercal}$, with *k* = 0, …, *T*, which we refer to as the *state of the system* (i.e., it describes the evolution of the BOLD signal across different regions). The evolution of the system’s state is mainly driven by (*i*) the cross-dependencies of the signals in different regions (not necessarily adjacent), and (*ii*) the external inputs that are either excitation noise or inputs arriving from the environment surrounding the regions captured by the state of the system (e.g., stimulus arriving from subcortical structures not accounted for during BOLD signal collection). Subsequently, a first step towards modeling the evolution of the system’s state is: $$\begin{array}{r} {x\lbrack k + 1\rbrack = Ax\lbrack k\rbrack + Bu\lbrack k\rbrack + \omega_{k},\quad k = 0,\ldots,T,} \\ \end{array}$$ where $A \in \mathbb{R}^{n \times n}$ described the autonomous dynamics, $B \in \mathbb{R}^{n \times p}$ is the input matrix that describes the impact of inputs (i.e., external drivers) $u\lbrack k\rbrack \in \mathbb{R}^{p \times 1}$ on the system state’s evolution, and $\omega_{k} \in \mathbb{R}^{n}$ is the internal dynamics noise (i.e., internal drivers) at sampling point *k*. Notice that $\left\{ x\lbrack k\rbrack \right\}_{k = 0}^{T}$ is the BOLD signal at the different ROIs and is the only known. However, the state of the underlying neural activity is unknown since we did not account for the hemodynamic response function (HRF) in our reduced model. Therefore, the input in the model captures the external drivers of regional BOLD and only indirectly, the underlying neural activity. In order to determine the *parameters of the system*, i.e., (*A*, *B*, $\left\{ u\lbrack k\rbrack \right\}_{k = 0}^{T}$), we need to solve an optimization problem that minimizes the distance between the system’s state *x*\[*k*\] and the estimate of that state given by $\hat{x}\lbrack k\rbrack$ driven by the unknown quantities. Specifically, we have the following optimization problem: $$\begin{array}{rc} {\left\{ \hat{x}\lbrack k\rbrack \right\}_{k = 0}^{T} \in \text{arg}\underset{z\lbrack 0\rbrack,\ldots,z\lbrack T\rbrack}{\text{min}}} & {\quad \parallel z\lbrack k\rbrack - x\lbrack k\rbrack \parallel}_{2}^{2} \\ {\mspace{720mu}\text{s.t.}} & {z\lbrack k + 1\rbrack = Az\lbrack k\rbrack + Bu\lbrack k\rbrack.} \\ \end{array}$$ Notice that this problem is more challenging than the usual least squares problem considered when the parameters of the system are known. Thus, similar to the method develop by, we perform the following steps: (i) we assume that the state *z*\[0\] = *x*\[0\], and $\left\{ u\lbrack k\rbrack \right\}_{k = 0}^{T}$ is identically zero, to find an approximation to *A*;(ii) assuming *A* is given by the initial approximation, we provide a sparse low-rank structure to matrix *B* and we find an approximation to both *z*\[0\] and $\left\{ u\lbrack k\rbrack \right\}_{k = 0}^{T}$, which suffices to obtain *z*\[0\], …, *z*\[*T*\] subsequently, $\left\{ \hat{x}\lbrack k\rbrack \right\}_{k = 0}^{T}$; and (iii) assume $\left\{ z\lbrack k\rbrack \right\}_{k = 0}^{T}$ and $\left\{ u\lbrack k\rbrack \right\}_{k = 0}^{T}$ are as approximated in step (ii) and determine an approximation to *B*. The process consists of executing step (ii) and (iii) iteratively. Our experiments reveal that the estimated parameters converge after a few iterations in both synthetic and fMRI time series (S18 Fig). Additionally, to force the inputs to be used as little as possible, since otherwise they could contain all the required information to obtain the sequence $\left\{ z\lbrack k\rbrack \right\}_{k = 0}^{T}$ (e.g., consider *A* to be zero and *B* to be the identity matrix), the optimization objective is rather given by ${\parallel z\lbrack k\rbrack - x\lbrack k\rbrack \parallel}_{2}^{2} + {\lambda \parallel u \parallel}_{1} + \lambda{\parallel B \parallel}_{1}^{2}$, which penalizes the use of the input with a weight *λ* \> 0.—See section SI1 in for algorithm details. We will demonstrate in the following results section that unaccounted external inputs result in error in estimation of system matrix *A*. Therefore, in a modified version of this algorithm, in step (i) we estimate *A* from *x*′\[*k*\] measured during an extended window without external stimulation (e.g., resting- state). Next, we repeat steps (ii) and (iii) iteratively—as detailed above. Since we did not know the true dimensionality of the external inputs, we approximated the dimensions of the input matrix *B* by performing principal component analysis on the residuals of the models. As seen in S19 Fig in, principal components 1–25 capture more than 80% of variance in the average residuals and more than average 60% of subject-level residuals’ variance across all tasks. In addition, we compared the goodness-of-fit of the LTI model with and without external inputs using Akaike information criterion (AIC). Our results demonstrate that incorporating external inputs does not results in overfitting and improves the model’s fit—an effect most pronounced in higher dimensional input matrices (S20 Fig). Finally, we demonstrate that we identify the external inputs during the motor task similarly at high-dimensional input matrices (S6 Fig), as indicated by the high correlation (\>0.8) of inputs estimated using input matrix dimensions higher than 25 (S6I Fig). Therefore, we select *p* = 25 for input matrix *B* to estimate the inputs from task fMRI time series. ## 2.2 Spectral analysis of an LTI system Provided an LTI description of the system dynamics, the autonomous evolution of the dynamical system can be decomposed in a so-called *eigenmode decomposition*. Briefly, consider the *n* eigenmodes (i.e., eigenvalues and the corresponding eigenvectors) associated with *A*. Each eigenmode corresponds to an eigenvalue- eigenvector pair (*λ*_*i*, *v*_*i*) satisfying *Av*_*i* = *λ*_*i* *v*_*i*, and it describes the oscillatory behavior for a specific direction *v*_*i*. Specifically, for any given eigenvalue *λ*_*i* represented in polar coordinates (*θ*_*i*, \|*λ*_*i*\|), we have that it captures the *frequency* characterized as $$f_{i} = \frac{\theta_{i}}{2\pi}\delta t,$$ where *δt* corresponds to the sampling frequency, and the *time scale* given by $$\rho_{i} = \frac{\text{log}\left( \middle| \lambda_{i} \middle| \right)}{\delta t},$$ which can be interpreted as the *damping rate*. In particular, we can re-write $A = V\lambda V^{\intercal}$, where *V* = \[*v*₁, …, *v*_*n*\] and *λ* = diag(*λ*₁, …, *λ*_*n*) are the matrices of eigenvectors and eigenvalues. Subsequently, we can apply a change of variable as *z*\[*k*\] = *V*\* *x*\[*k*\], where *V*\* is the transpose conjugate, which implies that $z_{i}\lbrack k\rbrack = v_{i}^{\intercal}x\lbrack k\rbrack$ is a weighted combination described by the *i*_*th* eigenvector associated with the *i*_*th* eigenvalue. Hence, this can be understood as the spatial contributions of the *n* ROIs at a given (spatiotemporal) frequency *f*_*i*. Additionally, we can revisit the damping rate of the process in such direction *v*_*i* by reasoning as follows: first, we can recursively obtain \|*z*_*i*\[*k*\]\| = \|*λ*_*i*\|^*t*\|*z*_*i*\[0\]\|. Therefore, we have the following three scenarios: (*i*) \|*λ*_*i*\|\<1; (*ii*) \|*λ*_*i*\|\>1; and (*iii*) \|*λ*_*i*\| = 1. In case (i) and (ii), we can readily see that \|*z*_*i*\[*k*\]\| → 0 and \|*z*_*i*\[*k*\]\| → ∞ as *k* → ∞, respectively. Lastly, in scenario (iii), or practically, when \|*λ*_*i*\|≈1, we have that the process oscillates between *stability* and *instability*, and therefore these dynamics are refer to as *meta-stable*. In summary, the dynamical process *z*(*k*) describes the spatiotemporal brain BOLD signal evolution. Specifically, the timescales are encoded in the eigenvalues and the spatial contributions of the different ROIs are described by the eigenvectors with a spatiotemporal timescale described by the associated eigenvalues. ## 2.3 Dataset and preprocessing We used data from the Human Connectome Project (HCP). As part of the HCP protocol, subjects underwent two separate resting-state scans along with seven task fMRI scans, both of which included two sessions. All data analyzed here came from these scans and was part of the HCP S1200 release. The fMRI protocol (both resting-state and task) includes a multi-band factor of 8, spatial resolution of 2 mm isotropic voxels, and a TR of 0.72 sec (for more details see). Subjects that completed both resting-state scans and all task scans were analyzed. Each of the scanning sessions included both resting-state and task fMRI. First, two 15-minute resting-state scans (eyes open and fixation on a cross-hair) are acquired, for a total of 1 hour of resting-state data over the two-day visit. Second, approximately 30 min of task-fMRI is acquired in each session, including 7 tasks split between the two sessions, for a total of 1 hour of task fMRI (for details see). Head-motion artifacts result in significant error in the functional connectivity estimates. Therefore, to minimize head-motion artifacts, we selected 100 subjects with the lowest mean frame-wise displacement in our study, where we utilized a cortical parcellation with *N* = 100 parcels that maximizes the similarity of functional connectivity within each parcel. Next to keep the same subjects across the resting state and task scans, we removed the four patients with missing either task or resting state scans. We preprocessed resting-state and task data using similar pipelines. For resting-state, the ICA-FIX resting- state data provided by the Human Connectome Project were utilized, which used ICA to remove nuisance and motion signals. For task data, CompCor, with five components from the ventricles and white matter masks, was used to regress out nuisance signals from the time series. In addition, for the task data, the 12 detrended motion estimates provided by the Human Connectome Project were regressed out from the time series. For both task and resting-state, the mean global signal was also removed in an effort to remove the auto-correlated non- physiological noise and reduce the model estimation error. ## 2.4 Statistics We performed student’s *t*-test and Welch’s *t*-test to test the statistical significance of the differences between the distributions of interest. Non- parametric Wilcoxon rank-sum test were utilized for comparisons of distributions with non-normal profiles. We corrected calculated test statistics for multiple comparisons using false discovery rate (FDR) method, as well as the more conservative Bonferroni method. To identify the task-specific fluctuations in the average estimated inputs, for each brain regions we compared task-related inputs to those estimated from resting-state time series (paired *t*−test, *p* \< 0.05, FDR). In addition we also generated phase-randomized null time series from each subjects’ BOLD times series for the task time series. We select the phase-randomized null model since it maintains most of the statistical properties of multivariate time series (e.g., autocorrelation, covariance). Next, for each brain region, we compared the average empirical and null estimated inputs for each time point (paired *t*−test, *p* \< 0.05, FDR). To identify estimated inputs that display changes that correspond to different task conditions in the motor paradigm, we first performed a principal component analysis (PCA) on all estimated inputs (*U*) concatenated over all subjects. Next, we identified a single input with the highest absolute principal component (PC) loading for every component. We then multiplied the selected inputs with negative PC loadings by −1. Next, we separately fitted a multiple linear regression model for each PC’s inputs (*U*) using the known task-regressors. We created task-regressors for different conditions by assigning every sample to baseline (0) or one of six events (i.e., visual cue, left hand, right hand, left foot, and right foot movements) based on their temporal proximity to events’ onsets and offsets. We repeated this analysis by shifting task-regressors by different lags (0–12 TRs) to identify the lag that produces the best fit (i.e., highest *R*² values) for each region. Finally, we performed *t*−tests on estimated coefficients at the group-level to identify task conditions similarly echoed in estimated inputs associated with each PC across participants. We also identified brain regions that correspond to the identified inputs by performing group-level region-wise *t*−tests on input matrix *B* elements that correspond to inputs *U* identified by PCs. We examined the estimated inputs’ non-stationarity using two methods. First, we used a sliding window approach to examine temporal fluctuations of estimated inputs’ means over resting-state scans for all brain regions, measured from the windowed-means’ standard deviation. Second, we used the nonlinear non- stationarity index introduced by, with *α* = 0.9 and *β* = 1 exponent parameters following their study, where *α* and *β* parameters control the relative weighting between the importance of long versus large excursions in time series. Therefore, non-stationarity indexes with our selected parameters give marginally greater weighting to excursions’ height. Finally, to test the group-level significance of both non-stationarity metrics, we first normalized the values across all brain regions. Next, we used the *t*−test (FDR corrected for multiple comparisons across all brain regions) to establish the statistical significance of the measured non-stationarities across patients. Traditionally, researchers have commonly used the 0.05 as the statistical significance level, though the choice is largely subjective. Therefore to convey the probabilistic nature of the statistical analysis and the proper interpretation of statistical test results, in the manuscript, we refer to results of the commonly accepted statistical threshold of 0.05 as “significant” and the more conservative thresholds of 0.0005 or lower as “highly significant”. ## 2.5 Ethics statement All subject recruitment procedures and written informed consents were approved by the Washington University Institutional Review Board (IRB). For more details see. ## 2.6 Retrieving the external inputs to a synthetic LTI system We use the proposed method to explicitly model the contributions of internal system dynamics and external inputs on the BOLD signal during rest and task. To build intuition, we begin by estimating the internal system parameters and unknown inputs using data simulated from a synthetic LTI model with four states representing four brain regions. We first simulate the dynamics of our model, where each region is driven by random internal noise, and only one region is driven by an additional square pulse train. For details regarding the simulation see section SI2 in. Next, we estimate internal system parameters (4 × 4 matrix of interactions) and unknown inputs from the simulated time series, and to recover the spatial and temporal profiles of the pulse train input. Although the estimated inputs (green line) fluctuate time-locked to the ground-truth input, their temporal profiles notably differ. We hypothesize that this divergence arises from the error in system matrices estimated during periods with external stimulations. In , we show that the LTI system parameters receiving time-varying external inputs can falsely appear to change and diverge farther from the ground-truth when examined over periods with external stimulation. Consequently, we hypothesize that system matrices estimated from periods without external inputs would improve our ability to capture the unknown inputs’ profile accurately. shows that using a fixed system matrix estimated from periods without external inputs significantly increases the similarity (correlation) to the ground-truth inputs. We also demonstrate that although estimated inputs contain noise, averaging inputs estimated over 100 simulations results in highly accurate estimations (correlation = 0.99). The significant (Wilcoxon rank-sum test, Bonferroni *p* \< 0.0001) changes in the input matrix *B*’s loading for estimation windows overlapping the external stimulation periods, reveals the unknown external inputs’ spatial profile (i.e., the blue input node). Together, these results demonstrate that external inputs can increase estimation error in system matrices, and consequently, input parameters. More importantly, these results also show that identifying system matrices from periods without external stimulation allow an accurate estimation of unknown external inputs’ spatiotemporal profiles. Next, we generate synthetic time series by stimulating LTI systems, parameters of which were estimated from subjects’ resting-state BOLD time series. We set external inputs’ magnitude such that the global average stimulus-induced changes in normalized simulated outputs match the largest average task-related changes in a sample (social) task. We confirm that similar to the low-dimensional example in, our approach is able to extract synthetic external inputs to high- dimensional LTI models of BOLD signal dynamics. Likewise, employing system parameters estimated from periods without external stimulation results in a significant (*t*−test, *p* \< 0.05, *p* = 6.6 × 10⁻⁶⁵ and *p* = 2.9 × 10⁻⁶⁶ for 1000 TR- and 250 TR-long estimation windows, respectively) increase in the similarity between the ground-truth and estimated inputs. The notably higher similarity between the average estimated to ground-truth inputs than that of subject-level estimated inputs suggests that profiles of external inputs are correctly approximated although with noise. Together these results demonstrate the utility of our framework in identifying external inputs to LTI systems, and highlight the importance of accurate estimation of model parameters. So far, we have examined the LTI system’s response in a low recording noise level (signal-to-recording noise = 1000). Next, we examine the accuracy of the retrieved model and input parameters at different recording and internal noise levels. The contributions of the recording and internal noise to the BOLD signal, for the most part, are unknown quantities. However, they play an essential role in our ability to capture external inputs accurately. Simulating the system’s response magnitude and variance (i.e., *t*-values) at various recording and internal noise levels show how different noise levels can lead to seemingly similar outputs. Moreover, at high noise levels, the error increases notably in the system parameters estimated from periods without external inputs, and consequently, in the estimated input parameters during stimulation periods. Interestingly, at such high noise levels, the system matrices estimated during stimulation periods more accurately recover external inputs than those estimated during periods without stimulation (S1 Fig in). These observations suggest that the choice of system matrices and the goodness-of-fit of the estimated inputs can further provide insight into the empirical noise levels. In the following, we consider the proposed methodology in the context of quantifying important spatial and temporal features of the internal system dynamics and external inputs estimated from the HCP resting-state and task fMRI scans. ## 2.7 Capturing external drivers of BOLD signal ### 2.7.1 Brain’s large-scale oscillatory modes display heterogeneous spatiotemporal profiles We begin by showing that the estimated system parameters during resting-state reliably capture and reproduce known brain functional organization. Further, because these parameters reside within a quantitative dynamical model, we simultaneously capture both spatial (regions that are co-active) and temporal (oscillation frequency) information through the *eigenmodes* of our estimated system. Specifically, each eigenvector indicates an independent pattern of co- active regions, and its corresponding eigenvalue determines both the oscillation frequency and the change in amplitude of the activation patterns. Intuitively, if we initialize our estimated system state to a pattern of activity corresponding to an eigenvector, then the system states would oscillate and dampen according to the associated eigenvalue’s characteristics (see more details in Materials and Methods section). To capture the spatial and temporal patterns of activity, we use our method to estimate the internal system parameters from the resting-state time series (1200 TR ≈14.5 min). The high stability of the (i.e., slow damping rate) low-frequency eigenvalues as seen in indicates that the system’s outputs are dominated by lower frequency oscillations. To identify the eigenmodes with similar spatial patterns across subjects, we aggregate all subjects’ eigenvectors and perform *k*-means clustering analysis. We used the elbow method (optimal *k* ≈ 4), Calinski-Harabasz, Davies-Bouldin, and Silhouette criteria (optimal *k* = 2) to identify the optimal clustering resolution—for details, see SI5 and S3 Fig in. The non-converging results across the different criteria suggest that the community organization of eigenvector clusters does not display a distinct optimal topological scale. We provide the course (*k* = 2) and finer scale (*k* = 4) clusters in S2A Fig and, respectively. To ensure that the image acquisition type (i.e., phase-encoding direction) or the scanning session does not affect these results, we provide statistical comparisons between the coarse-scale cluster’s stability and frequency in S4 Fig in. These results show that very similar distributions and clusters are identified regardless of phase-encoding direction or day of scans. Specifically, statistical comparisons (bootstrapping *n* = 50, 000, *p* \< 0.05) fail to find any difference between cluster’s means. To test the spatial inhomogeneity in the frequency and damping of these clustered eigenvectors, we performed a pairwise comparison between the distribution of eigenvalues corresponding to the eigenvectors in each of the clusters (bootstrap *n* = 50, 000, Bonferroni corrected *p* \< 0.05). We found significant differences in the frequencies and damping rates between all cluster pairs, except for the comparison between the frequencies in clusters 3 and 4). Together, these findings highlight the spatial heterogeneity in the frequency and damping profiles of brain oscillations. ### 2.7.2 Task-specific increases in the extra-cortical input’s power Up to now, we provided evidence that the system dynamics can capture the spatial and temporal behavior of resting-state brain networks. Next, we try to assess if the task-induced dynamics are driven by the external inputs, retrieved by the proposed method. The sensory inputs to the brain are some of the major drivers of cortical dynamics. Therefore, we hypothesize that the external inputs to the subjects’ brains, as estimated by the proposed method, will mirror real-time changes present in these task regressors (see S5 Fig in for details regarding the task regressors). To test this hypothesis, we apply our method to the fMRI activity to estimate the internal system parameters and external inputs for each subject during task performance (i.e., social, gambling, motor, working memory, language, and relational). Then, we compare the average estimated inputs’ frequency spectrum for each task. Statistical tests (Wilcoxon rank-sum test, FDR corrected, *p* \< 0.0005) reveal highly significant unique peaks, matching the expected external task-specific frequencies. Note that the distinct task-induced peaks are identified at low (\< 0.1 Hz) and high (\> 0.1 Hz) frequencies, even as high as 0.3–0.4 Hz. ### 2.7.3 Task-specific profiles of extra-cortical inputs Next, we consider an LTI framework to quantify spatial and temporal features of external inputs to the brain using HCP’s motor task dataset. The motor task comprises 3-second long visual cues, where participants are asked to either tap left or right fingers, squeeze left or right toes, or move their tongue over 12-second long periods following the visual cue’s offset. We select the motor task since the high dimensionality of input and various task conditions in this paradigm allows us to evaluate our framework’s ability to estimate external inputs’ complex spatiotemporal structure. We aim to assess if we can retrieve the external inputs that drive task-induced dynamics. We hypothesize that subjects’ estimated external inputs will mirror real-time changes present in known task regressors. Moreover, due to relatively lower levels of structured external stimulations during resting-state scans, we hypothesize that the system parameters estimated from subjects’ full-length resting-state time series will increase the accuracy of external inputs estimated from motor task datasets. demonstrates estimated inputs (input matrix *B* dimensions = 25, regularization factor = 0.5) to all brain regions (i.e., *B* × *U*) averaged across all subjects during the motor task. These results highlight the brain-wide significant task-specific changes in the estimated inputs when system parameters are estimated from the resting-state time series. We provide evidence of the robustness of these results to changes in the input matrix *B*’s dimension (S6 Fig in). Conversely, the identified inputs using the system parameters estimated from the subjects’ motor task time series notably reduces our ability to capture the task-related changes. We establish these observations’ statistical significance by comparing the external inputs estimated from task datasets against those from subjects’ resting-state scans (paired *t*−test, *p* \< 0.05, FDR corrected for multiple comparisons). Comparisons against the phase-randomized null time series also provide converging observations (S7 Fig). We also use multiple linear regression analyses to assess the estimated inputs’ similarity to the known temporal profile of the task regressors. Our results demonstrate that external inputs estimated using the full-length resting-state system parameters result in significantly (paired *t*−test, *p* \< 0.05, Bonferroni corrected for multiple comparisons) improved fit (measured by *R*² values), compared to system parameters estimated from the motor task (S8 Fig in). We also find similar results when resting-state system parameters were estimated from a short (250 sample) window that match task scans’ length (S8B Fig). Together these results highlight the importance of the modeled system’s accuracy in capturing a reliable picture of the brain’s external inputs. Next, we examine the temporal (i.e., *U* matrix) and the spatial (i.e., input matrix *B*) profiles of the external inputs (estimated using resting-state system parameters), to demonstrate how the estimated inputs reveal the dimensionality and the spatiotemporal dependencies of the task-related inputs. Prior works using univariate and multivariate analyses of HCP task datasets have demonstrated that activation induced by the hand, foot, and tongue movements can be localized over the somatomotor network. Therefore, we expect the dimensionality of the external inputs to roughly match or exceed those of task conditions (i.e., six dimensions). As mentioned in the Materials and Methods section, the principal component analysis reveals that in all HCP task conditions, principal components (PCs) 1–25 explain more than 80% of the variance in the model’s average residuals. Therefore, we choose *p* = 25 as the input matrix *B* dimension in. We performed principal component analysis on external inputs estimated temporal profiles (i.e., *U*) concatenated across all subjects to identify the input patterns similarly identified over the group. shows the temporal profile of the concatenated inputs’ PCs 1–15. As seen in, the first few PCs (≈ 9) explain a relatively larger portion of the variance. shows the high similarity between known task regressors and PCs’ temporal profiles. We quantify this similarity using subject-level multiple linear regression analysis of the estimated inputs using the known task (motor) regressors. We note apparent time lags between the known and estimated inputs. Therefore, we perform the multiple linear regression analysis using various lags. shows distributions of lags (samples) that yield the highest *R*² values for PCs 1–9. shows the group average coefficients estimated from external inputs associated with each PC (i.e., external inputs with highest PC weights). We used the group average optimal lag (based on *R*² values) identified in in. Estimated coefficients have significant values, only in PCs 1–9. These results demonstrate that the estimated inputs provide insight into the extra-cortical drivers’ dimensionality. Next, we examine spatiotemporal profiles of subject-level estimated inputs associated with these components to understand their relationship to the external stimuli. demonstrate that compared to other PCs, the inputs associated with PCs 1–4 and 6 fit task regressors relatively better, indicated by significantly (Wilcoxon rank-sum test, *p* \< 0.05, FDR corrected for multiple comparisons) higher *R*² values. reveals that PCs 1–4 and 6 are associated with the visual cue, hand and feet movements (maximum coefficient in left hand), all movements (maximum coefficient in right hand), feet movements (maximum coefficient in left foot), and tongue movements, respectively. S9 Fig in shows that the brain regions with the highest average absolute input matrix *B* values corresponding to PCs 2, 4, and 6 reveal the same regions identified in the somatomotor cortices using general linear model analysis of BOLD time series for hand, foot, and tongue movements. The input matrix *B* also captures the spatiotemporal relationship between the inputs across different conditions. For instance, S9A Fig in shows that hand or feet movements are associated with simultaneous positive and negative (e.g., inhibition or deactivation) inputs to the contra- and ipsilateral somatomotor cortices, respectively. also shows that PCs 5, 1, and 3 reveal the temporal order of inputs to visual, dorsal attention, and finally, somatomotor cortices following the onset of visual cue. Note that the spatial and temporal profile of PC 5 demonstrates the inverse relationship between inputs to visual and somatomotor cortices. This unexpected temporal profile contributes to the low similarity of PC 5 to task regressors in. We show that changing the delay between estimated inputs and task regressors changes the coefficient patterns with significant loading (S10 Fig). These results demonstrate an early positive relationship of PC 5 input with visual cue blocks, followed by a later positive (negative) relationship with left-hand movements (visual) blocks. Finally, in we demonstrate that PCs 7, 8, and 9 are primarily associated with the right foot movement blocks. However, the significantly smaller *R*² values of these PCs than other PCs in indicates the lower similarity of corresponding estimated inputs’ temporal profiles to those of task regressors. Closer examination of these inputs’ spatiotemporal profiles reveals that in addition to changes related to left-hand movements, these PCs capture the rapid sequence of inputs to frontal and somatomotor cortices following the motor task block’s offset and the baseline (i.e., no task) onset (S11 Fig). Together, these results suggest that an LTI model of cortical dynamics can reveal the unknown spatiotemporal profiles of the BOLD signal’s external task- related drivers. We provide additional analysis and discussion on model parameters and their effect on the reported results in the in document. We explored sparsity constraints on the system and input parameters in SI5. S12 Fig in demonstrates that increasing the system matrices’ sparsity reduces the model’s goodness-of- fit (measured using the AIC criterion). In the same vein, the increased spatiotemporal sparsity of the inputs overall reduces the accuracy (measured using the *R*² value of the linear regression) of the estimated inputs (S13 Fig). Nevertheless, estimated inputs’ group-level PCA reveals that the higher sparsity constraints can improve the accuracy of specific empirically identified input patterns (S14 Fig). In addition, we examined the effect of the estimation window’s size on the input’s accuracy in SI6. These results show that a smaller estimation window (3 min) provide comparable results to the full- length window, however overall it increases the accuracy of mean inputs to many brain regions (S8 Fig) and several main input patterns (S15 Fig). Finally, we explored the sensitivity of the identified input patterns to the factorizations method in the SI7. These results demonstrate that PCA decomposition of the model’s residuals reveals the analogous primary input patterns (S16 Fig) uncovered by our spatiotemporal regularization scheme. ### 2.7.4 Non-stationarity of inputs to resting-state networks So far, we showed that adopting a time-invariant model of the intrinsic relationship between large-scale brain regions allows us to extract the unknown external drivers of cortical dynamics. Our results demonstrate that the resting- state paradigm serves as a viable option for a more accurate estimation of internal system parameters. However, sensory and other extra-cortical inputs are still present during resting-state scans, resulting in system parameters and input estimation errors. Despite the estimation error in the external inputs’ profile, we hypothesize that quantifying the non-stationarity of the estimated resting-state inputs provides information on the external factors that contribute to resting-state BOLD signal non-stationarities. We quantify estimated inputs’ non-stationarity for every brain region (i.e., *B* × *U*) from the temporal fluctuations (i.e., standard deviation) of external inputs’ means, measured using a sliding window. shows brain regions that exhibit significantly high input means’ fluctuation across different sliding window sizes (see methods for details). We demonstrate the results for sliding windows of 6, 24, and 50 samples (TR = 0.72 sec) lengths and half window-length shifts. We also measure the non-stationary of external inputs during resting-state scans using the nonlinear measure developed by and find converging results. We find several brain regions within DMN consistently display high non-stationarity values. Statistical comparisons between the quantified non-stationarity of estimated inputs to identified brain regions in reveal the significantly (Welch’s *t*-test, *p* \< 0.05, Bonferroni corrected for multiple comparisons) higher non-stationarity of external inputs to identified DMN regions relative to several other resting-state networks (S17 Fig). Together, these results reveal that time-varying external inputs may partly contribute to the previously reported resting-state BOLD signal’s non-stationary, and the LTI model offers an avenue to determine the spatiotemporal profiles of these unknown external sources. # 3 Discussion Based on the theory of embodied cognition, the evolution and emergent function of the brain can be best understood in the context of the body and its interactions with the environment. In this view, the information does not exist in an abstract form outside the agent, instead, it is actively created through the agent’s physical interaction with the environment. Therefore, understanding the native structure of the external inputs to the brain, as well as the interaction between the brain and its exogenous drivers, is germane to understanding the functional dynamics of the embodied brain. What are the external drivers of BOLD signal? Current theories suggest that cortical outputs reflect changes in the balance between the strong recurrent local excitation and inhibition connectivity, rather than a feedforward integration of weak subcortical inputs. Changes in this balance heavily affects the local metabolic energy demands and consequently the regulation of cerebral blood flow and the BOLD signal, despite the net excitatory or inhibitory output of the circuits. Inhibition in principle can lead to both increases and decreases in metabolic demands. Moreover, cortical afferents and microcircuits can function as *drivers* by transmitting information about the stimuli, or alternatively as *modulators* by modulating the sensitivity and context- specificity of the response. Excitatory sensory information, transmitted mostly via glutamatergic or aspartergic drivers, combined with the strong evoked recurrent GABAergic interneurons are a major part of neurotransmission dynamics, which in turn affect the local cerebral blood flow (CBF). Likewise, regulation of cortical excitability mediated by neuromodulatory neurotransmitters including acetylcholine, norepinephrine, serotonin, and dopamine can also significantly effect CBF and the BOLD signal. What do input parameters of an LTI model capture in BOLD fMRI? We show that an LTI system acts predominantly as a high-pass filter and highlights the rapid transient fluctuations in the BOLD signal. We provide evidence that the influence of sensory inputs is identifiable in the estimated inputs to sensory cortices. More importantly, the task-related changes that are temporally decoupled from the sensory stimuli, such as the motor cortex’s activation following the offset of visual cues and onset of behavioral outputs, are also captured as external inputs to the LTI system. Prior research has reported brain-wide and heterogeneous task-related changes in the BOLD signal power spectrum and estimated system parameters. However, we provide evidence that the time-varying unknown exogenous (i.e., extra-cortical) inputs also likely contribute to non-stationarities in the cortical dynamics. Specifically, we demonstrate in silico that determining the LTI system’s parameters from periods with unknown stimuli can lead to high estimation errors in system and input parameters. We verify these observations empirically by showing that LTI system parameters identified from resting-state, instead of task BOLD time series, result in notably more accurate identification of unknown extra-cortical inputs’ spatiotemporal profiles in task scans. Our results have implications for the common interpretation of correlation-based functional connectivity changes as altered intrinsic relationships between regions. More importantly, our findings highlight the importance of modeling and interpreting the brain’s dynamic functional connectivity and non-stationarity as an open system. Can the brain during resting-state scans be fully described as a *linear* and *time-invariant* system? Prior studies demonstrate that temporal fluctuations in the BOLD signal (\< 0.1 Hz) cannot be fully attributed to linear stochastic processes, and suggest that the nonlinearities in the BOLD signal could be attributed to the presence of a strange attractor. Additionally, other neuroimaging studies using paradigms such as “temporal summation” have more directly probed the *system* and provide evidence of system nonlinearities. Model-based approaches such as work by have concluded that nonlinear transduction of rCBF to BOLD is sufficient to account for the nonlinear behaviors observed in the BOLD signal. However, care should be taken in the interpretation of these results as in the temporal summation framework, where the profile of input is assumed to be known and is approximated by an abstract stimulus representation. We believe our framework provides a novel avenue for testing the system linearities through the examination of the estimated unknown inputs in summation paradigms. Specifically, the delay between estimated and known external inputs can be further leveraged to tease out the nonlinear components of hemodynamic response function (e.g., vascular) from the neural impulse response function. Stationary signals are characterized by time-invariant statistical properties, such as mean and variance. To date, several tests have been proposed to examine the non-stationarity of BOLD time series and the presence of dynamic functional connectivity, including test statistics based on the variance of the FC time series, the FC time series’ Fourier transform, multivariate kurtosis of time series, non-linear test statistics, and wavelet-based methods, among others. These methods commonly compare measured properties between the time series of empirical data and a suitable surrogate or null time series that is designed to lack time-varying properties through non-parametric resampling, phase- randomization, or generative models, and the choice of measured properties and null models profoundly impact on the outcomes of stationarity tests in conflicting reports on BOLD signal. Notably, the presence of non-stationarity in the outputs does not directly imply the underlying system’s non-stationarity. An LTI system’s outputs, for instance, while receiving non-stationary external inputs, can also display time-varying properties. As mentioned earlier, using internal system parameters of an LTI system estimated over resting-state scans enables more accurate identification of exogenous inputs’ spatiotemporal profile task scans. These results suggest that a large-scale stationarity model of the brain with time-varying external inputs can, in theory, account for a large portion of the observed task-related changes in cortical dynamics. It is worth noting that any possible task-related changes in the underlying system parameters are also captured as external inputs in an LTI framework. Therefore, from the system identification and model-fitting perspective, it is likely that a linear switching system with higher degrees of freedom would improve the fit. Beyond the goodness-of-fit of the model, care should be taken in interpreting the epiphenomenal large-scale models’ parameters and their changes at the micro-scale biophysical level. However, the impetus for this work is to highlight the estimates’ notable sensitivity to the unknown, and thus, unaccounted external inputs. More practically, when simulated with a wideband unknown external inputs, our results suggest that an open LTI model estimated during resting-state allows us to uncover the influence of these unknown drivers of BOLD dynamics. Nevertheless, participants’ cortices receive external stimulation even during resting-state scans, contributing to estimation inaccuracy and the system’s outputs’ non- stationarity. In this work, we aim to disentangle the non-stationarity of the *system* from its *outputs* over resting-state by examining estimated inputs’ non-stationarity. Our results show that external inputs’ non-stationarity over resting-state scans are spatially inhomogeneous, with identified DMN regions showing the highest levels consistently across different analyses. These observations are in line with prior reports of higher dynamic functional connectivity of these brain structures over rest. The identified non-stationary inputs during resting-state scans also imply that we should expect more error in the estimated spectral profile of the aforementioned regions. Therefore, future work should explore leveraging other states of consciousness, such as sleep with lower global cortical activity, to address this limitation. Despite the presence of possible confounding factors such as unaccounted nonlinearities and non- stationarities in the recording noise, our framework and observations provide new insight into the external drivers of cortical dynamics and factors that contribute to their non-stationarity. Recent system-identification and control- theoretic work have also demonstrated the utility of a stationary system in explaining BOLD dynamics. Together these findings pave the way for principled model-based control of pathological brain dynamics, such as depression and schizophrenia, using open-loop external or closed-loop neurofeedback stimulation. Historically, a narrow band of slow frequencies between 0.01 to 0.1 Hz was thought to contain information relevant to underlying neural activity, and that the higher frequency (\> 0.1 Hz) BOLD activity considered mainly as an artifact. Our results also demonstrate that the primary oscillatory modes of the LTI model of the resting-state BOLD display similar slow frequencies heterogeneously over the brain. In addition, the hemodynamic response function (HRF) is also expected to dampen the higher frequency neural activity significantly. More recent evidence, however, portrays a broadband picture of BOLD signal fluctuations with frequencies up to 0.25 Hz and even higher. We also provide converging evidence that despite the expected low-pass filtering of HRF, information about the stimulus-related activity can still be extracted from the BOLD signal even as high as ≈ 0.4 Hz. Future work can leverage acquisition protocol with higher sampling rates than HCP and rapid stimuli capable of inducing brain-wide activations to accurately delineate the inputs’ attenuation profile by HRF at higher frequencies. In line with previous reports of intrinsic functional connectivity networks, our clustering analysis reveals the low dimensionality of the system eigenmodes as the eigenvectors can be roughly grouped in a small number of spatial patterns. However, we show that depending on the task, the dimensionality of inputs can be high; for example, in the motor task with multiple conditions, we identified task-specific inputs to different ROIs across motor cortices. Future work should use our proposed framework to identify the highest bound of input dimensionality using higher resolution parcellations or voxel-wise modeling of the BOLD signal. However, the HRF plays another critical role in biophysical models where it enables the approximation of the latent neural states from the BOLD signal. This is one of the main limitations of our simplified model, as it incorrectly assumes that the BOLD signal in one region (instead of the underlying neural activity) can cause changes in the BOLD signal in the connected regions. This assumption for spatially inhomogeneous HRF functions can, in theory, lead to incorrect identification of the external inputs’ focus and error in the direction and speed of the interactions within functional networks. We believe the overlapping patterns of inputs and the task activation maps identified using the conventional univariate general linear model analysis suggest that the above-mentioned error is likely tolerable. To improve the estimated unknown inputs’ accuracy, future work should leverage the formulated quantitative spatiotemporal models, or the more recent models informed by the precise mechanisms of neurovascular coupling. Nevertheless, care should be taken in these or other related deconvolution-based inferences, since as mentioned earlier, they rely on the assumption of a known profile of HRF or inputs. Future work can also leverage neural adaptation paradigms to influence the neural response timing and help tease out the neural and vascular components’ contributions to the modeled inputs. Comparing our identified inputs with those extracted from other neuroimaging modalities such as Magnetoencephalography (MEG) that are more direct measurements of the underlying neural activity will also us to further decouple the aforementioned mechanisms. Structured recording noise such as autocorrelated noise can negatively impact the modeled system, and the estimated input. Although we have included global mean signal regression (GSR) as a preprocessing step to account for the shared global noise that is present in many of the functional networks, our model is unable to account for other unknown structured (e.g., autocorrelated) and time- varying recording noise. Moreover, GSR may also introduce artifact, as in addition to the shared noise, it also removes any global activation patterns (e.g., vigilance or arousal) and can alter the correlation structure. These limitations are the source of ongoing controversy around this noise reduction method. Having weighed the potential drawbacks of GSR against the major concerns regarding the significant global artifacts such as the cardiac and respiratory noise, we adopted this preprocessing step. Nevertheless, it would be beneficial to investigate the spectral profile of the global signal and the impact of GSR on the estimated system and inputs’ spectral characteristics. One of the current limitations of our proposed framework is that the estimated inputs’ accuracy depends on the internal and recording noise levels. We show that group-level analysis and repeated measurement designs are effective strategies to increase signal-to-recording noise and to increase the estimated inputs’ accuracy. In addition, although we can not accurately tease out the contributions of internal noise from other sources of noise, our simulations and experimental results suggest lower levels of internal noise relative to external drivers in task fMRI. We draw this conclusion based on the relatively large input estimation errors associated with system parameters identified during external stimulation. We used individual subjects’ resting-state datasets to identify the system parameters for uncovering the unknown inputs from the BOLD signal. Although beyond the scope of our current work, it is critical to comprehensively examine the identifiability of the estimated system parameters across different scanning sessions, types, and individuals. Subsequently, it would be interesting to explore further the extent to which our proposed system identification framework can highlight the shared features across subjects or increase the accuracy of the subject-specific mapping of spatiotemporal dynamics. Finally, it is worth highlighting that model-based *data-driven* methods such as our proposed framework and the *hypothesis-driven* methods such as DCM are complementary approaches, suited for interrogation of different aspects of system and output dynamics. For instance, DCM can also be leveraged fruitfully for a more accurate estimation of the system, and consequently, external input parameters using highly controlled experimental designs with known external input profiles. Though, as mentioned before, care should be taken in the interpretation of the results produced by methods that incorporate priors, as the boxcar regressors commonly used to model the profile of external inputs are merely abstractions and do not account for other possible factors such as anticipatory responses, adaptation, or other unknown drivers that shape the profile of external inputs. However, data-driven approaches are particularly advantageous when the brain is driven by extensive complex inputs, for instance, during naturalistic stimuli (e.g., watching a movie), or in general, if we lack a priori information or hypothesis on the structure of external inputs—for instance, during the healthy resting-state or pathological brain activity such as epileptic discharges. # 4 Conclusion We show that the proposed framework provides an avenue to uncover the structure of the unknown drivers of BOLD signal fluctuations and shines light on factors that contribute to its apparent non-stationarities. However, more significantly, our results highlight the importance of modeling and interpreting the brain’s dynamic functional connectivity as an *open* system. Broadly, our approach provides a framework for understanding the brain’s large-scale functional dynamics and non-stationarities, mechanistically via the modeled system and its time-varying drivers. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction The prevalence of mouth breathing among children remains controversial but is at most reported to be 50–56%. Mouth breathing is defined as using the mouth alone or the mouth and nose instead of the nose alone for respiration for longer than 6 months. Mouth breathing is thought to be caused by mechanical factors such as septal deviation and adenotonsillar hyperplasia, inflammatory diseases such as allergic rhinitis, congenital malformation, and behavioral mouth breathing \[, \]. The functions of the nasal cavity are air-conditioning, olfaction, and defense, but mouth breathing causes environmental air to bypass these nasal functions, allowing air to directly enter the lower respiratory tract, which can cause airway hyperreactivity and chronic bronchial inflammation. Two case–control studies showed that children suffering from asthma exhibit more mouth breathing behaviors than controls. Meanwhile, a cohort study revealed that the risk of otitis media with effusion is 2.4 times higher in mouth breathers than nose breathers. In addition, mouth breathing might be associated with skin diseases, given its previously demonstrated relationships with periodontal disease and enlarged tonsils. Periodontal disease is associated with chronic skin diseases such as chronic urticaria, chronic pigmented purpura, and chronic nodular prurigo. Mouth breathers had an increased risk of gingivitis in a case-control study, while patients suffering from adenotonsillar hypertrophy, which is a cause of mouth breathing, have an increased risk of periodontal disease, which improves after adenoidectomy. In addition, Valera et al. reported that children aged 3–6 years with enlarged tonsils had a significantly increased risk of mouth breathing. Streptococcal tonsillitis is associated with psoriasis, while some reports indicate tonsillectomy improves psoriasis. However, no population-based studies have investigated the relationship between mouth breathing and the prevalences of pediatric diseases, including atopic dermatitis, which is a highly prevalent skin disease in children. Accordingly, this study investigated the relationship of mouth breathing with the prevalences of various diseases including atopic dermatitis by using a questionnaire targeting preschool children in day nurseries. # Materials and Methods ## Design, setting, and participants Aimed to have a total of 600 respondents, we targeted preschool children aged 2–6 years who attended day nurseries in Tokushima City. The questionnaire was distributed at 13 randomly selected day nurseries in Tokushima City. We distributed anonymous questionnaires to the parents or guardians from November 27 to December 16, 2013. The questionnaires were submitted through collection boxes in each day nursery. This was a population-based cross-sectional study performed with the permission of the Ethics Committee of the Tokushima University Hospital. After obtaining written informed consent from the head of each day nursery, we distributed the questionnaires to parents/guardians attached with an explanatory leaflet specifying that their submission of the questionnaire was considered consent. Within the leaflet, we also informed the parents/guardians that some, as yet unknown, behavioral habits might contribute to the development of some diseases; however, we did not provide information regarding the specific potential associations. ## Questionnaire The questionnaire included questions on the following: age, sex, smoking habits of family members, behavioral habits, present and previous diseases, and present and previous diseases of parents. A multiple choice question was used to collect information about present diseases, including allergic rhinitis, chronic sinusitis, asthma, chronic bronchitis, pollinosis, atopic dermatitis, tonsillitis, otitis media, chronic headache, proteinuria, and hematuria. In addition to the present diseases, previous diseases included acute sinusitis, acute otitis media, chronic otitis media, pneumonia, and meningitis. As some cases of allergic rhinitis are pollinosis, we treated pollinosis as allergic rhinitis. Excluding mouth breathing, which is described in the next section, we collected information about the following behavioral habits: regular bedtime and rising time, sleeping hours, sleeping posture, pacifier use, and dietary habits including mastication and food and drink preferences. ## Mouth breathing criteria Although there is no widely adopted questionnaire for evaluating mouth breathing, we prepared the following 3 items for detecting mouth breathers in daytime (MBD): “breathes with mouth ordinarily,” “mouth is open ordinarily,” and “mouth is open when chewing.” These items were developed with reference to generally used methods including the Glatzel mirror, lip closure, and the water test. “Breathes with mouth ordinarily” corresponds to the Glatzel mirror, which judges mouth breathing by vapor emanating from the mouth using a mirror placed below the child’s nose. “Mouth is open ordinarily” corresponds to lip closure, which is determined according to soft contact between the upper and lower lips. “Mouth is open when chewing” corresponds to the water test, in which children hold some water in their mouth while keeping their lips closed without swallowing for 3 minutes. The necessity of opening the mouth during mastication suggests the child is not in the habit of complete nose breathing. Children who met 0–1 and 2–3 of the above criteria were considered nasal breathers in daytime (NBD) and MBD, respectively. We adopted the following 3 items to detect mouth breathers during sleep (MBS): “snoring,” “mouth is open during sleep,” and “mouth is dry when your child gets up.” Snoring is significantly associated with MBD and MBS. Dry mouth is caused by dry air passing through the mouth unless a loss of saliva occurs. Children who met 0–1 and 2–3 criteria were considered nasal breathers during sleep (NBS) and MBS, respectively. For each item, we defined positive or negative choices for mouth breathing before distributing the questionnaire. As shown in, the underlined and non- underlined choices were defined as positive and negative for mouth breathing, respectively. We did not indicate in the questionnaire which questions were used to analyze mouth breathing or which choices were used to determine mouth breathing. Complete mouth breathers (CMB) were defined as children who met the criteria for both MBD and MBS, partial mouth breathers (PMB) met the criteria for either MBD or MBS, and complete nasal breathers (CNB) met the criteria for both NBD and NBS. Because nasal congestion, which is the main symptom of rhinitis and sinusitis, is an important cause of mouth breathing, we included an item for “nasal congestion”; it was assessed as negative only if “not blocked unless cold” was selected and assessed as positive if “often blocked” or “always blocked” was selected. Genetic factors were considered positive if the parents selected the corresponding disease from among present and/or previous diseases, because atopic dermatitis and asthma improve naturally with age and allergic rhinitis is classified as intermittent, seasonal, or persistent according to the nature of the allergen. Smoking was categorized according to whether family member(s) smoked around the children or not. ## Statistical analysis Statistical analyses were performed by using SPSS version 21 (IBM Corp, Armonk, NY, USA). Continuous variables are presented as mean ± standard deviation (SD) or median and interquartile range (IQR). For univariate analysis, categorical variables such as habits and disease presence were analyzed by Pearson’s *χ*² tests or Fisher’s exact tests where appropriate. The results are presented as odds ratios (ORs) and 95% confidence intervals (CIs). To adjust the ORs of MBD and MBS with respect to disease prevalence, the Mantel–Haenszel test was performed using variables with *p* \< 0.25 in univariate analysis as potential confounders. For multivariate analysis, forward stepwise multiple logistic regression was performed by using breathing pattern (i.e., MBD or MBS) as the dependent variable and categorical variables showing *p* \< 0.25 in univariate analysis as independent variables. Bonferroni corrections for multiple comparisons were made as a *post hoc* analysis. The level of significance was set at *p* \< 0.05 except in cases of multiple comparisons, in which *p* \< 0.016 was used. We used the phi coefficient to indicate the effect size in the Pearson’s *χ*² test. For statistical power, we performed *post hoc* power analysis using G\*power (ver. 3.1.9.2, Erdfelder, Faul and Buchuner, Germany). # Results We distributed questionnaires to 1036 subjects from November 27 to December 16, 2013 and collected 552 responses for a response rate of 53.3%. No reminder letters were distributed. Thirteen responses were excluded because of age: 11 were outside the target age range and 2 had no age given. A further 71 responses were excluded because of at least one unanswered question. Therefore, a total of 468 valid responses were collected for a response rate of 45.2%. The subjects’ characteristics are shown in. Mean age was 4.5 ± 1.2 years with a median of 4.5 years (IQR: 3.4–5.5 years). The subjects were divided into 3 groups according to the main diseases reported: atopic dermatitis, asthma, and allergic rhinitis. Children with atopic dermatitis were significantly associated with a history of asthma and/or allergic rhinitis. In contrast, asthmatic children were significantly associated with a history of allergic rhinitis, atopic dermatitis, and/or pneumonia. The most afflicted children were those with allergic rhinitis, who had a history of all documented conditions except tonsillitis. Each group had a family history of their condition, and the atopic dermatitis group was also significantly associated with a family history of allergic rhinitis. The results show that 57.5% of the children had substantial nasal breathing difficulties, including CMB (23.9%) and PMB (33.5%). The symptoms appeared to be more frequent at night; 45.9% were MBS compared to 35.5% who were MBD. Atopic dermatitis was significantly associated with both MBD and MBS (*p* = 0.001 and *p* = 0.002, respectively). Asthma was significantly associated with only MBD (*p* = 0.013), while allergic rhinitis was significantly associated with both MBD and MBS (*p* = 0.035 and 0.006, respectively). Nasal congestion was significantly associated with a risk of all 3 diseases, which was considered a confounder. Also, nasal congestion, which was present with \[4/97 (4.1%)\] and without \[1/371 (0.3%)\] chronic sinusitis, was associated with chronic sinusitis (OR: 15.9, 95% CI: 1.8–144.1, *p* = 0.007, Fisher’s exact test). According to the results of the univariate analysis, we adopted previous disease (i.e., asthma and allergic rhinitis), family history of disease (i.e., atopic dermatitis, asthma, and allergic rhinitis), and nasal congestion as confounding factors for atopic dermatitis. Meanwhile, previous disease (i.e., atopic dermatitis, allergic rhinitis, and pneumonia), family history of disease (i.e., asthma), and nasal congestion were adopted as confounding factors for asthma. Because nasal congestion can induce mouth breathing and is one of the main symptoms of allergic rhinitis, we excluded allergic rhinitis from subsequent analyses. After adjusting for confounders, atopic dermatitis was significantly associated with risks of both MBD (OR: 2.6, 95% CI: 1.3–5.4, *p* = 0.010) and MBS (OR: 4.1, 95% CI: 1.8–9.2, *p* = 0.001), although asthma was not significantly associated with a risk of MBD (OR: 1.3, 95% CI: 0.7–2.7, *p* = 0.508). Multiple logistic regression for atopic dermatitis was subsequently performed by including the categorical variables listed above. The ORs of MBD and MBS for atopic dermatitis were 2.2 (95% CI: 1.2–4.2) and 2.7 (95% CI: 1.4–5.3), respectively; these values were lower than those for family history of atopic dermatitis and history of asthma and allergic rhinitis. Nasal congestion and parental history of asthma and allergic rhinitis were not significant risk factors. The effect sizes of MBD and MBS for atopic dermatitis were 0.149 and 0.141, respectively, and 0.897 and 0.862, respectively, represented the statistical power in the *post hoc* analysis, indicating sufficient power. Atopic dermatitis was present in 7.0% of children with CNB (n = 199), 12.7% of children with PMB (n = 157), and 22.3% of children with CMB (n = 112) (*p* \< 0.001). The prevalence of atopic dermatitis in CMBs was higher than that in PMBs (*p* = 0.038), but the difference was not significant after Bonferroni correction. # Discussion The results of the present study indicate mouth breathing is significantly associated with atopic dermatitis and asthma. In particular, atopic dermatitis was significantly associated with both MBD and MBS after adjusting for confounding factors. Moreover, the prevalences of these diseases showed an increasing trend with an increasing extent of mouth breathing. To our knowledge, this is the first study to evaluate the impact of breathing patterns on the prevalence of atopic dermatitis. Although two studies report that mouth- breathing children with positive skin-prick test results have higher prevalences of asthma and sleep apnea than those with negative results, they did not compare mouth breathers with nasal breathers. No previous study has analyzed the prevalence of mouth breathing among Japanese children. In the present study, the prevalences of MBD and MBS were 35.5% and 45.9%, respectively; however, these results were based on a questionnaire that has not been assessed for validity. The studies from other countries report a wide range of the prevalence of mouth breathing, 4–56%. Brazil has the highest prevalence, which exceeds 50% based on clinical assessment. On the other hand, the lowest prevalence is in India, at 4–7% based on clinical assessment. The prevalences of mouth breathing in England and New Zealand are 23% and 19%, respectively. However, direct comparison among studies is difficult owing to varying criteria. Standardized criteria for mouth breathing using clinical assessment and questionnaires are required to more precisely investigate differences in mouth breathing prevalence. Atopic dermatitis is a chronic skin disorder characterized by pruritus and inflammation that mostly develops during childhood and is strongly associated with the allergic history of patients and relatives. Filaggrin, which is encoded by the *FLG* gene, is a crucial protein for skin barrier function. Two meta- analyses show that children with an abnormal *FLG* gene are 3.12–4.78 times more likely to have atopic dermatitis than normal subjects. On the other hand, the prevalences of chronic pediatric diseases, including atopic dermatitis as well as asthma and allergic rhinitis, vary greatly worldwide. The prevalences of these diseases have increased in the late 20^th century, although this trend has not held true in the last 10 years. These findings suggest that, in addition to genetic factors, environmental factors play important roles in these diseases. This was a population-based cross-sectional study; therefore, causal relationships cannot be determined. However, if mouth breathing is shown to contribute to atopic dermatitis in future cohort studies, guidance to avoid mouth breathing should be provided to children and parents/guardians to prevent atopic dermatitis. Furthermore, otolaryngologist help should be considered, as necessary, if children find it difficult to quit mouth breathing. It is possible that periodontal disease and/or tonsillitis might mediate the mechanism underlying any association between mouth breathing, as an environmental factor, and atopic dermatitis. Satoh et al. suggested that immune reactions mediated by bacterial–immune complexes, superantigens, or toll-like receptors might induce skin diseases. However, the present questionnaire did not collect periodontal information. Furthermore, although the present results showed no significant relationship between history of tonsillitis and the prevalence of atopic dermatitis, inquiring about the number of repeated tonsillitis episodes might have revealed an association. On the other hand, it is possible that children could have mild nasal congestion unnoticed by their guardians. Thus, case–control studies with otolaryngological diagnostic tests are required to confirm the relationship between mouth breathing and atopic dermatitis. Furthermore, sleep disturbance due to the intense pruritus of atopic dermatitis can cause daytime sleepiness, which might in turn cause behavioral mouth breathing. According to the Japanese Ministry of Health, Labour, and Welfare, the prevalences of atopic dermatitis in 3- and 6–7-year-old children are 13.2% and 11.8%, respectively. Meanwhile, the prevalence of asthma in a study of 34,699 children aged 4–5 years at randomly selected nurseries in Japan was 11.2%, and the lifetime prevalences of atopic dermatitis, asthma, and allergic rhinitis were 16.0%, 16.1%, and 17.6%, respectively. In the present study, the prevalences of atopic dermatitis, asthma, and allergic rhinitis were 12.6%, 9.8%, and 13.0%, respectively; the lifetime prevalences were 15.4%, 14.5%, and 18.2%, respectively. Thus, the present results are similar to those of previous studies, suggesting the subject group is representative of Japanese children aged 2–6 years. In this study, 45.9% of children had family members who smoked. This is similar to the prevalence of smokers among Japanese men and women 20–40 years old: 40% and 12%, respectively. However, having a family member who smoked was not associated with disease prevalence. Possible reasons for this are the low rate of smokers who smoke around their children (7.1%) and parents who stop smoking when their child develops a disease. There are some limitations in this study, as already mentioned. First, the diagnoses of atopic dermatitis and mouth breathing were dependent on the questionnaire results. Because we created the questions and criteria for mouth breathing, the lack of experimental data and questionnaire validity limits the strength of our findings. Second, the valid response rate was comparatively low at 45.2%. Third, as this was a cross-sectional study, causal relationships cannot be determined. Although we cannot exclude facial injuries affecting nasal breathing, such cases would be rare and are unlikely to affect the results. The effects of common cold and flu appear quite small, because we inquired about the normal state of the children in the questionnaire. Finally, regarding genetic factors, only the parents’ information was collected and not that of sibling or grandparents. Therefore, genetic factors may have been underestimated. # Conclusion Mouth breathing is significantly associated with atopic dermatitis in Japanese preschool children aged 2–6 years. Additional case–control and cohort studies are required to confirm this relationship. Furthermore, studies targeting school children and adults would also help clarify this association. # Supporting Information We thank the heads and staff members of the day nurseries for their cooperation. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: HY KT. Performed the experiments: HY ST YN SK RT SY NS MK. Analyzed the data: HY SK RT SY NS MK KT AT. Contributed reagents/materials/analysis tools: KT. Wrote the paper: HY TS KT.

# Introduction Theoretical work in social psychology frequently seeks to integrate phenomena that are otherwise studied separately. One recent effort in this regard, the Balanced Identity framework proposed by Greenwald and colleagues, builds on classic theories of cognitive consistency, by arguing that constructs such as attitudes, identification, and self-esteem reflect a coordinated set of associative relationships in semantic space. The framework makes formal predictions regarding how such constructs should relate, making it straightforward to test and extend in new directions. In particular, it provides a means of testing the ideas that intergroup bias emerges as a natural outgrowth of self-related positivity interacting with group identification. More specifically, Balanced Identity entails the claim that ingroup preference must be understood in the context of several interrelated cognitive constructs, namely self-esteem and ingroup identification. These three constructs are conceptualized as occupying a triangular constellation of mutual associative influence (depicted). This cognitive structure allows a specific set of predictions to be made regarding the expected relationships among the constructs: any one construct can be predicted from the product of the other two. That is, one can predict the strength of a *group attitude* from the product of *self-esteem* and *identification* with that group, *self-esteem* from the interactive effect of *identification* and *attitude*, and so on. Because the model purports to describe patterns based on an associative memory system, it is suggested that these relationships will be most prevalent when constructs are themselves measured at the implicit or associative level, and will not necessarily apply to more explicitly held, propositional forms of information. Recent meta-analytic evidence provides strong support for the presence of these relationships across multiple attitude and stereotype domains. So far, the Balance framework has been used to examine attitudes and stereotypes towards a range of groups, but primarily ones that are enduring and central to mature identity, such as gender and race/ethnicity. Capturing attitudes and identifications in these domains is a crucial test of model fitness, but because individuals have usually been members of these groups for a long time, we know very little about how the consistency processes postulated by Balanced Identity initially form. Is consistency the result of a protracted process of dissonance reduction in which inconsistencies are gradually reduced as related concepts are repeatedly co-activated? Alternatively, when a new attitude object (such as a social group) is encountered and evaluated, is the form of that evaluation immediately constrained by the strength and direction of existing cognitions (such as self-esteem), such that the new cognition is, from its genesis, in balance with its neighbors? Long-standing social affiliations cannot be used to address this question, because consistency observed long after their acquisition could be the result of either process. At present, the shortest time scale that has been investigated is about a week, in the specific case of attitudes towards residential colleges at a university. The current research seeks to explore this problem space by asking whether attitudes and identifications learned just minutes before, within the context of the experimental setting, conform to the predictions of the Balanced Identity model. Participants are assigned to previously unfamiliar ‘minimal’ social groups, and self-esteem, (minimal) group attitude, and (minimal) group identification are assessed. If these cognitions conform to the predictions of Balanced Identity, it would favor the possibility that cognitive consistency emerges immediately, without requiring a gradual period of dissonance reduction. At a broader level, this inquiry can be considered a test of the generality of the Balance framework: Does it appear only with respect to highly familiar, personally important and culturally salient social groups, or can we observe it from the earliest moments of social affiliation, even with previously unfamiliar and minimally meaningful social groups? The present work also addresses a second question regarding how the Balance Framework should be interpreted. All current investigations have been correlational designs, demonstrating that the posited pattern of relationships does exist with respect to real-world social groups. But the theory presumes to go beyond description; it posits that, for a given association, surrounding associations are, at least in part, *causally constitutive* of it. Testing this causal assumption requires moving beyond the correlational framework and directly manipulating constructs within the Balance framework to see if that manipulation affects neighboring constructs in the predicted directions. Thus, in addition to testing whether cognitive balance appears with newly encountered “minimal” social affiliations, in a follow-up series of experiments, each of the three “legs” of the Balance triangle is manipulated independently in either the positive or negative direction, and the effect of that manipulation on related constructs is measured. This design allows us to test two corollary questions. First, does cognitive balance survive the direct manipulation of one of the cognitive constructs? An affirmative answer would again suggest that the balance relationships are rapidly emergent. Second and relatedly, several studies have suggested that balance does not emerge when the ingroup in question is less positively evaluated overall, for example when the ingroup is stigmatized, lower status, or more ambivalently viewed with respect to valence. That raises the possibility that balance will be particularly disrupted when we manipulate constructs in the negative direction, e.g. by artificially reducing ingroup preference. Observing this in the present context would help to establish the generality of that previously observed phenomenon. The Minimal Groups Paradigm (hereafter MGP) has established that participants consistently show preferences for previously unfamiliar, randomly assigned ingroups, and that these preferences occur across substantial methodological variation and on both self-report and implicit measures. Because prior knowledge is controlled for through the novelty of the grouping dimension, the psychological consequences of “mere membership” can be directly assessed, providing a window into the generalized cognitive processes that underlie responses to group boundaries. This strategy has already proven valuable, showing, for example, that the tendency to show better recall for ingroup faces, to associate anger with outgroups, and to show signature neural responses to outgroup faces are all general intergroup responses that emerge in similar fashion for highly familiar groups as well as minimal groups and that appear early in development, suggesting that it is a reflection of core intergroup processes. In the current context, the question is whether the patterns of relationships described by Balanced Identity emerge outside the context of well- established and socially meaningful social collectives, i.e. in the minimal groups setting. ## Overview of the present research In Experiment 1, participants are randomly assigned to a minimal social group, implicit and explicit group attitudes, group identification, and self-esteem are assessed, and the predictions of Balanced Identity regarding the relationships between the three constructs are tested. I predicted that cognitive balance would be present, supporting the possibility that Balanced Identity can serve as a general explanatory theory of intergroup bias, and that the relationships it describes emerge immediately and automatically in newly formed cognitions. In Experiments 2–4, each of the three “legs” of the balance triad are manipulated prior to measurement. That is, each study experimentally manipulates one of group identification, group attitude, or self-esteem in either a positive (increasing strength) or negative (decreasing strength) direction. Manipulating all three constructs provides for an examination of the possibility that some of the links are more or less susceptible to manipulation than others. Most notably, at least within a minimal groups setting, self-esteem has a different status than identification with or attitudes towards a minimal group in that it is precedes the experimental setting. Indeed, positive implicit self-esteem is enduring, strong, and broadly present across persons and cultures. By contrast, within the minimal groups setting, identification and attitude are emergent phenomena not supported by a long history of activation and/or reinforcement. This might make self-esteem more resistant to influence via manipulations of the others constructs, or it could make self-esteem more difficult to manipulate more generally and thus a less reliable inroad into affecting the other elements of the balance constellation. Thus, manipulating the identification with a social ingroup, for example, might be more likely to affect attitudes towards that group than self-esteem itself. Across these three experiments, I predicted that manipulations of one component of balance would affect the others, providing support for a causal reading of Balanced Identity. # Experiment 1: Balanced Identity in the Minimal Group Paradigm ## Materials and Methods ### Participants All research reported in this paper was approved by the University of California, Merced Institutional Review Board. Ninety-seven participants, recruited from a university research study pool composed of undergraduate students participating for credit towards a course requirement in one of several psychology courses, participated in Experiment 1. Participants were highly diverse in terms of self-reported race/ethnicity (Asian = 40%, Hispanic = 33%, White = 19%, Black = 7%). Computer failure led to the elimination of data from seven participants prior to any analyses. ### Procedure Participants were greeted by a research assistant who secured written, informed consent and then escorted them to a private lab room. The participants were seated alone at a personal computer. The entire procedure took approximately 20 minutes, at which point they were debriefed and released. ### Minimal group induction procedure The experiment always began with the minimal group induction procedure, modeled after the recommendations provided by Pinter and Greenwald, who directly compared several minimal groups induction procedures. Participants read a short paragraph indicating that the study involved two groups (the “Copley group” and the “Dawson group”), one of which they belonged to. They would next be introduced to the names of the members of their ingroup, which they had to learn in order to complete the later portions of the experiment. Participants then observed a static display listing the names of six members of their ingroup for 45 seconds. Following that, participants completed a name assignment task in which names appeared in the middle of the screen and had to be assigned to either the Copley or Dawson group by pressing one of two keys (the ‘e’ or the ‘i’ key), which were indicated on the screen next to the appropriate group label. For the initial 40 trials, participants were provided with a color cue indicating the group membership of each name, making categorization straightforward, but in a subsequent 30 trials, the color cue was removed so that participants would be encouraged to remember the names. Error feedback was presented when a name was miscategorized, and participants had to correctly categorize the name before continuing to the next trial. This procedure produced acceptable accuracy at group categorization (mean accuracy across the last 30 trials in all studies was greater than 93%). While participants always learned ingroup names first, the ingroup/outgroup name and the pairing of names to groups were counterbalanced across participants. This completed the minimal group induction procedure. ### Measures The primary dependent measures were three Implicit Association Tests. The IAT is a dual categorization task in which, in the critical blocks, participants alternate between two categorization tasks involving four total target categories, but respond using only two response keys. The logic of the task is that if cognitively associated categories share a response key, responses will be facilitated, resulting in faster responding and less errors. For example, when categorizing positive words and words relating to the self with one key, and negative words and words relating to others with the other key, responding will generally be faster because of the cognitive congruence between positive words and a positively evaluated self. When the opposite pairings needs to be made (i.e., self paired with negative words, other paired with positive words), responding will generally slow down and the error rate will generally increase. The IAT is now the most well-validated measure of implicit attitude. The three IATs employed here were a group attitude IAT (contrasting Copley names and Dawson names as well as positive and negative adjectives), a group identification IAT (contrasting Copley and Dawson names as well as words relating to self and other, such as “me”, “my”, “I” versus “they”, “their”, and “them”), and a self-esteem IAT (contrasting self and other words as well as positive and negative adjectives). Because the strength of observed associations might depend on the temporal relationship between the IATs and the induction procedure, the two group-relevant attributes (attitude and identification) were measured prior to self-esteem. Thus, participants completed the attitude and identification IATs first (with task order counter-balanced across participants) followed by the self-esteem IAT. After the implicit measures, participants completed a short battery of self-report measures corresponding to the same three constructs of group attitude, group identification, and self-esteem. Following Pinter & Greenwald, these questions involved reporting on their liking for each of the two groups (e.g., “I like the Copley group”), their identification with each group (e.g. “I identify with the Dawson group”), and their liking for the self versus others (e.g. “I like myself”). To facilitate comparison with the IAT, difference scores were produced such that positive numbers indicate greater liking for and identification with the ingroup and greater liking for the self. The order of these measures was matched to the order of the IATs described above. ## Results All data reported in these experiments are available online via the Open Science Framework, hosted at <https://openscienceframework.org/project/Vvo9A>. ### Descriptive statistics Following prior work with the IAT, response latencies greater than 10,000 milliseconds (ms) or less than 400 ms were dropped, and participants with excessive extremely fast trials (\>10% of trials \<300 ms) were dropped entirely; such participants are generally rapidly pressing buttons without responding to task instructions. These criteria led to the exclusion of four participants from the attitude and identity IATs and eight participants from the self-esteem IAT. Response time data were then used to calculate an effect size, the IAT *D*, reflecting the relative speed advantage in one condition. *D* was used in all analyses reported below, but for ease of interpretation we also include Cohen's *d* of the simple effects when reporting descriptive statistics. There were no main effects of group (Copley or Dawson), name-group pairing, or explicit group assignment on any dependent measures, so these factors were dropped from preliminary analyses. Beginning with the implicit measures, preference for the ingroup was robust, *D* = .29 (.36), *t*(85)  = 7.37, *p*\<.001, *d* = .83, as was identification with the ingroup, *D* = .47 (.41), *t*(85)  = 10.50, *p*\<.001, *d* = .93. Implicit self-esteem was strong and positive, *D* = .53 (.28), *t*(81)  = 17.23, *p*\<.001, *d* = 1.18. Consistent with the notion that the MGP is essentially a manipulation of ingroup identification, identification as measured by the IAT was stronger than attitude as measured by the IAT, paired *t*(81)  = 3.56, *p*\<.001. Preference for and identification with the ingroup were modestly correlated, *r*(80)  = .35, *p* = .001, but self-esteem correlated with neither identification, *r*(78)  = −.15, *p* = .18, nor preference, *r*(79)  = .05, *p* = .65. Evidence for an effect of group membership was also present at the self-report level. Participants expressed more liking for the ingroup, *M* = .88 (1.7), *t*(89)  = 4.92, *p*\<.001, *d* = .52, as well as greater identification with the ingroup, *M*  = 1.56 (2.1), *t*(89)  = 6.95, *p*\<.001, *d* = .74 Participants also reported somewhat greater liking for the self as compared with others, *M* = .28 (1.2), *t*(89)  = 2.17, *p* = .03, *d* = .23. Correlations among self-report measures were consistently present, all *r*(88)\>.28, *p*\<.01. Strongest was the correlation between attitude and identification, *r*(88)  = .66, *p*\<.001. Implicit and explicit measures did not correlate, all \|*r*\|\<.18, *p*\>.12. ### Balanced Identity Analyses Following the original formulation of Balanced Identity, evidence of balance is assessed for each of the three possible regression models created by predicting each construct from the other two. This involves first modeling each criterion's association strength from the product of the other two, and then in a second step entering the two predictors as main effects. The prediction is that the addition of these two main effect terms will not increase the predictive power of the model because the relationship is wholly accounted for by the product. This produces four tests that each model can be assessed against: 1. The regression coefficient associated with the interaction should be numerically positive and statistically significant at Step 1. A failure at this step indicates that the primary prediction of the theory has not been confirmed, and testing often ceases at this point. However, if this prediction is upheld then: 2. the coefficient associated with the interaction should remain numerically positive at Step 2 (after the main effect terms have been added); 3. neither regression coefficient associated with the main effect terms should statistically differ from zero at step 2; and 4. the increase in criterion variance (*R²*) at Step 2 should not be statistically significant. Thus, these four tests can be used for each of the three regression models, providing 12 total tests that summarize the extent to which the predictions of cognitive balance are met in a given data set (independently for self-report and implicit data). Results of these analyses are presented in for both implicit and explicit data. For implicit data, evidence of balance was uniformly high, with all three models passing all four tests described above, except for one partial failure at Step 2 in which, for one model, one of the two main effect terms was statistically significant despite the presence (and continued statistical significance) of the interaction term. Nonetheless, 11 of 12 possible tests were passed, providing strong evidence of cognitive balance. For explicit results, evidence of balance was mixed, with all three models failing at least one test at Step 2 (in total, eight of 12 tests were passed). Thus, even though explicit measures were strongly correlated at the bivariate level (considerably more so than their implicit counterparts), they did not conform as well to the more complex pattern of relationships specified by the Balanced Identity model, replicating much prior work with this design. ## Discussion Upon inducing associations between participants and one of two novel groups, robust preference for and identification with the ingroup emerged, especially when measured at the implicit level. These findings add another replication to the large body of work employing the MGP. The novel contribution, however, is the demonstration that the predictions of Balanced Identity are satisfied by emergent cognitions in the MGP and therefore need not rely on an iterative or otherwise protracted period of enculturation or experience. The fact that new group-related attitudes and identifications formed in such a way as to be in balance with pre-existing associations (i.e., self-esteem) suggests that the Balanced Identity relationships can thought of as causal determinants of the strength and direction of newly formed associations such as group attitudes. The next set of studies directly tests causality by manipulating each construct and measuring the effect of this manipulation on related constructs. # Experiment 2: Manipulating Group Identification This experiment follows a nearly identical procedure to that described in Experiment 1, except that, following group assignment, the strength of the associative self-group relationship is manipulated. Because Balanced Identity emerges so much more consistently at the implicit level, a manipulation of group identification that also targeted associations (as opposed to the explicit reporting of a self-group relationship) was employed. The central question, then, was whether manipulating identification will affect group attitude and/or self-esteem, and what effect, if any, this will have on the emergence of cognitive balance. This experiment also begins to explore a secondary question. Some prior work has suggested that balance does not emerge as reliably for members of lower status groups, such as members of less prestigious residential colleges or ethnic minorities. This pattern can be formalized as the prediction that balance will not be as robust for groups that are less positively evaluated. The current manipulation involved artificially increasing group identification in half the participants, and artificially decreasing it in the other half. If the phenomenon described above is general, we could see a disruption of balance specifically in individuals in the latter condition. ## Methods ### Participants One hundred twenty-two participants were recruited from the same population and following the same procedure described in Experiment 1. ### Procedure The procedure was closely modeled after that employed in Experiment 1; participants were assigned to groups in the same way as described there. They completed a manipulation of group identification followed by the same dependent measures as in Experiment 1. ### Manipulation of group identification The self-group association was manipulated using a partial-IAT procedure. Participants were told that they were to imagine that they had the opportunity to spend a large amount of time with one of the two groups (the ingroup or the outgroup, as a between-participants factor). To simulate this experience, they would perform a categorization task in which they would categorize words related to the self with members of that group and words related to others with members of the other group. They completed two blocks of 60 trials in which they either repeatedly responded to self and ingroup names using one key and other and outgroup names using another key, or repeatedly responded to self and outgroup words using one key and other and ingroup words using another key. Thus, as a between-participants factor, the manipulation was designed to either bolster the self-ingroup relationship or weaken that relationship while bolstering the self- outgroup relationship. To avoid associating any response with a particular side of the screen, the pairing of group and self was counterbalanced across the two blocks of trials. ### Measures The same measures used in Experiment 1 were used here, namely implicit and explicit measures of group attitude, group identification, and self-esteem. ## Results ### Descriptive statistics Standard exclusion criteria for IAT results led to the elimination of data from 10 participants from the attitude and identity IATs and 16 participants from the self-esteem IAT. There was one significant effect of ingroup name (Copley or Dawson), with stronger ingroup preference for participants assigned to the Copley group, *t*(109)  = 2.84, *p* = .005. However, because this was the only effect of group name across the four experiments reported here and because it did not interact with other reported findings, it is not interpreted further. Overall patterns of results were highly similar to Experiment 1, with participants exhibiting robust preference for the ingroup, *D* = .30 (.37), *t*(110)  = 8.34, *p*\<.001, *d* = .78, robust identification with the ingroup, *D* = .49 (.36), *t*(111)  = 14.42, *p*\<.001, *d* = 1.06, and strong implicit positive self-esteem, *D* = .47 (.30), *t*(103)  = 15.91, *p*\<.001, *d* = 1.18. As in Experiment 1, identification was stronger than attitude, paired *t*(108)  = 4.65, *p*\<.001. Preference for and identification with the ingroup were modestly correlated, *r*(107)  = .35, *p*\<.001, but again self-esteem as not correlated with attitude, *r*(101)  = .13, *p* = .21, or identification, *r*(102)  = .13, *p* = .20. Evidence for an effect of group membership was also present at the self-report level. Participants expressed more liking for the ingroup, M = .70 (1.7), *t*(121)  = 4.50, *p*\<.001, *d* = .41, as well as greater identification with the ingroup, M  = 1.3 (2.4), *t*(121)  = 5.95, *p*\<.001, *d* = .54. Participants also reported somewhat greater liking for the self as compared with others, M = .47 (1.2), *t*(121)  = 4.39, *p*\<.001, *d* = .39. Correlations among self-report measures were moderate, all *r*(120) \>.39, *p*\<.001. Strongest was the correlation between attitude and identification, *r*(120)  = .59, *p*\<.001. Implicit and explicit measures did not correlate, all \|*r*\|\<.11, *p*\>.25. ### Effect of identification manipulation As noted above, half of participants completed a manipulation designed to strengthen the self-group relationship (the “match” condition), while the other half completed a manipulation designed to weaken it (the “mismatch” condition). The manipulation was effective; participants in the match condition exhibited markedly stronger ingroup identification as measured by the IAT, *M*_MATCH = .70 (.27), *M_MISMATCH* = .29 (.32), *t*(110)  = 7.38, *p*\<.001, *d* = 1.4, though identification remained positive and significant even in the mismatch condition, *t*(57)  = 6.98, *p*\<.001. This manipulation also affected ingroup attitude, *M*_MATCH = .37 (.37), *M_MISMATCH* = .22 (.39), *t*(109)  = 2.20, *p* = .03, *d* = .41, but did not affect self-esteem, *M*_MATCH = .49 (.27), *M_MISMATCH* = .45 (.33), *t*(102)  = .61, *p* = .54, *d* = .13. The manipulation had no effect on any of the explicit measures, all *t*\<.72, *p*\>.47. ### Balanced Identity Analyses provides a summary of results of the Balanced Identity analyses for implicit and explicit data, collapsing across condition. Overall, evidence for balance was strong at the implicit level but weak at the explicit level, with 12 and 7 tests passed, respectively. To examine whether the relative degree of balance differed depending on experimental condition, and in particular whether balance was less robust in the mismatch condition in which identification was reduced, these analyses were also conducted separately for each condition. However, one potential limitation of such an approach must be noted. By intervening on group identification in either the positive or negative direction, we may have artificially reduced the variability in that measure, which would work against finding evidence of balance on statistical rather than conceptual grounds. Thus, the following analyses should be interpreted with this in mind and in the context of supplementary analyses presented in the next section. provides these analyses sub-divided by condition (match or mismatch). However, when broken down by condition the results are more nuanced. Beginning with the implicit data, evidence for balance was once again strong in the match condition, with 12 out of 12 tests successfully passed. The results were quite different in the mismatch condition. All three models failed at Step 1, indicating that the interaction did not predict the criterion, nor did it do so at Step 2 when main effects were included. Thus, there was no evidence of balance when group identification was manipulated in the negative direction. The results from explicit data, while not providing strong evidence of balance, were relatively consistent across the two conditions, with 8 of 12 total tests passed for data from the match condition and 6 of 12 from the mismatch condition. ### Supplementary analysis: Comparison with Experiment 1 Because our design did not feature a control group for whom attitudes were not manipulated, the relative degree of change from baseline in the match versus mismatch condition is difficult to ascertain. However, Experiment 1 resembles such a control group, in that the design was identical to Experiment 2 except there was no manipulation of the identification construct. Indeed, the means across these two experiments (collapsing cross condition in Experiment 2) were highly similar, and did not differ significantly, all *t*\<1.37, *p*\>.17. Thus, under the assumption that the means in Experiment 1 are a reliable estimate of what we would have observed in Experiment 2 if there had been no manipulation, we can gain some insight into whether one condition was a more powerful elicitor of change by comparing Experiment 1 to the two conditions of Experiment 2. We can also use this analysis to explore the issue of reduced variability described above by comparing variances directly to see if variation was reduced with respect to any of the constructs. Results are presented in, showing the average change and associated standard errors for the comparisons between Experiment 1 and the match and mismatch conditions of Experiment 2. The results suggest equivalence in the strength of the positive and negative manipulation; i.e., it does not appear to have been easier to increase or decrease identification and attitudes. Supporting the worry that the manipulation might have reduced variance, the null hypothesis of equal variance was rejected in only the two cases in which ingroup identification in Experiment 1 was compared with ingroup identification in the match and mismatch conditions of Experiment 2. As expected, in both cases variability was lower in the Experiment 2 conditions. However, this finding does not appear able to explain the lack of balance found in the mismatch condition as compared to the match condition. This is because while variance was reduced relative to Experiment 1, it was reduced equivalently in the match and mismatch conditions, in which the null hypothesis of equal variance could not be rejected, *F*(57, 53)  = 1.39, *p* = .23. Thus, the differential degree of balance cannot be attributed to reduced variability in one of the criterion constructs. ## Discussion In Experiment 2, identification with a minimal ingroup was manipulated by either weakening or strengthening the associative relationship between self and group in a partial-IAT procedure. This manipulation was successful and constituted a large effect; participants in the “mismatch” condition were much less implicitly identified with their ingroup. This manipulation did not, however, affect self- reported identification, and participants continued to explicitly identify with their ingroup. This demonstrates that the manipulation did not simply *confuse* participants about their membership, but rather affected the associative relationship between self and group. Importantly, the manipulation of identification also affected implicit attitude, with participants in the mismatch condition showing weaker implicit preference for their ingroup. Thus, manipulating one leg of the “Balance triangle” exerts causal influence on at least one of its neighbors, lending support to a causal rather than merely descriptive interpretation of these relationships. It is interesting that implicit self-esteem was not similarly affected. In general, a causal interpretation of Balanced Identity merely predicts that changing one construct will change related constructs so as to preserve balance; it is agnostic with respect to exactly *how* the equation is balanced. That is, when group identification changes, the cognitive system could respond by adjusting self- esteem, group attitude, or both. In this case, the predominant operation was to decrease the strength of group attitude. A plausible interpretation of this pattern of results is that because the self-positive association precedes the experimental setting and generalizes far beyond it, it is more resistant to minor perturbations via associated constructs of the sort created here. More leverage will be gained on this issue after manipulating the other constructs and describing their downstream effects. Interestingly, the mismatch condition, which reduced group identification (and consequently group attitude), also disrupted balance; tests of balance in this condition all failed at the first step. By contrast, participants who had their group identification increased through the match manipulation continued to show robust balance, passing all 12 tests. This finding supports past work contending that belonging to a low status or otherwise non-liked group prevents cognitive balance, though the present findings allow us to generalize this considerably because “low status” in this case in no way reflects richer conceptions of enculturation but rather the results of a simple associative manipulation designed to reduce the self-group association. Supplementary analyses also suggested that the failure to find balance in the mismatch condition cannot simply be attributed to reduced variance in one of the constructs. The implications of this finding will be discussed further as data from the other experiments is presented. More broadly, however, we again saw clear evidence of balance on implicit but not on explicit measures. # Experiment 3: Manipulating Group Attitude This experiment followed a nearly identical procedure to that described in Experiment 2, except that group attitude was manipulated instead of group identification. In one direct sense, the MGP is in fact a manipulation of group identification; that is, it attaches participants to a group. By contrast, it does not provide evidence directly relevant to attitude or self-esteem. Thus, the manipulation in Experiment 2, above, could have been effective because it directly targeted the same factor implicated in the minimal group paradigm itself. If so, the causal effects found in that experiment could be thought of as simply strengthening or weakening the minimal group induction itself, rather than a consequence of causal relationships specified by the Balanced Identity model. However, if those effects were in fact the result of the consistency- generating processes described by Balanced Identity, results should be similar if another leg of the Balance triangle (in this case, group attitude) is manipulated. The central question, then, is whether manipulating attitude towards the newly assigned minimal group will affect the other two related constructs. A secondary question is whether artificially decreasing group attitude through our manipulation will disrupt the formation of cognitive balance. ## Methods ### Participants One hundred thirty-five participants were recruited from the same population and following the same procedure described in Experiments 1 and 2. ### Procedure The procedure was closely modeled after that employed in Experiment 2, except that the manipulation targeted attitude instead of identity. ### Manipulation of group attitude As in Experiment 2, a partial-IAT procedure was employed. Participants again completed two blocks of 60 trials in which, as a between-participants factor, they either repeatedly responded to ingroup and positive words using one key and outgroup and negative words using another key, or repeatedly responded to outgroup and positive words using one key and ingroup and negative words using another key. Thus, as a between-participants factor, the manipulation was designed to either increase or decrease the positive association with the ingroup (and affect the positive association with the outgroup in the opposite direction). ### Measures The same measures used in Experiment 1 were used here. ## Results and Discussion ### Descriptive statistics Standard exclusion criteria for IAT results led to the elimination of data from nine participants from the identification IAT and 11 participants from each of the other IATs. There were no effects of ingroup name or task order, and so these factors were dropped from subsequent analysis. Overall results again mirrored those described in the prior experiments, with participants exhibiting robust preference for the ingroup, *D* = .28 (.42), *t*(123)  = 7.60, *p*\<.001, *d* = .68, robust identification with the ingroup, *D* = .48 (.29), *t*(125)  = 18.45, *p*\<.001, *d* = 1.26, and strong implicit positive self-esteem, *D* = .51 (.31), *t*(123) = 18.33, *p*\<.001, *d* = 1.37. Identification was again stronger than attitude, paired *t*(123)  = 18.33, *p*\<.001. As in prior studies, implicit identification with the ingroup was stronger than implicit preference for the ingroup, paired *t*(122)  = 5.21, *p*\<.001. Turning to correlations between constructs, preference for and identification with the ingroup were modestly correlated, *r*(121)  = .38, *p*\<.001, and weaker correlations were observed between attitude and self-esteem, *r*(120)  = .23, *p* = .01, and identification and self-esteem, *r*(121)  = .19, *p* = .04. Participants self-reports also revealed more liking for the ingroup, M = 1.0 (1.8), *t*(134)  = 6.43, *p*\<.001, *d* = .56, as well as greater identification with the ingroup, M  = 1.8 (2.2), *t*(134) = 9.29, *p*\<.001, *d* = .82. Participants also reported somewhat greater liking for the self as compared with others, M = .50 (1.1), *t*(134)  = 5.18, *p*\<.001, *d* = .45. Correlations among self-report measures varied from weak in the case of self-esteem and attitude, *r*(133)  = .17, *p* = .051, moderate for self-esteem and identification, *r*(133)  = .30, *p*\<.001, and strong in the case of attitude and identification, *r*(133)  = .61, *p*\<.001. Implicit and explicit measures did not correlate, all \|*r*\|\<.15, *p*\>.10. ### Effect of attitude manipulation Half of the participants completed a manipulation designed to strengthen preference for the ingroup (“match” condition) while the other half completed a manipulation designed to weaken it (“mismatch” condition). This manipulation was effective; participants in the match condition exhibited markedly stronger ingroup attitude as measured by the IAT, *M*_MATCH = .56 (.31), *M_MISMATCH* = .03 (.32), *t*(122)  = 9.29, *p*\<.001, *d* = 1.66. Indeed, participants in the mismatch condition no longer preferred their ingroup following this manipulation, *t*(64)  = .74, *p* = .45. This manipulation also affected ingroup identification, *M*_MATCH = .58 (.29), *M_MISMATCH* = .39 (.27), *t*(124)  = 3.93, *p*\<.001, *d* = .68, though identification remained positive and significant in the mismatch condition, *t*(64)  = 11.61, *p*\<.001. This time the manipulation also affected implicit self-esteem, *M*_MATCH = .59 (.30), *M_MISMATCH* = .44 (.31), *t*(122)  = 2.91, *p* = .004, *d* = .49. The manipulation had no effect on explicit measures, all *t*\<1.32, *p*\>.19. ### Balanced Identity Analyses provides a summary of results of the Balanced Identity analyses for implicit and explicit data, collapsing across condition; again, balance was clearly observed in the implicit but not explicit data. provides these analyses sub-divided by condition (match or mismatch). Again results suggest that balance was disrupted in the mismatch condition. For implicit data in the match condition, evidence for balance was strong, with 10 out of 12 tests successfully passed, though it is important to note that one model (with self-esteem as the criterion) failed to show a significant interaction at Step 1, an important deviation from predictions. At the explicit level, evidence of balance was completely absent, with all models failing all tests. For participants in the mismatch condition all three models focusing on implicit data failed at Step 1, indicating little support for balance. For explicit data, there were some hints of balance, though not definitive, with 8 of 12 tests passed successfully. ### Supplementary analysis: Comparison with Experiment 1 As described in Experiment 2, we can estimate the relative degree of change affected by the match versus mismatch condition by comparing to the means from Experiment 1 (no manipulation). As in Experiment 2, the overall means from Experiment 3 (collapsing across conditions) did not differ significantly from those observed in Experiment 1, all *t*\<.34, *p*\>.74. Results of the comparisons between Experiment 1 means and the two conditions in Experiment 3 are presented in. The results again sugest equivalence in the strength of the positive and negative manipulation. Interestingly, we again observed some cases in which variance appeared to have been reduced by the manipulation, but instead of occurring with respect to the manipulated construct of group attitude, it again appeared only with respect to group identification (in both the match and mismatch conditions). However, as in Experiment 2, the variances between match and mismatch conditions themselves did not differ, suggesting that while manipulation appears to reduce variance, that reduced variance cannot account for the varying degrees of balance observed between match and mismatch conditions. The results of Experiment 3 lend further support to a causal interpretation of Balanced Identity by showing that intervening on implicit attitudes robustly affected both implicit group identification and self-esteem. Participants whose implicit preference for their ingroup was reduced also implicitly identified with that group less, and even showed less implicit self-related positivity. In addition, and as in Experiment 2, group identity, attitude, and self-esteem remained closely related in the manner predicted by Balanced Identity when group attitude was reinforced, but not when it was disrupted, suggesting that depressing ingroup preference prevents cognitive balance. # Experiment 4: Manipulating Self-Esteem A nearly identical procedure to that described in the previous two experiments was employed, except that this time the self-positive association was manipulated. This association differs in one profound respect from the group attitude and group identification associations, in that it is an enduring construct (i.e., implicit self-esteem) that exists before and endures after the experimental paradigm in which participants find themselves. That is, the minimal group paradigm that participants take part in here *produces* group attitude and group identification; it does not *produce* self-esteem in the same way, because the self-valence link is an enduring representation in semantic memory. Does this make it more resistant to the form of manipulation used above? And does it affect the influence it exerts on associated constructs? ## Methods ### Participants One hundred sixteen participants were recruited from the same population and following the same procedure described in prior experiments. ### Procedure The procedure was closely modeled after that employed in Experiments 2 and 3, except that the manipulation targeted the self-valence association. ### Manipulation of implicit self-esteem The same partial-IAT procedure used in prior experiments was used here. That is, as a between-participants factor, participants either repeatedly responded to self and positive words using one key and other and negative words using another key, or repeatedly responded to other and positive words using one key and self and negative words using another key. Thus, the manipulation was designed to either increase or decrease the positive association with the self while affecting the association with other-related concepts in the opposite direction. ### Measures The same measures used in Experiment 1 were used here. ## Results and Discussion ### Descriptive statistics Standard exclusion criteria for IAT results led to the elimination of data from three participants from the attitude IAT, seven from the self-esteem IAT, and 11 from the identification IAT. There were no effects of ingroup name or task order, and so these factors were dropped from subsequent analysis. Overall results again mirrored those described in the prior experiments, with participants exhibiting robust preference for the ingroup, *D* = .27 (.35), *t*(112)  = 8.27, *p*\<.001, *d* = .64, robust identification with the ingroup, *D* = .41 (.36), *t*(104)  = 11.69, *p*\<.001, *d* = 1.2, and strong implicit positive self-esteem, *D* = .47 (.35), *t*(109)  = 14.05, *p*\<.001, *d* = 1.0. Identification was again stronger than attitude, paired *t*(102)  = 3.05, *p* = .003. Unlike in prior studies, preference for and identification with the ingroup were not correlated, *r*(101)  = .06, *p* = .56, though weak correlations were observed between attitude and self-esteem, *r*(107)  = .20, *p* = .03, and identification and self-esteem, *r*(100)  = .23, *p* = .02. Participants self-reports also revealed more liking for the ingroup, M = .78 (1.4), *t*(114)  = 6.08, *p*\<.001, *d* = .56, as well as greater identification with the ingroup, M  = 2.0 (2.1), *t*(114)  = 10.38, *p*\<.001, *d* = .95. Participants also reported somewhat greater liking for the self as compared with others, M = .46 (1.2), *t*(115)  = 4.19, *p*\<.001, *d* = .38. Preference and identification were highly correlated, *r*(113)  = .59, *p*\<.001, but no other bivariate correlations reached significance, both *r*(113) \<.12, *p*\>.23. Implicit and explicit identity were moderately correlated, *r*(102)  = .28, *p* = .004, but no other implicit-explicit correlations achieved significance, both \|*r*\|\<.10, *p*\>.33. ### Effect of manipulation Half of participants completed a manipulation designed to strengthen association between the self and positivity (“match” condition), while the other half completed a manipulation designed to weaken it (“mismatch” condition). This manipulation was effective; participants in the match condition exhibited stronger implicit self-esteem as measured by the IAT, *M_MATCH* = .56 (.34), *M_MISMATCH* = .40 (.34), *t*(108)  = 2.53, *p* = .01, *d* = .47, though the size of this effect is considerably smaller than in the previous two experiments, suggesting that self-esteem is indeed more resistant to manipulation in this fashion. The manipulation also affected ingroup preference, *M*_MATCH = .36 (.33), *M_MISMATCH* = .20 (.35), *t*(111)  = 2.39, *p* = .02, *d* = .47, but did not affect ingroup identification, *M*_MATCH = .46 (.32), *M_MISMATCH* = .37 (.39), *t*(103)  = 1.38, *p* = .17, *d* = .25. Positive self-esteem, ingroup identification, and ingroup preference remained statistically significant in both match and mismatch participants, all *t*s\>4.5, *p*s\<.001. The manipulation had no effect on explicit measures, all *t*\<.43, *p*\>.67. ### Balanced Identity Analyses provides a summary of results of the Balanced Identity analyses for implicit and explicit data, collapsing across condition; overall, evidence for balance was strong at the implicit level but weak at the explicit level, with 12 and 6 tests passed, respectively. provides these analyses sub-divided by condition (match or mismatch). When broken down by condition the results of this study were somewhat less definitive than in the prior two investigations. For participants in the match condition, implicit data led to 9 of 12 tests being passed, but two of the failures were at Step 1, where for models with identification and self-esteem as criterion, the interaction was directionally consistent but did not reach significance. For explicit data, only 3 of 12 tests were passed, suggesting no balance at this level. In the mismatch condition, there was for the first time some evidence of balance with 10 of 12 tests passed, though again the two failures were at Step 1. For explicit data in the mismatch condition, 6 of 12 tests were passed. In one sense, these data are broadly consistent with prior experiments, in that there was better evidence of balance at the implicit than explicit level. However, evidence of balance was this time as good in the mismatch condition as in the match condition. Additionally, both match and mismatch analyses revealed failures at Step 1, which represent particularly dramatic divergences from predictions. One plausible interpretation of the maintenance of balance in the mismatch condition concerns the strength of our manipulation. The manipulation of self- esteem, while statistically significant, was considerably weaker than the manipulation of the other constructs in the previous studies (*d_{self- esteem}* = .44, *d_attitude* = 1.77, *d_{identification}* = 1.40; see also, introduced below). Thus, the manipulation might not have created a powerful enough perturbation to effect balance as dramatically, perhaps because self-esteem, an enduring and central aspect of personhood, is particularly resistant to negative perturbations. ### Supplementary analysis: Comparison with Experiment 1 As described in Experiment 2 and 3, above, we again compared results to the means from Experiment 1, which can serve as a no-manipulation baseline. As in the prior two experiments, collapsing cross condition in Experiment 4 reveals means that do not differ significantly from those in Experiment 1, all *t*\<1.3, *p*\>.20. Results for the comparisons between Experiment 1 and the match and mismatch conditions of Experiment 4 are presented in, showing the average change and associated standard errors. Unlike in the previous two experiments, visual inspection of the figure suggests that the manipulation of self-esteem was not particularly effective in the match condition (i.e., the mismatch bars are the only ones that are reliably displaced from 0). Thus, it appears that the manipulation was not powerful enough to inflate already robust implicit self- esteem but was powerful enough to (presumably temporarily) decrease it. As in Experiment 2, results again suggested a (in this case marginal) reduction of variance for the intervened upon construct (self-esteem). But as in the prior experiments, the variances between match and mismatch conditions in Experiment 4 did not differ. # General Discussion The current study investigated whether consistency between group attitude, group identification, and self-esteem—which has been widely demonstrated with respect to familiar and salient real-world groups—would emerge immediately following random assignment to minimal social groups. Four experiments provided a strong affirmative answer: cognitive balance was robustly present in every case, but (as has been noted with real-world groups) only at the implicit level. These findings strongly suggest that cognitive balance need not stem from a history of gradual revision or iterative dissonance reduction. Rather, the initial form that intergroup cognitions take is well described by the predictions of Balanced Identity, suggesting that new cognitions are constrained by existing ones as described by the Balance model. Relatedly, these studies provide direct evidence for a causal (as opposed to merely descriptive) reading of Balanced Identity by showing that manipulating any one leg of the balance triad (e.g., group attitude, group identification, self-esteem) affected related constructs while preserving the broader pattern of cognitive balance. Establishing these causal connections within the framework of the minimal group paradigm supports a reading of Balanced Identity as a generalized causal account of intergroup bias; intergroup bias can be conceptualized as the interactive effects of self-esteem and group identification. What's more, it suggests new pathways to intervention. Intergroup bias can be intervened with not only directly but also by manipulating the two related constructs of group identification and self-esteem. Finally, the present results bolster the prior observation that belonging to a group that is not positively evaluated inhibits balance for members of that group. When we intervened on group attitude or identification, participants who experienced the “negative” intervention that decreased the strength of that construct did not show cognitive balance. This was not merely the result of any artificial intervention, as participants who experience the “positive” intervention increasing the strength of the same construct did show cognitive balance. Nor was it the result of a reduction in variance in the manipulated construct; while manipulation did appear to reduce variance (as compared to Experiment 1, in which there was no manipulation), it reduced variance equivalently across the match and mismatch conditions and so cannot uniquely explain the different degrees of balance that were then observed. Results of a final experiment manipulating self-esteem were somewhat more equivocal, however, perhaps because the effort to change implicit self-esteem was less successful than the other two manipulations, and the effects it did have may have been asymmetrical, predominately deflecting implicit self-esteem in the negative direction. In any case, one suggestive interpretation of this broader pattern is that, when belonging to a group that is negatively evaluated, balance is disrupted so as to prevent that negativity from bleeding over into self-esteem. While broadly consistent with some motivational accounts of self-group relationships such as Social Identity Theory, it is striking that this phenomenon is seen most clearly at the implicit level, where we would expect that motivational concerns would be at least somewhat attenuated. Indeed, a question raised by these findings is how the implicit system “knows” to disrupt balance in the face of a negative perturbation while retaining it in the face of a positive perturbation. The microdynamics of these changes could be a fruitful topic for future study. Stepping back somewhat, some past research has postulated that intergroup bias is the result of self-related positivity that associatively spreads from the self to ingroups. However, support for this claim has been mixed. The present research suggests that the reason for this mixed support may be that, in trying to associate self-esteem with group bias, one crucial construct, group identification, has been left out. In a sense, it is attempting to impose a bivariate model (i.e., linkages between self-esteem and group attitude) on a multivariate and interactive phenomenon. Looking solely at the link between group attitude and self-esteem in the present studies, the correlations were a modest *r* = .29, and the relationship was significant in some but not other studies. In other words, the same mixed support found in prior investigations appeared in these data. By contrast, when predicting each construct from the product of the other two, implicit data yielded a statistically significant and positive relationship in the great majority of tests across these four studies (with the exception of the cases in which the constructs were artificially reduced through intervention). As with prior work employing the Balanced Identity framework, however, the predicted pattern appears to characterize *implicit* constructs much better than *explicit* constructs. Indeed, as compared with the 47/48 successful statistical tests for implicit data (collapsing for the moment across condition), only 24/48 tests were passed for self-report data. Indeed, it is possible that the relationships between self- report constructs may be better captured by bivariate relationships, which were generally sizable, as compared with the implicit data, which are better- described by the multiplicative relationships specified by Balanced Identity. One general pattern emerging from my manipulation of individual components of the balance triad was that, while each construct is easily manipulated, the influence of that manipulation on the more “downstream” constructs was not perfectly predictable. That is, it did not always affect *both* theoretically related constructs. Suggestively, the pattern was at least partially structured. Specifically, group attitude and group identification mutually affected one another, and group attitude and self-esteem mutually affected one another. However, self-esteem and group identification did not affect one another in either direction. Future work will be necessary to understand whether this pattern is replicable and, if so, why it occurs. One potential weakness of the present design is that the method used to manipulate the constructs in question was closely related to the method used to measure them following their manipulation. That is, a partial-IAT procedure was used to manipulate the constructs, which were then measured with a full IAT. While the left/right side pairing was balanced at the participant level such that a simple side-bias is unlikely, it is nonetheless possible that the partial-IAT procedure affected subsequent IATs to some extent. Future work could make the method of manipulation more distinct, for example through a conditioning paradigm, or could alter the method of construct measurement by using a different implicit measure. In closing, this study highlights the value of the minimal group paradigm as a methodological tool in intergroup social cognition research. By abstracting away from the variable knowledge and informational complexity that characterizes real-world groups, we can gain a clear perspective on how group affiliations emerge and change across contexts and can observe the emergence of such cognitions from their earliest moments. Of course, it is always important to verify that findings with minimal groups generalize to real groups. In the present context, this has already been accomplished, as cognitive balance has been widely observed with respect to real-world groups. But that observation left open several questions regarding the causal status and informational time course of such consistency. The minimal group paradigm became a powerful tool for addressing these issues in a controlled manner. At a higher level, by demonstrating that a phenomenon observed with respect to real-world groups also occurs with respect to novel social groups, we are able to reduce a set of specific observations about specific groups to a general principle governing intergroup cognition more broadly. This opens the door to interpreting Balanced Identity as a general account of intergroup attitudes (as well as, of course, intergroup identifications and self-esteem), offering a model for both prediction and intervention. Future work focusing on linkages between self- esteem and attitudes will benefit by incorporating these insights and ensuring attention is also paid to the closely related construct of group identification. [^1]: The author has declared that no competing interests exist. [^2]: Conceived and designed the experiments: YD. Performed the experiments: YD. Analyzed the data: YD. Wrote the paper: YD.

# Introduction Animal behavior has the potential to contribute substantially to conservation efforts, but the field of conservation behavior has been slow to develop. Only recently have attempts been made to develop unifying principles and overarching frameworks for the study of conservation behavior. Because behavior is one of the most flexible traits an animal may possess, it represents a promising mechanism by which species may cope with human-induced rapid environmental change (HIREC). Behavioral flexibility is the ability of an animal to modify its behavior under different environmental conditions. Behavioral responses are often rapid and reversible, and afford animals some degree of control over their external stimuli by choosing their surrounding environment. Appropriate behavioral responses to environmental change are vital for animals that live in naturally heterogeneous environments, and will likely be critical for the persistence of many species in the face of HIREC. Both natural environmental heterogeneity and the effects of HIREC occur across time and space, and this environmental variability affects species occurrence, distribution, abundance, population stability, and individual behavior. Some habitat types are relatively stable, while others are highly variable in their natural state. For example, wetland biota is influenced by a variety of both predictable (tides, seasonal flows) and stochastic (flood, drought) changes on a variety of time scales. Anthropogenic changes to wetlands, such as modification of tidal inundation regimes through the construction of earthen dikes/levees, can greatly interfere with these natural patterns of heterogeneity. Because coastal wetlands are both highly variable in their natural state and have been heavily modified by human activities, organisms that live in coastal wetlands present opportunities for the study of both behavioral flexibility in response to natural heterogeneity and behavioral responses to HIREC. We investigated the response of the endangered northern salt marsh harvest mouse (*Reithrodontomys raviventris halicoetes*) to both natural and anthropogenic environmental heterogeneity. The salt marsh harvest mouse is an ideal species in which to study behavioral flexibility and responses to HIREC, because it is endemic to changeable wetland habitats and currently occurs in both natural tidal (full tidal influence) and anthropogenically-altered (diked) wetlands. Features of salt marsh harvest mouse biology that are consistent with adaptation to tidal environments become clear when comparisons are made to the sympatric congener the Western harvest mouse (*R. megalotis*), which occurs in tidal marshes but also in a variety of other habitat types. Compared to Western harvest mice, the salt marsh harvest mouse is a stronger swimmer and is more capable of consuming sea water. They are also capable of entering torpor, a mechanism thought to aid small mammals in coping with osmotic stress, and are more active during warmer daytime temperatures which allows for a lower resting metabolic rate than activity during cool, damp nights. Finally, as a species that naturally resides in a highly changeable environment, the salt marsh harvest mouse may possess a high level of behavioral flexibility that predisposes it to alter its behavior to utilize resources in novel habitat types, such as diked wetlands. Understanding how salt marsh harvest mice use their habitat during daily tidal flooding is especially important because tidal restoration is the primary conservation measure for preserving this species which is endangered due to dramatic habitat loss (∼90%) and faces further loss due to predicted sea level rise. As the only terrestrial mammals that is entirely restricted to coastal marshes, the salt marsh harvest mouse must somehow avoid drowning during high tides. It has long been believed that salt marsh harvest mice spend the majority of their time in wetlands, and move upland only to escape tide and flood waters. Since the 1950s, researchers have attempted to characterize salt marsh harvest mouse behavior during tidal inundation to determine whether mice move vertically into tall vegetation or horizontally upland to escape the tide,. In tidal wetlands salt marsh harvest mice typically experience tidal flooding twice a day. In contrast, diked wetlands are flooded continuously during the rainy season (approximately October through March), and dried out for the remainder of the year. Thus, the patchwork of modified and unmodified habitat in the San Francisco Bay Estuary presents the opportunity to study habitat use of mice in both highly changeable natural (tidal) habitat and highly modified (diked, surrounded by levees with water control structures) habitat that is temporally more stable. Conclusions from previous studies of the behavior of the salt marsh harvest mouse at high tide have been split on the question of how mice deal with daily tidal inundation of their habitat. Some concluded based on trapping data or visual observations that animals move out of tidal wetlands and into upland areas or onto levees to escape rising waters, , and others concluded that they remain in tall, dense vegetation over water where they can easily move about in the thatch layer. However, although trapping methods allow researchers to determine that a mouse had been at a location at some point during the night, it is impossible to tell from these previous trapping studies where mice were during the time that the habitat was actually flooded. The contrasting conclusions of previous studies suggest that the only way to definitively understand salt marsh harvest mouse habitat use during high tide is through the use of radio telemetry. The objective of this study was to determine how salt marsh harvest mice respond to both natural environmental heterogeneity and HIREC, focusing on the effects of regular tidal inundation. We approached this investigation from a conservation behavior perspective which uses general principals of animal behavior to address conservation issues. We applied two questions from a recently developed conservation behavior framework. One focus of the Berger-Tal et al. framework stresses performing behavior-based management, which considers behavior in conservation decision making protocols. In our case, answering the question *Do salt marsh harvest mice move vertically (into tall vegetation) or horizontally (into upland areas) to escape the high tide*? will allow managers develop restoration and enhancement priorities in the little habitat that remains, and is a research priority explicitly identified in the recovery plan for this species. A second focus of the framework addresses if and how HIREC affects behavior, in the case of this endangered species, *Do mice in anthropogenically altered diked wetlands behave differently than those subject to tidal influence*?. To answer these questions, we tracked mouse movements during high tide events in natural tidal wetlands using radio telemetry to identify the type of refuge used to escape incoming water. During the same periods, we tracked mice in adjacent diked wetlands. This allowed us to determine space use and estimate movement rates in response to high tides, as well as independent of tidal influence in diked habitats. With regard to our first question, we predicted that during tidal flooding, mice would remain in tall vegetation over water. Our predictions for the second question depended on the results of the first. If the prediction that mice remain in tall vegetation during the high tide was confirmed, we expected that movement distances (and thus rates) in natural tidal and diked wetlands would be similar. In contrast, if mice instead moved long distances upland to escape the tide, then we expected that movement rates in natural tidal wetlands would be greater than those in diked wetlands. # Methods ## Ethical note The research reported here was conducted under an approved IACUC protocol from New Mexico State University (2011–013) and a Memorandum of Understanding between the California Department of Fish and Wildlife and the United States Fish and Wildlife Service for handing of endangered species, with direct involvement and supervision by the California Department of Fish and Wildlife. All procedures were consistent with the guidelines for the use of wild mammals in research from both ASAB/ABS and the American Society of Mammalogists. ## Study Area We conducted this study between May and August 2011 in the Suisun Marsh in Solano County, CA, USA (38° 11′ 15.86″ N, 122° 3′ 52.67″ W), which is a large wetland complex in the San Francisco Bay Estuary. We established six study areas which were grouped into three blocks, each containing one natural tidal and one diked wetland. Study areas were established on public lands with the permission of the California Department of Fish and Wildlife and the Suisun Resource Conservation District. Paired wetlands (one tidal and one diked) within each block were 100 to 600 m apart and separated by levee roads or sloughs, and the three blocks were 1 100 to 8 500 m apart. Although salt marsh harvest mice are capable of moving distances similar to the distances between our study wetlands, we did not observe movement between these wetlands during our study. All tidal wetlands were subject to full natural tidal action. All diked wetlands were cut off from water access for the summer and were essentially dry. Vegetation on the study sites included pickleweed (*Salicornia virginica*, also known as Virginia glasswort), Olney's threesquare bulrush (*Schoenoplectus americanus*, also known as chairmaker's bulrush), cattail (*Typha* spp.), saltgrass (*Distichlis spicata*), tule (*Sc. acutus*), common reed (*Phragmites australis*), Baltic rush (*Juncus balticus*) and other halophytic species. Plant community composition was broadly similar between natural tidal and diked areas, although tidal areas tended to be dominated by reeds and rush (mean height ±1 *SE*  = 138.13±27.84 cm), while the diked areas tended to be dominated by pickleweed (mean height ±1 *SE*  = 42.40±6.00 cm). Although the vegetation was similar overall, diked wetlands had a higher proportion of bare ground, which lead to a lower calculated mean height. ## Live Trapping During more than 3 500 trap nights, we captured small mammals using collapsible Sherman live traps (SFA, 7.62×8.89×23.50 cm; H.B. Sherman Traps Inc., Tallahassee, FL). We placed trap grids in vegetation types known to support large salt marsh harvest mouse populations, including pickleweed and Olney's threesquare bulrush. Traps were placed at 10 meter spacing within grids consisting of 50 or 100 traps, depending upon wetland configuration and mouse density. We placed traps flat on the ground in diked wetlands. In natural tidal wetlands, we placed traps above the water line, nested securely in the vegetation. We placed paperclips at the top of trap doors to leave a small space between the door and the body of the trap to prevent the tails of salt marsh harvest mice from getting caught when traps closed. We baited traps with ground walnut and mixed birdseed, and added cotton batting to the traps for warmth. Traps were opened at sunset, and checked and closed at sunrise. We trapped and tracked mice simultaneously in paired natural tidal and diked wetlands within each block during the same moon phases. At each block, trapping was initiated 4–5 days before the highest tide of the full and new moons. Tracking began 2–3 days before the highest high tide of the lunar phase, and typically continued until 2–3 days after the highest tide. ## Radio Tracking To maximize the potential effect of tidal inundation on movement, we tracked mice during several days surrounding the full moon, which is typically when the highest tides occur. To control for potential effects of bright lunar illumination during the full moon, we also tracked during the new moon when tides are also very high but lunar illumination is low. We tracked mice during sequential full and new moon phases for each block, with a different set of mice tracked during each lunar phase. This yielded a 2×2 factorial arrangement with four “treatments”: tidal wetland during the full moon, tidal wetland during the new moon, diked wetland during the full moon, and diked wetland during the new moon. To track movements of salt marsh harvest mice, we fitted individuals with BD-2NC radiotransmitters (Holohil, Inc., Carp, Ontario, Canada). Transmitters were equipped with the smallest battery possible, resulting in a total transmitter package mass of approximately 0.5 g. Because of the small size of salt marsh harvest mice (adult mass  = 7.6–14.5 g), only the largest individuals could be collared in order to meet the recommendation that collars not exceed 5% of body mass. Because females of sufficient size were commonly pregnant or lactating, we used only males weighing ≥10 g. We collared up to five of the largest individuals per wetland per lunar cycle (up to ten individuals per block per lunar cycle), yielding a total of 42 mice, 23 in the tidal wetlands and 19 in the diked wetlands. We collared mice in the field without using anesthesia and held them for observation for ten minutes following the collaring procedure; none showed any adverse reactions. Following the observation period, we released mice at their capture location and began tracking their movement during the second high tide following release (approximately 12 hours later). We tracked each individual for up to six days (mean; *SD*  = 4.3; 0.99 days, min  = 2 days, max  = 6 days), collecting location data approximately once an hour for 2 to 3 hours on either side of the peak of both diurnal and nocturnal high tides. When tracking was complete, we re-trapped mice and removed their collars. Due to some dropped collars and unexpectedly short battery lives, not all collared individuals yielded sufficient data for analysis. We used standard radiotracking methods to estimate individual locations, using triangulation from known coordinates, with two bearings recorded per location by two observers (a total of four bearings from different points were used to triangulate each location). By allowing the location of animals from a distance, triangulation minimized the influence of observer presence on mouse movement and reduced damage to habitat. It also allowed us to track mice in flooded areas that were inaccessible to humans on foot. Location estimates were generated using Lenth's technique for maximum-likelihood estimates (MLE) in the computer program LOCATE III. To ensure that mice retained their collars and were not caught in vegetation, we homed in on each mouse following each high tide. We then recorded location coordinates using a handheld GPS unit with submeter accuracy (Trimble GeoExplorer 3; Trimble Navigation Limited, Sunnyvale, CA, USA). Homing locations also allowed us to estimate the accuracy of triangulated locations. When a homing location was recorded within 15 minutes of a triangulated location, we calculated the distance between the homing location and the triangulated location to estimate triangulation error. The distance between triangulated and homed points was 15.51; 8.83 meters (mean; *SD*, N = 37). However, this error distance is probably an overestimate, because mice are capable of moving large distances in short periods of time, and because animals often move away from researchers during the homing process, which would increase the distance between triangulated and homed locations. ## Determining Refuge Use Our first goal was to determine whether salt marsh harvest mice moved vertically into emergent vegetation or horizontally into upland areas during the high tide. We examined mouse locations that were recorded within one hour of the high tide in relation to the tide height. We created maps of inundation for each tidal site using water surface data. To measure water surface level, we placed tide level markers throughout tidal areas within and around trap grids at 20 meter spacing, and measured the maximum height reached by water following each high tide. Tide markers were wooden dowels marked each centimeter and dusted with colored chalk. Tide height was estimated by examining dowels following high tide and noting the height to which chalk had been washed away. We recorded coordinates of all tide markers using a handheld GPS unit with submeter accuracy (Trimble GeoExplorer 3; Trimble Navigation Limited, Sunnyvale, CA, USA). We input all tide measurements as points into ArcMap 10 and interpolated a raster surface using an Inverse Distance Weight method (Inverse Distance Weight, ArcGIS 10.0; Environmental Systems Research Institute, Redlands, CA, USA) which overlaid well with aerial imagery and was consistent with observations in the field. Over this raster surface, we overlaid all mouse locations recorded “during” the high tide. Points occurring within one hour before or after the time of the highest point of the tide were categorized as “during” the high tide. If a mouse location fell in an area where water level was ≤1 cm we considered it “upland”. If a mouse location fell in an area where water level was \>1 cm it was considered “over water”. If a location was determined to be upland, we assumed the mouse moved horizontally to escape the tide. If a location was categorized as over water, we assumed that the mouse had remained in emergent vegetation to escape the tide. Refuge use was determined using the lowest high tide of both the full and new moon periods during which mice were tracked, thus represented the most conservative categorization of locations as over water. That is, locations were only considered to be over water if they would have been over water during the lowest high tide that we observed at a site. Locations that fell within 15 m of the boundary between inundated and unindated areas were excluded from analysis of refuge use. Because these “boundary” locations were within the average triangulation error distance, we could not definitively determine whether they were over water or not. We also performed an analysis using minimum convex polygons (MCP) of salt marsh harvest mouse locations to test for differences in space use between mice in tidal and diked wetlands. Individual locations were grouped as “before”, occurring more than one hour before high tide; “during”, within one hour before or after the high tide; and “after”, occurring more than one hour following the high tide. Using the Minimum Bounding Geometry-Convex Hull tool in ArcMap we created “before”, “during”, and “after” polygons for each mouse during each time period for which we recorded ≥6 locations. Then using the Intersect tool in ArcMap we calculated what percentage of the “before” and “after” MCPs overlapped the “during” MCP. ## Estimating Movement Rates Our second goal was to determine whether there were differences in movement distances between mice in natural tidal wetlands and mice in diked wetlands. However, because the time intervals between location estimates were not equal, it was necessary to first standardize movement data as rates. Movement rate is the Euclidean distance between two consecutive point locations divided by the time elapsed between those points. We calculated movement rates using all locations of an individual that were recorded less than three hours apart before, during and after high tide events (mean ±1 *SE*  = 83.94±1.40 minutes between successive points, min  = 21 minutes, max  = 180 minutes). Based on average times of sunrise and sunset during the study, we categorized movements occurring between 0601 and 2000 hours as “diurnal” and movements between 2001 and 0600 as “nocturnal”, and calculated average movement rates separately for day and night for each mouse. We omitted from analysis 18 movement rates that spanned the diurnal/nocturnal time cut-offs. The number of movement rates per mouse per time of day (diurnal vs. nocturnal) ranged from 1 to 17, with an average of 7.28; 3.87 (mean; *SD*) rates per individual per time of day (total movement rates  = 411). Mice with \<5 movement rates were excluded from analysis, leaving a total of 29 mice for which we analyzed movement rate data, 15 in tidal wetlands and 14 in diked wetlands. ## Analysis Statistical analysis of the MCP data was performed using paired t-tests to compare space use before and after the high tide to space use during the high tide. Statistical analysis of movement rates was conducted using PROC MIXED in SAS 9.3 (SAS Institute Inc., Cary, NC, USA). We used a replicated block design with blocks and the combination of wetland type and moon phase as fixed effects (four “treatments”: tidal full, tidal new, diked full, and diked new), and mouse identification number as a nested random effect, with time of day (diurnal or nocturnal) as a repeated factor. Tukey post-hoc tests were used to detect differences among treatments using an alpha level of 0.05 for statistical significance. Only locations from mice in natural tidal wetlands were used for the refuge use analysis, while locations from all mice with sufficient data were used in the space use and movement rate analysis. # Results ## Refuge Use For mice in natural tidal wetlands, we recorded a total of 167 locations within 1 hour of high tide from 16 mice (mean; *SD*  = 11.48; 4.02 points per mouse). Forty-two of these 167 locations (∼25%) were within 15 m of the boundary between areas that we considered “upland” versus “over water”. Because these locations were within our average triangulation error from the boundary, we could not classify them to either category with confidence, and they were eliminated from further analysis. Of the remaining 126 locations, 125 (\>99%) were definitively in vegetation over water and 1 (\<1%) was upland. It is worth noting that even if locations falling within 15 m of the boundary between inundated and unindated areas were included, only 6% (10/167) would have been categorized as “upland”. Because we used the most conservative definition of “over water” (e.g., that the position would have been over water during the lowest high tide during which mice were tracked), 6% is likely an overestimate of the frequency with which mice used upland refuges. When comparing the MCP of points for individuals before (t-test: t₁₅ = 1.327, P = 0.20; diked: n = 8, mean ± SE  = 0.61±0.09; tidal: n = 9, mean ± SE  = 0.44±0.09) and after (t-test: t₁₈ = −0.55, P = 0.59; diked: n = 10, mean ± SE  = 0.56±0.09; tidal: n = 10, mean ± SE  = 0.49±0.08) high tide to the MCP during high tides we saw no significant difference between tidal and diked wetlands. We found no evidence that salt marsh harvest mice in tidal wetlands shift their space use more or less than mice in diked wetlands during the same high tide periods. ## Movement rates Because mice remained over water instead of moving large distances to use upland refuges to escape high tides in natural tidal wetlands, it was not surprising that we found no significant difference in movement rates among treatments (combinations of wetland type and moon phase; \[repeated-measures ANOVA: F_3,24 = 0.92, P = 0.44\]). As expected for a primarily nocturnal species, we did find an effect of time: diurnal movement rates were lower than nocturnal movement rates (repeated-measures ANOVA: F_{1, 23} = 7.46, P = 0.01). Post-hoc comparisons of diurnal and nocturnal movement rates within treatments suggested that this difference was primarily due to diked habitats, in which diurnal movement rates during the new moon were considerably lower than nocturnal movement rates (Tukey post-hoc test: P = 0.012). In contrast, in natural tidal wetlands, diurnal and nocturnal movement rates were very similar. These data indicate that in natural tidal wetlands, mice are moving almost as much during the day as at night. # Discussion Understanding how animals cope with both naturally-occurring heterogeneity and HIREC in their environment can be crucial to the conservation of threatened species such as the salt marsh harvest mouse. Behavioral flexibility represents a coping mechanism likely to be crucial for many animals in the face of anthropogenic change in their environment. The purpose of this study was to determine how salt marsh harvest mice respond to both natural heterogeneity and HIREC. Following the framework of Berger-Tal et al., we answered two questions. First, we considered behavior-based management and asked *do salt marsh harvest mice move vertically (into tall vegetation) or horizontally (into upland areas) to escape the high tide*? Determining high tide refuge use by the salt marsh harvest mouse is a high conservation priority for this endangered species. We used radio telemetry to show that the vast majority of the time, male salt marsh harvest mice in the Suisun Marsh used emergent vegetation to escape tidal inundation, suggesting that maintenance of healthy intertidal vegetation is important in the conservation of small mammals and protection of biodiversity in this system. This conclusion is supported by extensive live-trapping data from a previous study that indicated that salt marsh harvest mice are only rarely trapped in upland areas, and is not surprising in the Suisun Marsh where vegetation is generally tall and thick providing extensive structure and cover. Our behavioral results will allow for improved habitat management, allowing managers to concentrate conservation efforts on maintaining intertidal habitat rather than expending efforts enhancing upland habitats that salt marsh harvest mice rarely use. Secondly, we addressed whether and how HIREC causes changes in behavior ; specifically, do salt marsh harvest mice in anthropogenically altered diked wetlands behave differently than those subject to tidal influence? The lack of a significant difference in nocturnal movement rates between natural tidal and diked wetlands would suggest that this anthropogenic change has not affected the behavior of this species. However, diurnal movement rates were consistently lower than nocturnal movement rates in diked wetlands, a pattern not observed in tidal wetlands. This result indicates that mice in human-influenced diked wetlands may have altered their diurnal movement behaviors in anthropogenically- modified diked wetlands, moving less during the day. Current telemetry efforts support these results and are revealing similar trends throughout the day, not just during high tides (K.R. Smith, unpublished data). There are two potential explanations for the observed decrease in movement rate in diked wetlands during the day. First, anthropogenic habitat modification may have had an adverse effect on salt marsh harvest mice. For example, differences in habitat structure or a lack of inundation may either allow for increased access by diurnal predators or decreased cover for mice, potentially leading to reduced diurnal movement rates in diked wetlands (mice are *forced* to seek refuge during the day). A second potential explanation is that removal of tidal influence has instead had a positive effect on salt marsh harvest mice. For example, mice in diked wetlands may have increased availability and/or accessibility to resources, or an increased ability to cache resources, which could allow mice to restrict their foraging to night-time hours, remaining in refuges during the day when predation pressure is high (mice are *choosing* to seek refuge during the day). This possibility is supported by our observations of salt marsh harvest mice using underground refuges, and taking cover in voids below a hummocks of cattails during the day in three separate diked wetlands, behavior that would not be possible in a tidal wetland. Habitat loss has been the primary threat to the salt marsh harvest mouse, and with the level of existing and continuing development in the San Francisco Bay Estuary, subsidence, and sea level rise, the marshes utilized by the salt marsh harvest mouse are vulnerable to additional habitat loss. Every remaining hectare of salt marsh harvest mouse habitat is crucial, and maximizing the value of this habitat through restoration and enhancement will be key to the persistence and recovery of the species. There are currently at least three plans that call for restoration of diked wetlands to tidal influence in the Suisun Marsh in the immediate future: Suisun Marsh Habitat Management, Preservation, and Restoration Plan, the Bay Delta Conservation Plan, and the Fish Restoration Program Agreement. Refuge use by salt marsh harvest mice is a behavior that must be taken into account when planning and implementing these tidal restorations. There is currently little research investigating the differences in value of various habitat types to the salt marsh harvest mouse. Our trapping was not designed to assess effects of wetland type on populations or densities of salt marsh harvest mice, but another study suggests that diked wetlands may be at least as valuable to conservation efforts as tidal wetlands, by supporting higher densities of mice. Additionally, large populations of salt marsh harvest mice have recently been found in diked wetlands, even during flooded winter months (K.R. Smith, unpublished data). This suggests that this species is capable of thriving in flooded diked wetlands by remaining in tall vegetation over standing water. Given that much of the remaining salt marsh harvest mouse habitat in the Suisun Marsh exists as diked wetlands, the behavior and ecology of individuals living in these modified habitats warrants further exploration. It is important to note that our findings may not necessarily apply to the entire range of the salt marsh harvest mouse. In the Suisun Marsh tidal vegetation is tall and marshes are wide, and there is an abundance of diked (21 044 hectares) and upland (11 210 hectares) habitat available, while only 21% of historic tidal habitat remains (2 550 hectares). The two other major strongholds for this species differ in marsh width and vegetation type: in San Pablo Bay many marshes are wide, but vegetation is short, and in the South San Francisco Bay marshes are narrow and vegetation is short. In these areas, mice may employ alternate strategies when escaping the tide; for example, if vegetation is shorter than the water height at high tide, mice may have no choice but to move upland where they may be exposed to predation. Finally, due to a number of limiting factors including minimum transmitter size and logistical constraints, we were unable to track female or young salt marsh harvest mice, so any speculation about their behavior should be made with care. Indeed, a recent study of another endangered wetland rodent, the New Mexico jumping mouse (*Zapus hudsonius luteus*), found different patterns of habitat use by males and females. As technology advances and radio collars become smaller, it would be prudent to repeat this study using both male and female mice, as well as sub-adults and juveniles. The results of this study provide insight to a question that researchers have been investigating for over 55 years: where do salt marsh harvest mice go during the high tide ? This study increases our understanding of the behavior of endangered salt marsh harvest mice in tidal and diked wetlands, suggesting that these small mammals may be capable of coping with the challenge of living in tidal wetlands with little to no emergent land as long as vegetation is sufficiently thick and tall. They also indicate that small mammals may exhibit different behaviors in diked wetlands than in tidal wetlands, highlighting the need to consider behavior when planning for conservation or management projects and shedding light on the need for comprehensive management strategies that account for potential behavioral differences in populations. Behavioral flexibility may be critical for threatened species in the face of HIREC. As we cause further changes to the environment both directly, through processes such as urbanization, and indirectly, though processes such as climate change, the work of behaviorists will be critical in understanding behavioral responses of threatened species, like the salt marsh harvest mouse, to HIREC. For invaluable field assistance we thank S. Estrella, D. Fidler, M. Riley, R. Pardee, L. Preteska, L. Patterson. For logistical support we thank S. Chappell and B. Wickland and the Suisun Resource Conservation District, and the California Department of Water Resources. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KRS LBT KEM. Performed the experiments: KRS LBT. Analyzed the data: KRS WG KEM. Contributed reagents/materials/analysis tools: KRS LBT WG KEM. Wrote the paper: KRS LBT WG KEM. [^3]: Current address: Wildlife, Fish, and Conservation Biology Department, University of California at Davis, Davis, California, United States of America

# Introduction Miniature legged robots are transportable, inconspicuous, and can pass through tiny openings and narrow corridors, which makes them excellent navigators for search and rescue missions at disaster sites. However, even state-of-the-art miniature legged robots have not yet matched the walking system of insects (ideal miniature legged systems in nature) in terms of power efficiency and motion controllability, despite many of them being inspired by the locomotion systems of living insects. What if a living insect itself could be controlled using an electrical stimulator? That is, what if we could create an insect–machine hybrid robot—a fusion of a living insect and a man-made device (electrical stimulator)—to elicit our desired motions and behaviors? Compared with the power consumption in entirely man-made miniature legged robots, which is on the order of 100–1000 mW, that of an insect–machine hybrid robot can be drastically reduced because electrical stimulation consumes power on the order of only a few hundred microwatts. Furthermore, insect–machine hybrid robots can be self-powered by energy harvesters, such as implantable biofuel cells, implanted into the living insect platform. In addition, these robots can be more robust in motion control than robots that are entirely man-made. Complicated algorithms and controls, which are necessary for entirely man-made robots to retain their postures or to avoid or overcome obstacles, would not be necessary for insect–machine hybrid robots because the insect's intrinsic control system can be utilized if needed. For example, when an insect–machine hybrid robot encounters an obstacle, the user shuts off the electrical stimulator and releases the insect from the user's control system to allow it to avoid or overcome the obstacle by itself. Hence, overall, insect–machine hybrid robots would exhibit high power efficiency and excellent motion controllability. Research related to insect–machine hybrid robots should be advanced from open- loop control systems to closed-loop control systems to allow these robots' legs to be regulated to follow the operator's predetermined motion path. Numerous research groups have investigated locomotion control or appendage motion response of various insects through electrical stimulation of insects' brain, ganglia, and nerve cords or muscles. They have developed protocols and methodologies of electrical stimulation of neurons and/or muscles to elicit desired motion and overall locomotion (e.g., left–right turns in walking or in flight) in an open-loop control (i.e., non-feedback control) manner. To date, researchers have focused on the development of stimulation protocols, i.e., determining what stimulation applied to which neuromuscular sites can elicit the desired motor actions and behaviors and evaluating the success rate and power consumption. Consequently, various stimulation protocols have been developed, but they have been demonstrated in an open-loop control manner. The questions arise as to what the next step should be in the research into insect–machine hybrid robots and how they can be practically used. Note that because of animals' intrinsically complicated motion control systems and the unavoidable physiological differences between individual animals, the elicited motions and behaviors vary from trial to trial and from animal to animal, even under identical stimulation protocols. As such, open-loop control techniques are insufficient for achieving precisely controlled motions; thus, closed-loop control must be introduced to reduce the deviations between the actually elicited motion and the user's predetermined and desired motion. Therefore, the next stage is the introduction of closed-loop control techniques to regulate insect legs to follow predetermined angular positions. When a closed-loop system is used with an insect–machine hybrid robot to control, for example, an insect leg actuator, the electrical stimulation must have two critical functions: (1) reliably control the direction of elicited leg displacement and (2) induce a graded response in the magnitude of the leg displacement to different electrical stimuli (e.g., different stimulation frequencies and amplitudes). Reliability in the direction of elicited leg displacement, as defined in this study, means that we can elicit leg motion in the desired direction at a success rate of almost 100%, even though the elicited magnitude of the leg displacement exhibits some variation. To achieve such reliability in the displacement direction, we stimulated leg muscles instead of neurons. Neurons are densely arrayed and stacked, which makes the separate stimulation of individual neurons difficult. Even if an implanted electrode is strongly fixed at the targeted neurons, small drifts of the electrode on the order of the size of a single neuron would be impossible to avoid, and even a tiny drift of the electrode could cause undesired motor action at unexpected muscles. Compared with neurons, muscles are easier to visually identify and are sufficiently large to be observed under a conventional optical microscope. Variation in the position of an implanted electrode within the target muscle and/or drift of the implanted electrode can cause small changes in the elicited motion magnitude but might not affect the displacement direction. As such, we expected to achieve a high success rate in eliciting leg motion in the desired direction by stimulating leg muscles. With respect to the other key operation needed for a closed-loop control, (2) graded response, some variables must be identified to grade the magnitude of leg displacement elicited by electrical stimulation. Such variables can be used as manipulated variables in a closed-loop control system. Suppose a leg is displaced by electrical stimulation of a leg muscle in the desired direction, but the magnitude of the displacement is smaller than our predetermined value. If we know a specific variable that governs the magnitude to be graded, such as a higher value of the variable inducing a greater magnitude of leg displacement, we update the electrical stimulation to generate a higher value of that variable and output (apply) the updated stimulation to the muscle. For many insect muscles, muscle contraction is enhanced by increasing the rate of neural input (monitored as a spike in an electromyogram (EMG)) or increasing the electrical stimulation frequency (summation or facilitation). Thus, the electrical stimulation frequency can be used as a manipulated variable to grade the magnitude of leg displacement and as an output from the final control element to the leg muscle in a closed-loop control system. We measured the displacement of leg motion elicited by various stimulation frequencies. We also measured the EMG of a leg muscle group and associated it with the leg displacement. We then confirmed the tendency for both a higher neural input rate and a higher stimulation frequency to elicit a larger-magnitude leg displacement, and consequently decided to use the stimulation frequency as the manipulated variable in our closed-loop control system. Overall, this paper reports the control of an insect's front leg motion by electrically stimulating multiple leg muscle groups in a closed-loop control manner. We successfully demonstrated a reliable biological microactuator with multiple degrees of freedom (DoFs). The three pairs of antagonistic muscle groups of the insect leg enable the leg to have three DoFs: protraction/retraction, levation/depression, and extension/flexion. The threshold stimulation voltage to elicit significant leg displacement in the desired direction was determined. We then observed that the electrical stimulation frequency is a variable that governs graded leg motion (i.e., the magnitude of leg angular displacement) by measuring the elicited leg angular displacement and velocity at various stimulation frequencies and the muscle EMG associated with the leg motion. We then developed a closed-loop control system in which the stimulation frequency was the manipulated variable from the final control element to the leg muscle and the angular displacement of the leg was the system response. This closed-loop control system can regulate the leg to set at predetermined angular positions. Because the muscle configurations are similar among all the six legs of an insect, the successful motion control of the front leg will aid in developing motion controls for all other legs and overall walking control in an insect–machine hybrid robot in the future. # Materials and Methods ## Study animal We used the beetle *Mecynorrhina torquata* (order Coleoptera; length: 62±8 mm; mass: 7.7±1.9 g for all the beetles used in the experiments. Unless otherwise stated, all data are represented as mean ± standard deviation) as our insect platform for a biological actuator. The beetles were kept in separate plastic terrariums (20 cm×15 cm×15 cm) with woodchips at the bottom. They were fed sugar jelly every 2–3 days. The temperature and relative humidity in the terrariums were maintained at 25°C and 60%, respectively. The use of this animal is permitted by the Agri-Food and Veterinary Authority of Singapore (AVA, HS code: 01069000, Product code: ALV002). Invertebrates, including insects, are exempt from ethics approval for animal experimentation according to the National Advisory Committee for Laboratory Animal Research (NACLAR) guidelines. ## Electrode implantation A beetle was anesthetized by placing it in a small plastic zip bag filled with CO₂ gas for 1 min. For immobilization, the beetle was subsequently placed onto a plastic or wooden plate and wrapped with dental wax (Cavex, Set Up Modeling Wax), which had been softened in hot water (80°C) for 10 s. Four small holes were made on its pronotum (positions indicated by red crosses) using an insect pin (Indigo Instruments, enamel-coated \#5). Using the same technique, eight more holes were made on the coxa and femur (positions indicated by red crosses). A thin Teflon-insulated silver wire (A-M Systems, 127 µm uncoated diameter, 178 µm Teflon-coated diameter) was used as the stimulating electrode. The insertion depth of the electrode was 2 mm from the outer surface of the cuticle. Both ends of the silver wire were heated in a flame to remove the insulation and enable electrical contact at the ends. ## Electrical stimulation To obtain a suitable threshold voltage, the non-implanted end of the wire was connected to the output channel of a function generator (Agilent, 33220A). The stimulation pulse width was fixed at 1 ms and the frequency was fixed at 30 Hz. The stimulation voltage was varied in increments of 0.25 V, starting at an initial stimulation voltage of 0.25 V. To investigate the elicited leg motion due to different stimulation frequencies, the three motion types of protraction/retraction, levation/depression, and extension/flexion were analyzed individually. For example, when investigating the protraction/retraction response to different stimulation frequencies, we restricted the levation/depression and extension/flexion motions by inserting an insect pin into the corresponding articulation. Two markers placed on the beetle's leg were recognized by a 3D motion capture system as a solid line segment, and the third marker placed on the beetle's body indicated the beetle's position. The 3D motion capture system recognized and stored the *X*, *Y*, and *Z* coordinates of all markers. Angular displacement was determined using the following formula for calculating the angle between two vectors:where *X*₁, *Y*₁, and *Z*₁ and *X*₂, *Y*₂, and *Z*₂ are the initial coordinates of markers 1 and 2, respectively, and *X*₁′, *Y*₁′, and *Z*₁′ and *X*₂′, *Y*₂′, and *Z*₂′ are the coordinates of the two markers as a consequence of the beetle's leg motion. Therefore, all angular displacement values were calculated with respect to the leg's initial (resting) position. The position of each leg segment at rest (before electrical stimulation) was defined as the initial position (the initial position varies from beetle to beetle; the variation is on the order of a few degrees). Each time after the leg muscle was stimulated, we manually positioned the leg to its initial position by checking the 3D coordinates of the markers placed on the beetle's leg. For all experiments, the stimulation voltage was fixed at 1.5 V and the pulse width was fixed at 1 ms. The electrical stimulation power consumption was measured for all six muscle groups present in the beetle's front leg. Current flow through a muscle was measured using an oscilloscope (Yokogawa, DL 1640), and a function generator (Agilent, 33220A) was used to supply a positive pulse train at 100 Hz, 1.5 V, and a 1-ms pulse width. ## Measurement of leg-muscle EMGs synchronized with leg motion A pair of thin silver wires (A-M Systems, 127 µm uncoated diameter, 178 µm Teflon-coated diameter) were implanted into the muscle group of interest using the technique described in the electrode implantation section. The electrodes were glued to the outer surface of the cuticle with dental wax (Cavex, Set Up Modeling Wax) to avoid potential artifacts due to the electrodes' drift. The EMG signals were amplified 500-fold using an amplifier (LT1920, Burr-Brown Products). A custom-programmed wireless microprocessor (Texas Instruments, CC2431, 6×6 mm², 130 mg, and 32 MHz clock) was used to collect the EMG signals from the muscles at a sampling rate of 2000 Hz. The input/output (I/O) pins of the microprocessor were set at inputs so that EMG signals from the muscles were collected as input electrical potentials. The beetle's leg motion was captured in the same manner as that described in the previous section (“Electrical stimulation”) using the 3D motion capturing system. We developed a customized software tool, BeetleCommanderEMG, which can simultaneously collect and store EMG signals from the microprocessor and the beetle's leg motion information from the 3D motion capturing system. The threshold voltage of the EMG signal was determined individually for each experimental result to identify the maximum number of EMG spikes captured. The EMG burst onset time was defined as the time at which the voltage of an EMG spike exceeds a certain threshold value. The EMG burst termination time was defined as the end time of the last detectable EMG spike. The mean EMG frequency was calculated as the average of the instantaneous frequencies within a single burst. The average angular velocity was calculated as the linear regression slope of the angular displacement during the time interval of motion. The onset time of leg motion was defined as the time of first detectable leg retraction motion. The motion offset time was defined as the beginning motion of the first detectable protraction motion. ## Closed-loop control system We developed a closed-loop motion control system to be introduced into BeetleCommander. An electrical stimulation signal was generated using a custom- programmed microprocessor (Texas Instruments, CC2431, 6×6 mm², 130 mg, 32 MHz clock). Electrical stimulation signals generated from two separate stimulation channels were used to control one pair of antagonistic muscle groups. The BeetleCommander system could extract instantaneous marker position information from the 3D motion capture system and calculate the immediate leg angular position. Update time intervals (i.e., the time interval at which the closed-loop system updated the instantaneous leg position and output stimulation frequency) were user-adjustable. The concept of proportional control was also used to adjust the magnitude of a step increment or decrement of the stimulation frequency:where is the last output stimulation frequency from the final control element, is the updated output stimulation frequency from the final control element (its value is limited to the 10–200 Hz range), is the proportional gain (user-adjustable), and is the instantaneous angular displacement error at time *t*. Protraction/retraction closed-loop motion control was used as an example of operation, which was controlled by two separate channels from the final control element. Electrical stimulation signals from the two channels were generated on the basis of predetermined angular positions set by the user. Both stimulation channels operated independently and concurrently. The electrical stimulation frequency from one channel was initially increased to elicit the leg to move to the desired position. If the actual leg angular position was greater than the predetermined angle, the stimulation from the current working channel started to decrease its stimulation frequency while that from the counter channel began to decrease the angular position by increasing the stimulation frequency. Likewise, if the actual angular position was less than the predetermined angle, the counter channel stopped generating its stimulation signal and the other working channel began to increase its stimulation frequency to increase the leg angular position. Therefore, the two channels used to stimulate the pair of antagonistic muscle groups operated simultaneously to ensure that the beetle's leg followed the predetermined angular positions. # Results and Discussion ## Threshold stimulation voltage to elicit leg displacement We fixed the stimulation frequency at 30 Hz and used a 1 ms monophasic pulse train with amplitudes ranging from 0.25 V to 2.50 V. shows the elicited protraction/retraction angular displacements at various stimulation voltages (number of beetles = 5 and 17≤ number of data points at each stimulation voltage ≤22). The threshold voltage required to elicit a beetle's leg movement was approximately 1.0 V, and the maximum angular displacement was reached at approximately 1.5 V (maximum protraction/retraction angle at 1.5 V = 18.09°±4.87°/12.73°±7.04°). When the stimulation voltage exceeded 1.5 V, the maximum angular displacement for both protraction and retraction remained relatively constant. For optimal results, we had to set the stimulation voltage as low as possible to minimize any possible damage to the beetle's muscle while simultaneously ensuring that the stimulation voltage was sufficiently high to reliably elicit the desired leg motion. Thus, we fixed the stimulation voltage at 1.5 V for all subsequent experiments. With a 1.5 V stimulation voltage, the success rate for inducing leg movement in the desired direction was 100% (number of beetles = 42). In addition, after repeatedly applying the electrical stimulation to a single muscle group more than 200 times within one day during a single experiment, we observed no obvious indications that the beetle's muscle was damaged by the 1.5 V stimulation voltage. ## Stimulation frequency as a variable to grade the leg displacement magnitude For all experiments, the stimulation voltage was fixed at 1.5 V and the pulse width was fixed at 1 ms. The elicited leg motion at various stimulation frequencies was studied. As shown in, the resulting angular displacements for all six motion types monotonically increased with the stimulation frequency (number of beetles = 10). The angular displacement of the beetle's leg reached its limit when the stimulation frequency exceeded approximately 80 Hz for the other five motion types, with the exception of levation motion (the maximum angular displacement for levation was reached at stimulation frequencies of approximately 40 Hz). This maximum angular displacement might be due to the mechanical limitation of the beetle's leg structure. Similarly, the average angular velocity also increased monotonically with the stimulation frequency. When the stimulation frequency was greater than 250 Hz, the average angular velocity of the leg motion reached the maximum value for the other five motion types, with the exception of levation motion (, maximum average angular velocity of levation was reached at a stimulation frequency of approximately 100 Hz). ## Comparison between muscle stimulation and sensory system stimulation As discussed in the, the performance of a closed-loop control system in stimulating a muscle is compared to that of a sensory system (e.g., a system stimulated by an antenna or compound eye) on the basis of two key performance requirements: (1) reliability in the direction of elicited leg displacement and (2) a graded response in the leg displacement magnitude. As demonstrated in the previous two sections, our leg-muscle stimulation satisfies both these key requirements. However, numerous researchers have demonstrated various protocols of stimulating sensory systems to control insect locomotion in walking and in flight. For instance, Holzer and Shimoyama stimulated the antennae of a cockroach (*Periplaneta americana*) to control its walking direction. Transient (100–200 ms) turning behavior was observed when the ipsilateral antenna of a walking cockroach was stimulated. However, the electrodes used to stimulate sensory systems are not well secured, and the two aforementioned performance requirements are not guaranteed. Unlike muscle stimulation, the stimulation of sensory systems does not result in a 100% success rate in eliciting the desired motor action or behavior. A graded response in a target muscle by stimulation of sensory systems is possible and has actually been demonstrated, however, this approach is relatively difficult and less reliable than muscle stimulation, as discussed in the. Overall, the adaptation of muscle stimulation in closed-loop motion control is preferred over the adaptation of sensory system stimulation. ## Comparison between the leg motions elicited by electrical stimulation and those by intrinsic neural input To understand how effectively our electrical stimulation (synthetic input to muscle) mimics an animal's intrinsic neuromuscular system (natural input to muscle), we recorded the EMG signal of the retraction muscle group during natural leg motion and synchronized the EMG with the leg displacement. A decrease in the angular displacement corresponds to a retraction motion in the figure. shows the average retraction velocity as a function of the mean muscle EMG frequency (number of beetles = 4, total number of data points = 43). The maximum EMG frequency recorded was approximately 70 Hz. A linear relationship was observed between the angular velocity and the mean EMG frequency. shows a plot of the average retraction velocity of the beetle's front leg as a function of the electrical stimulation frequency (number of beetles = 5, number of data points at each stimulation frequency = 25). Another linear relationship is observed in this figure, which indicates that a higher stimulation frequency resulted in higher angular velocity and hence elicited greater muscular force. Despite the expected result that the angular velocity monotonically increases with both average muscle EMG frequency and electrical stimulation frequency, the slope of the linear regression line for the former case is approximately five times steeper than that for the latter case; that is, provided that the EMG frequency is the same as the electrical stimulation frequency, the resultant leg motion is five times faster when the beetle moves voluntarily compared with when it is electrically stimulated. This difference might be due to the basic differences between the EMG signal (natural neural input) and the electrical stimulation signal (synthetic input) in terms of amplitude and signal forms. In addition, because the beetle might have resisted the electrically elicited motion by activating the antagonist muscles and produced an opposing force, the angular velocity of leg motion elicited by electrical stimulation could have been smaller than that expected, on the basis of the tendency of the motion associated with the EMG spike frequency. ## Demonstration of a closed-loop control system for insect leg motion As each muscle group was stimulated by independent, isolated outputs from the stimulator board, we could elicit the individual leg motion types either separately or simultaneously, as demonstrated in and. Moreover, on the basis of the findings that the leg angular displacement monotonically increased with respect to the stimulation frequency, we developed a closed-loop control system to make the leg move according to preset angular positions. shows typical closed-loop control results (number of beetles = 5) for protraction/retraction motion; these results were obtained by comparing the actual leg movement achieved (blue path) with the predetermined leg angular position (red path) at *K*_p values of 0.1, 0.5, 1.0, and 1.5 and at update time intervals of 100, 200, and 300 ms (the update time interval is the time interval at which the closed-loop system updates the instantaneous leg position and output stimulation frequency). shows the overshoot and reaching time of the closed-loop control experiment (number of beetles = 5, 35≤ number of data points at each experiment setting ≤49). The statistical information for the experiments in which *K*_p = 0.1 is omitted from because the rate of increase of the muscle stimulation frequency is not always sufficient to bring the leg to the predetermined angular position due to the relatively small *K*_p value (see, *K*_p = 0.1, update time interval = 100, 200, and 300 ms). Of the 243 experimental data points for *K*_p = 0.1, the beetle's leg reached the predetermined angular position 104 times (42.8% success rate). However, for *K*_p = 0.5, 1.0, and 1.5, the beetle's leg was always (100% success rate) brought to the predetermined angular position. As evident in and and, when the *K*_p values were increased, the leg response overshoot generally increased, whereas the reaching time (i.e., the time required to reach the predetermined angular position) decreased. For example, when the *K*_p value was increased from 0.5 to 1.5 using an update time of 100 ms, the protraction overshoot angle increased from 10.47°±3.66° to 22.12°±7.75° and the retraction overshoot angle increased from 11.03°±5.06° to 17.48°±4.94°; in contrast, the protraction reaching time decreased from 0.518±0.133 s to 0.299±0.188 s and the retraction reaching time decreased from 1.249±0.917 s to 0.488±1.111 s. This tendency was consistent because larger *K*_p values would result in greater changes in the output stimulation frequency (either increases or decreases) for the same instantaneous angular displacement error. This larger output stimulation frequency change would then elicit a greater angular displacement and a greater angular velocity of the beetle's leg (i.e., a larger force elicited in the muscle). As a result, the beetle's leg would move at a faster rate (decreased reaching time). However, at the same time, the beetle's leg would be more likely to exceed the predetermined angular position (increased overshoot). In general, as the update time interval was incrementally changed, the leg response overshoot decreased, whereas the reaching time increased (comparison across different *t* values, and). For example, for *K*_p = 1.0, as the update time was increased from 100 to 300 ms, the protraction overshoot angle decreased from 17.51°±5.79° to 16.01°±8.39° and the retraction overshoot angle decreased from 15.97°±5.38° to 6.52°±4.34°, whereas the protraction reaching time increased from 0.350±0.184 s to 0.564±0.150 s and the retraction reaching time increased from 0.492±0.217 s to 0.875±0.308 s. This tendency was consistent because a larger update time interval would result in a longer required period for the beetle's leg to respond to the electrical stimulation. This longer response period decreased the likelihood that the 3D motion capture system would capture the leg position before it reached its final position. Therefore, this longer response period decreased the likelihood that the BeetleCommander closed-loop control system would overly increase or decrease the stimulation frequency, which would result in reduced overshoot. However, at the same time, with an increasing update time interval, the reaching time would increase. The statistically obtained reaching times and overshoots at the different system settings (the proportional gain and update time interval) reveal certain constraints and limitations in the design of insect–machine hybrid legged robots in terms of the step cycle of the leg in the walking gait. We can refer to the reaching-time data to determine the appropriate stepping frequency (e.g., the duration of each step should be longer than the combined reaching time of all motions involved in that step). Given the angular overshoot, we can estimate the step-length error present in a given closed-loop control system. For example, when *K*_p and the update time interval *t* are set to 0.5 and 300 ms, respectively, the overshoot for retraction motion is 5.05° on average. This overshoot angle of 5.05° results in an estimated step-length error of approximately 0.22 cm if the beetle's leg length is assumed to be 2.50 cm. As such, even for future advanced close-loop control systems, we can refer to these indices (e.g., the reaching time and overshoot) to reduce the constraint and limitation of the step cycle and other relevant parameters in insect–machine hybrid legged robots. ## Potential experimental errors The use of living organisms in experiments can introduce several unavoidable errors, which is why closed-loop motion control is necessary. For example, the neutral or resting position of the leg differs within a few degrees from beetle to beetle. Although some researchers have defined the neutral position of a joint in terms of the angular position between the two leg segments (e.g., Guschlbauer et al. defined the neutral position of the femur-tibia joint of the stick insect as 90°), ensuring that the leg appendage of a living insect rests at the neutral position, as predefined by us, is difficult. The CO₂ anesthetization of the beetle before implantation of the stimulation wires (see the Electrode implantation section) can affect the leg response. Several side effects of CO₂ anesthetization on insects have been reported. CO₂ anesthetization can increase haemolymph acidity and cause the heartbeat to stop. As a result, exposure to CO₂ can impair oxygen delivery to tissues, thereby reducing oxidative phosphorylation and adenosine triphosphate (ATP) production in cell mitochondria. These effects can significantly influence the efficacy of a bio-actuator and are known to affect insects' locomotion. Nonetheless, CO₂ exposure is one of the most popular anesthetic methods in entomological research, even though its side effects have not yet been fully elucidated. Improved anesthetic methods might be helpful in future entomological research. Despite these potential experimental errors, the magnitude of the leg displacement is undoubtedly increased by increasing the stimulation frequency. Inspired by this graded response to the stimulation frequency, we successfully developed a closed-loop control system to regulate the leg to set at predefined angular positions; this controlled response is independent of the potential experimental errors in leg response because closed-loop control systems, in general, reduce such errors. ## Power consumption We confirmed that the power consumption of the insect leg actuator was remarkably low (on the order of 100 µW to a few milliwatts). shows the positive pulse train at 100 Hz, 1.5 V, and a 1 ms pulse width used as the muscle stimulation signal. shows a typical current-flow profile through the depression muscle group. The power consumption of electrical stimulation on all six muscle groups of the front leg was measured across five different beetles. shows the detailed numerical values of the mean and standard deviation in power consumption (number of beetles = 5, number of data points collected from each muscle group = 40). The average power consumption of the electrical stimulation of a single muscle group was 148 µW. Assuming that the stimulations of the middle and hind legs consume similar amounts of power, the power consumption of an insect–machine hybrid legged robot using a beetle is approximately 5.3 mW in the worst-case scenario, where all six muscle groups are stimulated simultaneously for all six legs (stimulation of 36 leg muscle groups). Note that the worst case scenario should be far from the actual case because just half of the muscle groups would be simulated simultaneously in actual insect walking control. Nonetheless, this power consumption is considerably low compared with the 100–1000 mW order of power consumption in entirely man-made miniature legged robots. # Conclusions The experimental results demonstrated that electrical stimulation with a threshold voltage of 1.5 V elicited significant displacement of the leg in desired directions (three DoFs, i.e., protraction/retraction, levation/depression, and extension/flexion) at a 100% success rate. The magnitude of the leg displacement was graded by the stimulation frequency: a higher stimulation frequency elicited a larger-magnitude displacement. We used the stimulation frequency as the manipulated variable in our closed-loop control system, and the controlled leg was successfully set at predetermined angular positions. In conclusion, coupled with the low power consumption compared with that of entirely man-made legged robots, the ability to regulate a beetle's leg motion under a closed-loop control system should contribute significantly to the future design of biological actuators and hence biological legged machines (i.e., insect–machine hybrid legged robots). Note that the leg-muscle configurations are common or similar (i.e., a pair of antagonistic muscle groups dominating the leg displacement in opposite directions) among many insect orders. In addition, various stimulation protocols to elicit leg displacement in the desired direction have been proposed and demonstrated for various insect orders, including moth, stick insect, and locust. Therefore, the methodology and experimental design demonstrated in this paper may be applicable to the development of a closed-loop control of the leg motion of other insect orders. # Supporting Information The authors offer their appreciation to Mr. Cheo Hock Leong, Mr. Ow Yong See Meng, Ms. Chia Hwee Lang, Mr. Poon Kee Chun, and Mr. Roger Tan Kay Chia at the School of MAE, NTU. The authors thank Professor Ryohei Kanzaki (University of Tokyo) and Professor Jiro Okada (Nagasaki University) for their helpful advice and Professor Kris Pister (UC Berkeley) for providing the wireless communication device. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: FC CZ TTVD YL DHS JSK NAH KI PA MMM HS. Performed the experiments: FC CZ TTVD YL DHS JSK NAH MFBA HS. Analyzed the data: FC CZ TTVD DHS JSK NAH MFBA HYC HS. Contributed reagents/materials/analysis tools: FC CZ TTVD YL DHS JSK NAH MFBA HS. Wrote the paper: FC CZ TTVD MMM HS.

# Introduction Half of UK workers are office based. Office workers spend approximately two- thirds of the working day in seated tasks and are estimated to accumulate around 10.5h of sitting per waking day. Growing evidence has associated prolonged sitting, characterised by very low energy expenditure (1.5 metabolic equivalents or less), with increased disease and premature mortality risk. Sedentary office work is an urgent public health issue. There is some evidence to suggest that some of the deleterious health effects of sitting time can be offset by engaging in moderate physical activity for at least one hour daily. However, based on self-report data–which typically overestimate true activity levels–a third of the UK population fails to meet the more modest target of 150 minutes of moderate activity per week. It may be unrealistic to expect office workers who are highly physically inactive and highly sedentary to adopt daily bouts of activity lasting an hour or more. Displacing sitting with standing may offer a more feasible sitting-reduction strategy. Standing at work is often proposed as a way of reducing sitting. Standing has been linked to lowered mortality rates, likely due at least in part to greater energy expenditure, reduced glucose variability and oxidative stress. While there are no government guidelines offering targets for standing and sitting time in the UK, a recent expert-consensus document recommended that office workers accumulate 2-4h of standing and light activity daily, and take regular breaks from prolonged sitting. Achieving this target will require developing strategies to displace sitting with standing. While the most commonly-evaluated strategy has been to restructure the environment to facilitate desk work while standing or moving, another commonly-proposed strategy is to promote standing in normally-seated meetings. There are multiple routes through which standing in meetings might be widely adopted within the workplace. At the organisational and environmental levels, managers could implement initiatives that explicitly support standing meetings among employees, such as introducing standing-permissive meeting room furniture, or enshrining standing meetings into workplace practices and procedures. Indeed, standing meetings are commonplace in the software development sector, where they are used for short daily team briefings. At the individual level, employees might drive organisational change by voluntarily choosing to stand in meetings, with the aim of normalising standing in typically-seated contexts. Such a strategy, however, depends on the acceptability of standing in meetings to those who attempt it. Interview studies suggest that office workers believe standing in meetings to be an acceptable sitting-reduction strategy *in principle*, but few studies have documented *lived experiences* of attempting to stand in normally-seated meetings. The scant available research into standing meetings to date has focused on responses to the topic matter and meeting frequency, not the standing experience itself. Initial experiences of novel behaviours are important determinants of maintenance; the sustainability of standing in meetings where all others are seated will depend not only on prior expectations, but also whether experiences match expectations. Employees are unlikely to continue to stand in seated meetings where they find the experience intolerable or unsuitable. This paper reports the first study to our knowledge to examine employees’ experiences of standing in seated meetings using qualitative data. Understanding how employees experience standing in meetings, and the organisational, social and psychological structures that surround such experiences, will help to inform development and implementation of worksite standing interventions that go beyond the provision of sit-stand workstations. Our study focused on university workers. Universities are complex organisational structures offering diverse office environments, with workers from across the socioeconomic spectrum, so university workers’ experiences may potentially be relevant to many other office workers and settings. Our research question was: “how do office workers experience standing in normally-seated meetings?” # Methods ## Participants Participants were recruited from three UK-based universities, between January- April 2016, using convenience sampling methods. Recruitment was conducted using print and online advertisements through university communication channels (internal email, poster, staff newsletter, Twitter). Participants were offered a £50 (\~US\$65) voucher for taking part. Potential participants returned by email an expression of interest form, and self-reported their demographics (work/job role, gender, age, disability, ethnicity, income;) and eligibility, via items assessing the following criteria. Eligible participants were: desk-based employees; aged 18+ years with no intention to leave the organisation before July 2016; able to identify three meetings, differing in size, with potential for standing; and willing to be observed at meetings and interviewed. Those who had engaged in standing up in meetings in the past 3 months, used a sit-stand desk in the past 3 months, were full time students, or were unable to stand, were excluded. Those meeting eligibility criteria took part in a face-to-face, telephone or email inception meeting to clarify study requirements and agree meeting dates. Due to deadlines imposed by the broader project within which this study was located, a six-month window was set for study planning, data collection and analysis, which included three months for recruitment. Twenty-seven participants were recruited during this time, of whom two (from Workplace A) dropped out, citing insufficient time for participation, after the first meeting. The final sample thus comprised 25 participants (7 from Workplace A, 10 from Workplace B, 8 from Workplace C). Of these, six (24%) were male, 18 (72%) female, and one (4%) self-declared as ‘other’. The modal age category was 25-29y. Twenty-three (92%) participants had degree level or higher qualifications, and two (8%) had technical or professional qualifications. ## Procedure and interview schedule As part of the pre-study information sheet, participants were given minimal information on the potential health benefits of standing (“public health researchers have suggested that office-based workers should stand up in meetings, to promote health”), and were told that “the study aims to explore what it is like for office-based employees to stand in meetings”. Participants were asked to select three group workplace meetings that they had been due to attend irrespective of study participation, and which differed in size (small: 3–10 attendees; medium: 11–19; large: 20+), such that each participant attended one meeting of each size (i.e., three meetings in total). We determined three meetings to be sufficiently conducive to variation in experiences for each participant, and feasible within the study period. We instructed participants to stand whenever they felt they wanted and for durations decided by them. No further instruction was given. Except for five meetings (covering three participants) to which access was denied to non- invitees, a researcher attended each meeting to observe participants’ standing behaviour, others’ responses, room layout and number of attendees, though these field notes were not deemed sufficiently rich for analysis. A semi-structured interview (face-to-face or telephone) was conducted as soon as possible (and no longer than 48 hours) after each meeting (i.e. 3 interviews per participant), to gather reflections on experiences of standing. One participant (Amelia) attended her third meeting immediately following the second meeting, and so completed one interview addressing both meetings. Interviews took place within the meeting venue or another setting within the participant’s workplace, according to participant preference. Nobody else was present during interviews. An interview schedule was developed to cover the fundamental determinants of standing (capability, opportunity, motivation), and piloted among three office- based colleagues of the researchers. In all interviews, questions focused on affective reactions to standing; others’ reactions; standing location, timing and duration; occupational identity and status; workplace culture and norms; and acceptability and feasibility of standing in meetings. In the third interview, views towards workplace standing interventions were also sought. Each interview was conducted by one of three female researchers, one of whom (JH) was a Social Sciences doctoral student with previous experience of interviewing in the workplace sedentary behaviour research domain, and two (MR, ER) were Health Psychology Masters students trained in interviewing by the senior authors (LM, BG). Each participant was interviewed by the same interviewer on three occasions. The only interviewer-participant contact prior to the first interview was via email or telephone, for the purposes of organising the first interview. Participants were told, prior to the interview, that the interviewer was conducting the study as part of a funded research project (JH), or as coursework (MR, ER). No other interviewer characteristics were shared with participants prior to the first interview. Interview duration ranged from 8-32mins (mean 20mins). The study was approved by the Research Ethics Committee of each institution (LRU-15/16-2533; 4385/002; 1875-LR-Jan/2016-1200). All participants provided fully informed written consent to participate. ## Analysis Interviews were digitally audio-recorded and transcribed verbatim, with transcript quality and completeness verified by the researchers. We offered to return transcripts, and our ongoing or completed analyses, to participants for comment, correction, or any other purpose, but none expressed a wish to receive them. Interview transcripts and analysis were managed via NVivo 10 software. A phenomenological methodological orientation was adopted, as this allows for description (rather than quantification or explanation) of participants’ experiences, and exploration of the subjective meanings that they ascribe to elements of these experiences. Data were analysed using Framework Analysis, which allows inductive co-creation between multiple researchers of an initial coding framework which guides subsequent analysis, and which is developed and iteratively refined as coding progresses and new insights emerge. Repeated and independent reading of a sample of interview data from one institution was undertaken by two researchers (JH, LM), to develop a preliminary coding framework. Next, selected interview transcripts (n = 9/25) were read and re-read by four other researchers (MR, LS, ER, BG), who refined the coding framework to ensure its accuracy and relevance. Discrepancies were resolved by discussion, and resultant codes and themes verified by all researchers. The coding framework was used (by JH, LM, and BG) to analyse the remaining data. Given that data collection was constrained by the study timeline, data saturation was not discussed prior to analysis. Quotes are provided below as evidence of the validity of our analysis. All participants were assigned pseudonyms. Punctuation was added to unambiguous quotes and where necessary, words added in parentheses to clarify intended meaning. # Results Four themes were derived from the data: physical challenges to standing; implications of standing for meeting engagement; standing as norm violation; and standing as appropriation of power. The first theme details experiences of physical discomfort and of attempts to negotiate the physical environment to permit standing. The second theme reports different ways in which standing impacted on participants’ involvement in meetings. The third and fourth themes describe discrete psychological challenges involved in negotiating the organisational and inter-person context that frames workplace meetings. ## Theme 1: Physical challenges to standing Some participants were aware of the detrimental impact of sitting and expected that standing in meetings would confer health benefits (“*you just perceive being standing up*, *it’s just better for your health and it just makes you feel better overall”*; James, Workplace A). Yet, the physical experience of standing in meetings rarely mapped onto participants’ expectations. Some reported unexpected and unaccustomed discomfort from physical sensations in the muscles of the feet, legs, back and shoulders, while others anticipated but did not experience physical discomfort: > *I just thought, oh your back will ache or legs ache or something, but > actually it wasn’t borne out in reality*. > > (Tom, Workplace B). The physical impact appeared partly due to the time spent standing, which was self-determined by participants. Some expected to be able to stand for the entirety of lengthy meetings, but on attempting to do so realised this was not possible, due to physical discomfort: > *After about twenty-five minutes … \[I was\] thinking oh my back is > killing me! There’s the realisation that, oh I can’t stand for very > long*! > > (Joan, Workplace B) Aspects of the physical environment–furniture design, spatial configuration, and the numbers and positions of attendees–also presented barriers and challenges to standing. Participants’ descriptions of their meeting room environments suggested that there were many chairs, but no standing areas, nor desks and tables to support standing (a “*lack of furniture for standing”*; Angela, Workplace C). The physical environment was felt to elicit sitting (*“the \[physical\] environment … kind of shapes social expectations \[about standing\] … we sit so much”*; Angela, Workplace C), which in turn reinforced perceptions that standing in meetings was neither acceptable nor feasible: > *Today, with the chairs all round, already set up, it makes it more of > a barrier to actually saying I’m going to stand*… *it’s so easy just > to go, oh I’m going to sit down* > > (Tom, Workplace B) A lack of perceived environmental support for recording information when standing posed practical difficulties for several participants, making engaging in the meeting non-ergonomic and potentially physically uncomfortable: > *Bending down and trying to take notes, it didn’t feel natural*. *It > was just a question of the desk or table not being at a certain height > really* > > (Charlie, Workplace B) Suggestions were made for overcoming barriers presented by ill-suited physical environments, including furniture adaptation, or using accessories to permit usual meeting engagement while standing; for example, using a tablet instead of pen and paper for note-taking. ## Theme 2: Implications of standing for meeting engagement Many participants reported that standing affected their engagement in the meeting. Some found that physical discomfort from standing motivated them to increase engagement to minimise meeting length: > *Because I was stood … \[I was\] kind of more, more ‘let’s get on with > it, let’s get to the point’. \[…\] Because it’s not so relaxed as > being sat back in a chair … the perception is the meeting is going on > for longer, or more frustration if there’s no action being taken. > \[…\] It’s more efficient*. > > (Tom, Workplace B) Others felt that shifting from a relaxed, seated position to standing sustained focus on the meeting, because standing prevented them from “*switching off*” (James, Workplace A) as they would while sitting. Many reported, at least in the first of the three meetings in which they stood, unanticipated feelings of psychological discomfort from standing while all others were sitting, together with a heightened awareness of the self, others, and the interpersonal context. Participants variously felt *“disconcerted”* (Joan, Workplace B), *“awkward”* (Brianna, Workplace C), or *“stupid”* (Charlie, Workplace B), possibly due to being more visible than others. Some reported that enhanced visibility made them feel more accountable to others: > *I tend to drift out or find them a bit boring. \[…\] Because I was > both at the front of the room and standing, I felt much more like I > had to, even if I wasn’t engaged, look like I was more engaged, which > then made me more engaged. So I actually listened to the whole thing! > \[…\] If I started drifting out … I’d then get really self-conscious > and think, oh God, what if somebody saw me with my glazed-over eyes or > something like that*! > > (Anne, Workplace C) For others however, self-conscious thoughts were a distraction from the meeting: > *When I did sit down, I was like (sighs). More relaxed. I could just > focus on the meeting, not focus on my standing*. > > (Brianna, Workplace C) Some participants sought to minimise psychological discomfort by standing in a position within the room that they perceived to be less visible, so avoiding obstructing others: > *I positioned myself right towards the back in the corner… it was OK > to stand because I wasn’t making a nuisance of myself to anybody, I > wasn’t in any body’s way*. > > (Anusha, Workplace A) However, physically removing themselves from others was experienced by some as isolating, and potentially limited involvement in interactions within the meeting: > *Standing up made me … \[feel\] like I wasn’t part of the group. … > \[Once I sat down\] I felt like I was then part of the meeting. And it > felt more like we were a team group coming to some decisions and > stuff, because we were all on the same eye level*. > > (Alisha, Workplace C) > > *I got missed out on the signing register. I didn’t draw attention to > myself because I wanted someone to notice me and give it to me, and > they didn’t*. > > (Selma, Workplace C) Some felt that other attendees were preoccupied with their standing, which might have compromised the engagement of others, and limited overall satisfaction with the conduct of meetings: > *They were kind of looking at me instead of looking at the director… I > think I just felt like I was being a distraction for them \[others in > the meeting\], I felt like I was taking away from the meeting*. > > (Brianna, Workplace C) Some were concerned that standing could be interpreted by others as unwillingness to fully engage in the meeting, and indeed, a minority of participants reported being asked by others to sit for this reason: > *\[In a meeting\] you expect someone to sit down and then \[if\] they > don’t you think, are you not staying or do you not really want to have > the meeting*? > > (Alisha, Workplace C) > > *She \[the chair of the meeting\] goes: “it’s really distracting with > you standing up. It feels like you’re getting ready to go, could you > sit please?” So then I sat*. > > (Brianna, Workplace C) Many felt it particularly inappropriate not to sit in formal meetings, or those addressing sensitive topics, as it risked diminishing the seriousness of the meeting: > *Knowing that \[the meeting topic\] is actually maybe quite > confidential or sensitive, I don’t want to be standing up. I need to > be sitting down*. > > (Else, Workplace C) ## Theme 3: Standing as norm violation Many participants found that standing in all-seated meetings made salient the prevailing implicit norm of sitting in meetings, and their deviation from this behavioural standard. > *As soon as it’s called a meeting, it formalises everything. There are > just certain social expectations and standing is not one of them. > \[…\] I felt like I was breaking the rule. \[…\] As with any social > norm, as soon as you’re in a position where you might be going against > it, you suddenly feel the weight of society’s expectations on you*. > > (Angela, Workplace C) Participants worried that standing would be seen as a deliberate attempt to challenge the sitting norm, and that they would be perceived by other attendees as “*an attention seeker*” (Ben, Workplace B), willfully detracting from the business of the meeting. Some participants were concerned that being seen to be violating the norm could potentially detrimentally impact other attendees, and so the progression and outcomes of the meeting: > *I would worry that I was making them \[other attendees\] > uncomfortable and worry that they wouldn’t be able to have the meeting > that they wanted, and that they wouldn’t get out of it what they > wanted or not be able to talk as freely as they would normally*. > > (Alisha, Workplace C) Characteristics of the meeting context–the perceived formality, purpose, type, length, and size of the meeting, and relationships between attendees–shaped the standing experience for many, often affecting the extent to which participants felt compelled to conform to the sitting norm. Meeting contexts that made participants’ contravention of the sitting norm more prominent were most aversively experienced. For example, meetings characterised by frequent interaction between attendees, larger meetings, and those where other attendees were unknown to the stander, were often cited as challenging. Many participants sought to deflect unwanted attention and avoid misconceptions from others by forewarning other attendees of their intention to stand, or seeking explicit permission from the meeting leader, in advance or at the outset of the meeting. Most felt that they had to explain their decision to stand to others, and while many truthfully cited involvement in our study, some felt that this provided insufficient justification, instead feigning ill-health to claim exemption from the sitting norm: > *I lied and told them I had a health reason for needing to stand*. > \[…\] *It’s just one of those things that*, *unless you have a good > enough excuse to stand*, *they’re going to assume that you’re just > being difficult*. > > (Anusha, Workplace A) Several participants recounted episodes in which their decision to stand was misinterpreted by others as reflecting a lack of opportunity to sit, which in turn was felt to obligate the participant to sit when such an opportunity was provided: > *I probably stood for about a minute … and then someone else looked at > me and they were gesturing that they’d saved me a seat! \[…\] I felt > super awkward and sat down*. > > (Angela, Workplace C) ## Theme 4: Standing as appropriation of power Psychological discomfort appeared to arise not only from being seen to violate sitting norms, but also because standing was felt to affect the power dynamics of the meeting. Standing in an all-seated meeting was felt to symbolize status and authority within the meeting: > *You tend to think the more authoritative person, or the person that’s > going to lead the conversation, might be the one that’s a bit higher*. > > (Brianna, Workplace C) Indeed, many reported that standing made them feel empowered: > *I probably addressed everyone and raised my voice a little, projected > it a bit more than I might do … if I was seated. \[…\] \[Standing is\] > a much more confidence-boosting posture*. > > (Joan, Workplace B) Where the participant was hosting the meeting, the additional power conferred by standing was deemed useful for denoting and exercising leadership (“*I was the lead … so it seemed natural that I had that authoritative position*”; Anusha, Workplace A). Where participants were not leading the meeting, however, they worried that standing would be misconstrued as a tacit attempt to appropriate power by challenging the authority of the meeting leader, or other, more senior attendees: > *If everyone’s sat down there and you’re up there*, *there’s*, *you > know*, *almost a visual representation of a hierarchy in a weird way*. > > (Joan, Workplace B) > > *I stood up while \[my manager\] sat down. I felt uncomfortable > because I felt like I was telling her what to do, like I was like a > teacher and she was a student because I was standing over her. \[…\] I > think she probably felt equally as uncomfortable*. > > (Angela, Workplace C) Meetings featuring presentations by one attendee to the group were typically less problematic, as were those in which senior attendees explicitly supported standing, because participants felt that there was little risk of standing being interpreted as an attempt to assert power in such situations: > *As a junior member in a meeting, I wouldn’t really be willing to > stand up and say, well I’m standing up because I want to stand up. > \[But\] I’d be more willing to, if … someone more senior said I’m > standing up, I’d say, great, I’m going to join you*. > > (Amelia, Workplace B) Meetings held in familiar social or physical settings, especially locations over which the participant felt they already had ownership and authority (e.g. the participant’s office), were also less psychologically uncomfortable: > *Just the familiarity of the people in the team now, my relationship > with the people in the team makes it easy to stand, I’m comfortable, > you know if anyone made, no one’s made really any negative comments, > but even if anyone did, I’d be comfortable just being like, well this > is what I want to do*. > > (James, Workplace A) # Discussion Standing in meetings is often proposed as a strategy by which to reduce workplace sitting, but no study to our knowledge has yet documented how people experience standing in meetings in which all other attendees sit. Interviews with volunteers who stood in planned workplace meetings revealed barriers that workers can expect to face if deciding to stand where others are sitting, and potential facilitators. For many, the experience of standing in meetings was uncomfortable in some way. Some participants experienced physical discomfort from standing for self-imposed lengthy periods and spoke of practical challenges posed by the lack of suitability of meeting room furniture to standing. Perhaps moreover, participants felt psychologically uncomfortable about standing. This apparently arose from concerns about being seen to violate a compelling social norm favouring sitting, or being seen to be challenging the authority of other attendees by standing while others sat. Many participants chose to reduce the potential impact of their standing on others by removing themselves to the edges of the meeting room, though this risked limiting their involvement in the meeting. For meeting hosts, standing was often found to confer greater power, and enhance confidence. Our findings provide a much-needed illustration of the broader interpersonal and organizational contexts that frame workplace behaviour, and the difficulties these pose for standing in meetings. Promoting standing in normally-seated meetings requires that office-based organisations and workers anticipate these challenges. It is important to mitigate potentially aversive consequences of standing in meetings; while office workers generally appear willing to try standing in meetings, they are unlikely to continue to stand if initial attempts fail to attain expected positive outcomes or yield predominantly negative outcomes. Our findings revealed several such negative outcomes, many of which participants did not foresee. Many reported physical discomfort, sometimes identified as pain, which appears to have been due to prolonged standing; many felt that they had to stand for the entirety of meetings, though we did not instruct them to do so. This likely reflects a misplaced belief among the public that the health risks of sitting can only be offset by prolonged standing. Yet, standing still for long periods can also harm health. A recent expert consensus statement on workplace sitting and standing advises that “prolonged, static standing postures be avoided” (p1360), and that sitting be replaced by frequent sit-stand transitions, standing, and light physical activity. While physical activity is unlikely to be feasible in meetings, workers in lengthy meetings could realistically be encouraged to build standing time gradually, stand only for as long as is comfortable, and regularly alternate between sitting and standing. Participants also described psychological discomfort resulting from ‘standing out’ from others in the meeting. This echoes previous research showing that people report feeling ‘weird’ or self-conscious from standing in normally-seated workplace contexts. For many, such discomfort arose from knowingly violating a strong perceived social norm exerting pressure to sit and not stand. These findings support the centrality of social norms as a predictor of action, in potential competition with one’s own attitudes, such that people sometimes act in a counter-attitudinal way to conform to social pressures. Indeed, some participants reported that, despite wanting to stand, they aborted their standing attempts early, in response to implicit or explicit pressure from others to sit. Experiences of standing appeared to vary in accordance with the extent and visibility of perceived norm violation. In meetings in which standing seemed to more strongly contravene the implicit sitting norm–such as larger meetings characterized by discussion among attendees–standing produced stronger feelings of awkwardness. Particularly where highly emotive topics–such as job losses–were under consideration, many participants felt that their decision to stand could be misinterpreted as a challenge to the seriousness of the meeting, or the authority of attendees. Standing also elicited psychological discomfort for some because they feared that their standing would be interpreted as an assertion of power and authority over other attendees. Indeed, for many animals, moving from sitting to standing can be a sign of dominance and aggression, and previous research has documented concerns among office workers about the potential for standing to be misconstrued as aggressive or threatening. Concern about such misperceptions was an especially powerful barrier to standing in meetings with more senior colleagues. In meetings in which participants felt that standing could not perceived to be an infraction–such as when standing to present information, or when participants considered they had a legitimate reason to stand–standing did not appear to elicit strong feelings. Interestingly, however, many felt that current health problems precluding prolonged sitting offered the only legitimate rationale for standing. This testifies to the perceived strength of the sitting norm, and of being seen to respect established hierarchical relations within the meeting; some believed that standing for the sake of health promotion (rather than management of ill-health), or to honour commitments to participate in our study, did not constitute sufficient reasons to disrupt norms or be seen to challenge the authority of others. Thus, where meetings were held in settings over which the participant perceived a sense of ownership and authority–such as where the participant was leading the meeting–standing was seen as less of a contravention of norms, and no threat to established power dynamic, so did not evoke psychological discomfort. Together, our findings suggest several potentially fruitful strategies for overcoming norm barriers. First, as many of our participants found, notifying others–especially meeting leaders–of intentions to stand, or relocating to a less visible position in the room, can alleviate perceived social pressure to sit. Relocation can, however, yield mixed consequences. In meetings characterized by interaction among attendees, some participants felt that standing physically and psychologically distanced them from collaborative discussions. Others reported that standing could prompt more efficient meetings, due to the desire to minimize anticipated physical discomfort from prolonged standing. Indeed, previous research suggests that standing meetings tend to be shorter in duration than seated meetings, with no impact on the quality of meeting outcomes. Second, and more broadly, office-based organisations should explicitly promote standing in meetings, to counter the perceived sitting norm, and thus empower those who wish to stand without fear of infringing social expectations. Meeting hosts should also encourage standing in meetings; participants often felt more confident standing when they had secured prior approval from those leading the meeting. Hosts might, for example, suggest that attendees must stand when speaking in contribution to a group discussion, a strategy shown to be acceptable in principle in a study of employees in Belgium. Managers may be encouraged to support standing in meetings through emphasizing potential benefits to productivity and staff time arising from shorter, more efficient meetings. Third, there is an urgent need to promote standing for health promotion purposes in the workplace, by developing messages that frame standing as a legitimate strategy for sitting less. Given also the pervasive culture of sitting cited by our participants, organisational buy-in, involving creating and making salient and explicit a standing-permissive culture, will be central to the effectiveness of promotional strategies for standing in meetings. Organisations can also facilitate standing in meetings by providing standing- appropriate infrastructure, such as meeting spaces with high tables and stools. While many participants adapted to standing in sitting-conducive environments by using accessories (e.g. using tablets to take notes), restructuring of the physical environment has been shown to be an effective upstream method for reducing sitting among workers. Study limitations must be acknowledged. We did not assess participants’ prior sitting time or the frequency with which they attended meetings, nor were measures taken of the length of time for which participants stood, all of which may have influenced responses to standing. Moreover, as we showed, the experience of standing in meetings is influenced by the perceived responses of other attendees, but we did not collect data from meeting attendees other than the standing participant. Standing may create psychological discomfort among non-standing attendees, in turn feeding negative perceptions of the stander. While we have focused our practical recommendations on how to minimize feelings of discomfort among standers, effectively promoting standing in meetings may perhaps also require understanding and assuaging potentially negative experiences of non-standing attendees. We sought to explore real-world experiences of standing in meetings, but study procedures may have influenced such experiences. Participants may at least in part have been incentivized to stand by the gift voucher incentive, or by the belief that participating in our study in this way would contribute to scientific knowledge. It is thus possible that their reasons for and experiences of standing among our sample may differ to those of employees who would stand in meetings in more naturalistic settings, thus questioning the representativeness of the experiences we documented. We sought to minimize our influence on the standing experience by giving minimal instructions to participants, asking only that they attempt to stand for a self-determined time. However, a lack of further instruction ironically appears to have had an important influence on experiences; the physical discomfort reported by many was apparently due to the misconception that participants should stand for as long as possible. Participants may have had more positive experiences had they, for example, been advised on how best to integrate standing into meetings, including setting realistic standing duration goals, and informing meeting hosts and other attendees in advance. Nonetheless, our methods have documented the potential importance of informing others of decisions to stand, and of the potential for people to misunderstand advice to stand *more* as a recommendation to stand for as long as possible. We are confident that our findings offer valid insights of importance for informing future guidance for incorporating standing into meetings. Our sample was small and, while university employees span a socioeconomically broad range, our participants were highly educated, which questions the generalizability of findings. However, participants were recruited from three office-based university organisations, and captured a diversity of meeting types, job roles and seniority. Moreover, our aim was not to identify a generalizable set of experiences, but rather to capture and explore a range of reflections on the experience of standing in meetings. Indeed, while previous research has suggested that office workers find the idea of standing in normally-seated meetings acceptable in principle, ours is the first study to document the rich complexity of the psychological, interpersonal and organisational contexts that frame the standing experience. # Conclusions Displacing sitting with standing at work requires an in-depth understanding of how to integrate standing into normally-seated work practices. While meetings offer but one workplace context in which sitting time might be reduced, our study demonstrates the complexity of this specific context, which should be acknowledged by future workplace sitting reduction initiatives. Specifically, we have highlighted some important physical, psychological and social barriers and facilitators that may determine whether someone feels sufficiently capable to break the mould and stand in normally-seated meetings. Office workers must acknowledge that standing in meetings will involve a period of acclimatisation to an unusual way of working. Many of our participants learned to adapt to standing over the course of the three meetings, and so reduced initial physical and psychological discomfort. Strategies that may enable office workers to sustainably adopt standing in meetings as a sitting-reduction strategy include building standing time gradually, and alternating between sitting and standing, to alleviate physical discomfort, and notifying attendees of intentions to stand, to avoid psychological discomfort from being seen to be challenging norms and social hierarchy. Office managers should seek to provide visible organizational support for standing, including the explicit promotion of the acceptability of standing in the workplace as a health promotion strategy, and provision of designated areas of standing-supportive furniture. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction The most widespread definition of underuse is *“the failure to provide a health care service when it would have produced a favourable outcome for a patient”*, while overuse occurs “*when a health care service is provided under circumstances in which its potential for harm exceeds the possible benefit*”, although recently it has been pointed out that overuse could include different dimensions in relation to risk-benefit, cost-benefit and patient preference approaches. While research has traditionally focused on identifying and reducing the underuse of appropriate services in patients with a specific condition (e.g. the use of antiplatelet agents in secondary prevention of ischemic heart disease), the overuse of health care services has become an increasingly recognized but understudied problem. Nonetheless, recent reviews have shown high rates of overuse for a range of diagnostic tests, imaging tests and therapeutic services in the US setting. Overuse and underuse require operational definitions for each set of the patient’s clinical condition and the service provided (**Fig A in**). Criteria for these definitions may come from clinical trials, but more frequently come from expert consensus or from the criteria established in clinical practice guidelines (CPG). However, the latter may exhibit some variability in their recommendations. In any case, and regardless of the method used, the assessment of overuse and underuse requires having sufficient information to apply the appropriateness criteria to each of the patients evaluated. This is particularly important in the identification of underuse because it requires population samples of non-treated patients with sufficient information to assess treatment appropriateness, and in many cases these patients have no specific information or may even have no contact at all with the healthcare system. Therefore, availability of information allowing underuse assessment is extremely valuable. Furthermore, studies have usually shown the overuse of specific services in a particular condition or the underuse of other services in a different condition (**Fig A in**). However, overuse and underuse may concur in the same healthcare service and in the same clinical condition. Regarding data on use of antiosteoporotic treatment, while Spain is one of the European (and worldwide) countries with a lower incidence of osteoporotic fracture, antiosteoporotic medications are widely prescribed. A recent report analyzing the variability in the consumption of several therapeutic drugs in 15 developed countries (including the U.S., Canada, and several European countries) identified Spain as the country with the highest utilization rates of antiosteoporotic drugs. In addition, temporal trends show a very rapid and disproportionate growth in osteoporosis drug consumption in recent years. Concerning the appropriateness of antiosteoporotic drugs, previous studies suggest that Spain and other countries are witnessing a massive use of these treatments in young women with a very low risk of fracture, while there is a significant underuse in women (and men) at a high risk of fracture, including those who have already suffered a major osteoporotic fracture. These estimations, however, could vary according to the criteria used to assess over and/or underuse. Given the wide range of clinical practice guidelines on osteoporosis existing globally, it would be desirable to determine the extent to which estimations change according to such guidelines, and a setting with high utilization rates of antiosteoporotic treatment could serve as a good example. The FRAVO study is a population-based cross-sectional study designed to estimate the prevalence of vertebral fracture and densitometric osteoporosis among post- menopausal women over 50 years old living in Valencia (Spain). The comprehensive information collected allows the estimation of the risk of fracture and the operationalization of the criteria for antiosteoporotic prescribing used in most CPGs, enabling the assessment of the impact on the population of using different guidelines, as well as estimating the population over or underuse of these treatments according to the criteria of each CPG. In this study, we aimed to: 1) describe the population prevalence of antiosteoporotic treatment among post-menopausal women of 50 and over and the possible associations with socioeconomic factors, individual fracture risk factors and the 10-year risk of hip fracture (assessed by FRAX), 2) estimate the impact on the population of using different international and national guidelines regarding antiosteoporotic treatments and, 3) estimate the over and underuse of antiosteoporotic treatments among post-menopausal women of 50 and over according to the criteria established by these guidelines. # Methods ## Design Population-based cross-sectional study conducted between February 2006 and March 2007, primarily designed to estimate the population prevalence of vertebral fracture and densitometric osteoporosis among post-menopausal women of 50 and over in the city of Valencia, Spain. ## Population and Sample The study population was post-menopausal women of 50 years old and over living in the city of Valencia, Spain, excluding women with cognitive impairment, physical impediment preventing a woman from going to the radiology center by her own means, race other than Caucasian and unwillingness to participate in the study. The methods and main results of the FRAVO study have been fully described elsewhere. Briefly, from an age-stratified random sample of 1758 women resident in Valencia, a total of 824 fulfilling inclusion and exclusion were included. Twenty cases for whom the X-Ray, the densitometry or the BMI was not available were excluded in some analyses. As the final sample did not exactly fit the population age distribution of the women of 50 and over in Valencia, some estimates were weighted according to that population age distribution in 2006. ## Variables and definitions Information about socio-demographic characteristics, lifestyle and risk factors for vertebral fracture collected using the interviewer-administered questionnaire included, among other variables, the subject's age, educational level, body mass index, early menopause (defined as menopause before the age of 40), history of parental hip fracture, prior non-vertebral osteoporotic fracture, treatment with glucocorticoids (use of oral glucocorticoid for at least 3 months in the previous year) or other drugs that decrease bone mass (at least one prescription of lithium, anticonvulsants, high dose thyroxin or immunosuppressive treatment in the previous year), smoking, dietary calcium intake, and secondary causes of osteoporosis (gastrectomy, bowel resection, inflammatory bowel disease, thyroidectomy, diabetes mellitus, chronic liver disease, chronic obstructive pulmonary disease, rheumatoid arthritis, transplantation, chronic kidney failure). Spinal radiographs were performed using standardized techniques and two radiologists, who were blind to all data concerning the patients, performing the semi-quantitative evaluation of the radiographs using the Genant method. Densitometric examinations were performed with two calibrated densitometers and the World Health Organization (WHO) osteoporosis classification criteria based on T-scores were used to classify bone mineral density (BMD) results as normal, osteopenia or osteoporosis. Using the FRAX tool calibrated for Spain ([www.shef.ac.uk/FRAX/index.htm](http://www.shef.ac.uk/FRAX/index.htm)) the 10-year risk of hip and major fracture was estimated for each patient. Regarding antiosteoporotic medication, information was recorded on current treatments (bisphosphonates, raloxifene, strontium ranelate, teriparatide, hormone replacement therapy and calcitonins, which were the antiosteoporotic agents available in Spain during 2006–7), duration of treatment and the specialty (general practitioner, orthopaedic surgeon, gynaecologist, rheumatologist, and other/unknown) of the prescriber of the first antiosteoporotic treatment. ## Selection of guidelines and operational criteria We revised the guidelines chosen for inclusion in a previous review and selected the closest to 2007 versions of four international guidelines (National Institute for Health and Care Excellence (NICE, UK), National Osteoporosis Foundation, (NOF, US); National Osteoporosis Guideline Group (NOGG, UK); and Osteoporosis Canada), and six Spanish guidelines (Spanish Society for Family and Community Medicine (semFYC), Spanish National Health System (SNS), Spanish Society for Bone Research and Mineral Metabolism (SEIOMM), Spanish General Medical Society (SEMERGEN), Spanish Orthopaedic Surgery and Traumatology Society (SECOT), and the Spanish Rheumatology Society (SER)). These CPGs are, from the authors’ point of view, the most well known and influential in the Spanish setting, and many of them are also highly influential globally. This selection was also based on a survey to around 75 professionals of different specialities who rated the different guidelines according to their influence in their clinical practice. This choice did not take into account the quality of CPGs development and does not involve any judgment about the quality or validity of these CPGs compared to other guidelines, nor any endorsement from the authors. The guidelines criteria are described in **Table A in**. Some vague criteria were unambiguously defined to allow their use in the study databases (the corresponding specifications are also included in the **Table A in**). ## Ethical Aspects The study was approved by the Institutional Review Board of the Primary Care Departments of Valencia and Castellon. All of the participating women were informed of the study’s characteristics and risks, and all gave signed informed consent prior to enrolment. ## Analysis First, we briefly described the clinical and demographic characteristics and treatment rates of the participating women and conducted bivariate analyses to determine which characteristics were related to osteoporosis treatment. We also described the drugs used and the medical speciality of the physician who prescribed the first antiosteoporotic treatment. Second, we used a multivariable logistic regression (backward-forward stepwise method, with p\<0.05 for entrance and p\<0.10 for removing variables) to retain the variables independently associated with receiving osteoporosis treatment. Third, we used the information from participants in the FRAVO study to estimate the percentage of women aged 50 years and over who would be recommended for treatment according to the respective guidelines (impact on the population), with the corresponding 95% confidence intervals (95% CI), calculated using the binomial approach. Finally, we assessed the inappropriateness of treatments: the proportion of treated women without a treatment recommendation according to the respective CPGs (overuse) and the proportion of non-treated women with a positive recommendation for treatment according to the above-mentioned guidelines (underuse). All the analyses were performed using the STATA 11.0 (Stata Corp) statistical software. # Results The study population included 824 post-menopausal women of 50 years old and over living in the city of Valencia, Spain. Of these, 186 (22.0%) were on antiosteoporotic treatment. After weighting the sample according to the population age structure of Valencia, the estimated prevalence of antiosteoporotic treatment in postmenopausal women of 50 years old and over in Valencia was 20.9% (95%CI: 17.6–24.4). The most commonly prescribed drugs were alendronate (36.6%), risedronate (24.7%) and raloxifene (22.5%), followed by HRT (9.1%), calcitonins (3.7%), strontium ranelate (2.2%) and PTH (1.1%). Regarding the origin of the initial prescription, orthopaedic surgeons were responsible for most of them (37.1%), followed by gynaecologists (32%), general practitioners (19.4%) and rheumatologists (8.1%). shows the antiosteoporotic drugs prescribed according to the medical specialty responsible for the initial prescription. Orthopaedic surgeons prescribed risedronate as their first choice (40.6% of their treatments), followed by alendronate (36.3%), but prescribed calcitonins and PTH more frequently than other specialties; gynaecologists prescribed primarily raloxifene (40.0%), followed by hormone replacement therapy (26.7%) and alendronate (21.7%); general practitioners preferentially prescribed alendronate (58.3%), followed by risedronate (22.2%) and raloxifene (16.7%); and for rheumatologists, their first choice was alendronate (40.0%), followed by risedronate (26.7%), raloxifene (20.0%), and strontium ranelate (13.3%), the latter being prescribed mainly by this speciality. Regarding the sociodemographic, lifestyle, and clinical characteristics considered, age at menopause, bone mineral density (BMD), morphometric vertebral fracture, BMI and fracture risk score, were associated to antiosteoporotic prescribing. Women with early menopause, densitometric osteoporosis and moderate or severe morphometric vertebral fractures were more likely to be treated (33.8%, 30.2% and 42.0%, respectively). Regarding the 10-year risk of hip fracture (assessed by FRAX), the proportion of women treated was higher for those with moderate (1–3) risk scores (28.8%). Obese women had lower treatment rates (15.0%). In the multivariable analysis, the factors independently associated with the prescription of antiosteoporotic drugs were: early menopause (2.6 times greater odds of having an antiosteoporotic drug prescription), morphometric moderate or severe vertebral fractures (2.7 times greater odds), and densitometric osteoporosis (1.5 greater odds). Women aged 65–69 years old were also more likely to have a prescription than women in the lowest age group (reference group). Obesity worked as a factor reducing the likelihood of being treated. Regarding the potential impact on the population of applying the criteria for treatment of the CPGs analyzed, between 8.7% (Osteoporosis Canada guideline) and 36.6% (National Osteoporosis Foundation guideline) of women would be recommended for treatment according to the international CPGs, while the percentage of women of 50 years old and over that would be treated according to the Spanish CPGs would range between 17.7% (Spanish Society for Family and Community Medicine guideline) and 44.3% (Spanish Rheumatology Society guideline). Regarding inappropriateness of treatments, between 56.4% and 77.8% of women under treatment did not meet any criteria to be treated according to the international CPGs; these figures ranged from 41.5% to 66.0% when we applied the Spanish CPGs criteria. According to all guidelines, the overuse of antiosteoporotic treatments in postmenopausal women of 50 years old and over was 45.3% (61.3 and 46.4% according to the international and Spanish CPGs, respectively). Concerning the underuse, between 6.6% and 34.6% of untreated women should have received treatment according to the international CPGs, whereas these figures were between 14.3% and 41.3% when the Spanish CPGs were applied. According to all guidelines, the underuse of antiosteoporotic treatments in postmenopausal women of 50 years old and over was 0.9% (3.4 and 2.0% according to the international and Spanish CPGs, respectively). # Discussion Our study shows that the prevalence of antiosteoporotic treatment in postmenopausal women ≥ 50 in Valencia was 20.9% and the type of antiosteoporotic drugs prescribed varied greatly depending on the medical specialty responsible of the initial prescription. The impact on the population and the proportion of treatments considered inappropriate when applying the most influential osteoporosis guidelines, varied strikingly, with the percentage of women 50 and over who should be treated ranging from less than 9% to over 44%. A large proportion of inappropriate treatments was found when applying these guidelines to the Spanish population, combining a high overuse (which ranged between 42 and 78%) and, to a lesser extent, underuse (ranging between 7 and 41%). In the city of Valencia, one out of five women aged 50 and older were treated with antiosteoporotic drugs. This treatment rate, lower than that reported for the Valencia region in 2010, may be related to the healthier characteristics of a real population sample (enrolled from a population registry, not from medical practices), increasing treatment rates during the time period between the two studies (the dispensing of antiosteoporotics in Spain rose by nearly 50% between 2006 and 2008) or real differences in treatment rates between the city of Valencia and the rest of the Valencia region. Nevertheless, our results show high treatment rates in women with a low to moderate risk of fracture (19% and 29% in women with≤1% and 1–3% 10-year risk of hip fracture assessed by FRAX, respectively), and a huge proportion of unnecessary treatments (between 42% and 78%) according to the criteria of the most influential CPGs. These figures suggest a striking amount of osteoporosis treatment overuse and an interesting opportunity to reduce costs (including those related to adverse events from unnecessary treatments) without compromising-and probably improving- patients’ health. Regarding treatment underuse (7–41%, depending on the guideline used), although lower than the treatment overuse observed in relative terms, it is based on a larger population (the untreated women, 79%), also becoming a major issue in osteoporosis management. Some of the figures described in the bivariate analysis were highly suggestive of underuse. Although some acknowledged risk factors are associated with an increased likelihood of being treated, other recognized risk factors did not show such an association, or the proportion of women treated is too sparse: only 33% of women with prior non-vertebral fractures and 42% of women with moderate-severe vertebral fractures were receiving osteoporosis treatment. Even some risk factors such as age or FRAX 10-year risk of hip fracture showed paradoxical behaviour, with a greater likelihood of treatment at intermediate values but no association with the highest-risk scores. These figures confirm the existence of a relevant “osteoporosis care gap” in the Spanish setting, something which has already been revealed in other countries,and has remained unchanged over time. Our study also shows the dramatic impact on the population treatment rates of applying the diverse CPGs recommendations, varying the percentage of women who should be treated according to different guidelines from less than 9% to over 44%. In real terms, from the approximately eight million women of 50 years old and over in Spain, the number eligible for treatment would range from 0.7 to 3.8 million, depending on the guideline used. These results are consistent with those reported by Bolland and Gray, showing treatment recommendations for 21% and 48% of women after applying the NOGG and NOF guidelines criteria, respectively, in a cohort of older women (mean age74 years) participating in a clinical trial in New Zealand (which would be 11% and 37% for such CPGs in our study population, which is 10 years younger). Treatment decisions and the choice of a particular drug could be influenced by patient characteristics, physician and organizational factors, pharmaceutical promotion and healthcare system characteristics. One interesting result of our study-limited by the small number of cases for analysis-is that the selection of the specific antiosteoporotic agent seems to be more dependent on the specialty of the physician starting treatment than on patient characteristics. Several studies have found that the pharmacological management of several conditions varies greatly by physician speciality; however, the quality and relevance of those studies has been criticized. Moreover, little is known regarding this variability in the pharmacological management of osteoporosis. In our study, the prescribing patterns of gynaecologists were particularly remarkable, treating only one third of patients with bisphosphonates and almost two thirds with raloxifene (40%) or hormone replacement therapy (25%)–five years after the publication of the Women's Health Initiative trial results. These results could be explained, at least in part, because these specialists often treat younger postmenopausal women. In this sense, it is likely that such prescription patterns could be a significant source of overuse, given the low risk of osteoporotic fracture in young women, although it could also be a strictly local finding. Furthermore, we found that orthopaedic surgeons were responsible for most of the initial antiosteoporotic prescriptions. This might be explained by the relative lack of rheumatologists within the Spanish National Health System, and the consequent tendency of primary care physicians to refer these patients to the orthopaedic surgeons. ## Limitations Our study has some potential limitations. First, we “applied” several U.S., U.K., and Canadian guidelines to a Spanish population that may have different characteristics (e.g. prevalence of risk factors, incidence of fracture, strength of the associations between risk factors and the incidence of fracture, etc.) than the populations where the decision rules have been developed. Second, although the dropout rate in the FRAVO study is similar to other population studies, it was higher in the oldest group (with higher expected morbidity) but also in younger working women, who were presumably healthier. Weighting for age should have partially reduced this limitation, but it is difficult to assess the effect and scope of any possible biases linked to missed cases in recruitment. Third, the selection of particular guidelines from among the myriad of existing documents (the International Osteoporosis Foundation website includes links to more than 50 guidelines; see: <http://www.iofbonehealth.org/guideline- references>) always has a subjective component, although we consider that those selected are among the most representative, widely used, and influential in Spain, Europe, the United States and Canada. Furthermore, these guidelines were also rated by physicians of different specialities as the most influential in their clinical practice. Fourth, some of the risk factors considered in the CPGs evaluated were not available or had different definitions in the FRAVO data or some CPGs include vague criteria that are impossible to operationalize unambiguously. We operationalized these criteria, as indicated in Table A in, using "reasonable" interpretations, but other possible interpretations may have led to different estimates of population impact and inappropriateness. # Conclusions and Implications The pharmacological management of osteoporosis in women of 50 and over combines an important overuse (mainly in young women with low risk of fracture) with an important underuse (in women who are older, at high risk or with previous osteoporotic fractures), although the level of inappropriateness varies dramatically depending on the CPGs used. In a recent paper we described the high variability among CPGs in recommending antiosteoporotic treatment, concluding that such variability limits the effectiveness of those recommendations and, given the heterogeneity of the criteria used, it should come as no surprise that doctors and health care providers could become confused to the point of inaction or misguided action. In the present study, we quantified the impact of this variability on the number of women who should be treated and on the inappropriateness of the prescribed treatments, and the overwhelming results should not be overlooked by public health care policies in Spain (and probably in other countries in similar situations) because osteoporosis is a frequent condition and even small variations in treatment indication can account for large differences in women treated and resources consumed. Targeting high-risk populations is a strategic element for developing cost- effective policies in the prevention of osteoporotic fractures. Predictive modelling of fracture risk factors seems to be the main instrument for stratifying the population into risk groups to which practical policies should be applied, and CPGs should help integrate this information to identify people who are more likely to benefit from treatment. The results of our study suggest that the current CPGs, although based on the same evidence, seem to interpret it differently and do not meet these clinical and policy needs sufficiently. The development of more accurate predictive tools (especially for the intermediate risks) could possibly contribute to the convergence of these interpretations, to a consensus on more homogeneous guidelines and, eventually, to the reduction of osteoporotic fractures. However, at present, it seems urgent to develop policies to reduce treatment overuse (at least in those cases where there is wide agreement) while reducing underuse should also not be neglected, especially in women in secondary prevention for whom, beyond their scores of predictive risk tools, a high risk of osteoporotic fracture has already been shown. # Supporting Information We are grateful to all doctors and nurses at the Valencia Health Agency primary healthcare centres participating in the ESOSVAL study for their collaboration, and to the Valencia Ministry of Health for its enthusiastic and continued support of the ESOSVAL research projects. [^1]: JSG has received speaking fees for symposia or other meetings from Amgen and GlaxoSmithKline (GSK). SP has received speaking fees for symposia or other meetings from Ferrer International and Health, Innovation and Society Foundation of Novartis Spain. GS has received speaking fees for symposia or other meetings from Ferrer and Boehringer Ingelheim International. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: JSG GSG SP. Performed the experiments: GSG IH CBP CLRB. Analyzed the data: GSG CBP IH CLRB JSG SP. Wrote the paper: GSG SP CLRB. Interpretation of data, critical revision of the manuscript for important intellectual content, and approved the final version submitted for publication: GSG IH CBP CLRB JSG SP.

# Introduction Huntington’s disease (HD) is an autosomal dominant heritable disorder and is characterized by progressive motor dysfunction, cognitive impairment and disturbances in behavior. The worldwide prevalence of HD is 2–5 per 100,000 people. HD is caused by an expanded CAG trinucleotide repeat at the 5’ end of the *Huntingtin* (*HTT*) gene and the length of the resulting polyglutamine (polyQ) repeat in the HTT protein is inversely correlated with the age of onset of the disease. CAG repeat lengths between 36 and 39 do not always cause signs and symptoms of HD (due to reduced penetrance) but *HTT* containing 40 or more CAG repeats leads invariably to HD. The expanded polyQ stretches result in misfolded mutant huntingtin (mHTT) proteins by the formation of β-sheets, and N-terminal mHTT fragments containing the polyQ expansion are aggregation prone. Lowering the levels of soluble mHTT prior to aggregation through increased degradation would be a therapeutic strategy to prevent or delay the onset of disease. The two main systems responsible for protein degradation in the eukaryotic cell are autophagy and the ubiquitin proteasome system (UPS) (reviewed by Soares et al.). Whereas autophagy functions mainly in the cytoplasm, the UPS is the main route for the degradation of misfolded proteins in both the cytoplasm and the nucleus. The UPS is thus a promising candidate to target both cytoplasmic and nuclear mHTT fragments. Initial studies suggested that the proteasome cannot degrade polyQ stretches and that proteasomes are irreversibly sequestered into mHTT aggregates. More recently, we showed that proteasomes are dynamically recruited to mHTT inclusion bodies, that they remain catalytically active and that they are accessible to substrates. Furthermore, when polyQ-expanded mHTT fragments were targeted towards the proteasome with an N-terminal degradation signal, they were efficiently and completely degraded. Together this indicates that the proteasome remains active in cells with HTT inclusion bodies and is able to degrade soluble mHTT when this is targeted for degradation. The proteasome consists of a latent 20S core particle, which is composed of four heptameric α- and β-rings (α₇β₇β₇α₇), of which the latter ones contain catalytic subunits that face the interior of the cylinder and are responsible for the proteolytic activity. Access to the central cavity is regulated by a gate formed by the N-terminal protrusions of the α-subunits. Modulation of the gate is required for substrate entry into the 20S core and is mediated by the proteasome activators (PA), such as the 19S complex and PA28. The 19S and PA28 can bind to both ends of the 20S proteasome core leading to single (referred to as 26S when one 19S complex binds 20S), double (referred to as 30S when two 19S complexes bind 20S) or even hybrid-capped proteasomes (with a 19S and PA28 complex on opposing ends of the 20S). The 19S is a complex of distinct subunits and is involved in the recognition, unfolding and de- ubiquitination of poly-ubiquitinated substrates in an ATP-dependent manner. PA28 is involved in ATP-independent degradation and has three homologues: PA28α and PA28β, forming heterodimers, and PA28γ, forming a homo-heptameric ring, which is only expressed in the nucleus. Although PA28αβ exists as a heterodimer, homomeric PA28α is sufficient to activate proteasome activity. By binding to the proteasomal α-ring, PA28 allows entrance of peptides into the 20S core and increases proteasome activity. PA28 overexpression also increases degradation of oxidized proteins in cells, with increased PA28αβ binding to the 20S proteasome immediately upon H₂O₂ treatment, followed by increased PA28αβ expression during oxidative stress adaptation. Interestingly, PA28αβ has been shown to prevent aggregation in the mouse hippocampus during aging. Furthermore, PA28γ overexpression relieved HD pathology and lowered the number of inclusion bodies in cells and *in vivo*. The role of PA28αβ, which is present in both the nucleus and cytoplasm, on mHTT turnover remains elusive. In this study, we examined whether changes in proteasome complex formation occur during disease progression in HD mouse models. To determine which consequences these alterations have on mHTT degradation, we modulated PA28αβ activated proteasomes in various *in vitro* models and determined the effects on both polyQ fragments and mHTT proteins. # Materials and methods ## HD mouse models All procedures were in accordance with the Animals (Scientific Procedures) Act 1986 and were approved by the King’s College London (KCL) Ethical Review Process Committee. In this study *Hdh*Q150 mice expressing endogenous full-length mouse *Htt* with an expanded CAG repeat and R6/2 mice expressing a human exon 1 *HTT* transgene were used. Animals were genotyped by PCR and CAG length was determined as previously described. The *Hdh*Q150 homozygous mice were on a CBA/Ca and C57BL/6J F1 background and had a CAG repeat size of 166±9. The R6/2 mice were bred by backcrossing R6/2 males (CBA/Ca x C57BL/6J) to F1 females (B6CBAF1/OlaHsd, Envigo, Netherlands) and had a CAG repeat length of 209±2. *Hdh*Q150 mice were sacrificed by cervical dislocation at 2, 16 and 22 months of age, and R6/2 mice at 4 and 14 weeks of age. The dissected brain regions were immediately snap frozen. ## DNA constructs PA28α and PA28β (kindly provided by Prof. PM Kloetzel, Charité Universitätsmedizin Berlin, Germany) were cloned into a pcDNA3 vector using EcoRI. GFP-Ub-Q54, DNAJB6 and m*HTT*(Q25/Q97)exon1-H4 were generated as described before, respectively. To create a stable cell line with inducible expression of mHTT, a pINDUCER m*HTT*(Q25/Q46/Q97)exon1-IRES-*GFP*-Q16 construct was generated. First, m*HTT*(Q97)exon1 was amplified with Xho on the 3’ end (fw (T7) 5’-`TAATACGACTCACTATAGGG`-’3, rv 5’-`GTTCTAGATTAAGGTCGGTGCAGAGGCTC`-’3) and cloned into pIRES2-*GFP*, using XhoI and SmaI, creating m*HTT*(Q97)exon1-IRES-*GFP*. Next, *GFP*-Q16 was amplified with BstX1 and Not1 on the 3’ and 5’ ends, respectively (fw 5’-`CGATGATAATATGGCCACAACCATGGCCACCATGGTGAGCAAGGGCGAGG`-3’, rv 5’-`TGATCTAGAGTCGCGGCCCCGCTTACCTGGGGCTAGTCTC`-3’) and cloned into m*HTT*(Q97)exon1-IRES-*GFP* using BstX1 and Not1, to replace *GFP*. Subsequently, m*HTT*(Q97)exon1-IRES-*GFP*-Q16 was amplified with EcoRI on the 3’ and 5’ ends (fw 5’-`GTCCAGTGTGGTGGAATTCTCGAGGTCGACCGCCATGG`-’3, rv 5’- `CTGGATATCTGCAGaattCCGCTTACCTGGGGCTAGTCTC`-‘3) and introduced into pENTR/D-TOPO by EcoRI. Finally, the construct was transferred from the pENTER Gateway vector into the pINDUCER20-Blast lenti-viral dox-regulated expression vector. To create cells expressing m*HTT*(Q25/Q46)exon1-IRES-*GFP*-Q16 m*HTT*(Q25/Q46)exon1 was amplified with BamHI on the 3’ and 5’ ends (fw 5’-`TGGTACCGAGCTCGGATCGCCACCATGGCGACCCTGGAAAAGCTG`-’3, rv 5’-`GAGGGAGAGGGGCGGATCTTAAGGTCGGTGCAGAGGCTC`-’3). Next m*HTT*(Q25/Q46)exon1 was cloned into pENTR m*HTT*(Q25/Q46)exon1-IRES-*GFP*-Q16 using BamHI, replacing m*HTT*(Q97)exon1. All plasmids were verified by sequencing before use. ## Cell culture HEK293 and ST*Hdh*^Q7/Q7 cells were cultured in DMEM (Gibco) supplemented with 10% FBS (Gibco), 1% penicillin/streptomycin (Gibco) and 0.2mM L-glutamine (Gibco) and grown in a humidified chamber with 5% CO₂ at 37°C or 32°C, respectively. ST*Hdh* cells expressing doxycycline inducible m*HTT* were generated by retroviral transfection of ST*Hdh*^Q7/Q7 cells with pINDUCER m*HTT*(Q25/Q46/Q97)exon1-IRES-*GFP*-Q16. ## Treatment and transfection HEK293 cells were stimulated with 100U/ml IFNγ (ProSpec) for 72 hours and proteasomes were inhibited with 250nM epoxomicin (Sigma) for 16 hours. HEK293 cells were transfected with jetPEI one day after plating according to the manufacturer’s instructions (Polyplus transfection). Neon Transfection System (Invitrogen) was used to overexpress constructs in ST*Hdh* cells. For silencing experiments with shRNA, the MISSION® TRC-Mm 1.0 (Mouse) library was used. The PA28α targeting sequence (5’-`CCCGATCCAGTCAAAGAGAAA`-3’, MISSION® TRC shRNA TRCN0000066420) was delivered through retroviral transduction. The pLKO.1-puro Non-Mammalian shRNA Control Plasmid (SHC002) was used as a control. SiRNA targeting PA28α (ON-TARGETplus Mouse Psme1 siRNA SmartPool, Horizon Discovery) or non-targeting siRNA (siGENOME Non-Targeting siRNA Control Pool, Horizon Discovery) was delivered to the cells by using Lipofectamine RNAiMAX Transfection Reagent (Invitrogen) directly with plating the cells. ## Native PAGE Brain tissue and cell pellets were suspended in TSDG buffer (10mM Tris/HCl pH7.4, 25mM KCl, 10mM NaCl, 1.1mM MgCl₂, 0.1mM EDTA, 10% glycerol, 1mM ATP fresh) and brain tissue was further homogenized by the use of Dounce tissue homogenizer. Lysis was performed by 3–5 freeze/thaw cycles in liquid nitrogen. After centrifugation (15 minutes, 14.000 rpm at 4°C), the protein concentration of the clarified lysate was determined by Bradford protein assay (Serva). After the addition of 4x native sample buffer (20mM Tris pH8.0, 50% glycerol, bromophenol blue) samples were separated on 3–12% NativePAGE Novex Bis-Tris gels (Invitrogen). For western blotting, native gels were transferred to PVDF membranes (Millipore, Bedford, MA, USA) in transfer buffer (25mM Tris pH7.5, 192mM Glycine, 20% MeOH) using the Criterion blotter (Biorad). After blocking in 5% milk, membranes were incubated with the antibody of interest and Odyssey detection system (LICOR Bioscienses) was used for scanning and analysis. ## Visualizing proteasome activity and peptide degradation in gel Activity based probe (ABP) labeling was performed either in the lysate or after running native PAGE. For in-lysate labeling, the samples were incubated with 0.5μM ABP for 30 minutes at 37°C before adding sample buffer and loading on native PAGE. In-gel ABP labeling or in-gel peptide degradation was performed after protein separation by native PAGE. For these overlay assays the wet gel slab was incubated in 10ml overlay buffer (20mM Tris pH7.5, 5mM MgCl₂, 1mM ATP fresh) with 25nM ABP or 400μM of the quenched peptide for 20 minutes at 37°C. Fluorescent intensities were measured directly on a Typhoon imager (Ge Healthsciences) using the 580 BP 30 filter. In order to inhibit proteasome activity in these assays, the lysate was incubated with 0.5μM epoxomicin (Sigma), or similar amounts of DMSO in control samples, for 1 hour at 37°C, prior to in lysate labeling or native PAGE separation. ## Western blot Cells pellets were lysed in Triton-x buffer (50mM Tris/HCl pH7.4, 150mM NaCl, 1mM EDTA, 1% Triton-X100, supplemented with complete mini protease inhibitor cocktail (Roche)). After centrifugation (15 minutes, 14.000 rpm at 4°C), the protein concentration of the supernatant was determined by Bradford protein assay (Serva). Samples were boiled in 6x sample loading buffer (350mM Tris/HCl pH6.8, 10% SDS, 30% glycerol, 6% β-mercaptoethanol, bromophenol blue) and separated on 12.5% SDS-PAGE gels. Proteins were transferred to nitrocellulose membranes (Biorad) with the use of the Trans-Blot Turbo Transfer System (Biorad). After blocking in 5% milk, membranes were incubated with the antibody of interest and Odyssey detection system (LICOR Bioscienses) was used for scanning and analysis. ## Filtertrap assay For filtertrap assay, the pellet obtained after centrifugation of the cell lysate was resuspended and treated with endonucleases for 1 hour at 37°C (1mM MgCl₂, 50mM Tris/HCl pH8.0, with 0.02U/μl DENARASE® (c-LEcta) added fresh). This reaction was stopped by adding 2x termination buffer (40mM EDTA, 4% SDS, 100mM DTT fresh) and samples were diluted in 2% SDS buffer (2% SDS, 150mM NaCl, 10mM Tris/HCl pH8.0). Cellulose acetate membranes (Schleicher & Schuell) with a pore size of 0.2 μm were pre-equilibrated in 2% SDS buffer. After sample loading through the Bio-Dot microfiltration apparatus (Biorad, Hercules, CA, USA), the membrane was washed twice with 0.1% SDS buffer (0.1% SDS, 150mM NaCl, 10mM Tris pH8.0) and further treated like western blot membranes. ## Antibodies The following primary antibodies were used: anti-PA28α, directed against RVQPEAQAKVDVFRED, (1:3000, kindly provided by Prof. M Groettrup, University of Konstanz, Germany) \[100\], anti-PA28α (1:1000, Enzo Life Sciences, BML-PW8185), anti-polyQ 1C2 (1:1000, Millipore, MAB1574), anti-polyQ (1:1000, Sigma-Aldrich 3B5H10, P1874), anti-β-actin (1:1000, Santa Cruz, SC-130656 and SC-47778), anti-α2 (1:1000, MCP236, kindly provided by Prof. Rasmus Hartmann-Petersen, Biologisk Institut, University of Copenhagen, Copenhagen), anti-α7 (1:1000, MCP72, Enzo Life Sciences, PW8110), anti-HTT (1:5000, Abcam, ab109115), anti- tubulin (1:1000, Cell Signaling Technology, CST2148) and anti-RPT (1:1000, Enzo Life Sciences, PW8825). IRDye 680 and IRDye 800 (1:10.000; LI-COR Biosciences) were used as secondary antibodies. ## In vitro degradation assays mHTT(Q25/Q97)exon1-H4 was purified as described before. 100ng purified HTT(Q97)exon1-H4 protein was incubated with 0.3μg mammalian open-gated 20S proteasomes in 1x 20S buffer (10mM Tris/HCl pH7.4, 30mM NaCl, 1mM MgCl2, 400μM DTT fresh) for 8 hours at 37°C. For proteasome inhibition, 1μM epoxomicin (Sigma), or similar amounts of DMSO in control samples, was added. For proteasome activation, 3μg isolated PA28αβ caps, 50μM RPT peptides (both kindly provided by Prof. M. Rechsteiner, University of Utah School of Medicine, USA) or 0.01% SDS were added to the reaction. ATP Regeneration solution (Enzo Life Sciences) was added to reactions involving the 26S proteasome. After the incubation period, 0.5μM ABP was added for an additional 30 minutes at 37°C. The reaction was stopped by boiling the samples in 6x sample loading buffer (350mM Tris/HCl pH6.8, 10% SDS, 30% glycerol, 6% β-mercaptoethanol, bromophenol blue). The complete reaction was used for 12.5% SDS PAGE to visualize ABP signal and for immunoblotting. ## Fluorescent microscopy and quantification of aggregate formation STHdh mHTT(Q97)exon1-IRES-GFP-Q16 cells were rinsed with PBS and fixed with 4% formaldehyde in 1x PBS for 1 hour at RT. Nuclei were stained with 0.01 mg/ml Hoechst 33324. Images were obtained using automated microscopy (ImageXpress Pico). To measure the aggregates, GFP-Q16 was used as a fluorescent reporter. The amount of aggregation was defined by using the MATLAB algorithm. In short, the algorithm defined the single nuclei on the pictures and it calculated the average intensity of the nuclei, which was later used to calculate the amount of cells on the images. Nuclei were defined based on size, intensity and shape. Aggregates were defined based on their size and intensity that needs to be above certain threshold based on the GFP overexpression in the cells. More information on the MATLAB script can be found as supplementary information. ## Data quantification Experimental results were normalized to the control condition and presented as mean ± SEM. Outliers were identified using Grubbs’ test statistic, and removed. Statistical differences between groups were determined by the one sample t-test or one-way ANOVA with Dunnett’s Multiple Comparison. Analysis was performed using GraphPad™ Prism v.9 (GraphPad Software, Inc.). An alpha level of 0.05 was used to define statistical significance. All data is available through the online platform FiglinQ (<https://create.figlinq.com/~k.geijtenbeek/92/collection>/). # Results ## Loss of PA28 activated proteasomes during HD progression in HD mice To explore whether proteasome complex formation changes during HD development, we investigated proteasome complex composition in two different HD mouse models: the *Hdh*Q150 mice expressing endogenous full-length mouse *Htt* with an expanded CAG repeat and the R6/2 mice expressing a human exon 1 *HTT* transgene. For the visualization of proteasome complexes, we used ABP labeling. These probes bind to the catalytic sites of 20S proteasomes and can be visualized via their fluorescence tag. Only proteasome complexes associated with proteasome activating complexes such as PA28 and the 19S are accessible for substrates. Therefore, ABPs will specifically label incorporated, active proteasome subunits. When the lysates are subsequently separated on native PAGE gels to keep protein complexes intact, and are analyzed for fluorescence, various bands can be distinguished representing various proteasome complexes. To represent different disease stages, proteasome complexes were analyzed at 2, 16 and 22 months of age for homozygous *Hdh*Q150 mice, and at 4 and 14 weeks of age for R6/2 mice , which show a faster disease progression due to the expression of the N-terminal exon 1 HTT fragment. Of these mice cortex, striatum, hippocampus, cerebellum and brain stem were dissected and analyzed, with HD pathology being more prominent in the first regions. The frozen sections were lysed, labeled with ABP and subjected to native PAGE analysis for fluorescence gel analysis. A decrease in ABP labeling at the height of PA28 capped proteasomes was observed in late disease stages in various brain regions, which suggests an alteration in these complexes (upper panels showing ABP labeling). Interestingly, immunostaining of PA28α subunits confirmed a decrease in PA28αβ activated proteasomes (middle panels, black arrowheads). In addition, an increase in the free pool of PA28αβ subunits (not bound to 20S proteasomes) was observed in late stage models (open arrowheads at the bottom of the gel), suggesting disassembly of the PA28αβ activator from the 20S core. Quantification of the fraction capped PA28αβ (black arrowheads) versus the total pool of PA28α (black and open arrowheads) showed a significant decrease in PA28αβ proteasome activation at late disease stages in cortex, striatum and hippocampus of *Hdh*Q150 mice, and in hippocampus of R6/2 mice (graphs in lower panels). ## PA28αβ overexpression improves degradation of polyQ peptides To examine whether the observed reduction in PA28αβ proteasome activation in HD mice would affect mHTT degradation directly, we first studied the effect of PA28αβ-activated proteasomes towards degradation of polyQ repeats using quenched Q8 (qQ8) peptides, which become fluorescent after cleavage by endopeptidases. When lysates are separated on native PAGE gels, proteasome complexes do not only remain intact, but also remain active. If these gels are subsequently incubated with the qQ8 peptides in an overlay assay, local fluorescence will appear at the height of the responsible enzyme when the peptide is cleaved. When HEK293 cell lysate were subjected to an overlay assay with the qQ8 peptide, a pattern of fluorescent bands was observed, demonstrating polyQ peptide degradation (left panel). Treatment with epoxomicin prevented cleavage of the polyQ peptides, indicating that the fluorescence was specifically generated by proteasomal cleavage. In addition, ABP labeled proteasomes showed a similar fluorescence pattern (middle panel), and merging both fluorescent channels demonstrated the ability of proteasomes to cleave polyQ substrates (right panel). To examine the effect of PA28αβ activation on the degradation of these polyQ-peptides in cells, HEK293 cells were transfected with PA28αβ. A native PAGE overlay with qQ8 peptides showed that PA28αβ overexpression led to an increase in qQ8 degradation. Together these data demonstrate that proteasomes are able to cleave within polyQ sequences and that activation by PA28αβ accelerates this degradation. Next, we examined whether overexpression of PA28αβ would enhance the degradation of polyQ peptides that exceed the pathological threshold. We transfected HEK293 cells with GFP-Ub-Q54, which generates pure Q54-peptides since GFP-Ub is separated by C-terminal hydrolases directly after synthesis. Cells were co- transfected with PA28α, PA28β or PA28αβ. The chaperone DNAJB6 was transfected as a positive control to detect soluble Q54, as it prevents aggregation of polyQ peptides. Although no significant decrease in soluble mHTT levels (arrowhead) was observed, overexpression of PA28α and PA28αβ led to a decrease in insoluble Q54 levels as shown by filtertrap analysis. To study whether the PA28αβ-induced effects on polyQ peptide levels were due to improved degradation through the proteasome, we treated the cells with the proteasome inhibitor epoxomicin. Epoxomicin slightly reduced the effects of PA28 overexpression. The fact that epoxomicin could not completely prevent the increased degradation of Q54 by PA28αβ can be explained by the relatively short incubation with the inhibitor compared to the long expression time of PA28 and Q54. Together this shows that PA28αβ improves degradation of polyQ peptides by the proteasome. ## PA28αβ hampers mHTT degradation by the 20S in vitro Since PA28αβ improved degradation of polyQ-expanded peptides, we next examined whether PA28αβ would also improve the degradation of polyQ-expanded mHTT protein fragments. First, we used isolated mHTT fragments to study the degradation *in vitro*. For this we isolated N-terminal mHTT(Q25/Q97)exon1 fragments from N2a cells. These proteins resemble the fragments that are also expressed in the R6/2 mice. The mHTT fragments were incubated with purified 20S proteasomes in the absence or presence of isolated PA28αβ. Both HTT(Q25) and mHTT(Q97) (arrowheads) were degraded by 20S proteasomes. However, when 20S proteasomes were activated by PA28αβ, as demonstrated by ABP labeling, the degradation of HTT(Q25) was reduced and mHTT(Q97) degradation was completely prevented. Interestingly, artificial opening of the 20S did result in increased mHTT(Q97) (arrowheads) degradation. We stimulated 20S opening by using RPT peptides that represent the C-termini of the 19S subunits RPT2 and RPT5, which are responsible for opening of the α-ring. Additionally, we used SDS to open 20S proteasomes. It should be noted however that SDS can also affect protein denaturation and thereby its accessibility into the proteasome. Toghether these results indicate that (artificial) opening the entrance of the 20S core can improve degradation of mHTT, but that PA28αβ only improves the accessibility of the 20S for small peptides, while entrance of larger protein fragments including mHTT exon1 is completely blocked. ## PA28αβ overexpression does not affect mHTT levels in cells Following the experiments with purified proteasomes, we examined the effects of PA28αβ activation on mHTT exon1 degradation in cells, which contain all other components of the proteostasis network that may affect the role of PA28αβ on mHTT degradation. HEK293 cells transfected with N-terminal mHTT(Q97) and PA28α, PA28β or PA28αβ for 72 hours showed an increase in activity of PA28αβ activated proteasomes, as shown by ABP labeling and immunoblotting, yet no statistically significant changes were observed in either soluble (arrowhead) or insoluble mHTT levels upon PA28αβ overexpression. Since cell-type specific differences were observed in mHTT aggregation and sensitivity and proteasome composition differs between different human cell lines, we next examined whether proteasomal complexes differ between HEK293 cells and striatal ST*Hdh*^Q7/Q7 cells, which are more relevant in HD. When proteasome complexes of both cell lines were analyzed by native PAGE ST*Hdh* cells showed higher levels of hybrid and 30S proteasome complexes, while the 26S proteasome is more abundant in HEK293 cells. Interestingly, PA28αβ seems to be more abundant in ST*Hdh* cells, both as a free pool and in complex with 20S proteasomes. Modulating proteasomal complexes could therefore have distinct consequences in these cell lines. To examine the effect of PA28αβ levels on mHTT degradation in ST*Hdh* cells, we generated ST*Hdh*^Q7/Q7 cells that express N-terminal mHTT(Q97)-exon1 under a doxycycline inducible promotor. These ST*Hdh*(Q97) cells were electroporated with PA28αβ and mHTT expression was subsequently induced for 48 hours. Resembling the data observed in HEK293 cells, no changes in soluble or insoluble mHTT(Q97) levels were detected after PA28αβ overexpression in ST*Hdh* cells. ## PA28αβ silencing in STHdh cells increases mHTT aggregation To examine the effects of PA28αβ silencing on mHTT(Q97) clearance in ST*Hdh* cells that express high PA28αβ levels, and thereby mimic the decrease in PA28αβ capped proteasomes observed in HD mice, we reduced PA28α levels by 80% using retroviral transduction of shRNA targeting PA28α. Native PAGE showed efficient reduction of PA28αβ activated proteasomes, also leading to less proteasome activity as shown by ABP labeling. Subsequently, mHTT expression was induced for eight hours (pre-aggregation state) or 48 hours and the effects of PA28α silencing on mHTT protein levels were determined. While reduced PA28αβ activated proteasome levels did not affect soluble mHTT(Q97) protein levels, levels of insoluble mHTT(Q97) increased significantly. These results were confirmed using fluorescence microscopy experiments, using ST*Hdh* cells expressing doxocycline inducible mHTT(Q97)-IRES-GFPQ16. Here the separately expressed GFPQ16 acts as a fluorescent reporter for mHTT aggregation, with the short polyQ sequence being sequestered into aggregates formed by untagged mHTT exon1, while in the absence of mHTT aggregation the GFP-Q16 reporter will be diffusedly distributed throughout the cell. This approach enables the visualization of aggregation of untagged mHTT in cells, which is preferred since a tag can influence a protein’s stability. Following siRNA transfection targeting PA28α and induction of mHTT expression, the cells were analyzed by automated microscopy and the percentage of cells with aggregates was determined by a generated Matlab script. This data showed that, similar to the filtertrap assay, PA28α silencing increased aggregation. When HTT(Q25) or mHTT(Q46) were used as wildtype or short polyQ- expanded HTT fragments, respectively, which do not form aggregates after 48 hours of expression, no effects of PA28α silencing on soluble wildtype HTT or mHTT levels were observed. To investigate whether PA28α silencing leads to accelerated aggregation HTT(Q25) and mHTT(Q46) samples were analyzed in a filtertrap assay. Although five times more protein was loaded for HTT(Q25) and mHTT(Q46) compared to mHTT(Q97), there was no signal above background visible. This indicates that there were no aggregates present, despite silenced PA28α levels. Altogether this data indicates that downregulation of PA28αβ accelerates aggregate formation, but does not affect soluble HTT turnover directly. # Discussion By studying changes in proteasome complexes during disease progression in HD mouse models, we observed that PA28αβ disassembles from the 20S core most obviously in the cortex, striatum and hippocampus during HD disease progression. This suggests that the changes in PA28αβ were most abundant in the HD-affected regions. When examining the consequences of alterations in PA28αβ activated proteasomes on mHTT turnover in cell models and using purified proteasomes, we observed that while the degradation of polyQ peptides is improved by PA28αβ activation of (purified) proteasomes, degradation of the mHTT protein fragment is hampered by the addition of PA28αβ to purified proteasomes. This shows that PA28 obstructs the entrance of the folded mHTT protein into the 20S core. Indeed it has been suggested before that PA28αβ selectively blocks the passage of larger protein fragments. However, PA28αβ overexpression did not affect mHTT levels when overexpressed in cells. This can suggest several things: PA28αβ activation is already sufficient; another limiting factor necessary for degradation is required; other proteasome complexes including the 26S proteasome mainly target mHTT; or mHTT is not efficiently targeted towards the proteasome. However, as reducing PA28αβ levels increased mHTT aggregation but did not affect the degradation of soluble (m)HTT, this suggests that PA28αβ is critical for overall proteostasis and only indirectly affects mHTT aggregation. The indirect role of PA28αβ affecting mHTT aggregation may be the result of its reported chaperone-like functions. The 90-kDa heat shock protein, HSP90, binds unfolded proteins to prevent aggregation and is, together with HSC70 (cytosolic HSP70), and HSP40 (cytosolic DnaJ homologue), involved in protein refolding. PA28 serves as a linker between HSP90 and HSC70 and is a necessary cofactor for protein remodeling. Since PA28 can act as a chaperone cofactor, the observed effects on mHTT(Q97) could be mediated by the ability of PA28 to interfere with aggregate formation. Indeed, hippocampal extracts from PA28α overexpressing mice, prevented aggregation of heat sensitive luciferase. In these samples activity of PA28α activated proteasomes was not increased and the total amount of damaged proteins was not altered, suggesting that chaperone-like activity is responsible for the observed effects. Overexpression of PA28γ in the striatum of HD mice reduced HD pathology as demonstrated by behavioral tasks. Furthermore, in these mice PA28γ overexpression led to reduced levels of mHTT aggregation, but no changes in mHTT levels were observed. This could again point towards a potential chaperone function of PA28 in obstructing mHTT aggregate formation. Another known function of PA28αβ is its role in the cell’s protection mechanism against oxidative stress. Rapidly after the induction of oxidative stress by hydrogen peroxide, PA28αβ and PA28γ bind to free 20S subunits. *In vitro* experiments show that PA28αβ activated proteasomes improve the selective degradation of oxidized proteins, but not their native form. In cells, overexpression of PA28α also improves the capability of the proteasome to degrade oxidized proteins. Reducing PA28αβ levels on the other hand, leads to an increase in carbonylated proteins in differentiating embryonic stem cells, without altering proteasome content. HD progression is associated with increased oxidative damage to DNA, proteins and lipids, which contributes to neurodegeneration (reviewed by Kumar and Ratan and Gkekas et al.). mHTT disrupts nuclear integrity and both mHTT protein and expanded CAG RNA impair DNA repair mechanisms. In combination with the increased oxidative damage to DNA, this contributes to neuronal pathology. Furthermore, HTT itself can be oxidized, which leads to stabilization of mHTT oligomers and accelerates aggregation, by facilitating aggregate interactions. By the dissociation of PA28αβ from the 20S core, the proteasome is less adapted to degrade oxidized proteins, which could accelerate oxidative stress and subsequent pathology in HD. The question remains whether dissociation of PA28αβ activators from the 20S core in the cortex, striatum and hippocampus of HD mice is a cause or consequence of increased mHTT aggregation. During normal homeostasis, proteasome complex formation is a dynamic process, which is altered upon several stimuli, including inhibition of catalytic activity, pro-inflammatory stimuli and competition between the different PAs. The pool of free PA28αβ heptamers, which are not bound to 20S core subunits, suggests that the cell has free PA28 activators that can be used to quickly increase PA28αβ-mediated proteasome activation. This underlines the ability of cells to dynamically regulate the number of PA28αβ activated proteasomes. Upon hydrogen peroxide treatment, levels of PA28αβ activated proteasomes increase within the first hour prior to synthesis of new PA28αβ complexes. Moreover, PA28γ activated proteasomes increase within hours after proteasome inhibition, without an increase in PA28γ transcription. Although the exact signaling pathways remain elusive, phosphorylation of PA28αβ is found to be involved in its binding to the 20S core subunit. Several studies, however, show that kinases are dysregulated in HD (reviewed by Bowles and Jones). This may imply that during HD, general dysregulation of the signaling pathways responsible for proteasome conformational changes may lead to the disassembly of PA28αβ activated proteasomes, after which normal functioning is impaired. However, PA28αβ-20S disassembly may also be a specific response to the increase in mHTT aggregates in affected cells. Since PA28αβ is involved in the clearance of aggregates though its chaperone like function, the increased need for free-PA28αβ may cause the disassembly from the 20S core. Based on this hypothesis, the PA28αβ-20S disassembly observed in mouse brain would not lead to increased aggregation but would rather be a coping mechanism to deal with the aggregates already present in the cell. Indeed, aggregates are already present at the disease stages at which we observe the change in PA28αβ activated proteasomes. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction In the perspective image of a slanted textured surface, oriented components of the texture that are aligned with the 3-D slant converge to form orientation flows, while components orthogonal to the slant increase in frequency. On casual observation, the horizontal component appears perceptually more salient than other components when a surface is slanted (top left and right) than it does when the surface is parallel to the frontal plane (top center). The increase in saliency is more pronounced in complex texture patterns, e.g. the octotropic plaid, which consists of eight gratings of the same frequency, equally spaced in orientation ( bottom). Since these converging orientation flows play a critical role in conveying the perceived 3-D slant and shape of the surface, an increase in their saliency should enhance the 3-D perceived slant. The goal of this work is to examine the neural mechanisms that enhance the visibility of orientation flows. Many surface textures contain components of roughly the same frequencies at many different orientations, with most of the frequencies in the higher frequency declining segment of the human CSF. Slanting the surface increases the frequencies of components not aligned with the slant, thus leading to a reduction in visibility. If different oriented components were processed independently by the visual system, the increase in saliency of the components parallel to the slant could be due just to the reduced visibility of the other components. However independent processing of different orientations is not a feasible premise. The response of oriented neurons in cat and primate striate cortex to a stimulus at a preferred orientation is suppressed by the superposition of a second oriented stimulus, even at the null orientation. Parallel to these results, psychophysical studies have reported that the contrast threshold of an oriented stimulus is increased in the presence of a superimposed orthogonal stimulus. Physiologically measured cross-orientation suppression (COS) is broadband for orientation and occurs over a wide range of spatial frequencies. Psychophysically measured COS appears to be broadband for orientation, but with mixed evidence for frequency-selectivity. Thus it is possible that psychophysically measured COS has components that are distinct from the COS measured in V1 neurons. In this study we identify the mechanism underlying the change in salience of orientation flows. In the first experiment, we show that the visibility of orientation flows increases as a function of surface slant. In the second experiment, we show that the increased salience results from the frequency- selectivity of COS and not the frequency dependent visibility of the masking components. # Methods All research followed the tenets of the World Medical Association Declaration of Helsinki and informed consent was obtained from the subjects after explanation of the nature and possible consequences of the study. The research was approved by the Queens College Institutional Review Board. ## 1. Apparatus and Presentation Stimuli were presented on a 22″ Mitsubishi Diamond Pro 2070 flat screen CRT monitor with an 1024×768 pixel screen running at a refresh rate of 100 Hz via a Cambridge Research Systems ViSaGe Visual Stimulus Generator controlled through a 3.2 GHz Pentium 4 PC. Observers' head positions were fixed with a chinrest situated 1 m from the stimulus monitor. All stimuli were presented so that the center of each image was level with the observer's eye. Viewing was monocular in a dimly lit room, and there was no feedback. ## 2. Stimuli and Procedure Planar surfaces were patterned with horizontal-vertical (h–v) and octotropic plaid patterns and projected in perspective. All stimuli were presented such that the horizontal grating component was interleaved with non-horizontal components in alternating frames at 100 Hz. This technique enabled us to alter the contrast of the horizontal component independently from the other components. For the h–v plaid, the contrast of the vertical grating was fixed at 50%, the highest possible for interleaved frames. Similarly, the contrast of each of the non-horizontal gratings in the octotropic plaid was fixed at the highest possible level, 7.1%. The phases of the plaid pattern components were randomized on each trial. Stimuli were presented in circular apertures spanning 6.5 deg against a grey background at the mean luminance of 58 cd/m². Contrast thresholds of the horizontal component were determined using a 2IFC paradigm. Each session was preceded by a grey screen with a central black fixation cross that remained onscreen for 1 minute. The fixation cross remained onscreen for the duration of the session. After the initial adaptation, a tone signaled the start of the trials. Each of the two stimulus intervals in each trial lasted 500 msec, separated by a 400 msec inter-stimulus interval. Audible beeps of different frequencies signaled the presentation of each of the two stimulus intervals. Test contrast was varied in interleaved 3-down/1-up double- random staircases to ascertain the 79% correct point. Each staircase completed two reversals at 1.8% contrast steps, then eight reversals at 0.4% contrast steps. Threshold was estimated as the average of the last six reversals. ## 3. Experiment 1: Orientation Flow Visibility as a Function of Surface Slant Surfaces were patterned with 3 cpd h-v and octotropic plaids. For each of the two plaid types, observers completed eight sessions which were grouped as follows. One baseline session measured contrast thresholds for the horizontal grating alone in the fronto-parallel orientation. Each of three other baseline sessions measured contrast thresholds of the horizontal grating alone at left and right slants of 25, 50, and 65 deg. In the other four sessions, contrast thresholds were measured in the presence of the non-horizontal components. Thus there were a total of 16 sessions per observer. For each pattern type, the four baseline sessions were run first in random order, then the remaining sessions were run in random order. Each session took approximately 10–15 minutes. ## 4. Experiment 2: Frequency-Selectivity of Cross-Orientation Suppression Mechanism Fronto-parallel surfaces were patterned with an iso-frequency h–v plaid or an h–v plaid consisting of a vertical grating of half the frequency of the horizontal grating. The same surfaces were also presented slanted at left or right at 60 deg which acts to approximately double the vertical frequency in the image. Consequently, the frequencies in the image of the slanted 6 cpd iso- frequency plaid become 6 cpd horizontal and 12 cpd vertical and the frequencies in the image of the 6 cpd horizontal and 3 cpd vertical plaid become approximately equal at 6 cpd. To test whether suppression is a function of the similarity of frequencies between the test and mask, or of the salience of the mask, we needed to select frequencies from the small set that are highly visible, effectively convey surface slant, and are significantly less salient when doubled. We used frequencies of 4 and 6 cpd which satisfy these requirements by being just past the peak of the human CSF. Contrast thresholds for the horizontal grating were measured using the interleaved staircase procedures. Observers ran four different sessions, three times each: one baseline session for the horizontal grating alone at fronto- parallel, rightward slanted by 60 deg and leftward slanted by 60 deg orientations, one session for fronto-parallel plaids and two sessions for slanted plaids. The slanted sessions were blocked in order to contain both types of plaids and both types of slants within each session, while keeping the length of sessions similar to the sessions testing the fronto-parallel stimuli. ## 5. Observers One of the authors and two experienced but uninformed observers participated in this study. All had normal or corrected-to-normal visual acuity. # Results ## 1. Experiment 1: Orientation Flow Visibility as a Function of Surface Slant Contrast thresholds of the horizontal components in the different conditions are shown for the three observers in in separate columns. The panels in the top row plot contrast thresholds of the horizontal grating alone (filled circles), in the presence of the vertical grating in the h–v plaid (open triangles), and in the presence of the seven non-horizontal components in the octotropic plaid (filled squares) as a function of surface slant. The axis along the top of each panel represents the frequency of the vertical grating component as it changes with surface slant relative to the frequency of the test. Thresholds of the grating alone (filled circles) are relatively unchanged by surface slant, reflecting the fact that the spatial frequency of this component is relatively unchanged. The presence of the vertical grating (open triangles) increases thresholds for all surface slants (except for the steepest slants for observer DT), reflecting an overall decrease in visibility of the horizontal component. Thresholds increased even more in the presence of the seven non-horizontal components of the octotropic plaid (filled squares). We quantified the suppression induced by non-horizontal components by dividing thresholds of the horizontal grating in the presence of other components by thresholds in the absence of other components. The suppression factors for the simple and octotropic plaids are plotted as functions of surface slant, with solid and dashed lines respectively, in the bottom panels of. Suppression for both patterns decreases as surface slant increases, with substantially greater and steeper changes in suppression for the octotropic plaid. We have previously shown that perceived orientation flows determine the perception of 3-D shape from texture. 3-D shape is not perceived when the flows are physically present if they are masked by other components (see, Figure 10). The results in indicate that orientation flows are more visible for the h–v than the octotropic plaid at shallow slants, but equally visible at steep slants. Hence, slants should be easier to see for the h–v plaid than the octotropic plaid at shallow angles, but the two should be equally perceptible at steep angles. This prediction is borne out in where the orientation flows and thus slants are easier to see in the h–v plaid than the octotropic plaid at ±40 deg, but are equally visible for the two plaids at ±70 deg. ## 2. Experiment 2: Frequency-Selectivity of Cross-Orientation Suppression It is clear from the results in that contrast thresholds are raised by orthogonal masks, which is a signature of COS. Since the frequencies in the fronto-parallel plane were 3 cpd which is near the peak of the human CSF, the question remains whether the peak suppression is a function of the similarity of frequencies between the test and mask, or of the salience of the mask. The four conditions of Experiment 2 provided two independent comparisons of these hypotheses. In (bottom), mean suppression factors averaged across the three observers are plotted for all conditions for the 4 cpd horizontal grating (left) and the 6 cpd horizontal grating (right). Error bars represent one standard error of the mean. Data in each panel are plotted in the same order from left to right as the four stimulus conditions shown above. First, the similarity hypothesis predicts that thresholds should be higher in the iso- frequency fronto-parallel plaid than for the unequal frequency fronto-parallel plaid, whereas the salience hypothesis predicts the opposite. Thresholds for both the 4 cpd and 6 cpd test gratings were raised more by the iso-frequency mask than the more salient unequal frequency mask. Second, the increase in suppression for the 4 cpd condition when the unequal frequency plaid is slanted (leading to an iso-frequency image pattern) also supports the similarity hypothesis over the salience hypothesis. In addition, in comparing the two slanted plaids, suppression was greater when the image pattern was iso-frequency than when the surface pattern was iso-frequency. Since we expected suppression to decrease with increasing slant for the iso-frequency condition and increase with slant for the unequal frequency condition, we tested for interaction between the frequency conditions and the slant conditions in a 2×2 ANOVA. The interaction was in the correct direction for both spatial frequencies, and statistically significant at the.05 level for the 4 cpd test (F(1,12) = 23.12, p = .0406) but not for the 6 cpd test (F(1,12) = 12.00, p = .0742). These results indicate that the COS from the vertical grating is greatest when the frequency in the projected image is equal to that of the horizontal grating, even when the frequency is one to which we are less sensitive. Previous measurements of the spatial frequency tuning of COS showed a decrease in masking for a 4 cpd test when the mask frequency increased from 4 to 8 cpd, but did not determine whether spatial-frequency mismatch or a decrease in mask saliency was the cause. ## 3. Feed-Forward Models of Cross-Orientation Suppression COS is well-documented in cortical area V1, the first site in the visual pathway containing orientation tuned cells. COS has been attributed to compressive contrast nonlinearities in LGN, but a cortical component has also been revealed. Although several electrophysiological studies examining the frequency selectivity of COS suggest that suppression mechanisms are broadly tuned, it is unclear whether this kind of tuning plays out psychophysically. It would be remarkable if the facilitation of 3-D shape perception occurs automatically through the neural processes that lead to COS, so to ascertain its locus, we have explored the possibility of frequency selectivity in an LGN based model. Although intra-cortical inhibition was the original suggestion for COS, the fact that suppression is not reduced by prior monocular or binocular adaptation to the masking stimulus, that suppression is robust for masks at temporal frequencies beyond the limits of cortical neurons, and that COS has an early onset led to the suggestion that the suppression results from the depression of thalamo-cortical synapses. More recent papers quantifying the fast recovery times of COS and the suppression of both synaptic inhibition and excitation by orthogonal masks challenge the notion of synaptic depression. Instead these models suggest that COS results from contrast saturation and rectifying nonlinearities in the LGN, and expansive spike threshold nonlinearities in the cortex. To test the frequency-selectivity of COS in the models of Li et al. and Priebe and Ferster, we computed cortical responses to a vertical test grating in the presence of superimposed horizontal masks of the same or different frequency. The model simulates responses of a simple cell as determined by excitation of LGN cells tuned to the spatial frequency of the test grating. The receptive field of each ON-center cell is modeled as the difference of two Gaussians:where σ_c is the variance of the central Gaussian, and σ_s is the variance of the surround Gaussian. OFF-center receptive fields were modeled as negatives of ON-center receptive fields. Linear outputs of LGN cells at each location of the stimulus were approximated by convolving ON- and OFF-center receptive fields with the stimulus (either a single vertical grating, or a vertical grating added to a horizontal mask). The outputs were then subjected to a compressive contrast nonlinearity in the LGN of the form:where *R* is the compressed response, L is the linear response, and the value of k dictates the strength of the compression (greater compression for greater values). Excitation and inhibition in the cortical simple cell receptive field has been modeled by summed responses of LGN cells in “push-pull” form. Excitation from ON-center LGN cells and inhibition from OFF-center LGN cells form an ON sub- region of the simple cell receptive field, while excitation from OFF-center LGN cells and inhibition from ON-center LGN cells form an OFF sub-region of the simple cell. Summed excitatory and inhibitory responses are then squared, representing an accelerating cortical spike-voltage non-linearity. (A range of different expansive nonlinearities yielded the same qualitative patterns in our simulation.) This model simple cell gives null responses to horizontal (mask) gratings in isolation. Responses of the model to the test grating plus the mask were computed for masks that were the same frequency as the test, or half, twice, and three times the frequency of the test. We defined response suppression as the response to the grating alone divided by the response to the grating plus the mask. The graph in the bottom of plots response suppression as a function of the frequency of the orthogonal mask relative to the frequency of the test. The points at zero mask frequency represent model responses to the vertical test alone. To test the generality of the simulations, we implemented two different center-surround variance ratios and two different compressive nonlinearities. Each of the four curves represents one combination of these variables: solid lines represent conditions in which the variance of the surround of the LGN cells is twice the variance of the center, dashed lines represent conditions in which the variance of the surround is three times the variance of the center. Square symbols represent NL1 conditions in which k = e in the compressive nonlinearity (Equation 2), and the triangles represent the more compressive NL2 conditions in which k = max(abs(L)). All combinations of receptive fields and nonlinearities lead to frequency selectivity, with suppression greatest when the frequency of the mask matches that of the test. Increasing the surround variance acts to slightly broaden the frequency tuning, and increasing the strength of the compression acts to increase the overall suppression and sharpen the frequency tuning. The magnitudes of suppression reported in Experiment 1 fell between the suppression values for the two model nonlinearities. # Discussion The suggested roles of COS in visual encoding have included orientation tuning, contrast gain control, and redundancy reduction in the coding of natural images. Here we postulate a potential role for COS in the decoding of 3-D slant. We have shown that when textured surfaces are slanted, the release of COS makes the critical orientation flows more visible, which correlates with better perception of 3-D slant. We have shown that COS is frequency specific, and that this specificity can arise in simple feed-forward models of COS. To our knowledge, feed-forward explanations of frequency selectivity of COS have not been suggested previously. The LGN models of COS were formulated on the basis of cat physiology, where almost all cell response functions are compressive as a function of contrast. In primate LGN, M-cells are compressive, but P-cells are fairly linear. Our model thus provides the M-cell component of COS. Since V1 cells get input from P and M-cells, some component of COS involves cortical interactions. Purpura et al. examined whether neurons in V1 and V2 facilitate the extraction of 2-D orientation patterns for the perception of 3-D shape. Of the 29 neurons in macaque V1 and V2 that were isolated from tetrode recordings, flat plaids induced significant suppression in 78% of the neurons compared to optimal single gratings. Suppression was significantly reduced in 45% of the neurons for plaids slanted along or orthogonal to the optimal orientation. In addition, 28% of V1 and 56% of V2 neurons showed enhanced responses to orientation flows per se, indicating that asymmetries may be more prominent in the 2-D structure of V2 receptive fields. Since COS and surround suppression significantly reduce responses to patterns in natural scenes, stimuli that undermine these sources of suppression may allow V1/V2 to mark areas that have a higher probability of containing 3-D shape. In particular, release from cross-orientation suppression serves to enhance the visibility of orientation flows that are the keys to decoding 3-D shapes as signaled by texture, shading, and specular reflection. This work was presented in part at the Vision Sciences Society meeting in Naples, FL, May 2008. The authors wish to thank John Zhu for programming. [^1]: Conceived and designed the experiments: AL QZ. Performed the experiments: AL. Analyzed the data: AL. Contributed reagents/materials/analysis tools: AL QZ. Wrote the paper: AL QZ. [^2]: The authors have declared that no competing interests exist.

# Introduction Atrial fibrillation (AF) is the most common cardiac arrhythmia, causing substantial cardiovascular morbidity and mortality. Although risk factors have been described, there are no available blood tests to predict AF risk. The activation of renin-angiotensin system (RAS) definitively plays an important role in the pathogenesis of atrial fibrillation. The angiotensin II is a primary mediator of RAS which is mainly produced from angiotensin I by angiotensin- converting enzyme (ACE, EC 3.4.15.1, CD143). Patients with AF are known to have increased level of ACE expression in the heart tissue, in particular, in atria. Study on the transgenic mice also demonstrated that the increased ACE expression in the heart might be a causative factor for AF and sudden cardiac death. Both ACE inhibitors and angiotensin receptor blockers reduce AF incidence and may prevent AF-related complications in patients and in experimental models. ACE is a Zn²⁺ peptidyldipeptidase which plays key roles in the regulation of blood pressure by producing angiotensin II and degrading bradykinin and in the development of vascular, including cardiac, pathology and remodeling. ACE is constitutively expressed on the surface of endothelial and some epithelial cells, as well as cells of the immune system (macrophages, dendritic cells) reviewed in. ACE expression in the normal and pathological human hearts has been studied previously. To date, several new substrates for ACE have been identified (AcSDKP, angiotensin 1–7) and new functions for ACE have been proposed, such as outside-in cell signaling, antigen presentation. The anti-fibrotic and anti-inflammatory actions of a substrate for ACE, AcSDKP, seem to be especially important for AF pathogenesis in patients with the enhanced ACE level in the heart. Despite the strong relationship of RAS activation to arrhythmias, plasma levels of ACE do not correlate to AF and ventricular arrhythmias. Generally, plasma ACE accurately reflects the level of tissue ACE. Blood ACE originates from endothelial cells, mostly lung capillaries, which exhibit nearly 100% ACE expression compared to only 5–15% ACE-positive capillaries in the systemic circulation. Based on that, we estimated that ACE shed from heart capillaries could represent not more than 1% of total ACE in the blood. This is why a total plasma ACE level in patients does not seem to be a predictive parameter for patients with AF. However, as atrial ACE increases 3-fold in patients with AF, the quantification of the definite heart-derived ACE (mainly, atria-derived ACE) in plasma theoretically could be used as a predictive test for the risk of AF. We proved the concept that the conformation of ACE is cell- and tissue-specific and stems likely from different glycosylation of the enzyme on the example of ACE from lung endothelial cells versus ACE from macrophages and dendritic cell of sarcoid granuloma and ACE from epithelial cells of prostate. The ACE conformational fingerprint based on the pattern of the binding of a set of mAbs to different epitopes on the surface of ACE has, therefore, a potential for the disclosure of the cells/organs from which ACE originates. We applied this approach here to demonstrate conformational differences of ACEs originated from human heart and lung and showed differences for purified ACEs, for ACEs in tissue homogenates, and ACEs in the blood after *in vivo* perfusion. We believe that such differences will allow the generation of monoclonal antibodies (mAbs) able to distinguish ACEs shed to the blood circulation from these two organs and, therefore, form the base for the blood test for predicting the risk of AF. # Experimental section ## ACEs from different sources The work was carried out in accordance with The Code of Ethics of World Medical Association (Declaration of Helsinki) and was approved by the Institutional Review Boards of the Bakulev Center of Cardiovascular Surgery, Moscow State University, and the University of Illinois at Chicago. None of the donors were from the vulnerable populations and all donors or next of kin provided written informed consent that was freely given. Human citrated plasma and tissue homogenates (1:9 and, in some cases, 1:3 w/v ratio) from 10 donors (five male donors, age 54–66, and five female donors, age 28–68), including three donors with atrial fibrillation, were used as sources of somatic ACEs. Lung and heart ACEs were purified from tissue homogenates using anion-exchange chromatography on DEAE-Toyopearl 650M and then lisinopril affinity chromatography—as in “”. Purified ACE preparations were proved to be homogeneous by electrophoresis in 7.5% SDS-PAGE “”. ## ACE activity assay ACE activity in blood plasma, homogenates of human organs or homogenates of heart chambers was measured using a fluorimetric assay with two ACE substrates, 2 mM Z-Phe-His-Leu (ZPHL) and 5 mM Hip-His-Leu (HHL). Inhibition of ACE activity with anti-catalytic mAb 5F1 to the N domain of ACE and mAb 1E10 to the C domain was performed at mAbs concentrations 100 μg/μl and 10 μg/μl, correspondingly, with 1mM ZPHL or 2.5 mM HHL as substrates. ## Substrate specificity of ACE The kinetic parameters of the hydrolysis of several synthetic tripeptide substrates and natural substrate decapeptide angiotensin I by purified human heart and lung ACEs were determined in 0.05 M phosphate buffer, pH 7.5, containing 0.15 M NaCl and 1 μM ZnCl₂, at 25°C. The rates of enzymatic hydrolysis of Z-Phe-His-Leu, Hip-His-Leu and angiotensin I were determined fluorimetrically, whereas the kinetics of the hydrolysis of FA- containing substrates, FA-Phe-Gly-Gly and FA-Phe-Phe-Arg, was studied spectrophotometrically. ## Immunological characterization of ACE (Plate immunoprecipitation assay) Ninety six-well plates (Corning, Corning, NY) were coated with anti-ACE mAbs via goat anti-mouse IgG (Pierce, Rockford, IL) bridge and incubated with different sources of ACE, which were equilibrated for ACE activity. After washing off unbound ACE, plate-bound ACE activity was measured by adding a substrate for ACE, Z-Phe-His-Leu, directly into the wells. Sixteen mAbs to human ACE were generated in our lab, while mAb BB9 was kindly provided by Paul J. Simmons (then Brown Foundation of Molecular Medicine, University of Texas Health Science Center, Houston, TX, USA). ## Dialysis and filtration of human plasma and heart and lung homogenates Dialysis of the heart and lung homogenates was performed in 10 kDa dialysis cassettes (Pierce, Rockford, IL) and in 100 kDa Biotech dialysis tubes (Spectrum Inc., Houston, TX) against 0.05 phosphate buffer, pH 7.5, 0.15 M NaCl and 1 μM ZnCl₂, at 4°C. Filtration of the homogenates was performed on Vivaspin filtration membranes (GE Healthcare, Sartorius Corp., Bohemia, NY) with 30 kDa and 100 kDa limits at 12 000g. ## Aminopeptidase activity Aminopeptidase activity in tissue homogenates at different dilutions was estimated by the rates of the hydrolysis of 0.01–0.1 mM His-Leu in PBS-BSA buffer, pH 8.3, at 37°C. ## ACE perfusion in rats All *in vivo* rat methods/experiments were approved by and performed in accordance with Moscow State University Committee on the Ethics of Animal Experiments guidelines and regulations that conform to the NIH guidelines (Guide for the care and use of laboratory animals). Adult male Wistar rats weighing \~300 g were put under pathogen-free condition in individual plastic cages with sawdust bedding in an air-conditioned room under constant temperature (23±1°C) and 70% humidity. Rats were provided with water and standard diet (LabDiet, 5053 (LabDiet; St. Louis, MO, USA)) *ad libitum*. All animals were monitored for body health on a daily basis for a week before an experiment. All efforts were made to minimize rats suffering. Purified human heart and lung ACEs, 1000 mU in 100 μl, pH 7.4, were injected into the tail vein of rats (two rats per group) with anesthesia with ketoprofen. After 30 min, euthanasia was performed by decapitation, the blood was collected and citrate plasma prepared. At that moment, about 30% of human ACE was still in the rat blood. Precipitation of human ACE from rat plasma by a set of mAbs to ACE was performed as described above but corrected for trace precipitation of the rat ACE by these mAbs. ## Statistical analysis All data are means ± SEM. Significance was analyzed using the Mann-Whitney test with STATISTICA 6 (StatSoft, Inc., OK). # Results and discussion ## ACE activity in the human heart We estimated ACE expression in different human tissues, namely, heart, lung, kidney and spleen, by the comparison of ACE activity in homogenates of these tissues as well as in human citrated plasma. ACE activity in the human heart homogenate (expressed in mU per gram of tissue) was about 3-fold more than in human plasma and 8-12-fold less than in human lung. This estimation correlates with the density of radioligand ACE inhibitor binding sites in human heart and lung, as well as with high ACE RNA transcription in the lung, while negligible in the heart, as in Human Protein Atlas. ACE activity measured simultaneously by two substrates in the homogenates of different whole heart chambers showed statistically significant difference in atrial and ventricular ACE activity, this activity being the highest in the right ventricle and the lowest in the right atrium (insert). Mammalian tissues and blood contain endogenous ACE inhibitors and ACE effectors, including putative ACE binding proteins. In order to demonstrate the presence of endogenous ACE inhibitors in human tissues we compared an apparent ACE activity in the heart and lung homogenates at serial dilutions. An apparent ACE activity significantly increased 3-4-fold during dilution of both homogenates. We excluded the presence of aminopeptidases in human tissues as a reason for that effect “”. Thus, an effect of dilution reflects the presence of endogenous ACE inhibitors in heart and lung tissues. Dialysis (10 or 100 kDa) and filtration (100 kDa) of the heart and lung homogenates also resulted in a similar (200–400%, p\<0.05) increase of an apparent ACE activity (data not shown). Therefore, in order to estimate correctly ACE activity in the heart (or lung) homogenate it is not necessary to perform time-consuming dialysis, as simple dilution of homogenates 10-fold successfully induces dissociation of low molecular weight (LMW) ACE endogenous inhibitors from their complex with ACE. We discovered that the effect of the dilution of tissues homogenates on relative ACE activity with two substrates, Z-Phe-His-Leu and Hip-His-Leu (ZPHL/HHL ratio), was substrate specific. We have shown previously that the selective inactivation/inhibition of the C domain increases ZPHL/HHL ratio for somatic, two-domain ACE, whereas selective inactivation/inhibition of the N domain decreases this ratio, as these substrates are cleaved by the two domains with different rates. Therefore, we can interpret the decrease of ZPHL/HHL ratio and increase of this ratio upon dilution of heart and lung homogenates, respectively, as an evidence of the presence of different sets of endogenous ACE inhibitors in the heart and lung tissues. ## Conformational fingerprinting of heart and lung ACEs We characterized the conformation of the heart and lung ACEs using a panel of mAbs directed against 17 different epitopes and mapped on the surface of the N and C domains of catalytically active human ACE–a method of “conformational fingerprint of ACE”. As apparent in, the immunoprecipitation profile of heart ACE significantly differed from that for lung ACE, the difference being observed both for the purified enzymes and ACEs in homogenates. This allowed us to conclude that heart ACE, originated from heart endothelial cells and perhaps myofibroblasts and lung ACE, originated from lung endothelial cells, exhibit different local conformations of their surface, probably, due to tissue- and cell-specific post-translational modifications (PTM). PTM, which are common for various proteins, can precisely regulate the functions of proteins by inducing conformational changes, which subtly or dramatically alter the protein surface or its overall tertiary structure. The most common PTM is glycosylation. Somatic ACE represents a type-I N-glycosylated membrane glycoprotein, exact glycan structures of which, as well as locations of actually present oligosaccharide chains on the surface of the protein globule, can vary with a protein source. The amino acid sequence of human somatic ACE contains 17 potential sites for N-glycosylation. The structure and exact positions of glycan moiety in human somatic ACE from different tissues were not investigated completely. Heart and lung ACEs possess rather equal masses about 170 kDa as was revealed by SDS-PAGE “”which indicates on the absence of major differences in the degree of glycosylation of these ACEs. It was shown earlier that common sialylated biantennary complex oligosaccharide can contact the surface of the enzyme within an area about 200–300 Å. As epitopes for mAbs are usually 600–900 Å, it is obvious that the presence of the oligosaccharide within an epitope (as well as its definite structure, i.e. the number of antennae, sialylation, fucosylation, etc.) can hugely influence mAbs binding. Previously, we showed that different glycosylation of ACE in endothelial and epithelial cells could be the main reason for the differences in mAbs binding to ACEs from lung and seminal fluid. Moreover, the difference in binding efficiency for some particular mAb can be attributed to the different glycosylation of the corresponding glycosylation site within epitope for this very mAb. So, the main reason for the difference observed for immunoprecipitation profiles of the heart and lung ACEs could be due to different glycosylation of ACE, even in endothelial cells but from different organs, induced by different environment of these cells in different organs. Conformational fingerprints of heart and lung ACEs allowed us to suggest that different glycosylation of these ACEs occurs in the following glycosylation sites on ACE: Asn25 in the epitopes for mAbs BB9, 3A5 and i1A8, and Asn117 in the epitope for mAb 5F1 on the N domain, as well as Asn666 in the epitope for mAbs 4E3 and 1E10, and Asn731, which is within the epitopes for mAbs 1B8 and 3F10 and close to the epitope for mAb 1B3 on the C domain of ACE. The purified heart and lung ACEs were isolated from different donors and strictly speaking the differences in mAbs binding to these purified ACEs could be attributed to the inter-individual differences in the protein glycosylation in these donors, in our case, ACE glycosylation. In order to exclude this possibility, we performed the conformational fingerprints of heart and lung ACEs in homogenates obtained from the same donor. For statistical reasons and reproducibility, we used 10 pairs of homogenates obtained from the tissues from 10 donors. As the general pattern of mAbs binding to ACE in pairs of homogenates was similar to the pattern of mAbs binding to purified heart and lung ACEs from different donors we can conclude that different conformational fingerprints of heart and lung ACEs demonstrate a real conformational tissue specificity of the enzyme. It is worth noting that ACE from donors with AF in anamnesis demonstrated similar differences in mAbs binding to heart and lung ACEs (not shown), that is this difference is not caused by the disease but by the nature of ACE-producing cells. In order to confirm that the differences in local ACE conformation of heart and lung ACEs retain in the blood we perfused purified ACEs from heart and lung (and seminal fluid ACE as a representative of ACE produced by epithelial cells) via alive rat circulation. Circulation of glycoproteins (including ACE) in blood after shedding from the cell membranes leads to the enrichment of serum glycoproteins with molecules with a higher content of sialic acid residues as a result of the selective removal of asialo-molecules (with terminal galactosyl residues) by hepatic lectin, while sialylated glycoproteins remain in the circulation. The results clearly demonstrate that the effect of such perfusion on the pattern of mAbs binding to heart and lung ACEs is significantly different, once again demonstrating that the patterns of glycosylation (sialylation, in particular) of the heart and lung ACEs are different. In addition, the binding of mAbs 5F1 and 1E10, which better bind to ACE molecules when glycan in the epitopes for these mAbs is less sialylated (Danilov and Trakht, unpublished observation), decreased after perfusion of the lung ACE into the rat in accordance with expected relative enrichment of ACE moiety with sialic acids. As a result of the perfusion of the heart ACE, however, the binding of these very mAbs unexpectedly increased, which indirectly confirms that the local conformation of heart ACE and its sensitivity to the structure of glycan differs from that for lung ACE. It is important, that even after circulation in the blood, the immunoprecipitation profiles of soluble ACEs from heart and lung remained significantly different ( and ““) proving the possibility to distinguish ACEs produced by different organs in a real blood. Therefore, an analysis of ACE conformation provides a structural base for the generation of mAbs able to distinguish ACE produced by a definite cell type or definite organ, in particular, by heart, and to generate a blood test for the detection and quantification of this particular ACE in the blood for the prediction of AF risk. ## ACE effectors in human heart tissue As we discussed above, tissue homogenates contain high amounts of ACE inhibitors/effectors. When we performed purification of heart and lung ACEs by a combination of anion-exchange and affinity chromatography, we noticed significant influence of this purification procedure on the ACE conformational fingerprint (and “”), that is a microenvironment significantly influenced on ACE conformation in the tested tissues. Purification of both ACEs from the corresponding homogenates led to a decrease of the binding of mAbs 1G12, 6A12, and i2H5 having overlapping epitopes on the N domain, which could be attributed to the dissociation of endogenous ACE inhibitors from the complex with ACE. In addition, several antibodies to the C domain, mAbs 1B3, 1B8, 3F10, 1E10, and 4E3, dramatically increased their binding to ACE as a result of the enzyme purification. It is important that we did not observe such remarkable increase in the binding of these mAbs neither after dialysis nor after filtration of homogenates through filters with 30 kDa limit (not shown) or even 100 kDa limit. These data allowed us to suggest that, in addition to the presence of endogenous easily-dissociating (due to dilution, dialysis or filtration) LMW ACE inhibitors (Figs) in both heart and lung tissues, these tissues contain some high molecular weight (HMW) ACE effectors/binding proteins, which could not be eliminated via filtration or dialysis but are eliminated during ACE purification procedures (“” and “”). ## Kinetic characteristics of heart and lung ACE In order to determine whether these differences in conformation of heart and lung ACEs are reflected in different functions of these ACEs, we compared kinetic characteristics of these enzyme. The pH-dependencies of enzymatic activity of heart and lung ACEs in the hydrolysis of Z-Phe-His-Leu at 0.15 M NaCl were quite similar with pH-optimum about 7.0 in accordance with previous results obtained for bovine ACE. However, the pH-optima for the hydrolysis of Hip-His-Leu were equal to 7.5 for heart ACE and 7.8–7.9 for lung ACE, thus confirming different conformations of the two enzymes. The inhibitory effect of anti-catalytic mAb 5F1 to the N domain and mAbs 1E10 and 4E3 to the C domain, while these mAbs differently bound to the heart and lung ACEs, did not show, however, significant difference in the extent of inhibition of these two ACEs (not shown). The kinetic parameters, *k*_cat and K_m, of the hydrolysis of several synthetic substrates, as well as a natural substrate angiotensin I, by purified heart and lung ACEs are presented in. Two of the substrates, Z-Phe- His-Leu and Hip-His-Leu, are C-terminal analogs of angiotensin I, while a substrate with Phe-Arg on its C-terminus could be considered as a C-terminal analog of another natural ACE substrate, bradykinin, as well as C-terminal analog of atriopeptin 2. The absolute k_cat values calculated for the hydrolysis of angiotensin I by the human ACE markedly differ in the literature, from 3.5 s^-1 to 40 s^-1 for recombinant human ACE expressed in CHO cells, to 250 s^-1 for human heart ACE, and to 66 s^-1 for human kidney ACE (Kost OA, unpublished data), likely due to different conditions of angiotensin synthesis, different conditions of its hydrolysis, and different conditions and extent of ACEs purification. So, to minimize the putative errors, we determined the kinetic parameters of the hydrolysis of definite substrates by the two ACEs, from heart and lung, simultaneously. The K_m values of the hydrolysis of any substrate by the two ACEs appeared to be similar, as well as the k_cat values for most substrates. However, for a short substrate Hip-His-Leu, the k_cat values of the hydrolysis differed two-fold for the heart and lung ACEs. Some differences in the catalytic properties of heart and lung ACEs were shown earlier for human and bovine ACE, as well as different ability of these enzymes to be inhibited by a set of inhibitors. There are some examples demonstrating the influence of glycosylation on the kinetic properties of the enzymes, and references herein, including ACE which activity could vary 2-fold due to different glycosylation. Thus, it is likely that different ACE conformations induced by different protein glycosylation in different tissues could be the reason for different kinetic characteristics of these ACEs. We could not even exclude the possibility that unique ability to hydrolyze atriopeptin II which was reported only for heart, but not lung, ACE could be partially explained by the differences in the conformations of the two ACEs due to different glycosylation. ## ACE phenotyping in different heart chambers In order to estimate putative heterogeneity of ACE expression in heart chambers we performed ACE phenotyping in homogenates of the whole heart chambers from 10 donors. The presence of endogenous ACE inhibitors in heart chambers was demonstrated by the comparison of the apparent ACE activity in atria and ventricles at different dilutions. The apparent ACE activity in heart chambers markedly depended on homogenate dilution, as was observed for the whole heart and lung homogenates. Effect of homogenates dilution confirmed the lower apparent ACE activity in atria in comparison with ventricles. The amount of inhibitors in heart chambers, however, varied in different donors. Bearing in mind that ventricles comprise the main part of the whole heart it is obvious that ventricle tissues are the main contributor to the amount of ACE inhibitors in the heart tissues. When we compared the amount of immunoreactive ACE protein determined by precipitation of ACEs by mAb 9B9 with an apparent ACE activity in different heart chambers, we have found that their ratio in the right atrium is significantly bigger than in other chambers, especially for undiluted homogenates, which is likely explained by the higher content of ACE inhibitors in the atrium. Conformational fingerprinting of ACE in heart chambers revealed statistically significant increased binding of mAbs 1B8 and 3F10 to ACE from the right atrium, while the difference in the binding of mAbs to ACE from the right ventricle or the left atrium was negligible. This difference was observed both in undiluted and diluted homogenates, which means that the reason is the conformational difference between ACEs expressed in atrium and ventricle, not the presence of different effectors. Moreover, conformational fingerprinting of ACE was separately performed on 10 sets of homogenates of heart chambers obtained from 10 donors for reproducibility and statistics. These two mAbs, 1B8 and 3F10, have highly overlapping epitopes on the C domain which contain potential glycosylation site Asn731. Highly probable that different microenvironment of ACE in the heart chambers (Figs) can lead to different glycosylation of Asn731 in ACE in the right atrium, compared to the glycosylation of this particular glycosylation site in ACE in other chambers, and, as a results of that, lead to different effectiveness of the binding of mAbs 1B8 and 3F10. It is worth noting that atrial tissues are likely to perform specific biochemical functions. As an example, only atrial bovine tissue was reported to contain a metallodipeptidyl carboxyhydrolase (EC 3. 4.15.4) which resembles ACE, as it is able to hydrolyze ACE common substrate Hip-His-Leu and be inhibited by ACE inhibitors. This enzyme, however, differed from ACE in a number of molecular and kinetic properties, the major difference being the ability to hydrolyze atriopeptin II, which could be considered as heart-specific substrate, and its analogue Hip-Ser-Phe-Arg, and references herein. Thus, there could be major differences in the biochemical processes within the heart chambers due to different enzymatic content. However, there could be also more subtle differences due to the changed properties of the same enzyme expressed in different heart chambers. ## Conclusions The significant differences in the local conformations of heart ACE (originated from heart endothelial cells and likely from myofibroblasts) and lung ACE (originated from lung endothelial cells) allow us to suggest that the properties and functions of ACE could be sensitive to the microenvironment and be regulated by accompanying constituents of tissues and blood, this regulation could depend on a set of possible ACE effectors in heart and lung tissues. Therefore, significant structural differences in ACE from heart and ACE from lung demonstrated in this study may be the base for the generation of mAbs, which will distinguish these two ACEs. Possible limitation of this study is that we did not generated yet these mAbs. We have to say that this is a pioneer and extremely difficult work because the discriminative power of such mAbs should be very high (about 100-fold) in order to detect only about 1% of heart-derived ACE in the blood where the majority of ACE molecules comes from lung. Therefore, the occurrence of hybridomas with required specificity would be low demanding to test thousands of hybridomas producing anti-ACE mAbs. However, such mAbs (if we will be able to select them) may have a potential for the development of blood-based assay for quantification of heart-derived ACE in the blood for identification of the patients with the increased level of heart ACE, i.e. with the increased risk of atrial fibrillation, which cannot be achieved by estimation of ACE activity in the blood. # Supporting information The authors are grateful to Prof. V. Pozdnev (Institute of Biomedical Chemistry, Moscow) for the substrate Fa-Phe-Phe-Arg and to Dr. V. Popov (M.V. Lomonosov Moscow State University) for the help with the experiments with rats. This work was supported by the Ministry of Science and Education of Russian Federation (grant number 14.Z50.31.0026). ACE angiotensin-converting enzyme AF atrial fibrillation FA- furanacryloyl- HHL (Hip-His-Leu) hippuryl-L-histidyl-L-leucine HMW high molecular weight LMW low molecular weight PTM post translational modifications RAS renin-angiotensin system ZPHL (Z-Phe-His-Leu) carbobenzoxy-L-phenylalanyl-L-histidyl-L-leucine [^1]: The authors have declared that no competing interests exist.

# Introduction Diabetes Mellitus (DM) refers to a group of common metabolic disorders that share the phenotype of hyperglycemia. It is one of the four common non- communicable diseases causing major morbidities and mortalities. World Health Organization (WHO) estimated that 422 million adults had diabetes in 2014. International Diabetes Federation (IDF) estimated this number to be 629 million by 2045. In Africa, the prevalence of diabetes adults between 20–79 years was 16 million in 2017 and projected to be 41 million in 2045. About 69.2% were undiagnosed. Africa attributes 77% of the deaths under 60 years to diabetes mellitus. The figure is the highest proportion in the world. International Diabetes Federation in 2015 estimated that 5.2% of Ethiopian adults had diabetes. Self-management is the ability of the patient to deal with all that a chronic illness entails, including symptoms, treatment, physical and social consequences, and lifestyle changes. Since health care cost for the treatment of acute and chronic complications of diabetes is high, self-management is compulsory. Ethiopia is one of the low-income countries. Thus, supporting patient self-management practice plays a key role in effective chronic illness care and improve patient outcomes. Effective management of diabetes requires predominantly self-directed practices, where the individuals become responsible for the day-to-day decisions related to controlling their disease. The emerging evidence supports the implementation of practice strategies that are conducive to patient self-management and improved patient outcomes among chronically ill patients. Implementation of self-management needs to be individualized, and people with diabetes should be assisted to understand the impact of medication, food and physical activity on blood glucose control. Adequate self-management can minimize the disease and disease-related complications, while the frequency of self-monitoring can be determined based on the individual’s self-management goals. The key goal of self-management is controlling blood glucose, improving quality of life and reduction of diabetes-complications. A systematic review conducted in 2018 in Sub-Saharan Africa on self-management of Type 2 diabetes shows that patients rarely self-monitored their glucose levels; had low duration/frequency of physical activity; moderately adhered to recommended dietary and medication behavior and had a poor level of knowledge about diabetes complications. Numerous studies conducted in different parts of Ethiopia shows slightly more than half of patients living with diabetes had good self-management practice. Different cross sectional survey indicated that dietary and medication adherence of diabetic patients was low. Internationally, only two-thirds of diabetic patients perform daily self-monitoring of blood glucose. However, self- monitoring of blood glucose was good in Ethiopia. Inadequate attention to diabetes center and lack of knowledge about self-management was reported as main factor in Ethiopia. Self-management practice in clients with chronic diseases is essential to maintain good health while taking the medications. However, the practice may vary from client to client due to several factors. Different researches have investigated that age, current occupation, lack of awareness, absence of self- practice health education, years of suffering from DM, having family members suffering from the illness and lack of knowledge about the illness were the factors that affect the level of self-management behaviors. Besides, belief in treatment effectiveness, family support, self-efficacy, awareness about the disease and social support were also the factors that affect self-management practice. Despite the knowledge about the factors that affect self-management practice, there was no comprehensive study conducted on self-management of chronic illness especially DM covering the different hospitals located in western Oromia, Ethiopia, as far as the researchers’ knowledge is concerned. This study thus focused on the assessment of self-management practice and its predictors in the study facilities. The findings of this study believed to give useful input for policymakers mostly in enforcing and establishing self-management practice interventions which, in turn, may encourage immediate health care providers to consider it in their routine care practices. # Methods and materials ## Study design and setting A hospital-based cross-sectional study was conducted among diabetic patients on follow-up at diabetic centers from November 2017 to February 2018 in public hospitals found in western Oromia, Ethiopia, namely Nekemte specialized hospital, Gimbi general hospital, and Nedjo general hospital. ## Source population All diabetic patients on follow-up attending hospitals were considered as source populations. ## Study population All diabetic patients attending diabetic centers and wards of the study hospitals during the study period were subjects of the study. ## Eligibility criteria Known diabetic patients who visited the diabetic centers for follow-up and wards to receive care were included in the study while patients with diabetic emergencies like diabetic ketoacidosis and diabetic coma were excluded. ## Sample size determination A single population proportion formula was used to determine the sample size. The proportion of patients who performed self-management practice (54.7% according to a study done at Nekemte referral hospital, Ethiopia in 2013) was considered in sample size calculation. Marginal error between sample size and population parameter of 5%, and 95% confidence level, and 5% non-response rate was considered. A total of 400 patients with known diabetes mellitus participated in the study. ## Sampling techniques All known diabetic patients visited the study hospitals for follow-up, and those admitted to the wards were taken into consideration. The average monthly client load was taken from the daily average DM client flow of the hospital and the registry book. Systematic random sampling was used. The interval was calculated at each hospital. The sample was allocated proportionally to the client flow of the respective facilities. The first client who arrived at the waiting area on the first day of data collection, and who met the eligibility criteria was taken as the first candidate for the study. This process continued until the desired sample size was attained. ## Data collection tools and methods The data was collected directly by interviewing diabetic clients after getting informed consent. The questionnaire was prepared in English by modifying from different literature sources with similar areas of interest. It was translated from English to Afaan Oromo, a local language, and re-translated back to English to ensure consistency. The questionnaire was pre-tested in Dambidolo hospital. Three trained diploma nurses were used as data collectors under the close supervision of one B.Sc degree nurse, and the data were collected in a face-to- face interview. For the presence of co-morbid, we had observed the patent’s folder. The only physician confirmed and recorded disease(s) was taken as a co- morbid disease. ## Data processing and analysis Data were entered into Epi Info 3.5.4 software package and cleaned first. Then, the analysis was made using a statistical package for the social sciences (SPSS) software package version 20. Analysis of overall self-management practice was done by transforming the scores on closed-ended questions related to self- management practices. Using the odds ratio (OR) with a 95% limit of the confidence interval, the association of dependent and independent variables was analyzed, and their degree of associations was computed. Potential confounding variables were controlled using binary and multivariate logistic regression. Statistical significance was considered at P\<0.05. Finally, the analyzed data were presented using frequency, percentage, and texts. ## Data quality control A pre-test was conducted at Dambidolo hospital on 5% of the total sample size to check clarity, understandability, and consistency of the data collection tool. Then, the necessary amendments were made to the questionnaire before the full- scale data collection was implemented. Data collection was conducted under the close supervision of supervisors and the collected data were checked for completeness. ## Study variables ### Dependent variable Diabetes self-management practice. ### Independent variables Sociodemographic characteristics like sex, age, occupation, marital status, religion, level of education, lack of self-management education, patient education, getting family support, presence of DM-related complications and other health problems. ### Operational definitions **Self-management**: The practice of diabetic patient’s self-initiated and performed activities to control disease and maintain life, health, and wellbeing. **Good self-management practice**:—Diabetic patients with average and above scores on closed-ended questions related to self-management practices. **Poor self-management Practice**:—Diabetic patients with less than average score on closed-ended questions related to self-management practices. **Hyperglycemia**—an abnormally increased concentration of glucose in the blood (≥ 126 mg/dl at FPG). ## Ethics statement Ethical approval was obtained from the Research and Ethics Committee of Wollega University. An official letter was written to each hospital to get official permission. Participants were informed that privacy and confidentiality were maintained. Written consent was taken from the study participants. # Results ## Sociodemographic characteristics of the respondents depicts the overall characteristics of the study participants. A total of three hundred ninety-eight diabetic patients participated in this study, raising the response rate to 99.5%. More than half of the respondents were male (225, 56.5%). The average age of all respondents was 41.33 ± 18.93 (SD) years. Majority of the participants were married (255,64.1%), Oromo people (377, 94.7%), living in urban area (204, 51.3%) and Protestant Christians (219, 55.0%). The remaining (106, 26.8%) were single, 20 (5%) Amhara people, (194. 48.7%) living in rural areas and (136, 34.12%) followers of Orthodox Christian religion. One hundred eighteen (29.6%) respondents had attended college/university, whereas about 92 (23.1%) can not read and write, and only 20.4% had attended secondary school education. Student patients were relatively high (21.6%) followed by housewives (20.4%) and government employees (20.4%). Based-on participants’ believe, more than half, 54%, of the participants had a middle-level income compared to their neighbors. About 30.7% of the participants were poor and a very small proportion of the participants were very rich (0.3%). The average family size was 5 ± 2 (SD). About three fourths of the participants (74.6%) were husbands and wives in the family, followed by sons and daughters (22.9%). Most of the study participants (91.0%) reported that they need family support for disease treatment. About two- thirds of the participants (65.8%) received support from their families. More than half of the participants, 220 (55.3%) did not know the type of diabetes they have. One hundred twenty-three (31.0%) of the participants had Type 1 diabetes, while the other had Type 2 diabetes. One hundred eleven (28.0%) had the disease for more than a decade, whereas about 72% live with DM for less than a decade. One-third of the study participants (66.8%) believe that they can cured of the disease, and the majority (71.6%) thought that the medication they were taking could cure them. Near half of the participants, 195 (49.0%) believe that health care providers had a good approach towards them. And two-thirds of the participants, 266 (66.8%) alleged that they would recover from the disease while others did not. The majority of the participants (93.8%) had either of the following chronic illnesses: hypertension, cancer, edema, Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome (HIV/AIDS), dyspepsia and bronchial asthma. About three-fourth (74.1%) of them had diabetes-related manifestations and complications. Again, more than half of the participants had shock (55.9%) followed by diabetes ketoacidosis (52.9%). Some of the respondents also reported nerve (32.9%) and eye problems (23.4%). Study participants had practiced different self-management interventions. They had also been taking medications and most of them had a monthly follow-up. About half (49.7%) of them had been taking insulin regimen followed by oral hypoglycemic agents (39.7%) for treatment and glycemic control. Only 42 (10.6%) were taking combined medications. The adherence to medication was varying. About the participants’ diet intake, the participants had been consuming food three times a day with the different food menu. However, only one-third of them used the food menu. About half of the respondents used to consume vegetables followed by starch at lunchtime. At dinner, 39.2% of them used starch, while the others (43.0%) used vegetables. Close to three-fourths of the participants (72.4%) did not use to consume snacks, and starch was the most consumable food item. About half (52.0%) of them performed regular physical activity while the remaining did not (48.0%). Again, more than half of the respondents (57.8%) performed regular foot care, whereas 168 (42.2%) did not. Participants did not perform annual foot care check-up. ## Predictors of self-management practice Overall, about 63.6% of the study participants self-management practice was good, while 36.4% self-management practice was poor. One hundred eighty-four (46.2%) of the participants knew about diabetes self-management management, of which 129 (32.4%) were practicing it. Cross-tabulation of the variables showed that those who are female, living in urban, married, rich and merchant tended to practice diabetes self-management. Females practice self-management more than males, 74.2% and 68% respectively. The magnitude of self-management practice was relatively higher among married clients (73.5%) than never married (66.7%). About 73.7% of participants from urban and 64.3% of rural had practiced self- management. No difference in self-management practice was observed between participants who had developed DM-related complications and who had not. Self- management practice was high in participants who believed DM is curable, than those who did not believe so. Participants who were merchants practiced more diabetes self-management, followed by daily laborers (79.2%). Those economically rich tended to practice more self-management than their poor counterparts. The logistic regression analysis results indicated that only two variables, namely occupation and having family support had shown statistically significant association with the practice of self-management management in both binary and multiple logistic regressions. Accordingly, merchants were observed practicing self-management about six times higher than clients with other occupations with AOR of 5.945 (1.177–30.027 at 95% CI). Clients who had family support in DM- related care were again observed practicing self-management 2.87 times more than those who had no family support with AOR of 2.835 (1.386–5.801 at 95% CI). No difference was observed among other variables entered into the regression model. # Discussion Among 398 study participants, 63.6% of them had good self-management practice. Self-blood glucose monitoring, physical activity, diet adherence, medication adherence, and foot care are the components of self-management practice. This finding is comparable with the finding from Northern Ethiopia. The result of this study demonstrates that self-management practice of people living with diabetes mellitus attending hospitals in western Oromia is higher than the result of previous studies conducted in Nekemte referral hospital, Jimma University teaching hospital, and Harar, Ethiopia, and Lesotho. This maybe due to day to day improvement of awareness of the disease management and cultural variation. Physical activity believed as essential in controlling blood glucose. The result of this study revealed that more than half of participants practiced regular physical activity. The participants level of physical activity is greater than the finding of the study conducted in Harar, Ethiopia. This discrepancy may be due to increased awareness regarding diabetic self-management and differences in the measurement tools. American Diabetes Association (ADA) recommends that physical activity performance five days/week that lasts for 30 minutes. In this study, only half of the respondents self-reported as they perform regular physical activity which is inconsistent with the recommendation of ADA. Under performance of physical activity perhaps a lack of awareness about recommendations on physical activity and lack of infrastructures. ADA recommends diabetic patients should perform a comprehensive foot evaluation at least annually to identify risk factors for ulcers and amputations. More than half of diabetic patients performed foot care. Foot care practice is less than the practice of community-dwelling Philippine diabetic patients, and adult patients attended Black lion hospital, Ethiopia. On the other hand, foot examination and foot self-management were not consistent with recommendations of ADA. This may be due to a variation in awareness creation, poor income, and increased age of the participants. World Health Organization recommends the consumption of a healthy diet. A healthy diet includes consumption of fruit, vegetables, legumes, nuts, and whole grains; less than 10% of total energy intake from free sugars; 30% of total energy intake from fats and less than 5g of iodized salt. American diabetes food pyramid recommends bread, rice, pasta, and rice as the first food menu, followed by vegetables and fruits. The majority of the participants were not using the food menu which is inconsistent with the recommendations of WHO and ADA. The result of this study is unlike the results of the studies from Northwest Ethiopia, and Harar, Eastern Ethiopia. Inconsistency on food menu may related with a level of awareness, the purpose of using food menu, absence of diabetes food education, and difference in data collection tools. This implies the participants had no awareness about food menu. The vast majority of diabetic patients (74.1%) had diabetes mellitus-related manifestations and complications. More than half of the participants had diabetes-related shock and diabetic ketoacidosis. The prevalence of DM-related complications was higher than the results of the study conducted at Jimma University teaching hospital. Nerve diseases were reported by one-third of the study participants which was slightly higher than the study done in Jimma, Ethiopia. However, the reported DM-related eye problems were more prevalent than the result of the study conducted in Jimma university referral hospital. High prevalence of eye problems may be be due to the public awareness about diabetes- related complication and lack of consistent health education. This implies that diabetes self-monitoring of blood glucose was poor in the participants of this study. Being female, living in urban, married, rich, merchant and having family support were associated with self-management practice. Occupation and family support were tended to be the predictors of self-management. The finding of this study shares a similarity with that of studies done in Iran and Malaysia Family support was reported to increase adherence to self-management. Increase in self- management practice is comparable with the finding of the systematic review and meta-analysis which reported the social support significantly improved self- management. However, these factors were inconsistent with factors reported by Amente, Belachew, Kaehaban, Hongsranagon and Formosa and Muscat from Ethiopia, Thailand, and Malta, respectively. This variation may be demonstrated due to day to day changes in diabetes education that increases awareness and culture variation. The present study has some strengths. First, the response rate was high. Second, it tried to find out the predictors of self-management practice using appropriate data analysis methods. Limitations of the study were using unstandardized and validated tools. # Conclusions Compared to the findings of previous studies, the diabetes self-management practice of the participants of this study was good. The study participants’ regular physical activity, food intake, medication adherence, and foot self- examination was moderate. Being a merchant and having family support were found to be the predictors of self-management practice. Randomized controlled trials involving the participants is needed to proof. Predictors of self-management should be considered to boost self-management practice. # Supporting information Our gratitude goes to the study facility administrators, service providers and participants for their collaboration and information. The authors would also like to thank colleagues who contributed their valuable suggestions throughout this research work. ADA American Diabetes Association AOR Adjusted Odds Ratio CDC Centre for Disease Control CI Confidence Interval DKA Diabetes Ketoacidosis DM Diabetes Mellitus IDF International Diabetes Federation OR Odds Ratio SBGM Self-monitoring of Blood Glucose SPSS Statistical Package for the Social Sciences WHO World Health Organization 10.1371/journal.pone.0232524.r001 Decision Letter 0 Assefa Woreta Solomon Academic Editor 2020 Solomon Assefa Woreta This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 31 Oct 2019 PONE-D-19-26947 Predictors of self-management practice among diabetic patients attending western Oromia hospitals, Ethiopia PLOS ONE Dear Dereje Chala Diriba Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== The topic appears to be the most studied in most part of Ethiopia including Oromia region. Abstract: The methods are very short sighted to provide proper information. The result looks fragmented. The conclusion drawn not based on the finding? Introduction: This section included literature review from different part of the world including Ethiopia. I found this necessary to make a strong argument in the discussion section. However, the write up looks very inconsistent and juggling from the beginning to the end. You should follow introduction writing guide line. Since, the topic is highly researched in various part of Ethiopia I recommend authors to incorporate as many literature from Ethiopia as possible. Methods: I have noticed similar study had been conducted one year prior to this study in Nekemet, Western Oromia region. What is your rationale to Nekemete Hospital in this study? How you operationally defined the study variables in this research? What is your dependent and independent variables? How you developed the questionnaire? How you validate your questionnaire? Where you pretested the questionnaire? Result: The socio-demographic section seems to have included junk of information and redundancy. You must take account of socio-demographic variables than unrelated information. In general, the result section is full of in consist, in comprehensive and fragments of data. Most of the paragraph appears to have a copy and paste. I would recommend the authors to give appropriate time to evaluate the manuscript from the beginning to the end before submission to make sure the information built-in is clear and vivid information to understand. Discussion: As it has been said this topic is one of the most studied, hence the chance to get appropriate literature to make a strong argument on the finding is likely very high even in Ethiopia. However, the arguments are shallow, not based on evidence and weak. Why you cited literature which doesn’t have any similarity to the study setting, for instance America and others. This makes the discussion very shallow and inadequate. What is the limitation of this study? Conclusion: How do you measure whether self-care is good or bad? There is no single statement in the methods section that clearly defined self-care and how it was measured. Overall, the conclusion not inferred from the analyzed result. ============================== We would appreciate receiving your revised manuscript by Dec 06 2019 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Solomon Assefa Woreta Academic Editor PLOS ONE **Journal Requirements:** 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> 2\. Thank you for including your ethics statement: Wollega University Institutional review Board Please amend your current ethics statement to confirm that your named institutional review board or ethics committee specifically approved this study. Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”). For additional information about PLOS ONE ethical requirements for human subjects research, please refer to <http://journals.plos.org/plosone/s/submission-guidelines#loc-human-subjects- research>. 3\. Our editorial staff has assessed your submission, and we have concerns about the grammar, usage, and overall readability of the manuscript.  We therefore request that you revise the text to fix the grammatical errors and improve the overall readability of the text before we send it for review. We suggest you have a fluent, preferably native, English-language speaker thoroughly copyedit your manuscript for language usage, spelling, and grammar. If you do not know anyone who can do this, you may wish to consider employing a professional scientific editing service. Whilst you may use any professional scientific editing service of your choice, PLOS has partnered with both American Journal Experts (AJE) and Editage to provide discounted services to PLOS authors. Both organizations have experience helping authors meet PLOS guidelines and can provide language editing, translation, manuscript formatting, and figure formatting to ensure your manuscript meets our submission guidelines. To take advantage of our partnership with AJE, visit the AJE website (<http://learn.aje.com/plos/>) and enter referral code PLOS15 for a 15% discount off AJE services. To take advantage of our partnership with Editage, visit the Editage website ([www.editage.com](http://www.editage.com)) and enter referral code PLOSEDIT for a 15% discount off Editage services. If the PLOS editorial team finds any language issues in text that either AJE or Editage has edited, the service provider will re-edit the text for free. Upon resubmission, please provide the following:     •    The name of the colleague or the details of the professional service that edited your manuscript     •    A copy of your manuscript showing your changes by either highlighting them or using track changes (uploaded as a \*supporting information\* file)     •    A clean copy of the edited manuscript (uploaded as the new \*manuscript\* file) Please note that PLOS ONE does not copyedit accepted manuscripts and that one of our criteria for publication is that articles must be presented in an intelligible fashion and written in clear, correct, and unambiguous English (<http://www.plosone.org/static/publication#language>). If the language is not sufficiently improved, we may have no choice but to reject the manuscript without review. 4\. Please include additional information regarding the survey or questionnaire used in the study and ensure that you have provided sufficient details that others could replicate the analyses. For instance, if you developed a questionnaire as part of this study and it is not under a copyright more restrictive than CC-BY, please include a copy, in both the original language and English, as Supporting Information. 5\. Please ensure you have thoroughly discussed any potential limitations of this study, e.g. the self-reporting nature of data collection. 6\. Thank you for stating the following financial disclosure: “No”   a) Please provide an amended Funding Statement that declares \*all\* the funding or sources of support received during this specific study (whether external or internal to your organization) as detailed online in our guide for authors at <http://journals.plos.org/plosone/s/submit-now>.     b)  Please state what role the funders took in the study.  If any authors received a salary from any of your funders, please state which authors and which funder. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." c\) If the study was unfunded, please state "The author(s) received no specific funding for this work." Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 7\. Thank you stating the following in your competing interests statement: "No" Please complete the competing interests section fully.  If NO authors have competing interests, please enter: "The authors have declared that no competing interests exist." If Authors have competing interests please enter competing interest details beginning with this statement: "I have read the journal's policy and the authors of this manuscript have the following competing interests: \[insert competing interests here\]" Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf. 8.  We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide. **Additional Editor Comments (if provided):** The topic appears to be the most studied in most part of Ethiopia including Oromia region. Abstract: The methods are very short sighted to provide proper information. The result looks fragmented. The conclusion drawn not based on the finding? Introduction: This section included literature review from different part of the world including Ethiopia. I found this necessary to make a strong argument in the discussion section. However, the write up looks very inconsistent and juggling from the beginning to the end. You should follow introduction writing guide line. Since, the topic is highly researched in various part of Ethiopia I recommend authors to incorporate as many literature from Ethiopia as possible. Methods: I have noticed similar study had been conducted one year prior to this study in Nekemet, Western Oromia region. What is your rationale to Nekemete Hospital in this study? How you operationally defined the study variables in this research? What is your dependent and independent variables? How you developed the questionnaire? How you validate your questionnaire? Where you pretested the questionnaire? Result: The socio-demographic section seems to have included junk of information and redundancy. You must take account of socio-demographic variables than unrelated information. In general, the result section is full of in consist, incomprehensive and fragments of data. Most of the paragraph appears to have a copy and paste. I would recommend the authors to give appropriate time to evaluate the manuscript from the beginning to the end before submission to make sure the information built-in is clear and vivid information to understand. Discussion: As it has been said this topic is one of the most studied, hence the chance to get appropriate literature to make a strong argument on the finding is likely very high even in Ethiopia. However, the arguments are shallow, not based on evidence and weak. Why you cited literature which doesn’t have any similarity to the study setting, for instance America and others. This makes the discussion very shallow and inadequate. What is the limitation of this study? Conclusion: How do you measure whether self-care is good or bad? There is no single statement in the methods section that clearly defined self-care and how it was measured. Overall, the conclusion not inferred from the analyzed result. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: No Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This is a study on self-management practice among diabetes patients in Ethiopian hospital setting. The authors do a great job in providing the big picture of diabetes in Ethiopian context and identifying those predictors of self-management practice. They followed a rigor procedure in their study design, analysis and reporting. This study adds a valuable information in the management of diabetes in western Oromia hospitals by providing some predictors for self-management practice. This study has several issues that need to be addressed before being considered for publication. • The introduction is well written but does not inform the reader about the importance of measuring the self-management practice among diabetes patient. Why we should seek to measure the self-management practice of diabetes among other chronic illness should be explained. • What self-management is not described to a reader in the introductory section, rather the definition happened in a latter section of the manuscript (at the beginning of the discussion). This is not good without introducing the concept at the beginning but presenting the result and defining the term in the discussion. • As the authors reviewed there are several studies conducted in Ethiopia (Harar, Nekemte, Addis Ababa). Therefore, why they are interested to do same topic in Ethiopia at western Oromia is not clear. The justification made as to why this study is conducted is not very convincing. For instance, 45% of participants at Nekemte referral hospitals had poor diabetes self-management though 54.3% of them had diabetic related knowledge, it worth doing a study why such level of practice with higher level of knowledge and what are the factors for such low self-care practice using other study design such as case-control than repeating same study in same place. • The last section of introduction mentioned the absence of diabetic self- management education program in the country. However, the relevance of mentioning this concept in this section is not explained. The authors know this idea in advance and makes them even to forward a recommendation on establishing the education center. The claims are not placed properly in the context of the previous literature. Generally, the introduction should be re-written signifying the importance of doing this study in the light of “effective patient self- management is necessary to prevent adverse health outcomes of diabetes” among diabetes patients in lower-middle income countries including Ethiopia. Reviewer \#2: Abstract: • Under method section number of study participants needs to be included • Under result number of those who have either good or poor self care management needs to be indicated. • The conclusion should be in line with the pertinent finding described under abstract • Recommendation is not in line with conclusion. Introduction • Definition of self care management needs to be included. • References needs to be in logical order. Methods The study setting is not well described. Tools used to assess self care management was not clearly indicated. Analysis method for self care management is not indicated. Under ethical clearance future tense was used. Result • Standard deviation of mean age should be reported in plus or minus, not only plus • Under table 1 you put asterisk on the other category of religion, but not define it below the table as footnote. • Eye problems which is considered as one complication of DM is indicated by two different numbers. • Rather than including other information better narration follows similar table. • There is redundancy of narration about patients believe of diabetic treatment before and after table 2 • Under statements which is written above the title “predictors of self management” under result section, there is a statement which says figure 2, but there is no figure in the document. • Better you include all variable you consider for logistic regression to table 3 with their respective confidence interval and odds ratio rather than duplicating it under table 4 • How you found at the end two predictor variable using logistic regression is not clearly explained. • Revise your interpretation of logistic regression especially your comparison of occupation. • How self care management is measured?????????? And classified as good and poor under the title “level of self care practice” are not clearly indicated. Discussion Your discussion needs revision. Implications given by the researcher is not satisfactory. Conclusion The conclusion is not in line with that of the title of the study and there are repeated copying and pasting of what is already described under result section. Reference Some of the references listed are not utilized for preparation of the manuscript Finally competing interest and list of abbreviations/acronyms are not included to the manuscript Reviewer \#3: The authors tried to make allignment between title, objective, results and conclussion. However the manuscript need further revision and modification particularly the reporting format and language used during reporting. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: Yes: Markos Desalegn Beyene \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0232524.r002 Author response to Decision Letter 0 24 Dec 2019 We really appreciate the genuine comments of the editors and reviewers. For the comments given by reviewers and editor, we reacted and corrected promptly. We are ready to correct if any further comments provided. The point by point response to reviewers commented was uploaded. 10.1371/journal.pone.0232524.r003 Decision Letter 1 Assefa Woreta Solomon Academic Editor 2020 Solomon Assefa Woreta This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 Feb 2020 PONE-D-19-26947R1 Predictors of self-management practices among diabetic patients attending hospitals in western Oromia, Ethiopia. PLOS ONE Dear Dr. Dereje Chala, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== I gain would like to applaud for putting together the effort to make the necessary improvement on the manuscript. Most of the comments and feedback that have been given by reviewers appears to be corrected and incorporated accordingly, but there are few issues need to be addressed in order to proceed to the next step. For instance, the definition used to describe knowledge has a preconceived notion. Did you evaluate knowledge based on individual question mean or the summation mean of the knowledge questions? Secondly, the argument used to discuss the finding seems to be so frail and not enough to provide vivid clarification. In general, this section needs to have literary supported evidence with clear reference or citation for your argument. For example, the first paragraph in the discussion section has compared and contrasted other studies with your current, but the statement used to ornate the argument read as “This could be related to change to diabetic education in different ways. Thus, sustainable scheduled diabetic education is very crucial to increase self- management practice.” Where did you get this information? Is this information research driven or based on the predetermined knowledge? Hence, all the argument you might reach need to have study based explanation or argument. Make sure you make the necessary changes in this section to eliminate the inconsistency. ============================== We would appreciate receiving your revised manuscript by Mar 21 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Solomon Assefa Woreta Academic Editor PLOS ONE Additional Editor Comments (if provided): I gain would like to applaud for putting together the effort to make the necessary improvement on the manuscript. Most of the comments and feedback that have been given by reviewers appears to be corrected and incorporated accordingly, but there are few issues need to be addressed in order to proceed to the next step. For instance, the definition used to describe knowledge has a preconceived notion. Did you evaluate knowledge based on individual question mean or the summation mean of the knowledge questions? Secondly, the argument used to discuss the finding seems to be so frail and not enough to provide vivid clarification. In general, this section needs to have literary supported evidence with clear reference or citation for your argument. For example, the first paragraph in the discussion section has compared and contrasted other studies with your current, but the statement used to ornate the argument read as “This could be related to change to diabetic education in different ways. Thus, sustainable scheduled diabetic education is very crucial to increase self- management practice.” Where did you get this information? Is this information research driven or based on the predetermined knowledge? Hence, all the argument you might reach need to have study based explanation or argument. Make sure you make the necessary changes in this section to eliminate the inconsistency. \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0232524.r004 Author response to Decision Letter 1 11 Mar 2020 Question 1: Did you evaluate knowledge based on individual question mean or the summation mean of the knowledge questions? Response: Thank you for thorough evaluation and observation. We didn’t evaluate the diabetes knowledge score; however, we reported the general knowledge about diabetes management. Thus, we have removed from operational definition part. Question 2. Secondly, the argument used to discuss the finding seems to be so frail and not enough to provide vivid clarification. In general, this section needs to have literary supported evidence with clear reference or citation for your argument. Response: Thank you for your comment. We noted that our discussion is weak, thus we made modifications with implications of the results. We also cited the literature we have used. 10.1371/journal.pone.0232524.r005 Decision Letter 2 Tu Wen-Jun Academic Editor 2020 Wen-Jun Tu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 17 Apr 2020 Predictors of self-management practices among diabetic patients attending hospitals in western Oromia, Ethiopia. PONE-D-19-26947R2 Dear Dr. Diriba, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Wen-Jun Tu Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0232524.r006 Acceptance letter Tu Wen-Jun Academic Editor 2020 Wen-Jun Tu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Apr 2020 PONE-D-19-26947R2 Predictors of self-management practices among diabetic patients attending hospitals in western Oromia, Ethiopia. Dear Dr. Diriba: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Wen-Jun Tu Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Despite the availability of highly effective COVID-19 vaccines to prevent hospitalization and reduce mortality, variants continue to fuel the surge of COVID-19 across the U.S.. High-quality diagnostic and serology tests are essential tools to better understand the epidemiology of COVID-19 and immunity after infection. Viruses and antibodies are primarily detectable within certain temporal windows. However, many individuals infected with SARS-CoV-2 are asymptomatic or may not seek medical care because of mild symptoms. In contrast to molecular diagnostic tests, serologic tests are informative even once the SARS-CoV-2 infection is no longer present. Currently, there are 90 authorized SARS-CoV-2 serology/antibody tests approved for Emergency Use Authorization (EUA). However, they have not undergone the same evidentiary review standards required for Food and Drug Administration (FDA) clearance due to the COVID-19 national emergency. There is a need to assess the real-world performance of these tests. Further, while large studies have shown that greater than 91% of people with active SARS-CoV-2 infection seroconvert, the factors associated with seroconversion (e.g., pre-existing conditions, the severity of COVID-19 presentation) remain elusive. From a public health perspective, confidence in the ability of serological tests to identify those with recent infections is critical for effective pandemic planning. Estimates of disease prevalence directly inform dynamic population estimates of susceptible, infected, and recovered, which are needed to understand the infectiousness of SARS-CoV-2. From a clinical perspective, an accurate understanding of SARS-CoV-2 exposure is necessary to understand disease presentation and a clinical course of action, especially when patients do not present with symptoms or present late in their disease course (e.g., post-acute sequelae of SARS-CoV-2). Additionally, identifying factors associated with seropositivity may elucidate potential mechanisms of action that may be foundational in the development of therapy and treatment plans. To address these gaps, we characterize the performance of serology tests by estimating the positive percent agreement (PPA) of serological samples obtained from people known to be positive for SARS-CoV-2 infection by molecular assay (e.g., PCR). We also sought to identify factors associated with seropositivity. Findings from this study may facilitate understanding of the real-world performance of serology tests, many of which were issued under EUA, and may help inform our understanding of the immune response to SARS-CoV-2. # Materials and methods ## Study population and setting Six health systems (i.e., datasets) collaborated on the Diagnostics Evidence Accelerator (EA): Health Catalyst, Mayo Clinic, Optum Labs, Regenstrief Institute, the University of California Health System, and Aetion and HealthVerity. The EA is a consortium of leading experts in health systems research, regulatory science, data science, and epidemiology, specifically assembled to analyze health system data to address key questions related to COVID-19. The EA provides a platform for rapid learning and research using a common analytic plan. Health Catalyst, Mayo Clinic, and the University of California Health System all utilized electronic health records (EHR) data from their respective healthcare delivery systems. The Regenstrief Institute accessed EHR and public health data from the Indiana Health Information Exchange, while Aetion sourced healthcare data from HealthVerity Marketplace encompassing medical claims, pharmacy claims, hospital chargemaster, and data collected directly from laboratories. Optum Labs data included de-identified medical, and pharmacy claims as well as laboratory results data utilized medical, and pharmacy claims from a single, large U.S. insurer as well as data directly from laboratories. We refer to these health systems as datasets A-F for the purposes of anonymity. Data sources included in the analysis are generally categorized as either payer (claims) or healthcare delivery systems. As illustrated in, data were drawn from across the U.S. with heavy representation in California, Illinois, Ohio, and Michigan. Characteristics of participating data sources and representative populations are described in the. ## Study design In this retrospective cohort study, we identified patients across different settings (e.g., inpatient, outpatient, emergency department (ED), or long-term care facility) who tested positive for SARS-CoV-2 ribonucleic acid (RNA) by molecular test between March–September 2020 and who received at least one subsequent serological test for SARS-CoV-2 immunoglobulin (Ig) G or Total antibody (Ab) from 14–90 days after the positive RNA test. We analyzed the first serology test in the 14–90-day follow-up period, which ended on December 31, 2020. “Date of RNA positive” served as the index (cohort entry) date and was defined hierarchically as either the date of 1) sample collection; 2) accession; or 3) result. Because the optimal time to observe a positive serology is at least two weeks after the index date, we only include patients who had at least one serology test 14–90 days after the index date. To minimize the effect of differential missingness between datasets, we applied the following rules: 1) included all persons with an office or telephone visit in the +/- 14 days around the index date to enable as complete an assessment of presenting symptoms as possible; 2) in claim systems, included only persons with at least six months of enrollment in the year before index; 3) estimated the proportion of patients at each site who had zero encounters in the prior year to contextualize our capture of pre-existing conditions, and 4) excluded variables from analysis if ≥30% of values were missing. The Western—Copernicus Group (WCG) Institutional Review Board (IRB), the IRB of record for the Reagan-Udall Foundation for the FDA, reviewed the study and determined it to be non-human subjects research. Additionally, all legal and ethical approvals for use of the data included in this study were submitted, reviewed, and/or obtained locally at each contributing dataset by an IRB and/or governing board. ## Measures ### Outcomes The primary outcome of interest for the validation analysis was the PPA of positive antibody (IgG or total) from serology tests with positive RNA from molecular tests (e.g., PCR), which served as the reference standard. Serology tests reported in this analysis included: Abbott Architect IgG, Euroimmun IgG, Diazyme DZ-Lite SARS-CoV-2 IgG CLIA kit, Beckman SARS-CoV-2 IgG, Ortho Vitros IgG, Diasorin Liaison SARS-CoV-2 S1/S2 IgG, and Roche Elecsys Total Ab. The Ortho Vitros was the only test used across multiple (3) datasets. We refer to these manufacturers—serological tests as Δ, Θ, Π, Λ, Ξ, Γ, and Ψ for anonymity. Molecular tests most reported in this analysis included: Hologic Panther Fusion, Hologic Aptima, Roche Cobas, Quest rRT-PCR, and Thermo Fisher Scientific Combo Kit. We refer to these manufacturers—molecular tests as Σ, Φ, Ω, X, Y, and j for anonymity. ### Covariates We collected demographic, behavioral, and environmental characteristics, baseline clinical presentation, key comorbidities, and test characteristics, including manufacturer, according to a diagram illustrating potential factors associated with serology testing. We identified comorbidities and clinical presentation using phenotypes defined by the International Classification of Diseases 10 (ICD-10), and/or National Drug Codes. We identified comorbidities (pre-existing conditions) in the 365 days before the index date through 15 days before the index date. We provided coding algorithms for groups to use, while some groups used existing algorithms generated by their site. The ICD-10 codes used to identify comorbidities are listed in the. We also stratified analyses by RNA tests conducted before June 15, 2020, which marked the beginning of the summer wave of infections in the first year of the pandemic, compared to on or after that date. ## Statistical analysis Each contributing dataset ran its analysis according to a common protocol. Results were reviewed as a group to ensure alignment with the protocol and to review any protocol deviations. We calculated PPA as: *(Number of positive antibody results ÷ Number of positive RNA results) x 100*. We calculated PPA based on the first eligible serology test in the follow-up period overall and by age, sex, race, ethnicity, U.S. region, pregnancy status, pre-existing conditions, including but not limited to cardiovascular disease, obesity, hypertension, kidney disease, asthma, dementia, chronic liver disease, and smoking status. We also report the PPA by presenting symptoms, and serology tests at the time of the first serology test. We examined variations in PPA by serology tests and time, and serology tests and symptom presentation. We also examined variations in PPA by geography and care setting over time. We calculated exact (Clopper-Pearson) 95% confidence intervals (CI). We report significant differences where 95% CI have complete separation—although we did not conduct formal statistical comparisons of PPA between groups. To study the odds of seropositivity, we estimated a model for the association to identify independent risk factors for seropositivity, assuming a binomial distribution for seropositivity status. Results are presented as the odds ratio (OR) and 95% CI that was calculated using score confidence intervals or exact CI. All variables were treated as categorical. Symptoms were reported as a binary variable: “1” if any of the following symptoms were present: fever \>100.4, abnormal chest imaging finding, high respiratory rate, low blood pressure, diarrhea, hypoglycemic, chest pain, delirium/confusion, headache, sore throat, cough, shortness of breath, pneumonia, acute respiratory infection, acute respiratory distress, cardiovascular presentation, renal presentation; and “0” otherwise. For datasets with data covering \>1 geographic catchment area, geography was included as either one of four U.S. Census regions, or nine U.S. Census divisions based on patient home zip code. Variables with \>30% missing/unknown values were excluded from models (except for pregnancy, pre- existing condition, or presenting symptoms, all of which were included). Each dataset used automated backward selection to remove non-significant pre-existing conditions while forcing all other covariates into the model. All analyses were performed using SAS software, version 9.2 or higher (SAS Institute, North Carolina, U.S.); or the Aetion Evidence Platform v4.13 (including R v3.4.2), which includes audit trails of all transformations of raw data and a quality check of the data ingestion process. # Results Samples sizes across datasets ranged from 660–7,115; a total of 15,615 people with at least one serology test 14–90 days after the index date were included in the analyses. Between 35–65% of patients identified from health care delivery systems had no documented encounter in the system between 365 and 15 days before the index date. In contrast, only 11% of patients from national insurers reported having zero claims in the baseline period. As shown in, the serotested population was primarily 45–64 years of age (\>40%), with a history of cardiovascular disease, including hypertension (8–70%). Race and ethnicity data were robust (\<30% missing) in four datasets. The serotested population in those datasets was primarily White (\>53%) and non-Hispanic (\>65%), From datasets with national representation, persons from the Northeast (New England and Mid- Atlantic) were most represented in this serotested population. In datasets that represent regionally-based healthcare delivery systems, their population reflected their locations: Pacific and Midwest. Information on manufacturer test names was provided in four datasets. Generally, 2–3 primary tests were utilized in each dataset; 4 of 7 tests reported were used in \>1 dataset. We did not observe any difference by age or sex in those for whom the test name was known versus unknown. In a single dataset with \<30% of missing data on race/ethnicity, we observe over-representation of White and Hispanic people in those for whom the test name was known. ## Positive percent agreement (PPA) of serology among molecularly confirmed SARS-CoV-2 The overall PPA ranged from 65–90% across analytic datasets. The real-world PPA met the EUA requirement of ≥87% in three datasets (A, B, D). Two of these datasets represented national administrative claims and associated results with the date the sample was collected or received by the laboratory; the third represented data from EHRs and associated results with the date the test was conducted, which is lagged further from the clinical interaction than the former. Overall PPA was likely influenced by the mix of serology tests represented in each dataset. Seven serological tests were reported in this analysis, of which two (Δ and Γ) met the EUA PPA requirements. Two tests were used across multiple datasets and performed similarly above the EUA requirement. PPA by serology test type varied across datasets; with three of five reporting significantly lower PPA from total antibody (PPA range: 69–90%) compared to IgG (PPA range: 87–92%); and two showing no difference. We observed no difference in PPA with antibody tests that target spike compared to nucleocapsid proteins. PPA was significantly higher in Black (PPA range: 86–92%), as compared to White (PPA range: 78–86%), persons in at least two of the four datasets reporting robust race/ethnicity data. PPA was significantly higher in Hispanic (PPA range: 79–96%), compared to non-Hispanic (PPA range: 60–86%), patients. PPA appeared highest in those with diabetes (PPA range: 75–94%) and kidney disease (PPA range: 75–95%), and lowest in those with conditions that leave them immunocompromised (PPA range: 56–93%). We observed higher PPA in the inpatient (PPA range: 70–97%) or ED (PPA range: 93–99%) setting compared to outpatient (PPA range: 63–92%). There was some evidence of higher PPA among patients with at least one COVID-19 related symptoms as compared to those with none (PPA range: 63–91%) among two datasets (B and D); and was particularly high for select conditions like pneumonia (PPA range: 82–97%). However, differences in the PPA by the presence of symptoms do not appear to be explained by the test. A stratified analysis by test comparing those with and without symptoms showed no significant difference in PPA. PPA trends by calendar time were not consistent across datasets. ## Factors associated with seropositivity In adjusted models (Figs –), the OR for seropositivity was significantly elevated in Hispanic compared to non-Hispanic ethnicity (OR range: 2.59–3.86); among those with pre-existing diabetes (OR range: 1.49–1.56) and obesity (1.63–2.23) as compared to those without pre-existing conditions; and among those observed in the ED compared to outpatient (OR range: 2.49–10.97). The OR for seropositivity was significantly lower in those with pre-existing immunocompromised or autoimmune conditions compared to those without such conditions (OR range: 0.25–0.70). In two of three datasets that included pre- existing cardiovascular disease in the OR model, the OR for seropositivity was significantly lower in persons with, compared to those without, such conditions (OR range: 0.49–0.57). The OR for seropositivity tended to be lower on or after June 15 compared to prior in half the datasets, but differences were not significant in the other half. # Discussion Serology tests are an important instrument in the toolkit to understand the epidemiology of COVID-19 because of their ability to identify persons with prior infection who may present too late in the infectious period due to mild symptoms, or no symptoms at all. Serology results may inform diagnoses of post- acute SARS-CoV-2 (PASC) and the appropriate treatment course, which may depend on whether patients are at increased risk for severe illness due to insufficient antibody response. The reported sensitivity of the serology tests included in this analysis that were submitted for EUA approval were all \>95%. Our analysis of multiple large datasets of patients with confirmed SARS-CoV-2 infection suggests that serology tests performed lower than = expected–with PPA ranges (a measure analogous to sensitivity) from 65–90%.—Our results align with results from smaller, detailed laboratory evaluations that suggest a lack of harmonization, including optimization of cut-off values, may contribute to decreased overall performance. Additionally, our results align with studies that include more representative samples of milder or asymptomatic persons. Two of seven tests reported across datasets achieved the EUA requirement of PPA ≥ 87%. As we did not have data on specific serology-molecular pairs or meta-information on the tests (including fidelity to protocols for serology and molecular test analysis), these results reflect more on the real-world implementation of the tests rather than the true quality of the tests. Specifically, where the same test was used across multiple datasets, they all performed similarly. For example, the serology test Γ performed similarly high (PPA \>90%) across three datasets. However, the overall PPA for tests performed in datasets A and B were higher than in dataset E. A major factor that may have contributed to this difference is that the other serological tests reported to datasets A and B performed above the EUA requirement. In contrast, the other tests reported in dataset E performed below the EUA requirement. Additionally, datasets A and B leveraged administrative claims data and associated RNA and serology results with sample collection or sample receipt date, while dataset E associated results with the date the test was run. Dataset E also represents those from a healthcare delivery system where serology tests were initially only used for symptomatic patients with at least 12 days of symptoms. This practice shifted after approximately two months (June 1, 2020) to a protocol that required both molecular and serological testing for SARS-CoV-2 as part of pre-procedure screening. This protocol was in effect for another three months (August 31, 2020), after which the healthcare system shifted to unrestricted testing for both molecular and serology tests and saw a substantial drop in the use of serological testing. We expected that procedural “lags” to serotesting, combined with additional lags due to associating results with a date downstream from the clinical interaction, may have further extended the time between infection/symptom onset and the actual time of serology sampling. The impact of this misclassification may be most important for serology samples at the upper bounds of 90 days; where samples were likely \>90 days from the point of infection and humoral antibodies more likely to have declined. Despite changes in the protocol over time, we observed no overall or test-specific difference in PPA before or since June 15, 2020 in dataset E. Nevertheless, administrative protocols create lags in serotesting that challenge our assumptions of whether the observed molecular “test date” is a good proxy for symptom onset. Absent any knowledge of such policy, it’s difficult to make broad assumptions regarding patterns in molecular or serology testing unless established clinical protocols are known. We observed that patients of Hispanic ethnicity compared to non-Hispanic patients, with pre-existing obesity and those who presented in the ED had a higher OR for seropositivity; and similarly higher PPA. These results further support what others have observed that persons with unmanaged diabetes, who are disproportionately people of color, are vulnerable to hyper-inflammation related to COVID-19. Furthermore, hyper-inflammation, including pro-inflammatory cytokine storm, has been associated with severe disease, reduced viral clearance, and sustained antibody production. Although a recent small study showed that while a low viral load is associated with lower antibody response, clinical illness does not guarantee seroconversion. Other studies have demonstrated people with cancer have a lower probability of mounting an immune response from the vaccine, as demonstrated by seroconversion, viral neutralization, and T-cell response. Our results demonstrating lower odds of seropositivity among those with cancer and other immunodeficiencies suggest that the same may be true regarding their antibody response to infection. ## Strengths Our study has many strengths. This was a large assessment of serotesting across the U.S. in diverse datasets leveraging either EHR or claims data. We developed a protocol that incorporated the unique characteristics of each data source and provided a forum to transparently communicate and collaborate on study design and interpretation. We also established a platform to rapidly collect and analyze data from various systems to evaluate process improvement and identify important trends over time. Such a platform may be used to evaluate process improvement and comparisons within data systems. We did extensive characterization of missing data to guide model development and help with interpretation. Additionally, this study was conducted before public availability of COVID-19 vaccines across the U.S., which minimizes the potential for confounding related to vaccine-induced antibodies. ## Limitations A major limitation in this real-world analysis is a large number of missing test names and relevant meta-data, including quality control measures adopted, for both molecular and serological tests. As such, we were unable to account for molecular-serology pairs when assessing PPA or the fidelity with which these tests were performed. A large amount of missing test name information limited our ability to describe trends by the manufacturer. Although, a thorough examination of missing data does not suggest differential missingness by age or sex. Importantly, the intent of this analysis was not to evaluate individual tests, but the performance of serology in the context of real-world implementation of test protocols and varying reference standards. As discussed in our prior manuscript, the sample included in this study included those who were more likely to be serotested for SARS-CoV-2: White, 45–64 years of age, with prior history of cardiovascular disease. Nevertheless, there was still sufficiently large number of people to assess PPA trends among younger ages and in those with and without other pre-existing conditions. Finally, this study was conducted before the surge of the Omicron variant, which has been shown to have a number of mutations on the N-gene and S-gene that reduce the sensitivity of some diagnostic tests. As such, our inference is limited to the SARS-CoV-2 variants prior to Omicron, primarily alpha. # Conclusion Across large samples of patients with molecularly confirmed SARS-CoV-2, serology tests did not consistently meet the EUA requirement of PPA ≥ 87% in the post- market setting. However, given the limited availability of test names, this analysis serves as a signal that further investigation into how serology and molecular tests are used, including protocol fidelity, is needed to understand ways to improve the real-world performance of serology tests. Despite differences in testing protocols and data availability, the similarity in performance of serology tests across datasets suggests that serology tests were robust to differences in care settings. However, the real-world PPA for several serology tests did not meet EUA requirements; and the exclusive representation and low use of such tests in certain datasets look to have impacted the overall performance of serology tests in those datasets. Where data were sufficiently robust, we observed that people of Hispanic ethnicity had a higher odd of seropositivity than non-Hispanics. Higher odds of seropositivity in those with pre-existing diabetes or obesity further support the hypothesis that these conditions are associated with more severe disease, reduced viral clearance, and the sustained presence of antibodies. Conversely, lower odds of seropositivity among those with cancer and other immunodeficiencies suggest that immunopathology in these groups associated with the vaccine may extend to infection. Interpreting results from real-world data collected from clinical and administrative databases is challenging. A clear understanding of testing protocols at the point of care is needed to validate assumptions regarding proxy variables and to interpret results. Incomplete information on race/ethnicity and test name limited our ability to address racial disparities in testing and real- world performance of serological tests. Nevertheless, implementing best practices for analyzing and reporting results from observational data across multiple datasets yields confidence in trends that are repeated. And where results are divergent, we were able to explore how differences in data sources may explain findings and target areas for future investigation. Improved data interoperability to link test names and clinical/demographic data is critical to enable rapid assessment of the real-world performance of in vitro diagnostic tests, particularly in the face of fast-mutating pathogens. # Supporting information Special thanks to our advisors on this project from the U.S. Food and Drug Administration: Aloka Chakravarty, Tamar Lasky, Gina Valo, Mary Jung, Stephen Lovell, Jacqueline M Major, Daniel Caños, Sara Brenner, and Wendy Rubinstein; and Duke-Margolis: Christina Silcox. We thank all members of the Evidence Accelerator Workgroup for their support and feedback: Roland Romero, James Okusa, Elijah Mari Quinicot, Amar Bhat, Susan Winckler, Alecia Clary, Sadiqa Mahmood, Philip Ballentine, Perry L. Mar, Cynthia Lim Louis, Connor McAndrews, Elitza S. Theel, Cora Han, Pagan Morris, and Charles Wilson. A special thanks and recognition for the contributions and sacrifice of Dr. Michael Waters, our dear colleague, and friend who will be forever in our thoughts. We thank Amir Alishahi Tabriz MD, PhD for his assistance with manuscript preparation. 10.1371/journal.pone.0279956.r001 Decision Letter 0 Banada Padmapriya P Academic Editor 2023 Padmapriya P Banada This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Jul 2022 PONE-D-22-11773Real-World Performance of SARS-Cov-2 Serology Tests in The United States, 2020.PLOS ONE Dear Dr. Rodriguez-Watson, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. The study offers important information on the reliability of the serodiagnosis. Please see the comments from the reviewers and I hope you will find it helpful to improve the quality of the manuscript overall. it is indeed difficult to comprehend some of the images in the pdf version of the article. Please submit figures as recommended by plosOne (authors instructions) and plosOne can generate a pdf including the figures which are generally high quality. Based on the reveiwer's comments I am recommending your article for major revision and will look forward for the revised manuscript. Please submit your revised manuscript by Aug 29 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Padmapriya P Banada, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well. 3\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match. When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. 4\. Thank you for stating the following in the Competing Interests section: “AJB is a co-founder and consultant to Personalis and NuMedii; consultant to Samsung, Mango Tree Corporation, and in the recent past, 10x Genomics, Helix, Pathway Genomics, and Verinata (Illumina); has served on paid advisory panels or boards for Geisinger Health, Regenstrief Institute, Gerson Lehman Group, AlphaSights, Covance, Novartis, Genentech, Merck, and Roche; is a shareholder in Personalis and NuMedii; is a minor shareholder in Apple, Facebook, Alphabet (Google), Microsoft, Amazon, Snap, Snowflake, 10x Genomics, Illumina, Nuna Health, Assay Depot (Scientist.com), Vet24seven, Regeneron, Sanofi, Royalty Pharma, Pfizer, BioNTech, AstraZeneca, Moderna, Biogen, Twist Bioscience, Pacific Biosciences, Editas Medicine, Invitae, Doximity, and Sutro, and several other non-health related companies and mutual funds; and has received honoraria and travel reimbursement for invited talks from Johnson and Johnson, Roche, Genentech, Pfizer, Merck, Lilly, Takeda, Varian, Mars, Siemens, Optum, Abbott, Celgene, AstraZeneca, AbbVie, Westat, several investment and venture capital firms, and many academic institutions, medical or disease specific foundations and associations, and health systems. AJB receives royalty payments through Stanford University, for several patents and other disclosures licensed to NuMedii and Personalis. AJB’s research has been funded by NIH, Northrup Grumman (as the prime on an NIH contract), Genentech, Johnson and Johnson, FDA, Robert Wood Johnson Foundation, Leon Lowenstein Foundation, Intervalien Foundation, Priscilla Chan and Mark Zuckerberg, the Barbara and Gerson Bakar Foundation, and in the recent past, the March of Dimes, Juvenile Diabetes Research Foundation, California Governor’s Office of Planning and Research, California Institute for Regenerative Medicine, L’Oreal, and Progenity.        CLB has intellectual property in and receives royalties from BioFire, Inc. She serves as a scientific advisor to IDbyDNA (San Francisco, CA and Salt Lake City, UT); and is on the Board of the Commonwealth Fund.     CK is a paid employee of Aetion and hold Aetion stock options. NES is an employee of Optum Labs and owns stock in the parent company UnitedHealth group. NDL was an employee of Health Catalyst at the time the work was performed. JLG is a full-time employee of Regenstrief Institute, which provides independent research services to entities including those within the pharmaceutical and medical device industries. SJG serves as Chief Medical Information Officer for the Indiana Health Information Exchange, and is a founding partner of Uppstroms, LLC.” Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing-interests>).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf. 5\. Thank you for stating the following in the Acknowledgments Section of your manuscript: “Financial support for this work was provided in part by a grant from The Rockefeller Foundation.” We note that you have provided additional information within the Acknowledgements Section that is not currently declared in your Funding Statement. Please note that funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form. Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows: “CVRW was funded by a grant from the Rockefeller Foundation. BDP, CK, GJ used funding provided by Yale University-Mayo Clinic Center of Excellence in Regulatory Science and Innovation (CERSI), a joint effort between Yale University, Mayo Clinic, and the U.S. Food and Drug Administration (FDA) (3U01FD005938).            CK, CMF, SJG, PJE, EHE, NDL, and JLG work was funded by a designated sub-grant from the FDA Foundation. AJB funded by award number A128219 and Grant Number U01FD005978 from the FDA, which supports the UCSF-Stanford Center of Excellence in Regulatory Sciences and Innovation. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the HHS or FDA. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 6\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. 7\. One of the noted authors is a group or consortium ”Evidence Accelerator Workgroup”. In addition to naming the author group, please list the individual authors and affiliations within this group in the acknowledgments section of your manuscript. Please also indicate clearly a lead author for this group along with a contact email address. 8\. We note that Figure 1  in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission:  a. You may seek permission from the original copyright holder of Figure 1 to publish the content specifically under the CC BY 4.0 license.  We recommend that you contact the original copyright holder with the Content Permission Form (<http://journals.plos.org/plosone/s/file?id=7c09/content- permission-form.pdf>) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (<http://creativecommons.org/licenses/by/4.0/>). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission. In the figure caption of the copyrighted figure, please include the following text: “Reprinted from \[ref\] under a CC BY license, with permission from \[name of publisher\], original copyright \[original copyright year\].”  b. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. The following resources for replacing copyrighted map figures may be helpful: USGS National Map Viewer (public domain): <http://viewer.nationalmap.gov/viewer/> The Gateway to Astronaut Photography of Earth (public domain): <http://eol.jsc.nasa.gov/sseop/clickmap/> Maps at the CIA (public domain): <https://www.cia.gov/library/publications/the- world-factbook/index.html> and <https://www.cia.gov/library/publications/cia- maps-publications/index.html> NASA Earth Observatory (public domain): <http://earthobservatory.nasa.gov/> Landsat: <http://landsat.visibleearth.nasa.gov/> USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): <http://eros.usgs.gov/#> Natural Earth (public domain): <http://www.naturalearthdata.com/> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The Rodriguez-Watson et al. manuscript describes a synthesis of real-world serology testing for SARS-CoV-2. It analyses agreements between SARS- CoV-2 PCR testing and antibody detection assays in 6 US health systems across different setting (inpatient, outpatient, ED, long-term care). The aims of the manuscript were to address gaps in understanding exposure to SARS-CoV-2 and identifying factors associated with seroconversion. The manuscript is well written and understandable even though presents a large amount of data. Although the odd of seropositivity according to demographics gives a good understanding of factors that might affect seroconversion, the authors failed to address the gaps in understanding exposure. Comments: 1- The manuscript describes agreement (PPA) between PCR testing and serology results at 14-90 days post PCR. The approach to analyzing agreement is not complete with no mention of kappa (Cohen, McNemar test, etc.). 2- Is there a loss of agreement depending on day serology was done? 3- When comparing PPA observed according to ethnicity or other factors, are the differences significant? 4- Line 284, the authors mention test quality. Many studies looking at the quality of several serology tests used in this manuscript have been published, can the authors discuss further and refer to these manuscripts? Does lower agreement observe in this study match the published data? 4- In the section results, no odd ratio is given, only 95%CI. Please add. 5- Factors associated with seropositivity, line 257-266. Are the differences described significant? 6- Line 135 change e.g. by i.e. or remove altogether 7-Line 135, remove comma after IgG, “IgG, \[21\],” 8- Table 1, “Na” in lowercase, all other table “NA” is capitalised. Please homogenise throughout. 9- Line 313, should read sustained antibody production. 10- Line 359, should read the sustained presence of antibodies. 11- Line 371, should be “in vitro”. Reviewer \#2: Summary: In this important study, Dr. Rodriguez-Watson and colleagues studied 6 large- scale datasets to understand the 2020 performance of real-world use of EUA- approved SARS-CoV-2 serology testing after a positive molecular test. The group demonstrates substantial real-world variance in the PPA of these tests across health contexts. Major comments 1\. What was done statistically regarding individuals with a 2nd serology test? Line 161 implies only the first test was used. What was the concordance/timing between 1st and 2nd tests? Did 2nd tests, where done, have a different PPA? 2\. Dataset C seems to have a broadly lower PPA vs the other datasets, has the smallest N, and is relatively geographically restricted. This dataset does not appear to have any manufacturer molecular test names available, but there is no PPA reported in the Unknown/missing category in table 2 for that variable. Is this dataset usable? It would seem that as both the serological and molecular test characteristics would contribute to the PPA, not knowing the molecular test name at all makes using this dataset problematic. 3\. What is known about the contribution of “other” molecular tests to this dataset, such as the adoption of “rapid” PCR testing and “in-house” testing that some institutions produced during this time period? Is it possible to address those tests where both the serological and molecular tests are known? As above, the confounding factor of molecular test characteristics could influence the PPA of the serological tests in question. 4\. Time between the molecular and serological test seems like a key point as well. Do you have that data? Does time between tests affect the results? Minor comments 1\. The figures are not showing well (scattered pixels) in my Adobe Acrobat Pro DC view of the PDF. Please ensure that high quality images are used during publication. I can access the tif files, which look right. 2\. I do not understand why figures 2-6 list the study period as starting in 2019. Is this a typo or is this correct? The methods list 2020 as start date, which would make sense given the dates of the pandemic. 3\. Please check line 380, there may be a comma instead of a period. 4\. Perhaps figures 2-7 should be combined into a single summary figure (with total N for the entire study) and the individual flowsheets by cohort might be moved to supplemental. The reader might better grasp the overall study with a simpler summary figure. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0279956.r002 Author response to Decision Letter 0 3 Nov 2022 Dear Dr. Padmapriya Banada: We thank you and the reviewers for your thoughtful comments on our manuscript. We appreciate the opportunity to respond and believe the revisions have improved the manuscript. Below, please find a table with the summary of reviewer comments and our responses. Please be advised that all line numbers referenced in the responses below correspond to line numbers in the tracked changes version of the manuscript. Please reach out with any additional questions, or if more clarification is required. We look forward to hearing from you. Regards, Carla Rodriguez-Watson, PhD, MPH Director of Research Reagan-Udall Foundation for the FDA Comments/Questions Response Journal Requirements 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. Please see the new version with the correct requirements. Additionally, we have listed manuscript requirements below with the acknowledgment that they are complete. Journal Requirement Acknowledgement of Completion Article Title YES Author Byline YES Affiliations YES Corresponding Authorship YES Contributorship YES Consortia or other Group Authors YES Level 1 Heading YES Figure Citations YES Figure Captions YES File Naming for Figures YES Display/Numbered Equation N/A Inline Equation N/A Level 2 Heading YES Level 3 heading YES Please submit your manuscript in double-space paragraph format. YES Tables and Table Citations YES Reference Citations YES Supporting Information Citations YES Acknowledgments- No funding or competing interest information YES References YES Supporting Information Captions YES File Naming for Supporting Information YES 2\. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well. We have listed our full ethics statement in lines 132-136 of the methods section: “The Western - Copernicus Group (WCG) Institutional Review Board (IRB), the IRB of record for the Reagan-Udall Foundation for the FDA, reviewed the study and determined it to be non-human subjects research. Additionally, all legal and ethical approvals for use of the data included in this study were submitted, reviewed, and/or obtained locally at each contributing dataset by an IRB and/or governing board.” 3\. We note that the grant information you provided in the ‘Funding Information’ and ‘Financial Disclosure’ sections do not match. When you resubmit, please ensure that you provide the correct grant numbers for the awards you received for your study in the ‘Funding Information’ section. Thank you for pointing this out. We have updated the sections with the correct grant numbers in the online submission. We have also included the correct grant numbers below. Funding Information Section: Financial support for this work was provided in part by a grant from The Rockefeller Foundation (HTH 030 GA-S). BDP, CK, GJ used funding provided by Yale University-Mayo Clinic Center of Excellence in Regulatory Science and Innovation (CERSI), a joint effort between Yale University, Mayo Clinic, and the U.S. Food and Drug Administration (FDA) (3U01FD005938) (<https://www.fda.gov/>). CK, CMF, SJG, PJE, EHE, NDL, and JLG work was funded by a designated sub-grant from the FDA Foundation. AJB funded by award number A128219 and Grant Number U01FD005978 from the FDA, which supports the UCSF-Stanford Center of Excellence in Regulatory Sciences and Innovation. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the HHS or FDA. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” 4\. Thank you for stating the following in the Competing Interests section: “CRW receives research funding from Novartis, Merck and AbbVie; and holds minor stock in Gilead. AJB is a co-founder and consultant to Personalis and NuMedii; consultant to Samsung, Mango Tree Corporation, and in the recent past, 10x Genomics, Helix, Pathway Genomics, and Verinata (Illumina); has served on paid advisory panels or boards for Geisinger Health, Regenstrief Institute, Gerson Lehman Group, AlphaSights, Covance, Novartis, Genentech, Merck, and Roche; is a shareholder in Personalis and NuMedii; is a minor shareholder in Apple, Facebook, Alphabet (Google), Microsoft, Amazon, Snap, Snowflake, 10x Genomics, Illumina, Nuna Health, Assay Depot (Scientist.com), Vet24seven, Regeneron, Sanofi, Royalty Pharma, Pfizer, BioNTech, AstraZeneca, Moderna, Biogen, Twist Bioscience, Pacific Biosciences, Editas Medicine, Invitae, Doximity, and Sutro, and several other non-health related companies and mutual funds; and has received honoraria and travel reimbursement for invited talks from Johnson and Johnson, Roche, Genentech, Pfizer, Merck, Lilly, Takeda, Varian, Mars, Siemens, Optum, Abbott, Celgene, AstraZeneca, AbbVie, Westat, several investment and venture capital firms, and many academic institutions, medical or disease specific foundations and associations, and health systems. AJB receives royalty payments through Stanford University, for several patents and other disclosures licensed to NuMedii and Personalis. AJB’s research has been funded by NIH, Northrup Grumman (as the prime on an NIH contract), Genentech, Johnson and Johnson, FDA, Robert Wood Johnson Foundation, Leon Lowenstein Foundation, Intervalien Foundation, Priscilla Chan and Mark Zuckerberg, the Barbara and Gerson Bakar Foundation, and in the recent past, the March of Dimes, Juvenile Diabetes Research Foundation, California Governor’s Office of Planning and Research, California Institute for Regenerative Medicine, L’Oreal, and Progenity. CLB has intellectual property in and receives royalties from BioFire, Inc. She serves as a scientific advisor to IDbyDNA (San Francisco, CA and Salt Lake City, UT); and is on the Board of the Commonwealth Fund. CK is a paid employee of Aetion and hold Aetion stock options. NES is an employee of Optum Labs and owns stock in the parent company UnitedHealth group. NDL was an employee of Health Catalyst at the time the work was performed. JLG is a full-time employee of Regenstrief Institute, which provides independent research services to entities including those within the pharmaceutical and medical device industries. SJG serves as Chief Medical Information Officer for the Indiana Health Information Exchange, and is a founding partner of Uppstroms, LLC.” Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing-interests>). If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. Please include your updated Competing Interests statement in your cover letter; we will change the online submission form on your behalf. Thank you for pointing this out. There are no conflicts between competing interests and PLOS One policy on data sharing. We have revised our competing interest in the cover letter to include the requested statement below: This does not alter our adherence to PLOS ONE policies on sharing data and materials. 5\. Thank you for stating the following in the Acknowledgments Section of your manuscript: “Financial support for this work was provided in part by a grant from The Rockefeller Foundation.” We note that you have provided additional information within the Acknowledgements Section that is not currently declared in your Funding Statement. Please note that funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form. Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows: “CVRW was funded by a grant from the Rockefeller Foundation. BDP, CK, GJ used funding provided by Yale University-Mayo Clinic Center of Excellence in Regulatory Science and Innovation (CERSI), a joint effort between Yale University, Mayo Clinic, and the U.S. Food and Drug Administration (FDA) (3U01FD005938). CK, CMF, SJG, PJE, EHE, NDL, and JLG work was funded by a designated sub-grant from the FDA Foundation. AJB funded by award number A128219 and Grant Number U01FD005978 from the FDA, which supports the UCSF-Stanford Center of Excellence in Regulatory Sciences and Innovation. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the HHS or FDA. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.” Thank you for bringing this to our attention. We have removed the funding information from the Acknowledgement section and include all funding information as updated above under “Funding Information.” 6\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. Thank you for bring this to our attention. The data that we used for our findings are not bound to some legal or ethical restrictions. All relevant data are contained within the manuscript or its supporting documents. Person-level data are unavailable. Qualified researchers interested in accessing deidentified person-level data should contact the corresponding author for more Information. 7\. One of the noted authors is a group or consortium “Evidence Accelerator Workgroup”. In addition to naming the author group, please list the individual authors and affiliations within this group in the acknowledgments section of your manuscript. Please also indicate clearly a lead author for this group along with a contact email address. Thank you for bringing this to our attention. The Evidence Accelerator Workgroup refers to our consortium. We only had included those who met the ICMJE authorship as co-authors, but wanted to acknowledge all those who worked behind the scenes to support the work. Given this advice, we will include their names as acknowledgements. We have added the following phrase “We thank all members of the Evidence Accelerator Workgroup for their support and feedback: \[list names\]” to lines 395-399. 8\. We note that Figure 1 in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. Thank you for bringing this to our attention. We had previously attached the Content Permission Form as part of the submission package. We attached the Content Permission Form as part of the submission package again and added the reprint information to the figure 1’s caption. Reviewer 1 comments: 1\. The manuscript describes agreement (PPA) between PCR testing and serology results at 14-90 days post PCR. The approach to analyzing agreement is not complete with no mention of kappa (Cohen, McNemar test, etc.). Thank you for the comment. This analysis focused on serology agreement with positive RNA tests. As such, we did not collect any RNA negative results, which are required to assess kappa. 2\. Is there a loss of agreement depending on day serology was done? Yes, there was loss of agreement depending on the day serology test was done, and one of the results we intend to present in a subsequent publication. The current manuscript is intended to discuss the overall agreement between serology and PCR during the window for which agreement is most expected (days 14-90 after PCR) and by key demographic and clinical factors. A subsequent publication will look at how that agreement changes over different time periods. 3\. When comparing PPA observed according to ethnicity or other factors, are the differences significant? We report differences in PPA by race and ethnicity only where there are sufficient data (i.e. missing \<30% and sample size ≥ n=40). We consistently observed higher PPA in Hispanic ethnicity compared to non-Hispanic ethnicity, as demonstrated by complete separation of confidence intervals (Clopper-Pearson). We did not conduct direct comparison of PPA across groups. We clarified the statistical analysis section (lines 171-173) to describe the meaning of 'significant differences’ outside of direct comparisons: “We calculated exact (Clopper-Pearson) 95% confidence intervals (CI). We report significant differences where 95% CI have complete separation - although we did not conduct formal statistical comparisons of PPA between groups.” 4\. Line 284, the authors mention test quality. Many studies looking at the quality of several serology tests used in this manuscript have been published, can the authors discuss further and refer to these manuscripts? Does lower agreement observe in this study match the published data? Thank you for the great suggestion. We have now included in the Discussion a comparison of our PPA results to those from sensitivities reported in original EUA submissions; as well as in context of other external evaluations (see lines 290-298): “The reported sensitivity of the serology tests included in this analysis that were submitted for EUA approval were all \>95% \[36\]. Our analysis of multiple large datasets of patients with confirmed SARS-CoV-2 infection suggests that serology tests performed lower than =expected – with PPA ranges (a measure analogous to sensitivity) from 65-90%. - Our results align with results from smaller, detailed laboratory evaluations that suggest a lack of harmonization, including optimization of cut-off values, may contribute to decreased overall performance. Additionally, our results align with studies that include more representative samples of milder or asymptomatic persons \[37–39\].” Although many of these evaluations are still limited by much smaller sample sizes than we report (though they are more detailed laboratory studies) and appear limited in the replicability of the same assay result across different laboratories, they do contain more diverse populations and note lower performance compared to initial certification related evaluations. 5\. In the section results, no odd ratio is given, only 95%CI. Please add. Apologies for the confusion, the results given in the parenthesis in this section are actually ranges of OR, not the 95% CIs. We have clarified this in reporting results in this version. 6\. Factors associated with seropositivity, line 257-266. Are the differences described significant? Thank you for the comment. The differences described in the text were significant. Significant differences can be observed in the figures as those whose 95% CI does not cross the ‘1’on the X axis. Across the board, we found age 20-44 yrs to have a lower odds of seropositivity than those 45-54 yrs; Hispanic Ethnicity to have higher odds than non-Hispanic; immunocompromised to have a lower odds of seropositivity than those with no pre-existing conditions. ORs for obesity and presenting with \>1 COVID symptom also were also significantly elevated in \>1 data source. We have clarified this in the text by adding the word “significantly” to lines 269 and 273. 7\. Line 135 change e.g. by i.e. or remove altogether Thank you for calling this out. To clarify, PCR was not the only molecular conducted. The list included NAAT, RT-PCR so we respectfully leave it as “e.g” 8\. Line 135, remove comma after IgG, “IgG, \[21\],” Thank you for bring this to our attention. We have removed the comma after IgG. Please see line 141. 9\. Table 1, “Na” in lowercase, all other table “NA” is capitalised. Please homogenise throughout. Thank you for bring this to our attention. We have updated Table 1 to read NA instead of Na to match the other table. 10\. Line 313, should read sustained antibody production. Thank you for bring this to our attention. We have updated the language in line 331. 11\. Line 359, should read the sustained presence of antibodies. Thank you for bring this to our attention. We have updated the language in line 377-378. 12\. Line 371, should be “in vitro”. Thank you for bring this to our attention. We have updated the language in line 389. Reviewer \#2 comments: 1\. What was done statistically regarding individuals with a 2nd serology test? Line 161 implies only the first test was used. What was the concordance/timing between 1st and 2nd tests? Did 2nd tests, where done, have a different PPA? The majority of person in our cohort had just one serology test done. In order to compare consistently, we picked the first test done on an individual occurring 14 or more days after their positive molecular test. This choice has the added benefit of avoiding bias that would occur if we counted the same individual more than once knowing that individuals are more likely to retest if they get a result which is unexpected. 2\. Dataset C seems to have a broadly lower PPA vs the other datasets, has the smallest N, and is relatively geographically restricted. This dataset does not appear to have any manufacturer molecular test names available, but there is no PPA reported in the Unknown/missing category in table 2 for that variable. Is this dataset usable? It would seem that as both the serological and molecular test characteristics would contribute to the PPA, not knowing the molecular test name at all makes using this dataset problematic. Keen observations! Not all partners reported the name of the molecular test and thus, did not estimate PPA by molecular test (reference). All partners did estimate PPA by serology test. We acknowledge in the limitations that we did not analyze molecular-serology pairs. As you note, characteristics of each test may influence PPA results, though we account for many other factors that may affect results. Because of this limitation, we chose not to report results by specific test name as it may suggest a deficiency that we could not accurately explain. 3\. What is known about the contribution of “other” molecular tests to this dataset, such as the adoption of “rapid” PCR testing and “in-house” testing that some institutions produced during this time period? Is it possible to address those tests where both the serological and molecular tests are known? As above, the confounding factor of molecular test characteristics could influence the PPA of the serological tests in question. Rapid tests were not included in this analysis. One site included an ‘in-house’ test that was not FDA approved or for whom an EUA was not issued; the majority used only FDA approved or EUA tests. As such, we did not conduct the suggested analysis. 4\. Time between the molecular and serological test seems like a key point as well. Do you have that data? Does time between tests affect the results? Yes, we agree with your comment. Analysis of PPA since the time of the molecular test is the focus of a subsequent manuscript. In the current analysis, we focus on tests 14-90 days from positive molecular test to maximize sensitivity of the test. 5\. The figures are not showing well (scattered pixels) in my Adobe Acrobat Pro DC view of the PDF. Please ensure that high quality images are used during publication. I can access the tif files, which look right. Thank you for letting us know. We will resubmit the figures as.tiff files. 6\. I do not understand why figures 2-6 list the study period as starting in 2019. Is this a typo or is this correct? The methods list 2020 as start date, which would make sense given the dates of the pandemic. Thank you for the question. This is correct. March 2019 represents the collection of baseline data such as comorbidities and socioeconomic data. The study follow-up period begins in March 1, 2020 and continues until September 30, 2020. 7\. Please check line 380, there may be a comma instead of a period. Thank you for bring this to our attention. We have removed the comma and changed it to a period. Please see line 399 8\. Perhaps figures 2-7 should be combined into a single summary figure (with total N for the entire study) and the individual flowsheets by cohort might be moved to supplemental. The reader might better grasp the overall study with a simpler summary figure. Thank you for the suggestion. We created a summary figure (Fig 2) that depicts the general study design and the sample size of each partners study cohort; thus renumbering the remaining figures and moving the individual study diagrams to the supplemental figures. Each of the final cohorts (A-F) include patients that have both molecular test and follow up serology test as indicated in the methods. The parallel analysis approach entails that each cohort was analyzed separately according to a common protocol. Since this was not an aggregated analysis, we did not aggregate the numbers across partners. 10.1371/journal.pone.0279956.r003 Decision Letter 1 Banada Padmapriya P Academic Editor 2023 Padmapriya P Banada This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Dec 2022 Real-world performance of SARS-Cov-2 serology tests in the United States, 2020. PONE-D-22-11773R1 Dear Dr. Rodriguez-Watson, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Padmapriya P Banada, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Thank you for resubmitting your article addressing the comments raised by the reviewers. Thank you for considering the comments constructive. The manuscript is greatly improved and is clear. Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Thank you for adressing all the comments. Two very minor typos if this can be changed before publication line 111 : Figs 2 - remove s line 161 add space between CI and reference \[33\] Reviewer \#2: Thank you for addressing my comments and questions. I have no additional concerns or questions at this time. I recommend acceptance and publication. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0279956.r004 Acceptance letter Banada Padmapriya P Academic Editor 2023 Padmapriya P Banada This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 24 Jan 2023 PONE-D-22-11773R1 Real-world performance of SARS-Cov-2 serology tests in the United States, 2020. Dear Dr. Rodriguez-Watson: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Padmapriya P Banada Academic Editor PLOS ONE [^1]: AJB is a co-founder and consultant to Personalis and NuMedii; consultant to Samsung, Mango Tree Corporation, and in the recent past, 10x Genomics, Helix, Pathway Genomics, and Verinata (Illumina); has served on paid advisory panels or boards for Geisinger Health, Regenstrief Institute, Gerson Lehman Group, AlphaSights, Covance, Novartis, Genentech, Merck, and Roche; is a shareholder in Personalis and NuMedii; is a minor shareholder in Apple, Facebook, Alphabet (Google), Microsoft, Amazon, Snap, Snowflake, 10x Genomics, Illumina, Nuna Health, Assay Depot (Scientist.com), Vet24seven, Regeneron, Sanofi, Royalty Pharma, Pfizer, BioNTech, AstraZeneca, Moderna, Biogen, Twist Bioscience, Pacific Biosciences, Editas Medicine, Invitae, Doximity, and Sutro, and several other non-health related companies and mutual funds; and has received honoraria and travel reimbursement for invited talks from Johnson and Johnson, Roche, Genentech, Pfizer, Merck, Lilly, Takeda, Varian, Mars, Siemens, Optum, Abbott, Celgene, AstraZeneca, AbbVie, Westat, several investment and venture capital firms, and many academic institutions, medical or disease specific foundations and associations, and health systems. AJB receives royalty payments through Stanford University, for several patents and other disclosures licensed to NuMedii and Personalis. AJB’s research has been funded by NIH, Northrup Grumman (as the prime on an NIH contract), Genentech, Johnson and Johnson, FDA, Robert Wood Johnson Foundation, Leon Lowenstein Foundation, Intervalien Foundation, Priscilla Chan and Mark Zuckerberg, the Barbara and Gerson Bakar Foundation, and in the recent past, the March of Dimes, Juvenile Diabetes Research Foundation, California Governor’s Office of Planning and Research, California Institute for Regenerative Medicine, L’Oreal, and Progenity. CLB has intellectual property in and receives royalties from BioFire, Inc. She serves as a scientific advisor to IDbyDNA (San Francisco, CA and Salt Lake City, UT); and is on the Board of the Commonwealth Fund. CK is a paid employee of Aetion and hold Aetion stock options. NES is an employee of Optum Labs and owns stock in the parent company UnitedHealth group. NDL was an employee of Health Catalyst at the time the work was performed. JLG is a full-time employee of Regenstrief Institute, which provides independent research services to entities including those within the pharmaceutical and medical device industries. SJG serves as Chief Medical Information Officer for the Indiana Health Information Exchange, and is a founding partner of Uppstroms, LLC. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction *Dichelobacter nodosus* is a gram-negative fastidious anaerobic bacterium and the causative agent of ovine footrot. The disease has a global presence and is endemic in many countries. In Switzerland, the true prevalence of virulent *D*. *nodosus* in sheep on animal level is estimated at 16.9% and on farm level at 16.2%. Clinical symptoms range from mild interdigital dermatitis in benign footrot to severe underrunning and separation of the hoof horn from the underlying tissue in virulent footrot. Clinical symptoms start as early as 2 weeks after first contact and the disease leads to pain, lameness, decreased meat and wool production as well as animal welfare issues. The economic and intangible costs of the disease are considerable. In Switzerland, costs for management and growth reduction without control measures were estimated at CHF 172.3 million for 2014–2030. To address these problems, various countries started to develop and implement footrot control or elimination programs. In Switzerland, a mandatory footrot control program started first in the canton of Grisons in 1994 whereas in other cantons sheep owners could voluntarily join the control program offered by the Swiss Consulting and Health Service for Small Ruminants. The successful control in Grisons, progress in laboratory diagnostics allowing PCR-detection and discrimination of virulent and benign *D*. *nodosus*, and the ongoing unsatisfactory situation in other cantons led to the political decision for a nationwide footrot control program. A cost-benefit analysis confirmed positive epidemiological and economic effects of this approach. The Federal Food Safety and Veterinary Office is currently preparing the nationwide footrot control program, which is planned to start in 2022. The goal is to reduce flock prevalence of virulent *D*. *nodosus* to less than one percent within five years. The control program consists of three phases: i) swab sampling for detection of virulent (*aprV2*-positive) *D*. *nodosus* by PCR, ii) treatment of *aprV2*-positive herds by claw-trimming and weekly footbaths in disinfectant solution, iii) surveillance of treated herds. Most frequently used disinfectants in footbaths in Switzerland are 4% formaldehyde, 10–20% zinc sulfate and 5–10% copper sulfate. In a recent proof- of-concept study, weekly footbaths in 10% zinc sulfate was shown to eliminate *aprV2*-positive *D*. *nodosus* from feet of sheep within 6–19 weeks. Despite their effectiveness, these substances have undesirable characteristics for use in a nationwide control program. Formaldehyde smells pungent and irritates airways. It is a known cause of allergic contact dermatitis and a carcinogen in both humans and animals. Repeated use of formaldehyde footbaths in sheep caused keratinization of the interdigital skin, which can lead to secondary infection and lameness. Zinc and copper are both essential trace elements, acting as catalytic or structural components of larger molecules and are therefore indispensable for live. However, similar to more toxic heavy metals they are a major contaminant of soil and groundwater, accumulating in water, sediment, aquatic plants and fishes, posing a potential health threat to aquatic, human and animal life. Hence, the aim of this study was to identify an alternative disinfectant solution, which is highly effective against *D*. *nodosus*, non-carcinogenic, environmentally acceptable, inexpensive, available as concentrate and suitable for licensing as biocide for treating footrot in Switzerland. The effects of different disinfectant solutions on virulent *D*. *nodosus* were investigated by culture based *in vitro* testing and *ex vivo* evaluation applying a newly established PMA (propidium monoazid) real-time PCR (PMA-qPCR) using the improved dye PMAxx^™ allowing discrimination of viable and dead *D*.*nodosus*. # Materials and methods ## *In vitro* disinfectant testing Disinfectants were selected from literature considering their toxicity, degradability and availability as concentrate. Three disinfectants already in use for treatment of footrot in Switzerland and 19 substances or products selected by expert opinion based on the results of a literature search were evaluated *in vitro* for their antimicrobial activity against virulent *D*. *nodosus*. Disinfectants, which resulted in a ≥ 5 log reduction of the number of colony-forming units (CFU), were further tested with organic soiling. Formaldehyde, copper sulfate, zinc sulfate, DESINTEC® Hoof Care Special D (Desintec) and its main compounds acetic acid, glycolic acid and glutaraldehyde were tested three times, all other substances once, with and without simulating soiling. The composition of Desintec is given in. Virulent (*aprV2*-positive) *D*. *nodosus* ATCC 25549^T was cultured on Brucella Blood Agar with Hemin and Vitamin K1 (Becton Dickinson) at 37°C in an anaerobic chamber (80% N₂, 10% CO₂ and 10% H₂; Scholzen Microbiology Systems AG). After 4–5 days, cultures were transferred with a cotton swab into Difco^™ LB Broth (Becton Dickinson) and suspended until McFarland 4 was reached in Densichek (bioMérieux). To simulate high-level soiling, a solution of 10% Bovine Serum Albumin (Sigma-Aldrich) and 10% yeast extract (Becton Dickinson) was filtered through a 0.2 μm Acrodisc syringe filter (Pall Corporation) and complemented with 10% defibrinated sheep blood (Thermo Fisher Scientific) resulting in a 10x soiling-solution. To determine the CFU/ml in the test mixture at the beginning of contact time (*N*_*0**)*, the number of CFU/ml in the test mixture after disinfectant treatment at the end of contact time (*N*_*D*), and in the positive control, 1 ml of LB or 1 ml soiling-solution were prepared in a 15 ml Falcon tube (Sarstedt). Then 8 ml of LB, 8 ml 1.25x disinfectant solution or 8 ml 5% formaldehyde (Sigma-Aldrich) were added, respectively. Finally, 1 ml of *D*. *nodosus* suspension was added before mixing the tubes. After 5 min contact time at room temperature, the tubes were centrifuged for 5 min at 4255xg, the supernatant discarded and the pellets resuspended in 10 ml LB. This was repeated twice, the last time with resuspension in 1 ml LB. Subsequently, tenfold dilution series down to 10⁻⁷ were prepared with 100 μl test suspension and 900 μl LB out of which 500 μl were plated on Brucella Blood Agar. The plates were incubated for 4–5 days as described before. After counting, plates with CFU in the range of 14 to 330 were included for calculation of the weighted mean count of *N*_*0* and *N*_*D*. If possible, two dilutions were evaluated, otherwise only one. *N*_*0* and *N*_*D* were calculated as follows: $$N = \frac{c}{\left( {n_{1} + 0.1n_{2}} \right)vd}$$ where *c* is the sum of CFU taken into account; *n*_*1* is the number of plates taken into account in the lower dilution; *n*_*2* is the number of plates taken into account in the higher dilution; *v* is the volume plated in ml; *d* is the dilution factor corresponding to the lower dilution. Example of two dilutions: $$N = \frac{244 + 33}{\left( {1 + 0.1 \cdot 1} \right) \cdot 0.5 \cdot 10^{- 5}} = \frac{277}{1.1 \cdot 0.5 \cdot 10^{- 5}} = 5.0 \cdot 10^{7}\ (in\ CFU/ml)$$ The reduction (R) was expressed as decimal logarithm: log₁₀R = log₁₀ *N*_*0* –log₁₀ *N*_*D*. Whenever *N*_*D* was zero, the value "1" was applied. Disinfectants demonstrating a ≥ 5 log₁₀ reduction at 5% disinfectant concentration with and without soiling were considered useful. ## *Ex vivo* experiments ### PMAxx^™ treatment conditions and linear range of PMA-qPCR A 1 ml McFarland 4 suspension of virulent *D*. *nodosus* was prepared as described before. Half of the suspension was heated at 99°C for 10 min. With both, the living and heat-treated *D*. *nodosus* suspension, tenfold serial dilutions were prepared down to 10⁻⁵. For determining CFU in the suspension, 10 μl of each untreated 10⁻³, 10⁻⁴ and 10⁻⁵ dilution were mixed with 500 μl 0.85% NaCl, plated on Brucella Blood Agar and incubated for 4–5 days. Killing of heat- treated *D*. *nodosus* was confirmed by plating 10 μl of the heated-treated suspension in the same way. For PMA-qPCR enumeration of viable and dead *D*. *nodosus* in the dilution series, 10 μl of dilution were mixed with 40 μl of 0.85% NaCl in transparent 1.5 ml tubes (Sarstedt). In a darkened room 100 μM (0.25 μl) of PMAxx^™ (Biotium) was added and the tube was vortexed and incubated for 3 min at room temperature in a metal box impervious to light. After incubation, tubes were placed on ice and exposed to LED light (Optonica LED floodlight Item No. FL5836, white light, 100 W, 6000 K, 8500 lm) at a distance of 20 cm for 5 min. Following light treatment, the tubes were centrifuged at 15'000xg for 5 min in a microcentrifuge. To remove the supernatant, the tubes were held in a horizontal position and twisted tissues were used to absorb the liquid. The pellets were resuspended in 500 μl SV-lysis buffer (4 M guanidine thiocyanate, 0.01M Tris–HCl, 1% β-mercaptoethanol). DNA extraction was done following established protocols including the VetMax^™ Xeno^™ IPC (Thermo Fisher Scientific) on an automated purification system (KingFisher^™ Duo-Prime, Thermo Fisher Scientific). Extracted DNA was used directly or stored at -20°C until further qPCR analysis was undertaken. Genomic DNA of *D*. *nodosus* ATCC 25549^T was used as an external standard in the qPCR with seven points corresponding from 10⁷ to 10¹ genome equivalents per well. Dilution stages of live and dead *D*. *nodosus* were analyzed in duplicate, the external standard in triplicate. Assay conditions consisted of a 25 μl reaction mixture containing 1 x TaqMan^™ Fast Advanced Master Mix (Thermo Fisher Scientific), 300 nM primers DnAprTM-L and DnAprTM-R, 100 nM Probe DnAprTm-v, 250 nM Probe DnAprTM-b, pyrogen-free water, 1 μl Xeno^™ LIZ Primer Probe Mix (Thermo Fisher Scientific) and 2.5 μl of DNA template. Amplification was done in a 7500 Real- Time PCR-System instrument (Applied Biosystems), using cycles of 2 min at 50°C and 10 min at 95°C followed by 40 cycles with 15 s at 95°C and 1 min at 60°C. Results were analyzed using the 7500 Software (v 2.3.) with the threshold set at 0.015. Mean cycle threshold (Ct) values of duplicates and triplicates were calculated. The Ct values of heat-treated and live *D*. *nodosus* dilution series were plotted versus the log quantity of standard DNA. Determination of linear range was done three times and average values calculated. The protocol for detection of live *D*. *nodosus* by PMA-qPCR is publicly available at <http://dx.doi.org/10.17504/protocols.io.bbh9ij96> ### *Ex vivo* disinfectant testing Clinically affected feet from sheep with footrot score ≥ 2 were collected at the abattoir in Thun and transported at room temperature to the laboratory within 30 min. Gross manure was cautiously removed from claws and the interdigital space, the open articulation of the carpal or tarsal joint was covered with gloves and an identification number was assigned to each foot. A cotton swab was soaked in 0.85% NaCl and a first sample (prevalue) was taken from the interdigital space. The swab was once rotated 360 degrees and subsequently soaked in 500 μl of 0.85% NaCl in a 1.5 ml tube. The feet were then attached to the cords of an in-house foot-dipping machine, which allowed foot-dipping of eight feet in parallel and had a frequency of five down movements per minute and a total contact time with the disinfectant of 50 s per minute. The machine was turned on, and feet were dipped and moved in plastic beakers containing 800 ml of disinfectant solution. After 10 min, the machine was stopped and the feet were left hanging for another 60 min outside the disinfectant solution at room temperature. The dipping machine is supposed to imitate the situation of the sheep treated alive as described by Greber et al.. In short, sheep are standing and moving their feet in the bath for 10 min. Foot movement is supposed to increase the contact with the disinfectant. After footbathing sheep are contained for 60 min on a clean and dry concrete floor. A second interdigital swab (postvalue) was afterwards collected from the same area in the same way as described above. Prevalue and postvalue tubes were centrifuged for 5 min at 15'000xg and the supernatants were discarded. The remaining pellets were resuspended in 50 μl of 0.85% NaCl. The PMA-qPCR was performed as described before. Each swab sample was analyzed in duplicate, and the external standard was used for quantification. For each foot, the reduction of live *D*. *nodosus* was calculated as follows: $${\mathit{\log}_{10}R} = {\mathit{\log}_{10}(N\ prevalue)} - {\mathit{\log}_{10}(N\ postvalue)}$$ With the exception of Desintec, all other substances from the *in vitro* experiment were not approved as biocide for use in footbaths with sheep. Therefore, only 3%, 6% and 9% Desintec was tested *ex vivo* and compared to 4% formaldehyde and 10% zinc sulfate. A 0.85% NaCl solution was chosen as negative control. The software NCSS12 (NCSS Statistical Software) was used for the statistical analysis of the logarithmized reduction values. Assumption of normal distribution was checked using histograms and Shapiro-Wilk test. The values were not normally distributed and equal variance was rejected. Therefore, Kruskal- Wallis Test with Dunn's Test for multiple comparisons was used for determination of significant differences among disinfectants. The significance level was calculated at z-value \> 1.9600. Post hoc power testing was performed using Two- Sample T-Tests allowing unequal variance. # Results ## *In vitro* disinfectant testing Results for each disinfectant with the corresponding reduction of virulent *D*. *nodosus* are given in for triplicate testing and for single testing. The disinfectants formaldehyde (4%; 7.2/6.7) and copper sulfate (5%: 5.7/6.1; 10%: 7.2/6.7) achieved a ≥ 5 log reduction without (x/) and with (/x) organic soiling whereas 10% zinc sulfate (4.9/4.7) failed to meet the envisaged log reduction. The 20% zinc sulfate solution could not be evaluated, because centrifugation in the solution failed due to its high density. The individually tested active ingredients of Desintec at 5% (acetic acid; glycolic acid; glutaraldehyde) showed log reductions ≥ 5 (7.2/7.2; 7.7/6.0; 6.2/6.6) without and with organic soiling, respectively. Desintec itself showed log reductions ≥ 5 in 1:10 and 1:100 dilutions without and with soiling (6.8/6.8 and 6.8/6.4, respectively). In a 1:1000 dilution, the product evoked a 5.8 log reduction without soiling and 0.3 log reduction with soiling whereas in 1:10 000 dilution both test conditions failed to meet the envisaged ≥ 5 log reduction (0.4/0.2). In single experiments also lactic acid, propionic acid, hydrogen peroxide, sodium hypochlorite, octenidine dihydrochloride, chlorocresol and Ampholyt 20 achieved the required ≥ 5 log reduction in the number of *D*. *nodosus* even under soiling conditions. While calcium magnesium tetrahydroxide, calcium hydroxide and calcium oxide showed the requested reduction without soiling but not with soiling, sodium benzoate, tartaric acid, calcium magnesium oxide and electrochemically activated water failed to meet the minimum ≥ 5 log reduction without and with soiling. ## *Ex vivo* experiments ### Linear range of PMA-qPCR Enumeration of viable *D*. *nodosus* by culture and PMA-qPCR revealed good correlation. Samples containing 10⁷ to 10³ of *D*. *nodosus* showed a decrease of DNA in the dead cell sample by PMAxx^™ treatment between 4 log and 2 log. At 10² CFU/ml, the decrease dropped to 1 log unit. Therefore, part of the DNA corresponding to approximately 10¹ to 10³ CFU/ml remained intact. ### *Ex vivo* disinfectant testing Efficiency of disinfectants is shown in. Medians of 10% zinc sulfate versus 0.85% NaCl were not different (z = 1.8705), whereas 4% formaldehyde evoked a significant reduction (z = 3.8794). For this reason, the more effective disinfectant formaldehyde was chosen for comparison with Desintec. Both 6% and 9% Desintec showed no significant differences compared to 4% formaldehyde (z = 0.8710 and 0.2646). Post hoc power testing between 4% formaldehyde and 6% Desintec with group sample sizes of 13 and 14 achieved 80.078% power to reject the null hypothesis of equal means when the population mean difference (delta) is 1 with standard deviations of 1.2 for group 1 and 0.8 for group 2, and with a significance level (alpha) of 0.050. A 4% formaldehyde solution is significantly more effective than 10% zinc sulfate (z = 1.9719). Unequal variances of disinfectants and a dose-responsive reduction of Desintec are visible in the box plot. # Discussion The aim of this study was to identify and test alternative disinfectants for future use as footbath solutions in a footrot control program. Based on the requirements for an effective, non-carcinogenic, environmentally acceptable, inexpensive and licensable concentrate, around 22 substances or products were selected by expert opinion. It became clear in discussions with representatives of veterinary drug companies that only an already registered product for footbaths would be considered for marketing since a new registration of a substance for the limited market would be too expensive. The product DESINTEC® Hoof Care Special D (Desintec) had already been registered as a biocide in Germany under the name PediSept G20 and became therefore the focus of the study. Thus, Desintec and its main components glutaraldehyde, acetic acid and glycolic acid, as well as the "gold standards" formaldehyde, copper sulfate, zinc sulfate and other chemical substances were tested. The *in vitro* study confirmed the effectiveness of formaldehyde, copper sulfate and zinc sulfate as disinfectants in footbaths. These have been used for a long time and are an effective treatment option for footrot in sheep. However, the 10% zinc sulfate solution was slightly below the requested ≥5 log reduction and the 20% solution could unfortunately not effectively be tested due to its high density, which prevented *D*. *nodosus* and erythrocytes from sedimentation by centrifugation. Desintec fulfilled the targeted reduction of ≥5 log at 1:10 and 1:100 dilution in accordance with the manufacturer recommendation to use it at a concentration of 3–5%. The product was even effective in 1:1000 dilution without organic soiling, however, organic soiling substantially reduced its effect. When testing the components of Desintec (glutaraldehyde, acetic acid, glycolic acid) individually at 5% concentration, each of them was able to reduce the number of *D*. *nodosus* at the requested scale even under soiling conditions. Glutaraldehyde has a broad spectrum of activity and a rapid microbial killing rate. It is supposed to destroy all forms of microbial life, including bacterial and fungal spores, tubercle bacilli and viruses. Glutaraldehyde is part of many disinfectant solutions for livestock, listed by the committee for disinfection of the German Veterinary Medical Society. Organic acids are known to be used as food preservatives due to their antimicrobial potential. Like that, acetic acid reduced the microbial load of foodborne pathogens on several fresh fruits and vegetables. In medicine, acetic acid has been used for wound disinfection. Even though microorganisms vary in their susceptibility, acetic acid solution proofed to be bactericidal for *D*. *nodosus* in the current study as well. Glycolic acid is well known for its keratolytic properties. The small molecular weight allows for easy penetration of the skin, targeting the corneosomes and resulting in desquamation of the stratum corneum. *D*. *nodosus* can be found in a depth of 2200 μm in footrot affected tissue. The keratolytic effect of glycolic acid may contribute to a deeper penetration of disinfecting ingredients of biocides into skin and hoof thus improving their effectiveness. Apart from the keratolytic effect, glycolic acid shows antimicrobial properties like other carboxylic acids although it is less commonly used for this purpose. It proofed to be an effective postmilking teat disinfectant, and the strong reduction of *D*. *nodosus* in our *in vitro* experiments is another example of its antimicrobial activity. Lactic acid, propionic acid, hydrogen peroxide, sodium hypochlorite, octenidine dihydrochloride, chlorocresol and Ampholyte 20 were effective in reducing the number of viable *D*. *nodosus in vitro* even under soiling conditions. On the other hand, calcium magnesium tetrahydroxide, calcium hydroxide and calcium oxide showed the requested reduction without soiling but not with soiling, while sodium benzoate, tartaric acid, calcium magnesium oxide and electrochemically activated water failed as disinfectant without and with soiling. Since all these components lack the chance of becoming registered biocides, they were only tested once and not considered in the *ex vivo* assays. Based on the promising use of Desintec this product was further investigated in *ex vivo* tests using sheep feet in a dipping machine. In contrast to the *in vitro* assay, culturing of *D*. *nodosus* from a swab sample of heavily contaminated feet is not a sensitive method and requires subculturing, which makes a quantification of viable *D*. *nodosus* as needed for assessing effectiveness of disinfectants in the *ex vivo* part of our study impossible. Therefore, an alternative approach able of detecting viable cells by PCR as e.g. presented by Nogva et al. was established. By this approach, the distinction between viable and non-viable cells is possible, based on membrane integrity. For that purpose, the samples containing *D*. *nodosus* were treated with the improved nucleic acid intercalating propidium monoazid (PMA) dye PMAxx^™ that selectively enters cells with compromised cell membranes, whereas the intact cell membrane presents a natural barrier for this molecule. After exposure to strong light, it covalently binds to the DNA, preventing DNA from being amplified by PCR, thereby enabling differentiation of viable from non-viable cells. At the same time when the cross-linking with DNA occurs, any unbound excess PMAxx^™ reacts with water. The resulting molecule is no longer reactive, preventing reaction of PMAxx^™ with DNA extracted from intact cells. The PMA-qPCR proved to be a valid method for comparison of antimicrobial efficiency in the *ex vivo* experiments. Thereby, 10% zinc sulfate did not show a significant difference to the NaCl control. Therefore, 4% formaldehyde that showed a significant reduction of live *D*. *nodosus* was chose to compare to Desintec. The Desintec solution diluted down to 6% was still effective not showing any significant difference to the "gold-standard" of 4% formaldehyde. Variability of reduction within disinfectant groups was observed and is explainable by various factors influencing the *ex vivo* experiment. *D*. *nodosus* loads vary naturally among feet due to individual differences and the different clinical status of footrot affected feet, leading to different prevalues. Furthermore, the total number of microorganisms can also affect the activity of a disinfectant. Higher inoculum levels can attenuate the efficacy of disinfection by adding to the level of soiling and by providing protection to other organisms at the site. In addition, disinfection inactivation follows first-order kinetics. Starting from a high prevalue, the absolute reduction is high whereas the relative reduction is small. On the other hand, starting from a small prevalue, the effect is reversed with small absolute and high relative reduction. Reduction values obtained from smaller prevalues possibly lead to higher relative reductions, widening the variability. Negative reduction can be explained with the low precision of the swab sample. Another reason for the variability could be the extent of soiling. Soiling can affect both the disinfection process and PMA-qPCR. Even though claws and interdigital space were cleaned and open articulations were covered, disinfectant solutions were soiled with manure and blood to varying degrees after the footbath. Soiling can affect microbial activity by direct interference with the disinfectant, by interaction and protection of the target organism and formation of microbial aggregates. Moreover, complex matrices found in environmental samples can negatively influence the PMAxx^™ treatment by chemical adsorption of the dye and interference with photoactivation. Glutaraldehyde contained in Desintec is not listed by the International Agency for Research on Cancer (IARC) and there is no evidence for carcinogenic activity, genetic or reproductive toxicity. However, glutaraldehyde is irritating and corrosive to the skin, eyes and respiratory tract and is a known cause of allergic contact dermatitis and occupational asthma. It is toxic to aquatic life and should not be discharged into water bodies. On the other hand, a smaller amount of glutaraldehyde in combination with acetic and glycolic acid is needed to achieve the same antimicrobial effect as 4% formaldehyde. In a 6% Desintec solution (recommended final concentration) there is 0.36% glutaraldehyde compared to 4% formaldehyde currently used in footbaths. Furthermore, glutaraldehyde is rapidly degradable in air, water, and soil, does not bioaccumulate and is less toxic than formaldehyde. Desintec applied as 6% solution in footbaths is therefore an environmentally acceptable biocide that after use is recommended to be disposed on the manure pile or in the slurry basin. # Conclusion The study showed that Desintec is an effective alternative to formaldehyde (4%), zinc sulfate (10%) and copper sulfate (5%) for the use in sheep footbaths to eliminate virulent *D*. *nodosus*. The product is not only effective but also non-carcinogenic, is biodegradable and available as concentrate, making it an improvement over the currently used disinfectants. The results of this study represent a step forward on the way to a footrot control program that will mainly be based upon herd-level footbathing. # Supporting information We would like to thank the various members of the expert opinion group (Yuval Tempelman, Giochen Bearth, Maria Christina Härdi-Landerer, Rita Lüchinger, Mina Gómez Góngora, Flurina Stucki, Alwin Meichtry, Richard Eicher, Heinz Burkhardt) for their advice concerning requirements, registration and practical implementation of an alternative disinfectant. We would also like to thank Beatriz Vidondo and Brian Friker (Veterinary Public Health Institut, Liebefeld, Switzerland) for assistance with the statistics used in this work. Furthermore, we are grateful to Isabelle Brodard and Simon Feyer (Institute for Veterinary Bacteriology, Bern, Switzerland) for excellent technical support and the abattoir Thun for providing the footrot diseased feet of sheep. [^1]: The authors have declared that no competing interests exist.

# Introduction A variety of animals, from nematode to human, show social behavior. The social behaviors allow for individuals to create an entity greater than the sum of the individuals and provide the key to successful adaptation to the environment. One of the most striking features of the highly-ordered animal society is the ability to share information among individuals. In turn, higher communicative ability is a fundamental basis that enables animals to maintain a more evolved society. Honeybees (*Apis mellifera* L.) organize a highly-ordered society and have a sophisticated communicative ability known as the ‘dance communication’. Worker honeybees that find a rich food source return to the hive and might transmit the information on the location of the food source to their nestmates using a symbolic dance. The dance information is decoded into the spatial information of the food source by the other worker bees (followers) that follow the dancers. During foraging flights, worker honeybees integrate the incoming sensory information: they estimate the distance of food sources based on the amount of optic flow they perceive, and direction based on the position of the sun, which are the essential bases for the expression of dance communication. Although there is a considerable amount of research concerning the sensory basis of these remarkable abilities, almost nothing is known about the underlying neural mechanisms. As a first step in elucidating the neural mechanisms of these remarkable abilities, it is important to identify active brain regions in dancing and foraging honeybees that might be involved in dance communication and/or information integration during foraging flight. Although methods to detect the expression of immediate early genes (IEGs) as markers of neural activity are widely used in vertebrates, neural IEGs have not yet been identified in insects. In the present study, we identified a novel IEG that can be used as a neural activity marker and found that the neural activity of a mushroom body (MB) neuron subtype is preferentially increased in foraging honeybees, suggesting its involvement in information integration during foraging flight. # Results ## A novel non-coding IEG, *kakusei*, can be used as a marker to visualize neural activity in the honeybee brain To identify IEGs, we used the differential display method to search for honeybee genes that are immediately induced in the brain by neural activity. To evoke strong neural activity in the brain, seizures were induced by awakening workers from ice-cold induced anesthesia, because some of the IEGs were identified by inducing seizures in the animals. When the workers are awoken from anesthesia, they show seizure-like movement with their legs and body shaking. Using differential display screening of approximately 6500 bands, which were derived from mRNAs extracted from the brains of seizure-induced and non-treated bees, 49 candidate bands were identified. Among them, we selected nine candidates that showed a pronounced difference in band intensity between the seizure-induced and non-treated bees. After preliminary Reverse transcription-polymerase chain reaction (RT-PCR) analysis of these candidates, we finally focused on a single candidate that showed the most prominent and reproducible seizure-induced transcript increase. As a result, we identified a novel IEG that we named *kakusei* after the word ‘awakening’ in Japanese (the whole sequence of *kakusei* was deposited as DDBJ accession number AB252862). To examine the size of the *kakusei* transcript, we performed Northern blot analysis using total RNA isolated from whole brains of bees anesthetized with CO₂ and bees awakened from CO₂-induced anesthesia. The results indicated that the induced *kakusei* transcript was approximately 7 kb long. There was no significant open reading frame in any of the three possible reading frames of the *kakusei* cDNA sequence, suggesting that the *kakusei* transcript functions as a non-coding RNA. RT-PCR experiments and sequence analysis also confirmed that the contig *kakusei* sequence is expressed as continuous transcripts. *Kakusei* expression was transiently induced in the whole brain after awakening the workers from anesthesia induced by either CO₂ \[; Sz (seizure)-induced\] or ice-cold (data not shown). Real-time RT-PCR revealed that *kakusei* is expressed predominantly in the brain, suggesting a brain-specific function. *In situ* hybridization revealed that *kakusei* expression can be detected in every brain region, including the MBs, optic lobes (OLs), and antennal lobes (ALs) in a seizure induction-dependent manner, suggesting that *kakusei* can be used as a marker in broad brain regions. In addition, *kakusei* signals (purple) were detected exclusively in the nuclei (green) of brain neurons, reflecting the characteristics of the *kakusei* transcript as a non- coding RNA. This notion was clearly demonstrated when the *kakusei* transcript localization was compared to that of *actin*, which is transported to the cytoplasm to be translated into protein and is detected as a broadly-distributed signal in the cytoplasm. This characteristic *kakusei* signal staining enabled us to count and quantify the number of *kakusei*-positive neurons. We next examined whether *kakusei* expression reflects physiological neural activity by testing whether *kakusei* expression was induced in the brain upon light exposure. First, workers were dark-adapted by keeping them in the dark overnight. The next day, experimental bees (light-exposed) were exposed to white light for 30 min, and then used for *in situ* hybridization. Dark-adapted bees were used as a control. *Kakusei* expression was detected in the OL and MB neurons of the light-exposed bees. The expression of *kakusei* was particularly prominent in the lamina neurons, which receive direct input from retinal neurons. In contrast, there was no strong *kakusei* expression in the AL neurons. These results indicate that *kakusei* can be used as a marker to visualize physiological neural activity in the honeybee brain. To further examine whether *kakusei* expression can be detected in the brains of bees that exhibit physiologic behaviors, we studied *kakusei* expression in the brains of bees exhibiting phototactic behavior. According to a previous report, we collected foragers that moved to the lighted side and nurse bees that did not move to the lighted side 30 min after the start of the phototactic behavior, and investigated *kakusei* expression in the brains (see for the experimental procedures). Almost the same *kakusei* expression pattern was observed in the brains of both the foragers and nurse bees: a large number of neurons in the optic lobes were *kakusei*-positive, which was similar to the findings in the light-exposed bees. These results suggest that *kakusei* expression is not so sensitive as to be induced by the neural activity specific to phototactic behavior, and that the activity in the optic lobes, which could be induced by visual inputs upon light illumination, is predominant even in the brains of bees that exhibited phototaxis. The fact that *kakusei* expression was observed in the optic lobes of bees that were not dark-adapted and had natural phototactic behavior strongly suggests that *kakusei* expression reflects neural activity under normal physiologic conditions, although we still cannot exclude the possibility that light-exposure is a stressor to the honeybee. ## Neural activity is increased in the small-type Kenyon cells of the dancer brain The workers shift their tasks from nursing their brood (nurse bees) to foraging for nectar and pollen (foragers) according to the number of days after eclosion. When foragers successfully find food sources, some of them perform a dance to enroll followers to forage. Therefore, we next examined *kakusei* expression in the brains of the dancers, followers, and nurse bees to identify the brain regions involved in dance communication. The individual workers were caught immediately after confirming their behaviors in the observation hives, and used for *in situ* hybridization. The bees caught from the observation hives were immediately anesthetized by ice-cold water and kept on ice until use for *in situ* hybridization to maintain the current state of the *kakusei* transcripts in the brain. There was a characteristic *kakusei* expression pattern in the dancer brains, especially in the MBs. The honeybee MBs consist of three types of intrinsic neurons termed large-type Kenyon cells (lKCs), small-type KCs (sKCs), and class II KCs. The most prominent *kakusei* expression was observed in the sKCs , whose somata are located in the center of the MBs. In contrast, only a small number of positive cells were detected in the brains of followers and nurse bees. *Kakusei* expression in the central complex neurons was not clear, as we could not identify the central complex neurons in our *in situ* hybridization experiments. Quantitative analysis revealed that the number of *kakusei*-positive cells in the sKCs was approximately 20 times higher in the dancers than in the followers or nurse bees. In addition, *kakusei* expression was also weakly induced in the lKCs and class II KCs, as well as in the dorsal and ventral OL neurons (dOL and vOL, respectively). A three-factor ANOVA \[F1: bee type; F2: brain region (repeated measure); F3: brain hemisphere (repeated measure)\] revealed that there was a significant difference between the bee type and brain region (F1 and F2: *P*\<0.0001, respectively; significant interaction between F1 and F2: *P*\<0.0001). In contrast, there was no significant difference between the right and left hemispheres (F3: *P* = 0.9455). Significant differences between bee types were observed in the dOL, vOL, lKCs, sKCs, and class II KCs (\*, *P*\<0.05; \*\*, *P*\<0.01; Tukey-Kramer's test). These results indicate that the neural activity in these brain regions, especially the sKCs, is increased in the dancer brain. The differences in *kakusei* expression between dancers and followers could not be due to their different developmental stages but only to their behavioral differences, because both the dancers and followers are thought to be of the same behavioral stage. Thus, the prominent neural activity in the sKCs observed in the dancer brains is likely due to their characteristic behaviors. ## Neural activity of the small-type Kenyon cells is also increased in the forager brain According to the expression profile, *kakusei* expression reflects neural activity that occurred 15 to 60 min prior to sampling of the bees. In our observation, dancers repeated the dance every 5 to 7 min and the foraging every 10 to 15 min. Thus, the neural activity detected in the dancer brains might be due not only to the dancing behavior, but also to the preceding foraging behavior. To address this question, we examined *kakusei* expression in the brains of foragers. Only some of the foragers that succeed in finding food display dance behavior. Thus, when we analyze foragers, only some of them are expected to be dancers. Therefore, we collected foragers with pollen loads (an indication that they were successful in finding food) in front of the hive entrance before we checked whether or not they danced in the hive. The results indicated that every forager (N = 12) had a *kakusei* expression pattern similar to that of the dancers (N = 6). In addition, there was no significant difference in the density of *kakusei*-positive cells in the MB neurons, including in the sKCs, between these bees \[; *P*\>0.05, two-factor ANOVA (F1: bee type; F2: brain hemisphere)\]. Thus, these results suggest that the increased sKC neural activity in the dancer brain is associated with foraging behavior rather than dancing behavior, although we cannot exclude the possibility that the foragers we examined also exhibited the dance behavior shortly before the observation period. ## Re-orienting bees showed different *kakusei* expression patterns from foragers and dancers Foraging behavior might involve multiple behavioral processes, such as visual, olfactory, tactile, and gustatory experiences, flying, landmark learning and memory, information processing for the dance communication, etc. We next examined whether the sKC-preferential *kakusei* expression is associated with forager-specific behavioral components or components that are common to the other behaviors, such as flying, visual experience, or landmark learning and memory. We investigated *kakusei* expression in the re-orienting bees, which were collected as the workers that fly around the hive to re-orient themselves to the hive when the location of the hive has changed. For this, we moved the hives at night with the entrance closed. The next morning, we opened the entrance for 5 min and then caught the bees flying around the hive 0, 15, and 30 min later. *Kakusei* expression was increased in the MBs in a time-dependent manner in the re-orienting bees, suggesting that this neural activity is induced by re-orienting behavior. One-way ANOVA among re-orienting bees revealed that the time effect was significant for every MB neuron type (: *P'*s\<0.0001–0.003). In contrast to sKC-preferential *kakusei* expression in the foragers, however, *kakusei* was induced in all KC types in the re-orienting bees. The proportion of *kakusei*-positive cells in the sKCs compared to that in the lKCs was significantly higher in the foragers than in the re-orienting bees. To exclude the possibility that the neural activity in the re-orienting bees is due to the increased light exposure when they leave the hive, a similar experiment was performed using re-orienting bees from transparent observation hives. In this experiment, a significant time-dependent increase in *kakusei* expression was observed in every MB neuron subtype in the re-orienting bees (*P*\<0.0001, one-way ANOVA), like in the re-orienting bees from the normal hives, indicating that the increased *kakusei* expression in the re-orienting bees is due to re-orienting behavior, and not merely to light-exposure. Here, the workers that performed re-orientation flights were probably foragers, also suggesting that the differences in the *kakusei* expression pattern between the brains of re-orienting bees and foragers/dancers are due to differences in behavior, but not age. These results demonstrate that the active brain regions are different between the re-orienting bees and foragers/dancers, suggesting that the sKC-preferential *kakusei* expression in forager/dancer brains is not due to behavioral components in common with those of re-orienting bees, like visual experience, flying, or landmark learning and memory. Rather, the forager- specific behavioral components are likely to be responsible for the sKC- preferential *kakusei* expression. # Discussion In the present study, using novel methods to visualize neural activity in the honeybee brain, we demonstrated that neural activity in the MBs, especially the sKCs, is prominently increased in the brains of dancer and forager honeybees. In contrast, the neural activity of both the sKCs and lKCs was increased in the brains of re-orienting workers. These findings strongly suggest that sKC- preferential activity, but not mere MB-preferential activity, is associated with behavioral components that are specific to foraging flight. The MBs are believed to be important for higher sensory integration in the insect brain. Partial ablation of the MBs impairs only complex olfactory learning without affecting simple olfactory learning. In addition, in the honeybee brain, all sensory modalities investigated (visual, olfactory, gustatory, and mechanosensory) project to the MBs, ,. Furthermore, foraging experience greatly influence the MB neuropile volume increase and the MB neuron outgrowth. Although these findings imply that the MBs are the appropriate brain regions to process complex information during foraging behavior, such as the calculation of distance and direction, there was no direct evidence whether the MB neurons are actually active in forager brains. The methods that we established in this study provide the first opportunity to investigate neural activity in the brains of naturally behaving honeybees and revealed that the MB neurons are active in the dancers and foragers, although these methods are limited to labeling only cell bodies and not the neuropile because *kakusei* encodes a non-coding RNA whose expression is confined to the nuclei. Our study demonstrated that neural activity of the sKCs, one of the MB neuron subtypes, was prominently increased in the dancer and forager brains. Although the inputs and outputs of the sKCs have been well investigated, the role of the neural circuitry composed by the sKCs is unknown. The calyx (input region) of the honeybee MB is divided into three zones: lip, collar, and basal ring. The sKCs project dendrites to the basal ring zone, whereas the lKCs project to the lip or collar zone. On the other hand, all sensory modalities investigated (visual, olfactory, gustatory, and mechanosensory) project to both the basal ring and lip/collar zones, which are then relayed to both sKCs and lKCs,. Studies of the anatomy of the bee brain suggest that the basal ring zone receives multi-modal (visual, olfactory, gustatory, and mechanosensory) inputs and extensive recurrent inputs from extrinsic MB neurons, whereas the lip and collar zone receive mono-modal (olfactory and visual, respectively) inputs. In addition, the sKCs are composed of neurons with different morphologies. Thus, we speculate that the sKC-preferential neural activity observed in the forager/dancer brains reflects the complex activity within the MB neural networks required for sensory integration during the foraging flight. Recent studies suggest functional differences among the KC types based on their gene expression patterns,. For example, activation of *Amfor*, one of the genes preferentially expressed in the sKCs, affects the onset of foraging behavior. These findings are also consistent with our notion that the sKCs play roles in higher integration of the complex information that accompanies foraging experience. Prominent *kakusei* expression was detected only in the forager brains, not in the follower brains, possibly due to the differences in their foraging experience during the hour before sampling. The followers often show reduced foraging activity before they start following the dance. Thus, considering the short half-life of *kakusei* expression, it is possible that this behavioral difference is reflected in the *kakusei* expression in the follower brains. In addition, there were strikingly different *kakusei* expression patterns in the MBs between the re-orienting and foraging workers. Because the re-orienting bees fly around to form spatial memory regarding their hive location, they have several behavioral components in common with the foragers. How then are the differences in *kakusei*-expression between re-orienting and forager/dancer bees explained? Foraging behavior is different from orienting behavior in some ways: for example, foragers need to calculate the distance and direction of food sources, memorize them in association with food information, and recall them repetitively to repeat foraging, which involves much broader and multi-modal sensory integration than just orienting. Thus, it is possible that the increased activity in the sKCs in the forager brains is associated with these behavioral components and reflects such sensory integration. Alternatively, it is also possible that the sKC activity is important for both the foraging and re- orienting behaviors and the relative repression of the lKC activity is important for the foraging behaviors. If this is the case, the integration of sensory information during foraging flight might result from interactions between active sKCs and inactive lKCs. Here we identified the transcript of novel IEG, *kakusei*, as a non-coding nuclear RNA. Although microRNA, which is expressed in response to neural activity, has been reported in vertebrates, *kakusei* is the first example of a long non-coding nuclear RNA that shows an immediate early response to neural activity. Long nuclear RNAs regulate gene expression, whereas microRNA post- transcriptionally regulates gene expression. In general, vertebrate IEGs encode transcription factors and have roles in modulating neural functions in an activity-dependent manner. Thus, *kakusei* might regulate gene expression as a long non-coding RNA to modulate neural function. Although methods using IEGs as markers of neural activity are widely applied in vertebrates, no IEG has yet been reported in the insect brain. Thus, this is the first use of an IEG to identify an active brain region in the insect. Future studies examining *kakusei* expression after a well-defined experience of various sensory modalities such as gustatory, tactile, olfactory, and visual (colors, patterns, etc.) should dramatically enhance our ability to interpret the present data. Moreover, in future studies, the link between *kakusei* expression and neural activity will need to be clarified to reveal the kind of neural activity that is reflected by *kakusei* expression. Due to limitations of the experimental methods, detailed behavioral components that induce *kakusei* expression in the sKCs remain to be examined. Nonetheless, the present study provides important insight into the neural basis of sensory integration during foraging flight, which might be related to the dance communication. It also describes a useful method for mapping active brain regions involved in behaviors of interest in the honeybee. # Materials and Methods ## Bees European honeybees (*A. mellifera* L.) were purchased from a local dealer and maintained at the University of Tokyo. Observation hives were made as previously described with some modification. ## Differential display Worker honeybee brains were dissected out from each of 10 bees anesthetized on ice and bees awakened from ice-induced anesthesia, which showed a seizure-like phenotype. Total RNA was isolated with TRIzol (Invitrogen), treated with DNase I, and reverse transcribed with SuperScript II (Invitrogen). The differential display method was performed as described previously using a Fluorescent Differential Display Kit and LA *Taq* polymerase with a combined total of 216 primer sets (Takara). Bands of interest were excised, reamplified, and subcloned into a pGEM-T vector (Promega). ## Northern blotting Whole brain total RNA was isolated from each of 10 bees anesthetized by CO₂ and bees awakened from CO₂-induced anesthesia. RNA was subjected to 1% formaldehyde-agarose gel electrophoresis, transferred to a nylon membrane, and hybridized with ³²P-labeled riboprobes. ³²P-labeled riboprobes were synthesized by T7 polymerase with Strip- EZ^TM RNA Kit (Ambion) from a template containing the fragment isolated by differential display (DD fragment; from+4511 to+5159). ## cDNA cloning To identify the whole length of *kakusei* cDNA, 5′-and 3′-rapid amplification of cDNA ends (RACE) methods were performed repeatedly using the SMART RACE cDNA Amplification Kit (Clontech). ## RT-PCR RT-PCR experiments were performed using LA Taq (Takara) according to the manufacturer's protocol and the SMART RACE cDNA as templates. Primers were designed to amplify the regions shown in ; (a) 5′-CACGCTCGTCGTCGTGCCTTGCTCAGATAA-3′ and 5′-TTCAGAGCACGTTGGAACTAATCTCGCG-3′, (b) 5′-ACCTTGGAACGTGAAAGCGCATTTTCGA-3′ and 5′-AACCGTGTCCTTCTGCAGACACCTGACA-3′ ## Quantification of the *kakusei* transcript The expression of *kakusei* was induced by awakening bees from CO₂-induced anesthesia. Control bees were kept in CO₂. Total RNA was extracted from three to five bees for each sample. Real-time RT- PCR was performed with Light Cycler-DNA master hybridization probes (Roche) according to the manufacturer's protocol, using gene-specific primers (*kakusei*; 5′-GGAAACAGGTGGTTTGATGACCATTG and 5′-CACGTTCCAAGGTTTAACGATGCG, *actin*; 5′-GAAATGGCAACTGCTGCATC and 5′-TCCACATCTGTTGGAAGGTG) and fluorescent probes (*kakusei*; fluorescein isothiocyanate (FITC) probe, 5′-CGCTGTAGTGCGTTTTCACTCGGATCGA, and LC-Red640 probe, 5′-TCCGAGGAAATCCGAGCAAAGTTCGTTC, *actin*; FITC probe, 5′-CCATGAAAATTAAGATCATCGCGCCAC, and LC-Red640 probe, 5′-CGAGAAGAAATATTCCGTATGGATTGGTG). The amount of *kakusei* transcript was normalized with that of *actin* and is shown as relative to the value of control bees at 0 min or to the whole brain. There was no significant difference in the levels of *actin* expression between control and seizure-induced bees. ## *In situ* hybridization and image analysis *In situ* hybridization was performed as described previously with some modification. Frozen coronal brain sections (10 µm thick) were fixed in 4% paraformaldehyde in phosphate buffered saline, pretreated, and hybridized with digoxigen (DIG)-labeled riboprobes. The DIG-labeled riboprobes were synthesized by T7 or SP6 polymerase with a DIG labeling mix (Roche) from a template containing the fragment isolated by differential display (from+4511 to+5159). After stringent washes, DIG-labeled riboprobes were detected immunocytochemically with peroxidase-conjugated anti-DIG antibody (1:500; Roche) and TSA Biotin System (Perkin Elmer). Sense probes were used as negative controls and the signals were confirmed to be antisense probe-specific in every experiment. Micrographs of fluorescent *in situ* hybridization were taken using an IX71 confocal microscope (Olympus). SYTOX Green (Molecular Probes) was used to stain the nuclear DNA. Intensity and brightness of the micrographs were processed with Photoshop software (Adobe). For quantification, the brain regions were defined as shown in. The rostral section was defined as the section containing the AL, and the caudal section was defined as the section containing the DL. The numbers of *kakusei*-positive cells were manually counted from rostral and caudal sections. For each animal, one rostral and one caudal section were analyzed. The area of each brain region was measured using ImageJ analysis software (NIH, <http://rsb.info.nih.gov/ij>). The amount of signal was divided by the area and the values from two sections were averaged, when there were two data points for one individual (e.g., MB and OL neurons). Sections to be counted were randomly selected from many sections. The density of *kakusei*-positive cells was shown as a value relative to 100000 µm². Micrographs were numbered and signals were counted by an investigator blind to the bee type. The number of examined bees is shown in the figure. Statistical analyses were conducted by *F*-test and Student's *t*-test using Microsoft Excel (Microsoft) or JMP (SAS) software. Multiple comparisons were performed using one-way analysis of variance (ANOVA) and Tukey-Kramer's test. Statistical comparisons were made within the same brain hemisphere. Data are shown means±standard error (SEM) throughout this paper. # Supporting Information [^1]: Conceived and designed the experiments: T. Kiya. Performed the experiments: T. Kiya. Analyzed the data: T. Kiya. Wrote the paper: T. Kiya, T. Kunieda, T. Kubo. [^2]: This work was supported by the Program for Promotion of Basic Research Activities for Innovative Bioscience (PROBRAIN). T. Kiya is the recipient of a Grant-in-Aid for Japan Society for the Promotion of Science (JSPS) Fellows.

# Introduction Defensins are important small, basic, cysteine-rich, antimicrobial, and cationic peptides that are abundant and widely distributed. Defensins are widely distributed in multiple tissues in the body, most notably leukocytes and epithelial surfaces. They are often present at high concentrations and play an essential role in the innate immunity of their hosts from insects and plants to amphibians and mammals. Membrane permeabilization is the crucial step in defensin-mediated antimicrobial activity and cytotoxicity. Defensins from different origins exhibit structural and functional similarities with phylogenetic relationships between different types of defensins. Mature defensins amino acids sequences are highly variable in each defensin family and subfamily. Accurately identifying the types of defensins will be helpful in analyzing their specificitis for various microbial targets, provide novel insights for understanding their function, and facilitate antimicrobial drugs targets discovery. Biochemical experimental methods are highly reliable for elucidating types of defensins, such as nuclear magnetic resonance (NMR) spectroscopy. However, such the experimental techniques are time-consuming and expensive. Bioinformatics methods can timely provide useful information and insights for both basic research and antibiotics design. Thus, better understanding the distinct functions of defensin proteins requires an automated method for timely and reliably annotating the families of many defensin proteins. In our previous study, four defensin families (vertebrate, plant, insect and other defensins) were successfully classified using the increment of diversity (ID) method. In another work the authors developed the DEFENSINPRED classifier to predict human defensin proteins and their types based on pseudo amino acid compositions. However, further work is necessary because the datasets constructed in those methods were too small to reflect a statistical profile and did not impose a rigorous cutoff threshold to exclude the redundant samples in the existing defensin datasets. Moreover, a better web-server for defensins is also needed. In the present work, we constructed a more rigorous benchmark dataset to train the program, and a support vector machine (SVM) classifier was further proposed to classify these five defensin families. An 4% improvement was obtained compared with the previous method. For the convenience of experimental scientists, a free online server **iDPF-PseRAAAC** was first established. A friendly guide was further provided to describe how to use the web server. # Materials and Methods ## Dataset With rapidly increasing interest in defensins, the Defensins Knowledgebase is available, which is a manually curated database and information source devoted to the defensin family of antimicrobial peptides. The benchmark data set $\mathbb{S}$ for the defensin proteins in this study was taken from the Defensins Knowledgebase, which currently contains more than 500 defensin sequences ranging from prokaryotes to eukaryotes. To prepare a high-quality dataset, the program CD-HIT was used to remove the defensin proteins with ≥ 80% pairwise sequence identity to any other protein. Highly similar data will surely lead to overestimation of the performance of the proposed methods. If the sequence identity cutoff is set to a lower percentage (such as 25%), the results will be more objective and reliable. However, in this study we did not use such a stringent criterion because the currently available data do not allow us to do so. The proposed method is a sequence-dependent predictor, the input feature vectors are only derived from the primary amino acids sequence. So there is needed enough amino acids and dipeptide compositions to train the multi- classifier module. For the defensin peptides are polypeptides of fewer than 100 amino acids (See). Besides, for ensure the data reliable, this dataset is a manually curated database. All of the families annotation are gathered from bibliographic databases and sequence databases literature sources. If the sequence identity cutoff is set to a lower percentage (such as 25%), the numbers of proteins for family subsets would have been too few to have statistical significance. And the imbalanced data cause classifiers to tend to overfit and to perform poorly in particular on the minority class. Finally, we obtained a dataset $\mathbb{S}$ composed of 333 defensin proteins classified into five families, as formulated by the following equation: $$\mathbb{S} = \mathbb{S}_{1} \cup \mathbb{S}_{2} \cup \mathbb{S}_{3} \cup \mathbb{S}_{4} \cup \mathbb{S}_{5}$$ where the subset $\mathbb{S}_{1}$ contains 60 insect defensins, $\mathbb{S}_{2}$ contains 34 invertebrate defensins, $\mathbb{S}_{3}$ contains 42 plant defensins, $\mathbb{S}_{4}$ contains 40 unclassified defensins and $\mathbb{S}_{5}$ contains 157 vertebrate defensins, and ∪ represents the symbol for “union” in set theory. The length distribution of the five families is depicted in. For the readers’ convenience, the 333 defensin proteins sequences and codes are in. ## Reduced Amino Acid Alphabet In this study, the reduced amino acid alphabet composition(RAAAC) clustered by Protein Blocks(PBs) are used to predict defensins family and subfamily, which is composed of 16 average protein fragments of 5 residues in length. The Protein Blocks have proven their efficiency both in description and prediction of longer fragments\[–\]. Once the databank was encoded in terms of PBs, sequence specificity was computed. Each PB was so associated with a set of enlarged sequence windows \[–w; +w\] of length l. An amino acid occurrence matrix of dimension 20 × l was computed for each PB. Then, each matrix was transformed into propensities matrix. Finally, all the matrices were compiled to create a matrix F of size 20 ×m with m, a vector of length 16l. The distance between two kinds of amino acids *i* and *j* was computed by using the D(aa*i*, aa*j*). Then a hierarchical clustering using all the amino acid occurrence matrices of the 16 PBs was performed, each resulting amino acid cluster represents amino acids that showed the same over- and under-representations upon all the PBs. Different defensin peptides usually have specific functional regions, such as β-sheet-rich fold and framework of six disulphide-linked cysteines. Based on the similarity of their functional and physicochemical features in proteins, the 20 amino acids can be clustered into some smaller groups. The reduced amino acids not only can simplify the complexity of the protein system, but also improve the ability in finding structurally conserved regions and the structural similarity of entire proteins. The reduced amino acid alphabet derived from Protein Blocks method has the ability for abstracting useful functional and conservative feature. And it also is helpful for simplifying the amino acids composition of defensin peptide and improving the ability in finding structurally conserved regions and the structural similarity of entire proteins. Up to now, the Protein Blocks method has successful been used to analyze long protein fragments and to predict functional regions, and the results have proven their efficiency both in description and prediction of longer fragments, such as protein structure mining, outer membrane proteins analysis and backbone structure prediction of proteins. Our previous researches have also demonstrated that this feature selection method will be useful for analyzing the conservative domain and understanding function evolution of defensin protein. ## Support Vector Machine (SVM) SVM is a powerful and popular method for pattern recognition that has been widely used in biology classification based on statistical learning theory\[–\]. In training an SVM classification system, proteins are represented by sequence- derived properties and are projected onto a hyperspace where the proteins in a family are separated from proteins outside the family by a hyperplane. By projecting a new sequence onto this hyperspace, the SVM system can determine whether or not the corresponding protein belongs to the family based on its location with respect to the hyperplane. In the current study, the LIBSVM 3.0 package was used to implement of SVM; it can be downloaded for free from the website (<http://www.csie.ntu.edu.tw/_cjlin/libsvm>). Four types of kernel functions, a linear function, polynomial function, sigmoid function and radial basis function (RBF), can be used for predictions in this software. Empirical studies show demonstrated that the RBF outperforms the remaining three types of kernel functions in nonlinear classification. Thus, the RBF kernel function was used in the current work. All the computations were performed using LIBSVM-3.0 standard package (Chang and Lin, 2001). The various user-defined parameters, e.g., kernel parameter γ and regularization parameter C were optimized on the training dataset. The predictor obtained via the aforementioned procedure is called **iDPF-PseRAAAC**, where “i” stands for “identify”, “DPF” for “defensin peptide family”, “Pse” for “pseudo”, “R” for “”reduced”, “AAA” for “amino acid alphabet”, and “C” for “composition”. ## Multi-class SVM Prediction of defensin family classes is a multi-classification problem. SVM is regarded as a typical binary classifier. The methods of applying SVM to solve multi-class classification problems have one-against-one(OAO), one- against-all(OAA) and directed acyclic graph SVM (DAGSVM). In the present study, we adopt the “One-against-one” approach to transfer it into a two-class problem. This method involves construction of individual binary SVM classifier corresponding to each pair of the classes. Hence, if there are K classes, OAO will construct a total of K(K-1)/2 classifiers. Each classifier plays a role in classifying of one class and another class. Classifier *i j*, named *f*_*ij*, is trained using all the patterns from class *i* as positive instances, all the patterns from class *j* as negative instances, and disregarding the rest. The classifiers then are combined using majority voting scheme. Predictions are made with each binary classifiers and label is assigned to a class with maximum number of votes. In case when tie arise, i.e. two classes have identical votes, label assignment to the class is made on the basis of smallest index. More details for one-against-one(OAO) of SVM classification can be found in can be found in. ## Performance Evaluation This method's performance was measured based on sensitivity (Sn), specificity (Sp), Matthew’s correlation coefficient (MCC) and overall accuracy (OA), which were defined as follows: $$\left\{ \begin{array}{l} {\mspace{13mu}\text{Sn}(i) = \frac{\text{TP}(i)}{\text{TP}(i) + \text{FN}(i)}} \\ {\mspace{13mu}\text{Sp}(i) = \frac{\text{TN}(i)}{\text{TN}(i) + \text{FP}(i)}} \\ {\quad\text{MCC}(i) = \frac{\text{TP}(i) \times \text{TN}(i) - \text{FP}(i) \times \text{FN}(i)}{\sqrt{\lbrack\text{TP}(i) + \text{FP}(i)\rbrack\lbrack\text{TP}(i) + \text{FN}(i)\rbrack\lbrack\text{TN}(i) + \text{FP}(i)\rbrack\lbrack\text{TN}(i) + \text{FN}(i)\rbrack}}} \\ {\quad\text{OA} = \frac{1}{N}{\sum\limits_{i = 1}^{M}{\text{TP}(i)}}} \\ \end{array} \right.$$ where TP(*i*), TN(*i*), FP(*i*), and FN(*i*) represent true positive, true negative, false positive and false negative of family *i*; *M* = 5 is the number of subsets while *N* the number of the total samples in $\mathbb{S}$. # Results and Discussion ## Cross Validation Three cross-validation methods, namely the sub-sampling (or K-fold cross- validation) test, independent dataset test and jackknife test, are often used to evaluate the quality of a predictor. Among the three methods, the jackknife test is the least arbitrary and most objective as demonstrated in and can always yield a unique result for a given benchmark dataset hence. The jackknife test has been widely recognized and increasingly adopted by investigators to examine the quality of various predictors\[–\]. Accordingly, the jackknife test was used to examine the performance of the model proposed in the current study. ## Defensin Family Prediction The jackknife results obtained using the **iDPF-PseRAAAC** and the benchmark dataset $\mathbb{S}$ based on different sizes(S) and N-peptide compositions(N) are depicted in. shows the prediction results for the overall accuracy of the defensin families based on N-peptide composition with S size alphabet (N, S). As the dimensions increases, N-peptides provide progressively more detailed sequential information. However, the predictive ability did not increased linearly with dimension increase; for example, when the tripeptides composition (3, 20), 8000 dimensions, was selected as the input parameter, the overall accuracy for predicting five defensins families was only 79.28%. The results reflect the notion that a larger dimension does not necessarily result in better performance, and the prediction ability is not always better when the feature dimensions increase. Excessively large dimensions typically lead to information redundancy or noise, which results in bad prediction accuracy. The heatmap shows the adjacent correlation of 13 reduced amino acids for five different defensin families. From the prediction performance based on different vector dimensions depicted in, we observed that the overall accuracy reached a maximum 85.59% based on 2-peptide composition of 13 reduced amino acids (N = 2, S = 13). shows the Jackknife results obtained using **iDPF-PseRAAAC** to identify defensin family with dipeptide (N = 2) composition based on different reduced amino acid alphabet approaches. As shown in, 2-peptide compositions with alphabet of 13 (N, S) outperformed the other reduced amino acid alphabet sizes. The largest defensin family, vertebrate defensins, yielded the best success rate at 99.36%. The 10-fold cross-validation has been performed to examine the comparability of our method. The prediction results are similar to the jackknife test (Total accuracy: 83.78% Vs 85.59%). For further comparison, the amino acid (i.e., N = 1) and tripeptide (i.e., N = 3) results were also calculated and were in, which shows that none of the results exhibit a higher success rate than N = 2. The data indicat that the reduced amino acid composition provides a greater weight of compositional bias to proteins with a signal at different sequence regions. Subsequently, the adjacent correlation for 13 reduced amino acids was analyzed and depicted using a heatmap plot. ## Vertebrate Defensin Subfamily Prediction Vertebrates include three distinct defensin subfamilies, Alpha-, Beta-, and Theta-defensins, which exhibit a broad spectrum of antimicrobial activities against bacteria, fungi, and viruses. Subsequently, the proposed method was used to predict the vertebrate defensins subfamily. The prediction results in show that the best overall accuracy was 98.39%, and the Mathew's correlation coefficients (MCC) for the Alpha-type, Beta-type and Theta-type are 0.97, 0.96 and 0.89, respectively. Such high accuracies demonstrate that the proposed method is an effective and powerful approach for predicting defensin subfamilies. ## Comparison with Previous Methods To further demonstrate the performance of the proposed method, it is necessary to be compared with other existing methods. However, directly comparing the results is not objective and strict examination due to the different benchmark datasets used. Therefore, we repeated the feature selection and prediction process on the previous dataset. The jackknife cross-validated accuracies are depicted in. Obviously, our proposed method yields the highest predictive success rate. According to, when the 169 reduced dipeptides are used, our method can achieve a maximum overall accuracy (OA) of 95.10% for the defensin family, and 98.39% for the vertebrate subfamily, which is higher than the maximum accuracy obtained using other methods. Although the success rate for others family and subfamily obtained using our method is not better, the accuracy of the other families and subfamilies are dramatically better than using other methods, and clearly indicate that the proposed method is more powerful than our previous method. Subsequently, it is instructive to compare the overall success rate from **iDPF- PseRAAAC** with the success rate for weighted random guess (WRG). The overall success rate based on the OA to identify the defensin proteins among their four subfamilies WRG is given by. $$\text{OA}(\text{WRG}) = \frac{{(N_{1})}^{2} + {(N_{2})}^{2} + {(N_{3})}^{2} + {(N_{4})}^{2} + {(N_{5})}^{2}}{N^{2}}$$ where *N* is the number of defensin proteins in the benchmark dataset $\mathbb{S}$, *N*₁ the number of defensin proteins in the subset $\mathbb{S}_{1}$, *N*₂ the number of defensin proteins in the subset $\mathbb{S}_{2}$, and so forth. Substituting these data into, we obtain the following. <img src="info:doi/10.1371/journal.pone.0145541.e021" id="pone.0145541.e021g" /> OA ( WRG ) = ( 60 ) 2 \+ ( 34 ) 2 \+ ( 42 ) 2 \+ ( 40 ) 2 \+ ( 157 ) 2 333 2 = 29.55 % In contrast, the best overall success rate for **iDPF-PseRAAAC** was 85.59%. Compared with the results in, the overall success rate for **iDPF-PseRAAAC** is approximately 56% higher than using a weighted random guess, which indicate that **iDPF-PseRAAAC** may be an easy and useful tool for timely identifying defensin proteins families. ## Web-Server Guide For the convenience of most experimental scientists, below, we provide a step- by-step guide on how to use the **iDPF-PseRAAAC** web-server to achieve the desired results. **Step 1.** Open the web server at <http://wlxy.imu.edu.cn/college/biostation/fuwu/iDPF-PseRAAAC/index.asp> and you will see the top page for **iDPF-PseRAAAC** on your computer screen, as shown in. Click on the 'Read Me' button to see a brief introduction about the predictor and the caveat for using it. **Step 2**. Either type or copy/paste the query defensin peptide sequence into the input box at the center of. The input sequence should be in the FASTA format. A sequence in the FASTA format consists of a single initial line beginning with a greater than symbol (“\>”) in the first column, followed by lines of sequence data. The words immediately following the “\>” symbol in the single initial line are optional and only used for the identification and description. The sequence ends if another line starting with a “\>” appears; this indicates the start of another sequence. Example sequences in FASTA format can be viewed by clicking on the 'Example' button right above the input box. **Step 3**. Click on the 'Submit' button to see the predicted result. For example, if you use the query defensin protein sequences in the 'Example' window as the input, after clicking the 'Submit' button, you will see the "Result page" shown on the screen of your computer. All these results are fully consistent with the experimental observations. It takes approximately a few seconds for the above computation before the predicted result appears on your computer screen; for query sequences and longer each sequences, more time is typically required. **Step 4**. Click on the 'Citation' button to find the relevant papers that document the detailed development and algorithm of **iDPF-PseRAAAC**. **Step 5**. Click on the 'Data' button to download the benchmark datasets used to train and test the **iDPF-PseRAAAC** predictor. # Conclusions Defensins also play important regulatory roles in the immune systems of animals and plants, acting as a bridge between innate and adaptive immunity in vertebrates. In this study, a promising method, **iDPF-PseRAAAC,** was developed to improve prediction performance for defensin proteins. The use of reduced amino acid alphabets not only provides an efficient and accurate means of protein vectorization for sequence-based protein classification systems but also remarkably improves computational efficiency. High predictive accuracies demonstrate that our proposed method is a potentially useful tool for classifying defensin family. # Supporting Information We wish to express our gratitude to the editor and three anonymous reviewers whose constructive comments were very helpful in strengthening the presentation of this paper. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YCZ GLF. Performed the experiments: YCZ YL. Analyzed the data: LY GLF. Contributed reagents/materials/analysis tools: ZYW GPL. Wrote the paper: YCZ YL GLF.

# Introduction The immunoglobulin G (IgG) class of antibodies plays an important role in the adaptive immune defense of the human host against pathogens. IgG consists of two identical heavy chains and two identical light chains, which in turn are composed of variable and constant domains. Papain treatment of the IgG molecule generates two separate monovalent Fab fragments recognizing antigens and an intact Fc fragment, a recognition site for host receptors and a site of interaction with a number of effector molecules, including the classical complement pathway starting with factor C1q. IgG is a glycoprotein containing a conserved complex carbohydrate structure attached to the asparagine 297 residue in the CH2 domain of each IgG Fc part. It consists of a biantennary core of *N*-acetylglucosamine and mannose with added terminal and branching carbohydrate residues such as *N*-acetylglucosamine, fucose, sialic acid, and galactose. The presence of this carbohydrate is crucial for proper antibody structure and for interactions with cellular immunoglobulin G Fcγ receptors (FcγRs) and the complement system. Altered glycosylation of IgG have been associated with autoimmune disorders like rheumatoid arthritis (RA), systemic lupus erythematosus (SLE) and Crohńs disease. Several forms of autoimmune vasculitis show a differentiating glycosylation with decreased Fc galactosylation and sialylation. Additionally, it has been suggested that sialylated IgG *in vivo* switch from its anti-inflammatory activity with subsequent reduced antibody effector activity, to a pro-inflammatory/toxic activity upon decreased Fc sialylation. IgGs, classified into four subclasses, IgG1, IgG2, IgG3 and IgG4, are described to interact with different types of FcγRs giving different activation profiles. FcγRs provide a linkage between the humoral and cellular immune responses. Phagocytic cells express members of three classes of IgG-Fc receptors, FcγRI, FcγRII and FcγRIII, characterized by structural and functional homology and by the specific recognition site on the CH2 region of IgG. Binding of pathogen-IgG complexes to FcγRs mediates an essential response from the host against pathogens by initiating a cascade of signals causing antibody-dependent- cellular-cytotoxicity (ADCC), complement-dependent-cellular-cytotoxicity (CDCC), endocytosis, phagocytosis, oxidative burst, the release of inflammatory mediators, etc.. Complexed IgG-FcγR can besides activation of the C1q component of complement also activate other ligands e.g. mannan binding lectin (MBL), the neonatal receptor FcRn, the mannose receptor (MR), etc.. FcγRs may be expressed constitutively on haematopoietic cells and may also be induced or up-regulated by cytokines and other agents. FcγRs are responsible for balancing activation (FcγRI, FcγRIIa and FcγRIIIa) and inhibitory signals (FcγRIIb) of the immune system with the ability of both activating and inhibiting the IgG mediated effector stimulation. *Streptococcus pyogenes* is one of the most common human pathogens causing pharyngitis, scarlatina and more severe infections like necrotizing fasciitis and sepsis. Like other bacteria it expresses several virulence factors and uses several immune evasion strategies to successfully invade its host. The recently discovered enzyme Endoglycosidase S (EndoS) is secreted by *Streptococcus pyogenes* and has a specific endoglycosidase activity on native IgG by hydrolyzing the conserved asparagine-linked glycans on the heavy chains of IgG. This 108 kDa-enzyme is encoded by the gene *ndoS* that is highly conserved and is present in virtually all examinated isolates. EndoS is the first known bacterial enzyme with a unique specificity for native IgG. This is in contrast to other related endoglycosidases as EndoF1-3 from *Elisabethkingia meningoseptica* (previously *Flavobacterium meningosepticum*), which show enhanced hydrolytic activities on the denaturated forms of basically any glycoproteins with the appropriate N-linked glycan, or EndoE from *Enterococcus faecalis* that in addition to activity on the glycan of native IgG also hydrolyzes high-mannose glycans on other denatureted glycoproteins. EndoS is N-terminally processed by the cysteine proteinase SpeB that could be of importance in regulating EndoS activity. Furthermore, the molecular requirements for EndoS glycosidase activity have recently been elucidated revealing the importance of glutamic acid 235 (Glu-235) and tryptophans. EndoS activity affects the functionality of opsonizing IgG by decreased binding to Fc-receptors on a monocyte-like cell line and impaired classical complement activation *in vitro*. In the present study we elucidated the effect(s) of EndoS on IgG subclasses and IgG-FcγR interactions. The results revealed that EndoS hydrolyses the heavy chain of all four human IgG subclasses (IgG1–4), both soluble and in a plasma environment. Additionally, we found that EndoS hydrolysis of the IgG glycan dramatically influences the binding of IgG to soluble, immobilized FcγRIIa and FcγRIIb as well as to FcγR-expressing cells. Moreover, IgG pre-bound to these cells dissociates due to treatment of cells with EndoS. Furthermore, an inactive form of EndoS generated by site-directed mutagenesis binds with high affinity to IgG1–4, while the active form only transiently interacts with its substrates. These results provide novel information about the mechanisms behind enzymatic modulation of the host immune defense by bacteria, provide novel information about the molecular interactions between an IgG glycan-hydrolyzing enzyme and IgG, and emphasize the importance of IgG glycosylation for correct antibody effector functions. # Results ## EndoS has glycosidase activity on all four IgG subclasses It has previously been shown that EndoS hydrolyzes the chitobiose core of the conserved N-linked glycan on the γ-chain of human polyclonal IgG. It was therefore of interest to elucidate whether EndoS has activity on all four subclasses of human IgG (IgG1–4). Purified recombinant EndoS was incubated with purified human IgG1–4. SDS-PAGE analysis revealed that EndoS-treated IgG of all subclasses migrated at an apparent molecular weight of approximately 3 kDa lower than untreated IgG, which is consistent with hydrolysis of the chitobiose core of the IgG glycan. To confirm glycan hydrolysis, samples were also analyzed by lectin blot using a *Lens culinaris* agglutinin (LCA) lectin recognizing α-linked mannose residues. Lectin blot analysis of the samples revealed that all IgG subclasses lose the reactivity with LCA after incubation with EndoS consistent with complete or nearly complete hydrolysis of the glycan. Additionally, the glycosidase activity of EndoS on IgG1–4 in a plasma environment was investigated. In this experiment human plasma was incubated with purified EndoS or buffer, followed by affinity purification of the IgG fraction. These fractions were subsequently subjected to a LCA ELISA using immobilized monoclonal antibodies against IgG1–4 to capture IgG. This revealed that all four IgG subclasses reacted with lectin when plasma was incubated with buffer, indicating presence of the glycan (data not shown). In contrast, when plasma was treated with EndoS, a dramatically reduced IgG1–4 reactivity with LCA lectin was observed. IgG1 was hydrolyzed to 87±11%, IgG2 was hydrolyzed to 81±13%, IgG3 was hydrolyzed to 74±23%, and IgG4 was hydrolyzed to 72±3%. Taken together, these results clearly show that EndoS has the ability to hydrolyze human IgG of all subclasses, in purified form as well as in whole plasma. ## The inactive form of EndoS binds IgG We have previously partly elucidated the molecular requirements for EndoS hydrolysis of IgG. Site directed mutagenesis of glutamic acid 235 to glutamine (EndoS(E235Q)) at the proposed orifice of the catalytic tunnel abolishes enzymatic activity. In addition, chemical blocking of tryptophanes revealed that these amino acid residues are important for activity. To further investigate the physical interaction between enzyme and substrate, the binding of EndoS and EndoS(E235Q) to immobilized polyclonal IgG and IgG1–4 subclasses was studied using slot-binding experiments with immobilized IgG probed with EndoS and EndoS(E235Q). Purified, soluble IgG subclasses 1–4, each immobilized onto a nitrocellulose membrane, were probed with EndoS and EndoS(E235Q) followed by incubation with antibodies against EndoS. This experiment revealed a strong binding of EndoS(E235Q) to polyclonal IgG, IgG1 and IgG2, and a weaker association to IgG3 and IgG4, while only very weak interactions between active EndoS and all subclasses could be seen. To calculate the affinity constants between EndoS and immobilized IgG1–4, surface plasmon resonance technology was used. Similarly to slot-binding results, this showed that EndoS(E235Q) binds all IgG subclasses with high affinity, while there is no detectable binding of EndoS to IgG.. The kinetic parameters of EndoS (E235Q) binding to immobilized IgG subclasses were of similar character and the strongest interaction was demonstrated between IgG1 and EndoS(E235Q) with a binding affinity constant (*K_D*) of 0.42 µM. No binding of either EndoS or EndoS(E235Q) to IgG1–4 subclasses, which were hydrolysed by EndoS before immobilization, was detected. These findings indicate that the intact IgG glycan is necessary for the interaction between EndoS and IgG. Furthermore, the experiments comparing the interactions between EndoS, EndoS(E235Q) and IgG indicates that EndoS binds IgG with a high affinity, but the active enzyme is instantly released after glycan hydrolysis in a “touch and go” manner. ## EndoS influences the binding of IgG1–4 to FcγRs Since the nature of the interactions between FcγRs and the Fc domain of IgG is highly dependent on the IgG glycosylation state, we explored the effects of EndoS activity on IgG interactions with FcγRs. Thus, in an ELISA experiment the soluble FcγRIIa, FcγRIIb and FcγRIIIa were immobilized and probed with purified IgG1–4 subclasses. In line with other observations it was here seen that FcγRIIa and FcγRIIb binds IgG1. This binding was nearly abolished after treatment of IgG1 with EndoS. The binding of the other IgG subclasses to these receptors was weak and was even more reduced after treatment with EndoS. In general, ELISA studies revealed the IgG subclass binding affinity pattern IgG1\>IgG3\>IgG4\>IgG2 for FcγRIIa and IgG1\>IgG4\>IgG3\>IgG2 for FcγRIIb. Furthermore, we observed that the EndoS hydrolysed IgG2 had a different outcome regarding the binding to FcγRIIa/FcγRIIb with more extensive binding ability, compared to the untreated IgG2. FcγRIIIa was negative in binding of all IgG subclasses (data not shown). The interaction between IgG1–4, with or without EndoS treatment, with FcγRs was further analyzed by surface plasmon resonance. Each IgG subclass was tested for binding to a surface with an immobilized FcγRIIa, FcγRIIb or FcγRIIIa. Consistent with the ELISA data, the results showed that IgG1 had the strongest affinity for both FcγRIIa and FcγRIIb with similar binding affinity constants, 97 nM and 170 nM respectively. In agreement with our previous findings, no binding of IgG1 to these receptors was detectable when EndoS treated IgG1 was used. There was no detectable interaction between FcγRIIa/FcγRIIb and IgG2 or IgG3, or between IgG4 and FcγRIIa. No binding of soluble IgG1–4 subclasses to immobilized FcRIIIa could be detected. These results indicate that EndoS hydrolysis dramatically decreases IgG's affinity for FcγRs. ## EndoS decreases IgG binding to blood cells Based on results from ELISA and surface plasmon resonance, we continued to analyze the effect(s) of EndoS glycosidase activity on the interaction between FcγRs and IgG. For this purpose we used an erythroleukemic cell line (K562) exclusively expressing FcγRIIa. Since soluble FcγRI was not available to us, we also investigated human monocytes that predominantly bind IgG through this receptor. Thus, IgG was purified from plasma treated with EndoS or PBS, labeled with ¹²⁵I and incubated with the K562 cells. The radioactivity of the cell pellets was measured. This revealed significantly decreased binding of radioactive IgG, originally purified from plasma treated with EndoS, to K562 cells. In a control experiment, the specific IgG binding to these cells was calculated by addition of cold human IgG, which inhibited the binding of radioactive IgG to 93% (data not shown). A strong binding of IgG to K562 cells after incubation of cells with human plasma was confirmed by Western blot and the reactivity of cell lysates with antibodies against human IgG. In contrast, there was a clear decrease in binding of IgG to K562 cells incubated with plasma pre-treated with EndoS. Likewise, the binding of ¹²⁵I-IgG to monocytes as analyzed by SDS-PAGE was totally inhibited when IgG was treated with EndoS. To further analyze EndoS' influence on the interaction between FcγRs on monocytes and IgG, flow cytometry analysis of whole blood was performed. Human blood was pre-incubated with EndoS before addition of the leukocyte activator fMLP. Monocytes were gated based on forward and side scatter and the reactivity of monocytes with monoclonal anti-human IgG was evaluated. The result revealed that 87% of monocytes were positive for IgG binding, while only 43% of monocytes in blood incubated with EndoS were positive. These results indicate that EndoS-hydrolyzed IgG is significantly reduced in its binding capacity to human cells expressing different sets of FcγRs. ## IgG dissociates from FcγRIIa upon treatment with EndoS Our experiments this far have revealed that EndoS hydrolysis of IgG inhibits binding to FcγRs on cells and surfaces, but it remained unclear if EndoS has activity on IgG already bound to FcγRs and if such activity could release the IgG bound to FcγRs. Therefore, we investigated the effects of EndoS on IgG bound to K562 cells that had been exposed to human plasma and subsequently treated with EndoS. The cell lysates were analyzed by SDS-PAGE and Western blot using an antibody against human IgG. There was a significant binding of IgG to K562 cells as judged by the results presented in. Interestingly, no IgG-band was visible on a blot when cells were treated with EndoS, suggesting a total IgG dissociation from the cells. A control experiment, using EndoS(E235Q), revealed an IgG signal on the surface of K562 cells comparable to the untreated cells. These results strongly suggest that IgG dissociates from the cell surface due to N-glycan hydrolysis of IgG by EndoS. Similarly, the effect of EndoS on IgG bound to monocytes was analyzed. This showed that most of the monocyte-bound IgG dissociated from cells due to the treatment with EndoS as compared to untreated cells. As expected, monocytes treated with EndoS, in contrast to control cells, showed no reaction with the LCA lectin, indicating that the minute amounts of IgG remaining on the cells as detected in the IgG blot had most likely been hydrolyzed by EndoS. The results demonstrated above were further confirmed by surface plasmon resonance experiments. Soluble IgG1 and FcγRII receptor were chosen because of our earlier observation that IgG1 is the strongest binder of FcγRII. After binding of IgG1 to pre-immobilized FcγRIIa and reaching a steady- state dissociation phase, the IgG1 injection was aborted and replaced by EndoS injection or running buffer. This revealed that EndoS injection causes the dissociation of IgG1 from immobilized FcγRIIa receptor while the IgG1 dissociation from FcγRIIa was unaffected when adding running buffer. Taken together, these results clearly demonstrate that EndoS by IgG glycan hydrolysis can release IgG bound to FcγRs on cells and surfaces. # Discussion In the present study we attempted to elucidate the physical interaction between EndoS and IgG and the physiological relevance of EndoS IgG *N*-glycan hydrolyzing activity for IgG-FcγR interactions. We present for the first time that EndoS specifically acts as an endoglycosidase on all human IgG subclasses, both in purified form and in a plasma environment. As expected, there is a physical interaction between the enzyme and all IgG subclasses, that we successfully demonstrated using an enzymatically inactive, mutated form of EndoS. In this study we could not separately investigate the EndoS effects on the isolated binding of IgG to FcγRI. However, we observed that EndoS-hydrolyzed IgG did not bind to monocytes, and that there was a nearly complete dissociation of IgG from monocytes upon hydrolysis by EndoS. Since monocytes express FcγRI and this receptor has the highest affinity for IgG, we conclude that EndoS influences IgG binding even to FcγRI because the effects of EndoS observed must be predominantly due to involvement of FcγRI. EndoS seems to have an effect on both isotypes of FcγRII receptors, thus influencing both activating and inhibiting IgG mediated effector stimulation. Interestingly, IgG2 treated with EndoS, in opposite to what was observed for the other subclasses of IgG, showed increased binding to FcγRIIb, and slightly also to FcγRIIa, immobilized to microtiter plate. One possible explanation for this could be aggregation of IgG2 upon hydrolysis by EndoS leading to increased binding to FcγRs. However, no binding of IgG2 to FcγRs was detected using surface plasmon resonance which is in agreement with earlier publications. This could be explained by the constant flow of IgG2 over the immobilized receptors in the case of surface plasmon resonance, in contrast to ELISA where IgG2 is allowed to aggregate and interact with the receptor. The activity of EndoS on IgG has obvious benefits for *S. pyogenes* expressing this enzyme with potential modulation and/or evasion of an IgG-mediated response against the bacteria. We have previously demonstrated that EndoS treatment of human opsonizing IgG antibodies directed towards the cell-wall anchored M protein significantly enhances the bacterial survival in blood. Therefore, in the context of an intact infecting *S. pyogenes,* EndoS is a potentially harmful molecule to the human host that contributes to the bacterial virulence. In contrast to this, the purified form of EndoS has substantial potential as a biotechnological tool and/or a therapeutical agent that could be beneficial for future experimental science and possibly also health care. Our results reveal that EndoS possesses a capacity to inhibit the IgG binding to FcγRs and detach IgG bound to FcγRs on cell surfaces. We have recently been able to show that pre-treatment of arthritogenic antibodies abrogates development of arthritis in a mouse model of collagen-induced arthritis. This suggests that EndoS may have potential for being further developed as a therapeutical agent in other antibody-mediated autoimmune disorders. We suggest two principally different biotechnological uses of EndoS, one based on the IgG-glycan hydrolyzing activity of the wild-type enzyme, and the other based on the high affinity IgG-binding of EndoS(E235Q). The active enzyme could be used for *in vitro* treatment of whole blood or purified blood cells in order to remove IgG already bound to various FcγRs on these cells. This could facilitate the analysis of effects of specific IgG preparations added to the cells, regarding receptor binding and cellular activation, without the interference of pre-bound IgG. The inactive form of EndoS (EndoS(E235Q)) has a great potential as a specific IgG purification and detection tool. In this study we have demonstrated that EndoS(E235Q) interacts equally well with all subclasses of IgG. This is comparable to what can be seen for protein G, one of the major molecules currently used for IgG preparation and detection, but advantageous compared to protein A that does not bind IgG3. Protein A also binds IgM and IgA to a certain extent. We have previously shown that there is no interaction between EndoS and IgM or IgA. Furthermore, we could show here that EndoS(E235Q) does not interact with IgG lacking its heavy chain glycans. This is in contrast to both protein G and protein A that bind IgG irrespective of its glycosylation state. This could be especially important when only intact IgG with a certain functional effector region is required. When using currently available reagents like protein G, a second purification step using for instance a lectin column is required to obtain only the glycosylated fraction of IgG. This property of EndoS(E235Q) could be used in combination with for instance protein G to assess the glycosylation state and the functional quality of an IgG preparation. In conclusion, EndoS is a bacterial immunomodulatory protein with a great potential. Our results provide novel information about bacterial pathogenesis, i.e. how the pathogens evade the immune system of the host organism by affecting the functions of IgG/FcγRs. Moreover, EndoS could be used as an important biochemical tool for specific IgG *N*-glycan hydrolysis and IgG purification/detection, or perhaps as a potential immunosuppressing agent that could be used to interfere with antibody-mediated pathological processes. # Materials and Methods ## Proteins and reagents Blood was drawn from healthy individuals and collected in heparin-containing tubes. Full-length EndoS with glutathione-S-transferase (GST) as a fusion was recombinantly expressed and purified from *Escherichia coli* harboring the plasmid pGEXndoS. When appropriate, the GST-tag was removed using Factor Xa as previously described. The mutation of glutamic acid 235 of EndoS into glutamine was performed using QuickChange II Site-Directed Mutagenesis Kit according to manufacturer's instructions (Stratagene, La Jolla, CA) with following verification of the mutation by sequencing. Soluble purified Fc-receptors were generated by co-transfection of CHO-K1(CHO) cells with pNT-neo-FcγRII or pNT- neo-FcγRIII plasmids with subsequent selection in 1 mg/ml genetecin. IgG- subclasses were produced by transient transfection in 293T cells. RPMI 1640 medium and Hank´s balanced salt solution (HBSS) were from GIBCO, Paisley, U.K. All other reagents were purchased from Sigma-Aldrich unless indicated otherwise. ## EndoS-treatment of antibodies Purified human IgG1–4 was hydrolyzed with GST-EndoS purified as previously described. Enzyme/substrate molar ratio was 1∶20 in PBS and samples were incubated for 2 h at 37°C. GST-EndoS was removed from the samples by passing three times over a glutathione-Sepharose column (Amersham Biosciences, Uppsala, Sweden). 1 µg of treated and untreated IgG1–4 was separated on 10% SDS-PAGE followed by staining with Coomassie Blue or *Lens culinaris* agglutinin (LCA)-lectin (Vector Laboratories, Burlingame, CA, USA) blot analysis (se below). ## Treatment of human plasma with EndoS and purification of IgG A volume of 2 ml human plasma was incubated with 20 µg EndoS or a PBS equivalent for 1.5 hours at 37°C. The IgG fraction was purified using Protein G Sepharose (GE Healthcare Bio-sciences AB, Uppsala, Sweden). Briefly, 200 µl Protein G Sepharose suspended 1∶1 in PBS (phosphate-buffered saline; 10 mM phosphate buffer, pH 7.4, 120 mM NaCl, 3 mM KCl) was added to plasma samples and incubated at 4°C for 2 hours or over night. After centrifugation for five minutes at 8000×*g*, the supernatant was discarded and the pellet washed three times with PBS. IgG was eluated with 0.1 M glycine pH 2.0 and neutralized with 1 M Tris-HCl pH 8.0. The IgG concentration was determined to 8 mg/ml using the Advanced Protein Assay (Cytoskeleton, Denver, CO, USA). ## Cell preparations The K562 cell line was cultured in RPMI 1640 medium supplemented with Glutamax-I, 100 µg/ml antibiotics (penicillin and streptomycin) and 10% fetal calf serum at 37°C in an atmosphere containing 5% CO₂ and 95% humidity. Nunclon flasks for cell culture were used (Nunc A/S, Roskilde, Denmark). Cells were cultured in a serum free medium for 20 hours before being used in experiments. Monocytes were isolated from human whole blood using the Polymorphprep kit (AXIS-SHIELD, Oslo, Norway) or Ficoll-Paque Plus (Amersham Biosciences, Uppsala, Sweden) according to instructions provided by the manufacturers. After isolation, the cells were counted and resuspended in PBS or RPMI-medium. ## Enzyme linked immunosorbent assay (ELISA) For glycan detection, microtiter plates (NUNC, Roskilde, Denmark) were coated with 100 µL monoclonal mouse anti-human IgG1, IgG2, IgG3 or IgG4 (SIGMA®, Saint Louis, MO, USA) diluted to final concentrations of 1.5–0.5 µg/ml in a coating buffer containing 16 mM Na₂CO₃ and 35 mM NaHCO_3, pH 9.6 and kept at 4°C overnight. The plates were washed three times with lectin buffer containing 10 mM HEPES, pH 7.5, 0.15 M NaCl, 0.01 mM MnCl₂, 0.1 mM CaCl₂ and 0.1% v/v Tween 20 and blocked in the same buffer for one hour at room temperature. In the next step purified IgG fraction (dilution 1∶100) was added and the incubation proceeded for another 2 hours at 37°C. After three washes with lectin buffer, 1 µg/ml biotinylated LCA-lectin was added and incubation continued for 1 hour at 37°C. Following three more washes, 0.1 µg/ml peroxidase-labeled streptavidin (Vector Laboratories) was added and the plate was incubated for 1 hour at 37°C. The color reaction was developed with 0.1 M citric acid monohydrate, 0.1 M Na₂HPO₄×2H₂O buffer pH 4.5 containing 0.012% v/v H₂O₂ and 1.7 mM 2,2′-azino- bis(3-ethylbenzthiazoline-6-sulphonic acid) (ABTS). The absorbance was read on a model 550 micro plate reader (BIO-RAD, Hercules, CA, USA) at 415 nm. For detection of binding of human IgG subclasses to FcγRs, the plate was coated with soluble FcγRIIa or FcγRIIb or FcγRIIIa at a concentration 5 µg/ml for 20 hours at 4°C. Next day, the plate was blocked with PBS supplemented with 0.05% v/v Tween 20 (PBST) and 2% w/v bovine serum albumin for 2 hours at room temperature. After this step, the purified IgG subclasses, 0.1 µg of each, were added. The plate was washed three times with PBST after the coating step and between each of the following incubation steps. A peroxidase-conjugated protein G (dilution 1∶5000) (BIO-RAD, Hercules, CA, USA) was used for detection. The color reaction was performed as above. All experiments were made in triplicates. ## Radioactive labeling Proteins were labelled with 0.2 mCi Na ¹²⁵I (PerkinElmer, Upplands- Väsby, Sweden) using the IODE-BEADS Iodination reagent kit (PIERCE, Rockford, IL, USA) according to the manufacturer's instructions. The unbound radioactivity was removed by desalting the proteins on PD-10 Sepharose (Pharmacia, Sweden). The activity of the labeled proteins was estimated to 4 µCi/µg protein. ## Detection of IgG binding to cells IgG was purified from human plasma treated with EndoS or PBS as described above and thereafter labeled with ¹²⁵iodine (¹²⁵I). For detection of ¹²⁵I-IgG binding to K562 cells, 2×10⁶ cells were incubated with 0.5×10⁶ cpm of ¹²⁵I-IgG or ¹²⁵I-deglycosylated IgG for 30 minutes at room temperature. After five washes with PBS and centrifugations at 1000×g for three minutes, the radioactivity of the cell pellets was detected using Wallac Wizard™ 1470 Automatic Gamma Counter (PerkinElmer, Waltham, MA, USA). To evaluate the specificity of the binding of radioactive IgG to K562 cells a control experiment was performed. The cells were incubated with 20 µg human IgG in addition to radioactive IgG during the similar incubation conditions as mentioned above. In another experiment, 1×10⁶ monocytes were incubated with 0.5×10⁶ cpm of ¹²⁵I-IgG or ¹²⁵I-EndoS treated IgG for 30 minutes at room temperature. After five repeated washes of cells with PBS and final pelleting of cells by centrifugation at 1000×g for five minutes, the cells were resuspended in lysis buffer containing 20 mM Tris-HCl pH 7.4, 0.150 M NaCl, 1% v/v Triton-100 and 0.25% v/v NP40 for ten minutes at 4°C. Next, the samples were centrifuged for ten minutes at 14000×g and supernatants applied on a polyacrylamide gel. After separation, the gel was dried and samples analyzed by phosphoimaging in a Fujix BAS 2000 Bioimaging analyzer (Fujifilm Sverige AB, Stockholm, Sweden). In an experiment where the binding of IgG to cells was analyzed by Western blot, 0.5–1×10⁶ cells were incubated with plasma treated with either EndoS or buffer (as described above), at 37°C for 1 hour. Afterwards, the cells were washed three times with PBS or RPMI medium, resuspended in 100 µL lysis buffer and the bound IgG in cell lysates analyzed by Western blot. ## Incubation of cells with EndoS K562 cells or monocytes, 2×10⁶ and 8×10⁶ respectively, were incubated with two ml of human plasma for 30 minutes at 37°C. The cells were washed five times with PBS and centrifuged at 1000×g for ten minutes after every wash. EndoS, 40 µg in PBS or PBS alone was added to cells and incubation followed for one hour at 37°C. Cells were washed three times with PBS and resuspended in 100 µL lysis buffer. Samples were centrifuged for five minutes at 14000×g, pellets discarded and supernatants analyzed for IgG and glycan contents using SDS-PAGE and Western blot. ## Slot-blotting analysis IgG1–4, 0.3, 015 and 0.075 µg of each in PBS were applied to PVDF membranes using a slot-blot apparatus from Schleicher and Schuell, Inc., Kene, NH 03431, USA. The membranes were incubated with PBST and 5% skim milk for 1 hour, washed with PBST and incubated with EndoS or EndoS (E235Q), 0.05 mg/ml in PBST and 5% skim milk for 1 hour. After washing, the membranes were incubated with rabbit EndoS-antiserum and subsequently with peroxidase conjugated goat anti-rabbit antibodies. The color development was made using ABTS as peroxidase substrate. All incubation steps were performed at room temperature. ## Surface plasmon resonance interaction analysis Receptors, IgGs and deglycosylated IgGs were diluted with 10 mM sodium acetate pH 4 and immobilized via amine coupling to different flow cells of CM5 sensorchips (BIAcore, Uppsala, Sweden). Immobilization levels were optimized to around 8000–10000 response units. After determining EndoS(E235Q) as a non-binder to all deglycosylated IgG variants, these flow cells were considered as controls for bulk refraction index changes for EndoS(E235Q) binding to IgG1 throughout IgG4, respectively. In experiments determining IgG1-IgG4 affinity for the receptors FcγRIIa, FcγRIIb and FcγRIIIa, a flow cell subjected to the immobilization protocol but without addition of protein was used as control. For affinity measurements, the binding and dissociation phases were monitored in a BIAcore 2000 instrument. In control experiments for possible mass transfer limitations, the IgGs were injected over the receptors and the EndoS variants over the IgG sub-classes at different flow rates. No differences in initial binding were observed at 5 µl/min or above indicating no limitations to any combinations. Interactants were injected in different concentrations (typically 10–1.25 µg/ml) at 35 µl/min and 25°C over the different coated surfaces (flow cells) (in running buffer: 10 mM HEPES, pH 7.5, 150 mM NaCl, 0.005% surfactant P20, and 3.4 mM EDTA). Between experiments, the surfaces were strictly regenerated with pulses of running buffer containing 2 M NaCl followed by an extensive wash procedure after reaching baseline. For EndoS digestion of IgG bound to pre-immobilized FcγRIIa, an IgG1 concentration (10 µg/ml) was chosen to give a suitable steady-state dissociation phase at a time point were the IgG1 injection was aborted and replaced by running buffer. This experiment was considered as a control and as such compared to an EndoS injection at the same time point after IgG1 binding to FcγRIIa. After X and Y normalization of data, the blank curves from control flow cells of each injected concentration were subtracted. Where applicable, the association (*k_a*) and dissociation (*k_d*) rate constants were determined simultaneously using the equation for 1∶1 Langmuir binding in the BIA Evaluation 4.1 software (BIAcore). The binding curves were fitted locally and the equilibrium dissociation constants (*K_D*) were calculated from mean values of the obtained rate constants. ## Flow cytometry analysis of whole blood A volume of 15 ml blood was incubated with 0.4 mg EndoS or PBS for 35 minutes at 37°C. An activator of leukocytes, formyl-methionyl-leucyl-phenylalanine (fMLP), (dilution 1∶10000) (SIGMA, Saint Louise, MO, USA) was then added and the incubation continued for 10 minutes at 37°C. Next, blood samples were centrifuged 1000×*g*, five minutes. Plasma and buffy coat were transferred to another tube and centrifuged for five minutes at 1000×*g*. The cells were then washed three times with HBSS containing 30% v/v RPMI and finally resuspended in 100 µl of the same medium. Monoclonal mouse anti-human IgG was prepared by mixing equal amounts of mouse anti-human IgG1 (51 mg/ml), IgG2 (22 mg/ml), IgG3 (16 mg/ml) and IgG4 (24 mg/ml). Five µl of this mixture was added and samples incubated for ten minutes at room temperature. In the next step 5 µl of FITC- conjugated goat anti-mouse IgG (DakoCytomation, Glostrup, Denmark) was added before erythrocytes were lysed using the DakoCytomation Uti-Lyse erythrocyte kit (Carpinteria, CA). Signals were analyzed on a FACSCalibur flow cytometer (Becton Dickinson, Franklin Lakes, NJ, USA). Monocytes were identified by forward scatter and side scatter characteristics (FSC/SSC). ## SDS-PAGE and Western blot analysis Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) was performed using Mini Protean II cell equipment from BIO-RAD (Hercules, CA, USA) or equipment from LKB (Bromma, Sweden) using the buffer system described by Laemmli. Samples were mixed 1∶1 (v/v) with sample buffer supplemented with 5% mercaptoethanol, and boiled for five minutes before loading onto the 10% polyacrylamide gel. PageRuler™ Protein Ladder Plus (Fermentas, Burlington, Canada) was used as high-molecular-mass standards. The polyacrylamide gels were stained with Coomassie Brilliant Blue R-250 and in some cases dried. For immunoblotting, the gels were transferred to polyvinylidenefluoride (PVDF) membranes (Immobilon P, Millipore, Bedford, MA) as described by Matsudaira (18). After blotting, membranes were blocked in PBS supplemented with 0.05% v/v Tween 20 (PBST) and 5% w/v skim milk (DIFCO, Detroit, MI) for 20 minutes at room temperature. For detection of IgG, the blots were subsequently washed in PBST and then incubated with rabbit anti-human IgG (diluted 1∶3000) (BIO-RAD, Hercules, CA) for one hour at 37°C. After a washing step, membranes were incubated with horseradish peroxidase-conjugated goat anti-rabbit IgG (BIO-RAD) (dilution 1∶1000). For lectin blot analysis, membranes were blocked for 20 minutes in lectin buffer (10 mM HEPES, pH 7.5, 0.15 M NaCl, 0.01 mM MnCl₂, 0.1 mM CaCl₂ and 0.1% v/v Tween 20) at room temperature and incubated with biotinylated LCA lectin (diluted 1∶5000). After repeated washes in lectin buffer the membranes were incubated with peroxidase- labeled streptavidin (Vector Laboratories) (diluted 1∶10000). All membranes were developed using SuperSignal West Pico (PIERCE, Rockford, IL) according to the manufacturer's instructions before analyzing by the Chemidoc XRS imaging system and Quantity One image analysis software (BIO-RAD). The authors are indebted to Dr. Jeffrey V. Ravetch's laboratory for preparation of Fc-receptors and IgG-subclasses. [^1]: Conceived and designed the experiments: MA MC AO FN. Performed the experiments: MA MC AO. Analyzed the data: MA MC AO. Contributed reagents/materials/analysis tools: FN. Wrote the paper: MA. [^2]: A patent application on the in vitro use of EndoS has been filed with MA, AO, FN, and MC listed as inventors. The application is pending.

# Introduction Periodontal disease is an infectious disease of the supporting structures of teeth that affects over 47% of American adults. In the elderly, 65 and older, the prevalence rate increases to over 70%. The cost that is associated with the treatment of PD represents a significant fraction of all of the expenses related to dental care, which totals on average, \$113 billion a year in the US. Periodontitis is characterized by an acute inflammatory process, osteoclast activity, as well as connective tissue destruction, which is then characterized by a pro-inflammatory profile of cytokine release, known as Th1. Among the cytokines with a Th1 profile, we can cite IL-1β, IL-6 and TNF-α. Periodontal disease is often associated with other chronic systemic conditions, such as cardiovascular diseases, diabetes, and asthma.. Asthma is defined as a chronic inflammation of the airways, with recurrent and reversible episodes of dyspnea, chest stiffness, coughing and wheezing. Its prevalence can vary from 1% to 18%, depending upon the studied population. According to the World Health Organization (WHO), it estimates that 235 million people worldwide suffer from asthma. The socio-economic implications are considerable, when one considers work absenteeism, hospitalization costs, medicines, a decrement in the quality of life, and premature death. The respiratory system becomes hyperresponsive and the major consequence is a reversible mechanical obstruction of the airways. Although the primary cause may vary, it can be triggered by several factors, classified as predisposing, causal and contributory. After the sensitization phase, the asthmatic patient presents with eosinophil infiltrate, activated mast cells on the airways surface, together with activated T lymphocytes, with a profile of cytokine release, known as Th2. Among the cytokines with a Th2 profile, we can highlight IL-4, IL-5 and IFN-γ, due their effects during the allergic responses. In addition, the Th2 profile has also been described as being involved in the progression of periodontitis. Some studies linking periodontitis and asthma have been proposed. The lack of standardization for a periodontitis diagnosis and the inclusion of patients with different ages make it difficult to compare studies. As for asthma, a diagnosis is often made by a self-reporting of the disease. Further studies are needed in order to elucidate the link between these two pathologies. In a recent study with 5.976 patients, a positive association was found between periodontitis and asthma, while there is an inverse association when the patients have been taking antiasthmatic medication. The causal relationship between them is still unclear. Allergies have been negatively associated with clinical attachment loss hypothesizing that a periodontopathogenic colonization of an oral cavity could have a protective effect on an allergic disease. Other authors have found an inverse association between a clinical attachment loss and asthma. Corroborating with human studies, some authors evaluated the immunoregulatory mechanism of asthma during a periodontitis. This study determined if a subcutaneous infection with *Porphyromonas gingivalis* exerted a regulatory effect on the allergic airway inflammation. The authors showed a reduction of lung inflammatory cells, as well as cytokines, after a *P*. *gingivalis* infection, prior to an allergen sensitization with ovalbumin (OVA). The standard PD treatment aims to reduce the biofilm microorganisms through scaling and root planning, followed by an oral hygiene control by the patient. However, it is not able to eliminate the subgingival pathogens and the calculus. There may be a bacterial re-colonization at inaccessible periodontal sites. *Aggregatibacter actinomycetemcomitans* and *Porphyromonas gingivalis* have been found in infected pulmonary fluids. Nowadays, antibiotics are indicated for a restricted group of patients, as an adjuvant to a periodontal treatment, due to the risk of a bacterial resistance development. Antimicrobial photodynamic therapy (PDT) is an adjuvant to a periodontal treatment, in order to reduce the amount of microorganisms in the localized infections of the subgingival sites. The principal advantage of this adjuvant is that the light and the photosensitizer reach places where the conventional treatment with curettes only has partial access. The mechanism is based upon the activation of a photosensitizer drug, by light at a suitable wavelength, generating oxidative species, such as hydroxyl radicals, superoxide and singlet oxygen. These species act in the bacteria organelles, damaging structures and loosing essential functions for survival. However, there is an absence of literature reports regarding a bacterial resistance to PDT. Cationic photosensitizers bind to bacteria due to electrostatic interactions, since there is a negative potential in the cell’s surface. Regarding this concept, phenothiazinium dyes have been widely studied as photosensitizers, with methylene blue and toluidine blue being the main ones. Methylene blue presents a broad spectrum of activity against bacteria, being effective in the inactivation of *P*. *gingivalis*, *P*. *intermedia*, and *A*. *actinomycetemcomitans* bacteria, with a key role in a periodontitis. Despite the advantages of PDT, there is a lack of well-designed clinical studies for a proper evaluation of this therapy. There are no reports of a bacterial resistance or side effects, with a preservation of the oral microbiota and a low toxicity, unlike a treatment with antibiotics or a mouthwash with chlorhexidine. Notwithstanding the advantages of PDT, well- designed studies on this subject are needed. There is an important gap in the literature, evaluating whether a periodontitis is capable of influencing the development of a pulmonary disease. In addition, if the suppression of the infectious agent of the periodontium is able to influence the inflammatory parameters of asthma. Therefore, the hypothesis of this study has been to evaluate whether periodontitis modulates lung inflammation in an experimental model of asthma as well as the photodynamic therapy (PDT) is associated with a reduction of lung inflammation # Material and methods This study was approved by the Animal Ethics Committee of the University Nove de Julho (UNINOVE), São Paulo, Brazil, under \#020/2015. The mice were maintained in a constant temperature of 22°C to 25°C, with a 12-h light/dark photoperiod, under artificially controlled ventilation, with a relative humidity ranging from 50% to 60%. Rations (NUTRILAB CR-1^®) and water were provided *ad libitum*. Seventy-two BALB/c male mice (\~2 months) were randomly divided into 8 groups (n = 9): 1) Basal—without any induction of a disease, 2) P—Induction of Periodontitis, 3) P+TP—Induction of Periodontitis + Standard Periodontal Treatment, 4) P+PT+PDT—Induction of Periodontitis + Standard Periodontal Treatment + Photodynamic Therapy, 5) Asthma—Induction of Asthma, 6) Asthma +P—Induction of Asthma + Induction of Periodontitis, 7) Asthma+P+TP—Induction of Asthma + Induction of Periodontitis + Standard Periodontal Treatment, 8) Asthma+P+TP+PDT—Asthma Induction + Induction of Periodontitis + Standard Periodontal Treatment + Photodynamic Therapy. All of the manipulations of the animals (periodontitis induction, periodontal treatment, PDT) were performed by only one operator (Candeo LC). Seventy-two animals were identified. The randomization of the animals was performed (Microsoft Excel, Version 2013) by separating them into 8 blocks (groups) of 9 animals. For the induction of asthma, the animals were injected subcutaneously with 4 μg ovalbumin (OVA) (SIGMA^™), together with aluminum hydroxide solution, on the first day of the experiment (sensitization), and then 14 days thereafter (booster). From the 14^th day, the animals were submitted to a nebulization (challenge) with 10 μg OVA, 3 times a week, for 2 weeks. Periodontitis was induced by using a modification of a previously described method of ligature-induced periodontitis in mice. The ligature method is detailed at. The method of the periodontitis induction was realized by one operator when using an ophthalmological nylon silk ligature of 6–0 (SHALON^®, São Paulo, Brazil). It was gently introduced into the interproximal area between the first mandibular molar and the second mandibular molar with two curved needle-holders (DLMICOF^®, Sao Paulo, Brazil) that were developed for this study. The silk ligature was gently tied in order to avoid damaging the periodontal area around the first molar. In order to perform these procedures, the mice were anesthetized with an intraperitoneal injection of ketamine (100 mg/kg) (DOPALEN^®, VETBRANDS, São Paulo, Brazil) and the muscle relaxant xylazine (10 g/kg) (ANASEDAN^®, VETBRANDS, Brazil). They remained sedated for approximately 90 minutes. After 15 days, the ligatures were removed and the periodontal treatments were performed in a standardized manner in the groups P+PT, P+PT+PDT, Asthma+P+PT, and Asthma+P+PT+PDT. No antibiotics or anti-inflammatories were administered. A Mini Five 5/6 curette was used and crown-root scrapings/ planning were performed on the vestibular lingual, mesial and distal faces. PDT was then employed as an adjuvant therapy to the periodontal treatments in the P+PT+PDT and Asthma+P+TP+PDT groups. For such, the photosensitizer methylene blue (0.005%—CHIMIOLUX, DMC, São Paulo, Brazil) was administered with a syringe, cartridge and needle (with a stop and without a bevel) at the two sites (vestibular and lingual). After three minutes, the periodontal pockets were irradiated with a red laser (λ = 660 nm ±10nm) (THERAPY XT, DMC, São Carlos, São Paulo, Brazil—ANVISA 80030810157). The radiant power of the appliance is 100 mW. The spot (area) was 0.02827 cm². The radiant energy delivered per point was 9J in 90 seconds. The radiant exposure was 318 J/cm² and irradiance was 3.5W/cm². In the vestibular face 9J were applied and another point in the lingual face (9J) of the right first molar also applied. followed by abundant rinsing with water for complete removal of the methylene blue. We irradiated with a small spot (0,02827mm²) because the tooth size of the 1^st mandibular molar of mice is about 1.44 ± 0.011 mm at vestibular face and 0.80 ± 0.019 mm in the buccolingual length.. All periodontal treatment (with or without PDT) was performed in a single session Euthanasia and posterior analyzes of the material were performed, 43 days after the beginning of the study The complete timeline of the induced ligature (15 days), the periodontal treatment (7 days), the PDT (7 days) and the euthanasia (43 days), for the 8 groups of this study, is detailed in. After the experimental period, the mice were euthanized with an overdose of ketamine (1.6 g/10 ml of solution) and xylazine (3g/100ml of solution). The blood samples were collected from the aortic artery of the mice by exsanguination, for a cell count by using Sysmex^™ C9.0 Software for the hemogram test. For an evaluation of lung inflammation in the BAL (bronchoalveolar lavage), the animals were tracheostomized and cannulated. Their lungs were washed with 3x0.5ml phosphate buffered saline (PBS). The volume of recovered BAL was centrifuged (1600rpm, 5 min at 4°C). The supernatant was collected and stored at -70°C for the cytokine analyzes by enzyme-linked immunosorbent assays (ELISA). The cell button was resuspended in 1 ml of PBS and was used for the total cell count. Ten microliters of the samples were added to Trypan Blue for a total cell counting with a Neubauer chamber. One hundred microliters were used to prepare the laminae for a differential counting of the cells (5 min, 1900rpm, 4°C) (Cytospin II—Shandon Instruments, Sewickley, PA, USA). The staining of the slides was performed with Instant-Prov. Three hundred cells were counted per laminae. The lung fragments were fixed in a 4% solution of paraformaldehyde, with 0.1 M of Sorensen’s phosphate buffer, at pH 7.4, at 4°C, for 24 h in anhydrous alcohol, followed by dehydration in alcohol and by diaphanization in xylol. The fragments were embedded in paraffin, sectioned to 5 μm with the aid of a microtome (HYRAX M60, Zeiss, GR), de-paraffinized, cut to a thickness of 5 μm, and then stained with periodic acid-Schiff for analyzes of the mucus. The protocol was detailed in. The internal and external limits of the respiratory epithelium was delimited (Image Pro-Plus 7.0) The mucus area was determined by area of glycoprotein component relation to total area of the respiratory epithelium. The results are expressed as the percentage (%). The measurements were performed in the five airways of each animal at ×400 magnification. The hemi-jaws were dissected and fixed in 10% buffered formalin solution (Merck & Co. Inc, New Jersey, USA) at pH 7 for a period of 24–48 hours. They were decalcified in 4% EDTA for 2 months. The mandibles were then dehydrated in alcohol solutions and diaphanized in xylol. The fragments were embedded in paraffin, sectioned to 5 μm with a microtome (HYRAX M60, Zeiss, GR), de- paraffinized, cut to a thickness of 5 μm and stained with hematoxylin and eosin. Serial cuts were used involving the best specimens. The morphometric analyzes were performed by a single examiner. Kappa values were used in order to measure the intra-observer agreement of measures. The observer was an experienced pathologist, França CM, and he had an intra-observer agreement of 0.85. This pathologist followed some references in order to proceed with the morphometric analyzes. The histological sections were photographed by using an Olympus Bx43 microscope with the help of Olympus cellSens^™ software. The measurements of bone loss were performed with Image J Software Version 1.45. The distance between the cementum-enamel junction (CEJ) to the alveolar bone (the distal region of the first molar) was measured in millimeters. The linear measurements of each slide were summed in order to obtain an average value for each animal. The concentrations of cytokines were determined in the supernatant samples of the lavage fluid and the serum. The results were expressed as picograms of cytokine produced per mL. The Interleukin Th2 (IL-10, IL-4, IL-5, IFN-γ) and the Th1 (IL-1, IL-6, INF-γ) profiles were quantified by using ELISA (BioLegend, San Diego, USA). The determinations were performed in duplicate for each sample by using standard curves and following the manufacturer’s specifications. The quantification of serum alkaline phosphatase was also quantified by using ELISA (BioLegend, San Diego, USA). Statistical analyzes were performed by a blind statistician using the GraphPad Prism program (GraphPad Software, Inc). The Kolmogorov-Smirnov test was used in order to determine the data distribution. Since the data was parametric, a one- way analysis of variance (ANOVA) was employed, followed by the Student-Newman- Keuls test. A p-value \< 0.001 was considered indicative of statistical significance. # Results The quantification of the cells that were recovered in the bronchoalveolar lavage (BAL) of the asthmatic mice shows that the total count of the inflammatory cells (x10⁴/ml) in the bronchoalveolar lavage (BAL) was increased in the periodontitis (P) group (p \<0.05) and in the asthma group (Asthma) (p \<0.001) when compared with the basal group (B). The periodontal treatment (P + PT) (p \<0.05) and the periodontal treatment associated with PDT (P + TP + PDT) (p \<0.05) were able to decrease the total amount of inflammatory cells in the bronchoalveolar lavage when compared to the periodontitis (P) group. The association of the periodontitis with asthma (P + Asthma) reduced the number of the total cells that were recovered in the BAL when compared to the group with asthma (Asthma) (p \<0.001). There were no differences in the total number of BAL cells after the standard periodontal treatments in the asthmatic mice (Asthma + P + TP) when compared with the asthmatic mice in the periodontitis group (Asthma + P). On the other hand, when photodynamic therapy was associated (Asthma + P + TP + PDT), there was a significant increase in the total number of cells (p \<0.001) in relation to the periodontitis with asthma group (Asthma + P). In the quantification of the differential number of cells that were recovered from the BAL there was an increase (p\<0.001) in eosinophils for the (P) and (Asthma) groups in relation to the Basal group. The treatment with PDT decreased (p \<0.001) the amount of eosinophils in the (P + TP + PDT) group when compared to the (P) group. When associated with asthma, the two modalities of TP (Asthma + P + TP) and (Asthma + P + TP + PDT) increased (p\<0.001) the number of eosinophils when compared with the (Asthma + P) group. There was a reduction (p\<0.05) in the macrophage differential count (x10⁴/ml) of the BAL for the (P+PT) group in relation to the (P) group. When we associated asthma with P (Asthma + P), there was a reduction of cells (p\<0.001) in relation to the Asthma group. The periodontal treatment (Asthma + P + TP) and PDT (Asthma + P + TP + PDT) in the asthmatic mice increased (p\<0.001) the amount of macrophages when compared to the (Asthma + P) group. The amount of lymphocytes increased (p\<0.001) for the (P) group and for the (Asthma) group (p \<0.01) when compared to the (Basal) group. For the (P+PDT) group, there was a decrease (p \<0.001) in relation to the (P) group. For the (Asthma + P) group, there was a decrease (p \<0.05) in relation to the (Asthma) group. When the treatments were associated, the (Asthma + P + TP) group and the (Asthma + P + TP + PDT) group had increased (p \<0.001) lymphocytes in relation to the (Asthma + P) group. There was an increase in neutrophils (p\<0.001) for the (Asthma + P + TP) and (Asthma + P + TP + PDT) groups when compared to the (Asthma + P) group. For the quantification of mucus in the airway (Figs) there was an increase (p\<0.001) in the mucus production in the asthma (A) and periodontitis (P) groups (p \<0.01) when compared to the basal (B) group. The association of P with asthma (Asthma + P) decreased the mucus production when it was compared to the (Asthma) group. For the analyzes of the IL-1β, IL-6 and TNF-α cytokines in the BAL supernatant, the inflammatory cytokines related to the Th1 mechanism were analyzed. An increase (p \<0.01) of IL-1β production in the periodontitis (P) group was observed when compared to the basal group (B). There were no differences in the other groups that were analyzed (p\> 0.01). There was an increase (p\<0.001) in the IL-6 levels in the BAL of the (P) group when compared to the basal (B) group. On the other hand, there was a reduction (p\<0.001) in the IL-6 production in the treatment groups (P + TP) and (P + TP + PDT) when compared to the (P) group. We also observed an increase (p\<0.001) in the production of TNF-α for the (P) and (A) groups when compared with the basal (B) group. It was observed that the standard periodontal treatment (P+TP) alone or in a combination with the photodynamic therapy (P + TP + PDT) decreased the production of TNF-α in the BAL, respectively (p\<0.001 and p\<0.01), when compared with the (P) group. The association of P with asthma (Asthma + P) decreased (p \<0.05) the TNF-α production when compared to its control group, the (Asthma) group. For the analyzes of the IL-4, IL-5 IL-10 and IFN-γ cytokines in the BAL supernatant–the cytokines related to the mechanism (Th2) of asthma were evaluated in the BAL supernatant. There was an increase (p \<0.001) in the production of IL-4 in the (A) group when compared to the basal (B) group. The association of P and asthma (P + A) decreased (p\<0.001) the production of IL-4 in relation to the (Asthma) group. In addition, the periodontal treatment (Asthma+P+TP) and the periodontal treatment with PDT in the asthmatic mice with P (Asthma+P+TP+PDT) decreased (p\<0.01) the level of IL-4 in the BAL when compared to the (Asthma+P) group. There was an increase (p\<0.01) in the IL-5 levels of the BAL in the (P) and (Asthma) groups when compared with the basal (B) group. On the other hand, there was a decrease (p\<0.01) in IL-5 when the periodontal treatments of (P+TP) and PDT were performed, (P+TP+PDT) when compared to the (P) group. It was observed that the PDT treatment (A+P+TP+PDT) in asthmatic mice increases the IL-10 production in the BAL when compared to the Asthma + P group (A+P). A decrease in the IFN-y levels (p\<0.05) in the periodontitis (P) and asthma (Asthma) groups when compared to basal (B) group was observed in. The standard periodontal treatment (P+TP) increases IFN-γ. Quantification of serum alkaline phosphatase—there was an increase in the production of alkaline phosphatase (U/L) in the (P) group when compared to the basal (B) group (p\<0.001). PT that was associated with PDT (P+ PT+PDT) decreased the alkaline phosphatase production (p\<0.01) in relation to the (P) group. An increase (p\<0.001) in bone resorption was observed in the (P) group when compared to the basal (B) group. PT that was associated with PDT (P+PT+PDT) was able to decrease the bone resorption (p\<0.01) in the alveolar bone region when compared to the (P) group. The asthmatic mice in which P was induced (Asthma+P) presented a greater resorption (p\<0.001) than did the group in which only asthma (A) was induced. The PT (Asthma+P+PT) and PDT (Asthma+P+PT+PDT) groups had lower values of bone resorption (p\<0.01) than did the asthmatic mice with periodontitis (Asthma+P). These values are shown in and illustrated in. # Discussion Periodontitis is defined as a chronic inflammation that destroys the supporting tissues of the teeth. It was characterized based on high levels of bone resorption, and alkaline phosphatase which are well-established methods in the literature. As we described before, periodontitis is characterized by an acute inflammatory process, which is then characterized by a pro-inflammatory profile of cytokine release, known as Th1. Among the cytokines with a Th1 profile, we can cite IL-1β, IL-6 and TNF-α. Periodontitis group (P) increased the amount of alkaline phosphatase and bone resorption, thus validating the experimental model of periodontitis and increase all the Th1 cytokines profile. The asthma model involved the induction of an allergic reaction using ovalbumin, which is recognized as mimicking signs and symptoms found in asthmatic patients. The experimental model for the induction of asthma has increased cellularity in the BAL represented by eosinophils, lymphocytes and neutrophils, as well as an increase in cytokines with a Th2 profile (IL-4 and IL-5), together with the production of mucus in the airways. This was an expected result for the asthmatic animals. To better understand the mechanism linking these two diseases, we studied the Th1/Th2 immunoregulatory control through its different cytokine profiles. Interestingly, the combination of asthma and periodontitis led to a significant reduction in the inflammatory cells (eosinophils, lymphocytes, macrophages) as well as the IL-4 cytokines release in the lungs. This decrease was accompanied by a reduction in the amount of mucus. It can therefore be inferred that periodontitis exerted some influence on pulmonary inflammation. In this group (A+P), bone resorption remained high, showing that periodontitis was still active. These findings are in agreement with data described previously by researchers studying the association between periodontitis and allergy. This study revealed an inverse relationship between periodontitis and allergy in a sample of 2.837 individuals. The authors discussed whether exposure to periodopathogens may influence the asthma progression. In other words, the regulation of the Th1-Th2 balance was deviated to the Th1 response in the asthmatic mice when periodontitis was associated with asthma. For these reasons, we have inferred that our results have an inverse relationship. Conversely, in a recent study with 5.976 patients, a positive association was found between periodontitis and asthma, while there is an inverse association only when the patients have been taking antiasthmatic medication. The causal relationship between them is still unclear. Further studies are needed to elucidate the link between these two pathologies. Considering the controversial data regarding the correlation between periodontitis and asthma, we have evaluated the effects of its treatment with conventional periodontal treatment and with PDT. When comparing the effectiveness of the two periodontal treatments in the mice without asthma, we noticed that the conventional treatment decreased the macrophages, as well as IL-6 and TNF-α, corroborating with the literature. However, it was interesting to note that the treatments appeared to act slightly differently, since PDT decreased the number of eosinophils and IL-5 in the mice without asthma and caused less bone resorption as well. Further studies are necessary to better understand the influence od PDT in Th2 mechanisms. In the asthmatic mice, after conventional periodontitis treatment, an increase in macrophages, lymphocytes, neutrophils and eosinophils were found. This variation in cellularity was followed by decreased IL-4 and TNF-α production. Moreover, standard treatment was able to interrupt the bone resorption process, confirming what occurs in clinical practice. Interestingly, PDT in asthmatic mice also increased the total cell count (macrophages, lymphocytes, neutrophils and eosinophils) in the BAL. Although the increase in defense cells is a disadvantage of this form of treatment, the cytokine release pattern was favorable, as the anti-inflammatory cytokine IL-10 was released in a greater quantity only for this group. These results suggest that PDT could stimulate the cellular immune response as demonstrated by some authors. Further studies could investigate if both M2 (macrophages) and Treg (regulatory T cells) are implicated in these processes producing anti-inflammatory cytokines. Although the mechanisms of the action are different, when photobiomodulation was directly applied in the lungs, has reduced the TNF-α expression after an acute immunocomplex lung injury in rats. They tested different doses of laser and they found a dose-dependent reduction of TNF-α levels in acute inflammation. In our study we also found a decrease in TNF-α levels in the BAL after a PDT. One limitation of this study was not to study the systemic alterations (serum) of cytokines release. More studies are needed in order to understand these systemic interactions. Photodynamic Therapy is known as the first order of antimicrobial action in a periodontal treatment and has been largely studied. Photosensitizers affect bacterial cells, damaging structures and impeding essential functions for bacterial survival. We selected methylene blue because it is a widely studied photosensitizer for antimicrobial PDT, due to its absorption around 650nm and its photophysical and photochemical properties. Besides that, it has a positive charge compound which binds to the bacteria due to an electrostatic interaction. Some studies have also shown an effective antibacterial action. Methylene blue is commercialized by different companies and it may be used in different formulations/concentrations. Some well-known brands are HELBO^® and Chimiolux. In Brazil, we have selected methylene blue because is approved by the Brazilian Health Regulatory Agency (ANVISA), and is widely used for periodontitis treatment. Besides, different chemical names for this molecule are also used, such as 3,7-Bis(dimethylamino)phenazathioniumchloride, Basic Blue 9, Tetramethyl Thionine Chloride and Phenothiazine Chloride. The nomenclature “phenothiazine” is not the most suitable of products since phenothiazines are tricyclic yellow compounds and not photosensitizers. Finally, it was interesting to note that we had an unexpected result in one of our control groups (P+PT). Our data has shown that periodontitis group (P) *per se* caused an increase in the total number of cells, mainly in the eosinophils and lymphocytes, as well as in the IL-5 cytokine release and mucus production. Thus, this work has contributed in elucidating, in part, the relationship between P and asthma. Additional studies involving functional measurements of airway hyperreactivity and the investigation of structural changes in the airways are needed to gain a better understanding of the effects of periodontal treatment in patients with asthma. In conclusion, periodontitis in the asthmatic mice reduced the inflammatory migrated cells in the BAL (eosinophils, lymphocytes, macrophages), as well as in reduce the levels of the IL-4 and TNF-α cytokines, which was additionally accompanied by a decreased mucus production. After the removal of the causative agent of the periodontal inflammation/infection (PDT treatment), the total cell counts increased, but this increase was not accompanied by a pro-inflammatory cytokines release. Only in the PDT group an anti-inflammatory cytokine (IL-10) was increased. More studies are needed in order to understand these mechanisms of action. # Supporting information We would like to thank Alana Dias dos Santos for her laboratorial assistance [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction The neurotrophins (NT) are a family of proteins comprising nerve growth factor (NGF), brain-derived neurotrophic factor (BDNF), Neurotrophin-3 (NT-3), and NT-4/5, initially identified through their crucial function in nervous system development, growth control, and neuron and astrocyte apoptosis. However, the effects of these growth factors are much more broad, and extend to a wide range of cell types, including immune cells. Indeed, growing evidence suggests that NGF, BDNF, and NT-3 participate in inflammatory responses, including the modulating and regulating immune function in inflammatory and autoimmune diseases. Neurotrophin function in immune regulation has been assessed in several reports that demonstrate that immune cells both secrete and are targets of the three major NTs (NGF, BDNF and NT3). Indeed, after activation, B cells, plasmocytes and T cells express NT receptors (TrkA, TrkB, TrkC and p75^NTR) and produce functional NGF, BDNF, and NT-3, which is involved in lymphocyte maturation, proliferation, and activation. Concerning B lymphocytes, NGF is secreted during B cell activation, which triggers their proliferation and differentiation into plasma cells. Immunoglobulin secretion (IgG and IgM or IgE) is enhanced by NGF or BDNF. In addition, BDNF also plays an important role in B cell development. This was demonstrated in BDNF^-/- mice, which show a developmental arrest in B cell maturation at the pre-BII stage. Lastly, both NGF and BDNF are autocrine factors for mature B and plasma cell survival, whereas this function for NT-3 has only been reported during inflammatory conditions. NTs also exert important functions in T lymphocytes: they may promote T cell activation of a Th2 or Th1 profile. Indeed, NGF enhances Th2 cell proliferation and modulates T-cell-dependent antibody synthesis and T cell production of gamma-interferon (INF-γ). Moreover, CD40L, interleukin-1β (IL-1β), IL-4 and *tumor necrosis factor-alpha* (TNF-α) upregulate production of NGF in lymphocytes. In contrast, INF-γ decreases NT synthesis. Th2 cytokines can also upregulate BDNF production in immune cells. NT3 synthesis is enhanced in Th1-activated human lymphocytes. Thus, these data support a potential crosstalk between NTs and Th1 and Th2 cytokine profiles during the inflammatory response. Data on immune cell expression of NT-4/5 are sparse. NT-4/5 is expressed by 25% of human circulating peripheral blood mononuclear cells (PBMC), activated human T cells, and murine alveolar macrophages. However, the function of this neuropeptide, known to interact with the TrkB receptor in neural cells, remains unknown in immune cells. The relationship between NT-secreting immune cells and the resulting tissue damage has been evaluated in some chronic inflammatory-autoimmune diseases. During rheumatoid or psoriasis arthritis, synovial CD3+ T lymphocytes and monocytes/macrophages produce high levels of NGF, which enhance both fibroblast- like cell proliferation and synovial T cell activation via TrkA ligation and Akt phosphorylation. In sarcoidosis, epithelioid and multinucleated giant cells of the granuloma, alveolar macrophages and T cells produce NGF, BDNF and NT-3. CD4 and CD8 NT expression correlates with the sarcoidosis radiological damage index. In contrast, in Crohn’s disease, local secretion of NT, especially NGF and BDNF by mast cells, reduces enteric glia cell apoptosis induced by pro-inflammatory cytokines. Together, these findings suggest that NT, excessively produced by immune cells in autoimmune diseases, may participate in disease progression by modulating both immune cell function and tissue lesions. Based on this foundational data, other studies have evaluated serum NT levels in various autoimmune and pro-inflammatory diseases. However, these reports have mainly dealt with NGF. Indeed, serum NGF concentrations are increased in juvenile arthritis, Kawasaki disease, Behçet’s disease, systemic sclerosis and primary Sjögren’s syndrome. Increased BDNF levels in sera have also been reported in primary Sjögren’s syndrome, which correlates with systemic activity and B and T cell activation. In contrast, serum BDNF levels are decreased in systemic sclerosis, reflecting the vascular aspect of the disease. It has also been reported that NT-3 is upregulated only in autoimmune diseases strongly affecting the joints. Serum NT-4/5 levels are upregulated in mood disorders but have not been yet evaluated in autoimmune disease. There is little data on lymphocytic NT expression in human inflammatory disease. BDNF-secreting T cells are reduced in untreated multiple sclerosis patients and increased after interferon beta treatment, while NGF, NT-3 and NT-4 production by PBMCs in multiple sclerosis patients is enhanced in the post-relapse phase. In contrast, BDNF production is unchanged in B and T cells in systemic sclerosis patients compared to healthy controls. In SLE, few studies have focused on NT expression and its relationship to disease activity. In NZB/W mice, serum NGF concentrations are significantly increased, correlating with an accumulation of NGF-containing cells in the kidney and spleen. NGF levels are higher in the sera of SLE patients than healthy controls and reflect systemic activity of the disease as assessed by the SLEDAI (SLE Disease Activity Index) score. However, reports on serum BDNF concentration in SLE are contradictory and limited to neuropsychiatric forms of the disease. Though serum BDNF levels are decreased in neuro-SLE according to one case report, they are increased in two other studies. The aim of the present study was to evaluate serum and lymphocytic levels of NGF, BDNF and NT-3 in SLE patients and identify their relation to clinical features (systemic activity assessed by SLEDAI score, joint, skin, neurological and kidney involvement, vasculitis), SLE-related immunological activity (anti- native DNA antibodies, complement activation via CH 50, C3 and C4 levels), and anti-phospholipid antibodies. Furthermore, we evaluated B cell activation parameters that could be modulated by SLE (serum BAFF levels and autoantibody production) and their association with enhanced levels of NT in sera. Additionally, we analyzed the cytokine profiles and T-regulatory cell population that could be modified by SLE activity. IL-10 and IFN-γ, two cytokines belonging to the TH2 and TH1 profile, respectively, were measured in sera from controls and patients. Several studies have previously demonstrated that IL-10 levels in sera are significantly higher in untreated SLE patients than in healthy controls and strongly reflect SLE activity. Their dynamics are closely linked to those of autoantibody synthesis. Strikingly, circulating IL-10 levels decrease after treatment and correlate with a change in the SLEDAI score, indicating that sera IL-10 level is a biological marker of SLE activity. Moreover, the evaluation of IL-10 levels in SLE is also supported by the direct effect of IL-10 on NGF secretion in a murine astrocyte model. Likewise, a relationship between IFN-γ and BDNF cell activation has been established in microglial cells, in that BDNF inhibits IFN-γ-induced activation. *Considering that no reported experimental data support the involvement of NT4/5 in B and T cell activation or autoimmune disease, this study mainly focused on the relationship between serum levels, B-lymphocyte expression of NGF, BDNF and NT-3, and SLE activity.* # Materials and Methods ## Patients and control population Twenty-six successive SLE patients, including 24 women (median age 44±12 years), were included in a one-year cross-sectional study (2011) in Limoges University Hospital. All participants fulfilled the revised American College of Rheumatology (ACR) criteria for SLE. Disease activity was evaluated using the SLEDAI score. Neurological complications were ruled out by clinical examination combined with normal brain MRI (n=11) and brain 18-F-fluorodeoxyglucose positron emission tomography (n=2). Patients with neoplastic disorders or depression were excluded in order to avoid confounding effects on serum NT levels. The control population consisted of 26 healthy age- and sex-matched volunteers. Patients were excluded if they were pregnant, under the age of 18, or unable to give valid consent. The “NeuroLED” study received ethical approval from The Limoges University Hospital Research Ethics Committee (N°I06023), and was carried out in accordance with the Helsinki Declaration. Written consent was obtained from all patients and control subjects. ## Clinical features The clinical features of SLE present at the time of blood sampling were cutaneous (n=22, 85%), articular (n=21, 81%), renal (n=4, 16%), neurological (n=2, 8%) and pleural (n=1, 4%), as well as pericarditis (n=1) and vasculitis (n=9;). The mean SLEDAI score was 7.2 ± 4.1 (range 4-20). Six patients (23%) presented a severe systemic SLE flare defined by a mean SLEDAI score over 9. Six patients (23%) presented an associated anti-phospholipid syndrome and 7 presented secondary Sjögren’s syndrome (27%). Out of the 26 patients, 14 (54 %) were treated with glucocorticoids (mean 18.1 ± 22.2 mg/day), 20 (77%) with anti-malarial drugs, and 4 (16%) with immunosuppressants (azathioprine n=2, mycophenolate mofetil to maintain remission of a previous renal flare n=2;). Treatment was increased directly after the blood sampling for systemic flare in 61% of cases (n=16) with the initiation of hydroxychloroquine (n=5, 19%), glucocorticoid (n=7, 27%) or immunosuppressants (n=7, 27%), or with an increase of previous corticosteroid dosages alone (n=3, 11%;). Another blood sample for serum NT levels was taken one month after the systemic flare in 9 cases. ## Additional assessment of quality of life Patients completed the Medical Outcomes Study, a 36-item short form health survey (SF-36), independently of their visit to the physician. The questionnaires were returned to the research assistant. ## Measurement of autoantibodies and neurotrophin levels Antinuclear antibodies were characterized by immunofluorescence in HEp2 cells (The Binding Site, Saint Egrève, France) and anti-native DNA (nDNA) antibodies by ELISA (Phadia, Saint-Quentin Yvelines, France). Anti-cardiolipin and anti-β2 Glycoprotein 1 (β2-GP1) antibodies were measured by ELISA (Ingen, Chilly Mazarin, France). Serum NGF, BDNF and NT-3 levels were measured using commercial ELISA kits according to the manufacturer’s instructions (NGF E_max^® ELISA, BDNF E_max^® ELISA, NT-3 ELISA, Promega, Charbonnières, France). All assays were performed in duplicate and the data are presented as pg/mL. Detection limits were 15 pg/mL for BDNF and 4 pg/mL for NT-3 and NGF. ## B cell activation analysis and determination of Th1 and Th2 profiles Serum BAFF, INF-γ (Th1) and IL-10 (Th2) levels were measured with an ELISA kit according to the manufacturer’s instructions (Quantikine^® Human Immunoassay R&D system, Lille, France). All assays were performed in duplicate and the data are presented as pg/mL. Detection limits were 4 pg/mL for BAFF and 2 pg/mL for INF-γ and IL-10. ## Determination of neurotrophin expression by B and T cells Expression of intracellular NGF, BDNF and NT-3 in T and B lymphocytes was assessed by flow cytometry. Whole blood cells were stained with either phycoerythrin (PE)-cyanin (Cy) 7-conjugated anti-CD3 or anti-CD19 antibodies for 15 min at room temperature. After lysing the red blood cells (Immunoprep, Beckman Coulter, France), white blood cells were fixed, permeabilized (Intraprep, Beckman Coulter) and incubated at room temperature for 30 min with rabbit anti-NGF, anti-BDNF and anti-NT-3 antibodies (all 1/100; Santa Cruz Biotechnology, France) in Phosphate-Buffered Saline (PBS) containing 1% Bovine Serum Albumin. After two washes in PBS, antibodies were detected with Alexa Fluor 488-conjugated goat anti-rabbit IgG antibodies (10 μg/mL; Invitrogen, France) for 30 min at 4°C. Cells stained with rabbit isotypic immunoglobulins (Santa Cruz Biotechnology, France) were used as controls to determine background and positive result thresholds. After washing twice in PBS, cells were suspended in PBS and analyzed with a flow cytometer (FacsCanto^TM II, Becton Dickinson, Le Pont-de Claix, France). ## Quantitative analysis of circulating T regulatory cells Cells were stained with Cy7-conjugated anti-CD3, FITC-conjugated anti-CD4, Cy5-anti-CD25 (Beckman Coulter), and PE anti-FOXP3 (eBioscience) antibodies or isotype controls, and FACS analysis was performed as previously described. ## Statistical analysis The results were expressed as means ± standard deviation. *P* values ≤ 0.05 were considered significant. One-way analysis of variance (ANOVA), Chi-square tests and Mann-Whitney tests were used when appropriate. To detect correlations between serum NT levels, clinical and other biological data, linear regression analysis was used and *p*-values were determined by Spearman’s rank correlation test. # Results ## Variations in serum NT expression in SLE ### The NT levels in sera of SLE patients, NGF, BDNF and NT-3 were determined by ELISA Serum NGF levels were higher in SLE patients (426.13 ± 70.85 pg/mL) than in healthy controls (373.9 ± 52.3 pg/mL, *p*=0.003). BDNF levels were also increased in SLE patients (598.9 ± 129.8 *vs* 326.1 ± 60.5 pg/mL in controls, *p*\<0.0001). Average serum NT-3 levels were similar in the SLE group and controls (2911.7 ± 1248.8 *vs* 2553.7 ± 879.7 pg/mL, NS,). The increases in serum levels of NGF and BDNF were statistically independent (r=0.28, NS). Therefore, we examined the correlation between enhanced NGF and BDNF levels and lupus systemic activity. ## Serum neurotrophins and SLE activation profile NGF, BDNF and NT-3 serum levels were not correlated with initial SLEDAI score (NGF: Γ=0.19, *p*=0.34, BDNF: Γ=0.16, *p*=0.42, NT-3: Γ=0.35, *p*=0.07). However, there were higher NT-3 levels in a subset of patients with severe systemic flare (SLEDAI ≥10; 4171.6 ± 1013.17 *vs* 2533.7 ± 1062.9 pg/mL, *p*=0.002). In contrast, concentrations of NGF (426.2 ± 62.6 *vs* 425.9 ± 74.6 pg/mL, NS) and BDNF (612.6 ± 73.6 *vs* 594.7 ± 143.7 pg/mL, NS) were similar in patients regardless of SLEDAI score. SLEDAI score reflects the global systemic activity of SLE, which can correspond to the involvement of various organs. The presence of cutaneous, neurological or renal SLE manifestations did not influence serum levels of NGF, BDNF and NT-3. However, NT-3 levels were significantly increased in patients with articular manifestations (3185.6 ± 1213.3 pg/mL *vs* 1761.7 ± 581.2 pg/mL, *p*=0.02). In contrast, there was no difference in NGF and BDNF levels between patients with and without joint involvement. We then investigated the correlation between NT serum levels and immunological parameters associated with lupus flare. Serum NT-3 levels alone were dramatically increased in patients with complement activation (n=8, 3749.1 ± 1433.25 *vs* 2457.9 ± 950.7, *p*=0.01) and these levels correlated negatively with serum CH 50 levels (Γ=-0.28, *p*=0.032). In contrast, anti-nDNA antibodies are independent of the serum levels of NGF (Γ=0.09, *p*=0.68), BDNF (Γ=0.29, *p*=0.18) and NT-3 (Γ=0.35, *p*=0.07). Considering that SLE may be associated with secondary Sjögren syndrome or anti- phospholipid syndrome, we examined if there was a correlation between NT serum levels and SLE-associated autoimmune disease. ## Serum neurotrophins and SLE-associated disease NGF, BDNF and NT-3 concentrations were similar in SLE patients with or without secondary Sjögren’s syndrome (NGF 430.7 ± 113.2 *vs* 424.5 ± 51.8 pg/mL, BDNF 624.27 ± 102.31 *vs* 589.4 ± 139.89 pg/mL, NT-3 3094.86 ± 1262.33 *vs* 2860.87 ± 1274.52 pg/mL, all NS). However, the presence of anti-SSB antibodies (n=3) was associated with higher serum NGF levels (507.2 ± 23.4 *vs* 417.4 ± 69.1, *p*=0.03). In patients with anti-phospholipid autoantibodies, lower serum BDNF levels were observed: anti-cardiolipin (n=9, 529.2 ± 142.9 *vs* 635.7 ± 109.2 pg/mL, *p*=0.04) or anti-β2GP1 antibodies (n=4, 438.8 ± 185.4 *vs* 627.9 ± 96.87 pg/mL, *p*=0.004). Furthermore, BDNF levels negatively correlated with the titers of both IgG and IgM anti-cardiolipin antibodies (Γ=0.46, *p*=0.01 and Γ=0.41, *p*=0.04, respectively). In addition, we examined a possible correlation between systemic flare, quality of life and serum NT profile. ## Serum neurotrophins and quality of life Scores from both the Physical Component Summary (PCS) and Mental Component Summary (MCS) were lower in SLE patients than healthy controls (PCS, 36.5 ± 6.9 *vs* 54.1 ± 7.4, *p*\<0.0001; MCS, 38.6 ± 10.1 *vs* 50.2 ± 6.5, *p*\<0.0001) and did not correlated with SLEDAI score (PCS: Γ=0.03, *p*=0.87, MCS: Γ=0.22, *p*=0.32). In SLE patients and the controls, neither PCS nor MCS scores were related to serum NT levels. *In order to evaluate the influence of immunomodulating drugs on the increased circulating levels of NT, we examined systemic NT levels before and after SLE flare treatment.* ## Dynamics of neurotrophin serum profile and SLE treatment Serum NGF, BDNF, and NT-3 concentrations were similar in SLE patients at the time of the blood sample whether they were untreated (n=4) or previously treated by corticosteroid, hydroxychloroquine or immunosuppressants. Interestingly, serum BDNF concentrations decreased one month after the systemic flare of the disease (407.7 ± 65.6 pg/mL) but were still higher than those in healthy volunteers (326.1 ± 60.5 pg/mL, *p*\<0.001). In contrast, serum levels of NGF (426.12 ± 70.8 *vs* 462.6 ± 47.1 pg/mL, NS) and NT-3 (2986.9 ± 954.5 *vs* 2911.7 ± 1248.8 pg/mL, NS) remained unchanged after treatment of the SLE flare. NGF, BDNF and NT-3 levels remained uncorrelated with the SLEDAI score (mean 4 ± 1.8) assessed after flare treatment (Γ=0.35, Γ=0.46 and Γ=0.27 respectively, all NS). In order to determine if there was a correlation between NT and SLE activity at the level of immune cells, we tested if there was a relationship between serum and lymphocytic NT expression, T and B cell activation, and cytokine profiles. ## Lymphocytic neurotrophins and clinical and immunological SLE profile ### NGF and BDNF-producing B cells are increased in SLE The numbers of B cells producing NGF and BDF were greater in the SLE group than in healthy controls (NGF, 29.3 ± 31.3 *vs* 11.3 ± 20.9, *p*=0.02; BDNF, 71.2 ± 30.9 *vs* 47.9 ± 27.8, *p*=0.03). Cell numbers were independent of serum NGF, BDNF and NT-3 concentrations. In contrast, the numbers of NT-3-CD19-positive cells were similar in the two groups (72.4 ± 29.7 *vs* 64.7 ± 29.1, NS). The numbers of NGF and BDNF-producing B cells were independent of SLEDAI score (Γ=0.22 and Γ=0.2 respectively, NS). Although numbers of NGF-producing cells were independent of clinical SLE profiles, CD19+ BDNF-producing B cells were dramatically decreased in patients with an associated anti-phospholipid syndrome (40.2 ± 8.9 *vs* 80.5 ± 22.05, *p*=0.03). NGF and BDNF-producing B cells were not influenced by corticosteroid (NGF: 36.5 ± 33.8 *vs* 20.8 ± 27.11, NS, BDNF: 75.4 ± 23.7 *vs* 88.8 ± 17.5, NS), hydroxychloroquine (NGF: 33.6 ± 23.6 *vs* 14.7 ± 16.8 NS, BDNF: 79.9 ± 23.3 *vs* 84.3 ± 19.2, NS) or immunosuppressants (NGF: 25.3 ± 19.8 *vs* 30.1 ± 33.3, NS, BDNF: 76.06 ± 30.2 *vs* 81.6 ± 21.7, NS). Interestingly, NGF production by B cells tended to be higher in SLE patients with cryoglobulinemia (mean fluorescence intensity, Xmean, 8.7 ± 7.7) than in patients without cryoglobulinemia (Xmean 4.1 ± 2.7, *p*=0.06). A similar relationship was observed between CD19-related NGF production and complement activation (Xmean 7.1 ± 5.41 *vs* 3.5 ± 1.8, *p*=0.05). CD19-BDNF-producing B cells were also decreased in patients positive for lupus anticoagulants (28.5 ± 14.3 *vs* 79.0 ± 23.9, *p*=0.01). ### NGF and BDNF-producing T cells in SLE The numbers of T lymphocytes expressing NGF (17.2 ± 21.2 *vs* 9.8 ± 21.3, NS), BDNF (55.1 ± 33.5 *vs* 55.1 ± 33.5, NS) and NT-3 (57.5 ± 33.7 *vs* 52.9 ± 24.2, NS) were similar in the SLE and control groups and were independent of patients’ clinical and immunological SLE profiles. ### NT and B-cell activation parameters As expected, serum BAFF levels were increased in the SLE group (1980.7 ± 1315.9 *vs* 1019.9 ± 193.4 pg/mL, *p*=0.01). However, they did not correlate with serum concentrations of NGF, BDNF, or NT-3 (Γ=0.12, Γ=0.03 and Γ=0.04, respectively) in the SLE group. In contrast, in the control group, serum concentrations of NGF positively correlated with BAFF levels (Γ=0.64, *p*=0.02), which were also independent of the serum levels of both BDNF and NT-3 (Γ=0.02 and Γ=0.6, respectively). Serum levels of BAFF were also independent of B and T cell NT expression in both the SLE and control groups. ### NT and T regulatory cells Only serum NGF levels and T regulatory cell counts were negatively correlated in SLE patients (Γ=0.48, *p*=0.01), whereas there was no significant correlation between BDNF or NT-3 levels and T regulatory cell numbers (Γ=0.06, NS, Γ=0.28, NS, respectively). T regulatory cells and NTs appeared to be independent in healthy controls (NGF Γ=0.27, BDNF Γ=0.12, NT3 Γ=0.08, NS). BDNF-positive CD19 cells and T regulatory cells were negatively correlated (Γ=0.39, *p*=0.04). ### NT and IL-10 production IL-10 concentrations were higher in the SLE group (29.4 ± 81.1 *vs* 0.19 ± 3.3 pg/mL, *p*=0.06), especially in the subgroup of patients with severe systemic flare (91.28 ± 162.6 *vs* 10.9 ± 11.26 pg/mL, *p*=0.03) or positive anti-SSA antibodies (77.25 ± 152.48 *vs* 12.24 ± 13.27 pg/mL, p=0.07). In the SLE group, NGF, BDNF and NT-3 serum levels were statistically independent of serum IL-10 levels (Γ=0.02, Γ=0.12 and Γ=0.12 respectively), which did not influence NT production in T or B cells. The same results were observed in the control group. ### NT and Interferon-*γ* production Serum INF-γ levels were higher in the SLE group (INF-γ 136.7 ± 230.9 *vs* 6.9 ± 10.5 pg/mL, *p*=0.03), particularly in patients with severe systemic flare (258.9 ± 422.32 *vs* 79.58 ± 98 pg/mL, *p*=0.08). NGF, BDNF and NT-3 levels were also statistically independent of INF-γ concentrations (Γ=0.21, Γ=0.17 and Γ=0.35, respectively) in the SLE group. In contrast, serum BDNF levels negatively correlated with INF-γ concentration in the control group sera (Γ=0.55, *p*=0.03). While NT production by B and T cells was independent of INF-γ concentration in the control group, INF-γ serum levels correlated with NGF production in T lymphocytes (Γ=0.64, *p*=0.01). # Discussion The present study provides new evidence of the significant variation in NT levels in both serum and circulating B lymphocytes in SLE patients. Previous studies mainly focused on serum NGF levels, and serum BDNF and NT-3 had not been studied in a SLE cohort. Furthermore, NT expression in circulating B cells had not yet been documented. We have identified an increase in NGF and BDNF levels in both serum and circulating blood CD19-B cells in patients. Interestingly, NT-3 levels were only increased in severe forms of SLE and correlated with complement activation. We have identified a significant increase in sera NGF and BDNF levels in SLE patients compared to a healthy control group. We confirm that serum NGF levels are increased in SLE patients, a result previously demonstrated. However, we did not find any correlation between serum NGF levels, systemic complications, and SLEDAI scores in this SLE population, characterized by previous treatment with either corticosteroid or immunosuppressants. These results differ from those of a previous study reporting a correlation between serum NGF levels and SLE activity. However, the inclusion criteria of this study was different, involving only untreated children (and not adults) with an over-representation of renal involvement (60% of patients *vs* 16% in the present study). Moreover, a link between renal involvement and enhanced NGF levels has been described in SLE mice and in patients with SLE glomeronephritis-related renal insufficiency, which could explain this discrepancy. However, we did not find any correlation between NGF serum levels and the presence of renal involvement (only 4 adults) or other systemic complications. Therefore, the effect of immunomodulating drugs on serum NGF levels in previously treated patients needs to be discussed in regards to other reports. In fact, NGF serum levels have been found to be independent of immunomodulating treatments in other diseases such as systemic sclerosis, rheumatoid arthritis, and primary Sjögren syndrome. Concerning BDNF levels, this study reports for the first time an increase in serum BDNF levels in SLE patients, occurring independently of central neurological involvement (absent in all but two patients). Serum BDNF levels did not correlate with SLEDAI score. Only one previous case report has been described of psychotic involvement in SLE with an increase of serum BDNF. Likewise, in multiple sclerosis and rheumatoid arthritis, a decrease in serum BDNF levels occurs with clinical improvement, independently of other inflammatory markers. After treatment, a decrease in BDNF levels is detected in these SLE patients. The significance of this decrease with treatment is debated and also appears to be independent of both corticosteroid and immunosuppressive drugs in psychiatric form of SLE patients, as well as in systemic sclerosis and rheumatoid arthritis. In contrast, BDNF serum levels are enhanced in primary Sjögren syndrome patients (pSS) treated with either corticosteroid or immunosuppressants for severe systemic involvement. Interestingly, we have shown that BDNF levels decrease after treatment of systemic flares, signifying that this decrease could be a biological marker for improvement and a therapeutic response in SLE. Strikingly, we find an increase of NT-3 serum levels only in severe (SLEDAI ≥ 10) and articular forms of SLE. Serum NT-3 has been infrequently studied in inflammatory diseases, although NT-3 immune function has been demonstrated in asthma: autocrine NT3 autocrine secretion leads to plasma cell survival. NT-3 involvement in autoimmune articular symptoms is indicated by its presence in the synovial fluid of patients with spondyloarthritis and by increased levels in the sera of systemic sclerosis patients with articular complications that require hydroxychloroquine treatment. Enhanced NT-3 serum levels in severe forms of SLE could reflect a link between NT-3 and lupus flare, in that complement activation (decrease of CH 50 levels) correlates with elevated NT-3 levels. This direct relationship between NT-3 synthesis and complement (C1q) has been shown in neuronal cells. In another cell model, NT-3 production by endothelial cells is upregulated by local ischemia. Therefore, NT-3 secretion could be a biological marker of severe SLE forms, potentially associated with vascular damage related to an active vasculitis process. The direct impact of NT on SLE physiopathology is also underlined by the significant overproduction of both NGF and BDNF by B cells, and could reflect an activation of circulating B cells in SLE. The activation of circulating B-cells in SLE patients has been reported during nephritis and SLE activity. Moreover, in some experimental conditions, a direct link between *in vitro* activation of B cells and their secretion of NGF and BDNF has been established. Overall, the overexpression of these NTs in SLE patients’ circulating B cells (compared to healthy controls) could be a hallmark of activation that correlates with the disease. To further evaluate the impact of NTs on SLE pathogenesis, we have looked for correlations between both serum and lymphocytic NT levels and immune characteristics, especially antibody production and T regulatory cell profiles. NGF, BDNF and NT-3 levels did not correlate with anti-nDNA levels. This correlation has not been previously examined in the 3 published studies on NGF and BDNF serum levels in SLE. This absence of relationship between NT serum levels and autoantibody production was also found in pSS and systemic sclerosis. Concerning rheumatoid arthritis, to our knowledge, the correlation between NGF, BDNF and anti-cyclic citrullinated peptide has not been previously studied. Interestingly, serum BDNF levels were the lowest in the group positive for anti- phospholipid. This negative correlation could be due to anti-phospholipid- associated vasculopathy and oxidative stress, an association previously identified in systemic sclerosis, atherosclerosis and diabetes. Similarly, we identified for the first time a striking decrease of BDNF-producing B cells in patients with an associated anti-phospholipid syndrome. These results suggest that a reduction of both circulating BDNF levels and BDNF-producing B cells are involved in the vascular damage associated with SLE. In addition, we report a negative correlation between serum NGF levels and T regulatory cells in SLE patients, which could reflect either SLE activity or a direct impact of serum NGF level on T regulatory cell survival. Also, NGF is an inducible survival growth factor for T cells, depending on the cytokine profile; anti-NGF treatment enhances Foxp3+ regulatory T cells and decreases Th17 cells in a murine model of asthma. Moreover, IL-17 enhances NGF production in human T cells. Together, this data suggests that NGF could regulate the balance of T-regulatory and Th17 cells. Nevertheless, no reports have shown that either p75^NTR or TrkA NGF receptors are present on human T regulatory cells, which are known to produce high levels of BDNF in HIV-associated neurodegeneration models. IL-10 serum levels were enhanced in the SLE group, reflecting systemic activity and anti-SSA production as previously described, whereas NT and IL-10 levels are were not correlated. Indeed, IL-10 release by immune cells, which is enhanced by both NGF and BDNF in normal conditions, is dramatically reduced in cases of allergy. As expected, INF-γ serum levels were increased in SLE. Interestingly, in our control group only, we found a negative correlation between serum BDNF level and INF-γ, which is known to reduce BDNF production in neurons, glia, and bronchial smooth muscle cells. Moreover, a cytokine balance between INF-γ and BDNF has been described in bowel mucosa, as a regulating factor of enteric glia cell apoptosis. The deregulation of the INF-γ and BDNF balance in SLE may reinforce the importance of BDNF over-secretion in SLE pathogenicity. In conclusion, the present study finds that the expression of the neurotrophins NGF and BDNF, overexpressed by circulating B cells, are increased in the sera of SLE patients independently of Th1/Th2 profile. NT-3 is upregulated only in severe flares of the disease, and BDNF levels are closely related to anti- phospholipid syndrome. ## Thus, we hypothesize that evaluating NT in both sera and circulating-B lymphocytes could be a new biological marker of SLE activity and systemic complications To test this hypothesis, prospective large cohort studies, including naive SLE patients, need to be performed. In addition, further studies of NT-secreting subpopulations of B and T cells, in association with their NT receptors (i.e. TrkA, TrkB, TrkC, p75^NTR and sortilin), will be conducted to define their fine-tuning functions during SLE disease. This study supports the idea that neurotrophins are involved in SLE physiopathology and thus could be a potential target of systemic treatment. We thank *Cornelia* Wilson for the English correction of this manuscript. We especially thank the Limoges Clinical Research Center (DRCI CHU Limoges) for its help in the completion of this study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ALF MOJ PM MCL FL. Performed the experiments: MCL PM HB FXL CM KL. Analyzed the data: ALF MOJ FL. Contributed reagents/materials/analysis tools: AS ME EV. Wrote the manuscript: ALF MOJ EV. In charge of the clinical protocol study: SD.

# Introduction Honeybees are key pollinators of crops and wildflowers. They contribute globally more than €53 billion annually to the economy. Still, honeybees are presently facing multiple stressors such as habitat destruction, pests, and pathogens. Across the Northern hemisphere the spread of the ecto-parasitic mite *Varroa destructor* (*Varroa*) and the key viral pathogen it transmits (Deformed Wing Virus) is the number one threat faced by beekeepers over past decades, causing an increase in over-wintering losses. In the UK, this was evidenced in the very high losses of managed colonies and the near eradication of feral colonies within a few years of *Varroa* becoming established. Over-wintering colony losses have been declining in the UK during the past decades since *Varroa*-treatments became widespread. Although, colony losses remain consistently high in the USA, which is unexpected due to the high levels of treatment carried out by beekeepers to control *Varroa* populations in both countries. Typically, a variety of acaricides are primarily used to control *Varroa* populations, but with the development of acaricide resistance alternative commercially based treatments established around thymol and organic acids (oxalic and formic) have become increasingly popular. More time consuming and often less effective biomechanical treatment methods can also be used, such as powdered sugar and drone trapping. The application of Integrated Pest Management (IPM) techniques is suggested to be the most effective way to treat mites since it involves a variety of methods applied in rotation throughout the year determined by mite-infestation levels through routine observations and mite monitoring carried out by beekeepers. In contrast, some countries like South Africa and Cuba took a decision not to treat *Varroa* and allowed mite-resistance to appear naturally. In both countries 1,000’s of colonies were lost initially, but losses declined after several years as resistance to the mite developed. Whereas, in Brazil the evolution of mite-resistance by Africanised bees (*Apis mellifera scutellata*) was not observed and any losses were probably masked by the losses from untreated non-resistant European honeybee colonies, although there is no published evidence to support or contradict this idea. Despite this *Varroa* resistance rapidly became widespread in these countries, hence treatment-free beekeeping has already been established for several decades. More recently in mainland Europe, the UK, and USA it appears an increasing number of mite-resistant populations are being managed treatment-free. Allowing honeybees to develop natural resistance will foster a long-term solution to the *Varroa* problem. Resistance is defined as any situation in which *Varroa* populations are maintained at a suitable level for the long-term survival of the honeybee colonies. Recently, three key traits (cell recapping, mite infertility and brood removal) have been associated with natural resistance in almost all resistant populations studied in different countries. In the UK the government advice is to treat *Varroa* with either biotechnical methods, with registered varroacides or a combination of them. Despite this, anecdotal evidence from beekeepers suggests that numbers of non-treating beekeepers are more than expected. The aim of this study was therefore to conduct an online survey of UK beekeepers via their associations to assess individual treatment habits, as it was the most efficient way to reach beekeepers easily. The survey will provide crucial empirical data to support or refute common perceptions around treatment and non- treatment beekeeping practices. # Materials and methods ## The survey The survey was constructed using Google Forms as it allowed an unlimited number of responses whilst also allowing the incorporation of the University logo to add credibility to the survey. The survey consisted of a brief description outlining the study, its aims, and our definition of treatment, which was, “any form of external or additional control administered to bees by beekeepers aimed at reducing *Varroa* numbers”. Then followed six questions: 1) association name, 2) number of colonies, 3) if they treat or not, 4) number of times a year they treat, 5) number of years since last treatment, 6) type of treatment (see for survey). Answers were either multiple-choice questions or open questions that all helped assess the beekeeper’s treatment habits. The survey was kept short in duration to increase the response rates. The contact details for 325 beekeeping area associations across the UK were obtained via the four UK Beekeeper Association websites (British, Scottish, Welsh and Northern Ireland). Initial contact was made by email with 303 of them; outlining the study aims to determine general interest. The responding associations were then sent the survey online link and asked to forward it to their respective members. Beekeepers were given ten weeks from 23^rd Dec 2020 to complete the survey. A small number of beekeepers did not complete every section of the survey fully, so for each analysis the precise number of responses obtained from that part are given. Amateur bee-associations were chosen as they contain around 30,000 members in the UK. ## The survey analysis After the survey closed, all 2,897 responses from 243 associations were exported into an excel spreadsheet. The 25 responses from the Isle of Man, which remains a *Varroa*-free island were excluded along with the 16 beekeepers with zero colonies leaving 2,856 for further analysis. A total of 67 beekeepers did not either belong to or failed to put down an association. As over 20 types of *Varroa* treatment were reported, we grouped them into single methods (Chemical, Biomechanical and Natural) and mixed methods. To estimate the number of colonies surveyed we multiply the median colony group size, or 30 for the 30+ group by the number of beekeepers in that group. ## Spatial analysis While conducting previous research into *Varroa-*resistance colonies it appeared that they existed in many parts of the UK. To investigate this maps were created using ArcGIS Pro (v. 10.8.1) software, showing the location of all UK area associations and those that participated. To investigate if any spatial patterns in treatment-free vs treated existed, the proportion of beekeepers falling into either category were calculated from each association that had at least five responses to avoid illegible pie diagrams on the figure. ## Statistical analysis A GLM was conducted in R using treatment status as the response variable, association as the fixed effect and colony number as the random effect. A binomial family was used to fit the model due to only two variants of the response variable (i.e., treating or treatment-free). The median group size or 30 for the 30+ group were used for colony group size in the GLM. # Results The 2,856 beekeepers who responded represents almost 10% of the estimated 30,000 members belonging to the four UK beekeeping associations and were widely distributed across the UK. We estimated that we surveyed around 21,200 colonies that is again around 10% of the 220,000 colonies estimated by the Center for Ecology and Hydrology to be in the UK. The majority (67%) of beekeepers that responded managed between 1 to 5 colonies and only 3% had over 30 colonies. Our data indicated that across the five colony size groups the proportion of treatment-free beekeepers was greatest in the 1–5 group, but all other groups were within 5% of that group. The GLM showed that treatment status was not affected by association but did indicate a significant effect of colony group size on treatment status. However, as the 95% CI (-19.559–15.493) contains 0 then we can conclude that there is not a strong significant relationship. Due to such high variability in the data, we conclude that there is not a statistically significant relationship between colony number and treatment status. ## Treatment-free beekeeping A total of 596 (21%) beekeepers stated they were not treating, and 2,260 (79%) beekeepers were treating their colonies against *Varroa*. When asked about the duration elapsed since their last treatment was applied, 72% had treated within the last year and 173 (6%) responders had not treated for 6 years or more. The spatial distribution indicates Scottish beekeepers were almost all treating and only four associations had more than 75% of treatment-free beekeepers as members, otherwise treatment free beekeepers were spread throughout England. This indicates the widespread distribution of beekeepers attempting treatment free approaches particularly in England. ## Treatments used in the UK An estimated 4,093 treatments per year were administered by 2,238 beekeepers with the aim of reducing *Varroa* numbers. The majority (70%) are treated once or twice a year using a single chemical method. The most popular chemical treatment method is oxalic acid, followed by commercially produced thymol, and amitraz. The current study found 78% of beekeepers use chemical treatments (oxalic acid, thymol etc), 3% use biomechanical methods (drone brood removal, sugar dusting etc) and less than 1% use other methods (rhubarb leaves, etc.). Overall, 80% use a single method of treatment and only 20% use a combination of treatment methods. # Discussion Based on 2,856 beekeeper responses from 243 UK beekeeping areas, the proportion of beekeepers not-treating ranged between 21–28% e.g., if the beekeeper can only choose between treatment or treatment-free (21%), or 28% of beekeepers if you include those that had not treated in the last year. Those not-treating for over 6 years represented 6% of responders. Based on this survey, that would mean 1,800 of the 30,000 estimated UK beekeepers are truly treatment-free having not treated their colonies for *Varroa* for six years or more. Around 100 of these beekeepers are in a single region in North Wales with many belonging to the Lleyn and Eifionydd Beekeepers’ Association. Likewise, in Swindon, a small beekeeper group have kept treatment-free colonies since 1995. In the UK there are active and growing “treatment-free” communities. For instance, the Westerham beekeepers are approaching their 5^th year of becoming treatment free and starting to bring in neighboring bee clubs. Finally, the ‘Natural Beekeeping Trust’ was established in 2009 and now has links to 35 like-minded groups. We had 18 responses from their members all falling into the ‘never-treated group’, along with 16 beekeepers that failed to provide an association. These ‘natural beekeepers’ and those not-treating for *Varroa* have not always been welcomed by those adopting treatment regimens. The ‘never-treated group’ represent 4% of the respondents but the length of time they have not been treating is unknow. These treatment-free beekeepers occur across all colony group sizes. The annual BBKA (British Beekeepers Association) overwinter survival survey found that in 2020/21, 27% and 37% of 2,950 randomly selected members from a pool of 26,407 did not treat from August to September or from October to April respectively. In 2019/20 the BBKA survey found 25% and 38% of treatment-free beekeeping during Aug-Sept and Oct-Apr, respectively. In both years the BBKA survey found treatment-free beekeeping was present in all English regions, as was found by this study. This is important, since the two surveys used different sampling methods. The BBKA and this study surveyed similar numbers of beekeepers, except the BBKA survey targets members randomly selected from their membership each year, while this study approaches the beekeepers via their associations using email, so they are self-selecting. Despite this the two outcomes are similar in many aspects indicating that our survey has not disproportionally been returned by treatment-free beekeepers that may be more active online. Nonetheless, the majority (72–79%) of UK beekeepers are still treating their colonies to control *Varroa* numbers. We found beekeepers in the UK are predominantly using a single or bi-annual chemical treatment regime. The order of popularity starting with the most common is oxalic acid, thymol, amitraz and formic acid. Only 20% of beekeepers in this study are adopting a combination of methods approach. The popular chemical *Varroa* treatments like formic and oxalic acids reported in this study and other UK surveys are also the preferred methods used in Europe and the USA. These compounds have a high efficacy but without the stigma of synthetic compounds or mite resistance, which could explain why many beekeepers are choosing to adopt these methods. However, the impact of these “natural” treatments should not be ignored. Whilst thymol is thought to be unproblematic at temperatures between 5°C-9°C, high mortality levels have been observed when temperatures exceed 27°C. The success of any treatment method or treatment-free beekeeping is determined by their overwinter losses. Over the last 14 years in the UK over-wintering annual colony losses have ranged from over 30% to 8% with an average of 18%. This falls to 14% when the three spikes related to unusually cold and wet winters are removed (data derived from BBKA annual survey data collected by D. Aston that is based on 2,500–3,500 beekeepers each year \[e.g. 6,25\]). Principal causes of colony losses reported by beekeepers in 2021 were queen- related problems (24%), isolation starvation (21%), weather-related 17%, and *Varroa* was just 4% reflecting the situation that the mite is no longer considered a major problem by beekeepers currently in the UK. However, *Varroa* could be contributing to some of the other colony losses indirectly. In 2019–2020 the BBKA annual survey reported both losses in the region as well as the percentage of colonies not treated in each region. Beekeeper observations by the Oxford shire Natural Beekeeping Association (ONBA) and the treatment-free group in Northwest Wales both found their colony survival rates were on par with those from the BBKA, but independent scientific studies are required to see if *Varroa*-resistance effects the rate of colony losses. One of the dangers of switching over to an intervention free-treatment regime involves a period of high winter losses whilst the bees develop resistance. This is due to re-invasion from collapsing colonies to surviving surrounding colonies or by visits from infested non-natal bees. Several strategies have been adopted at both local and national level to encourage mite-resistant honeybees. Catching free-living swarms from locations where they appear to have persisted for many years has been a successful strategy in the UK and Hawaii. Others have reduced the frequency of mite-treatments by selectively treating colonies with high mite levels, or to use less efficient biomechanical methods only when required. Already many countries (South Africa, Brazil, Mexico, Cuba etc), along with beekeepers in the UK and elsewhere have been able to stop treating as their honeybees have learnt to detect mite infested cells and remove the pupa to prevent mite-reproduction, which leads to decreased mite fertility and population size. However, the majority (72–79%) of UK beekeepers are still treating and it will be many years before most beekeepers in the UK and elsewhere can stop treating for *Varroa*. In the USA a recent survey found only around 63 (3%) of 2275 respondents stated no advantages to *Varroa* management and 92% of this group did not treat for *Varroa*, a very different situation than found currently in the UK. Most beekeepers in the Northern hemisphere have long wished for a silver bullet for the *Varroa* problem, however, it turns out that the bullet is their own bees. Beekeepers just need to give their bees time to develop mite resistant as has been done so successfully elsewhere in the world. # Supporting information We thank all the UK beekeepers that submitted a survey, help from their associations and G. Webb of University of Salford for proofreading the original submission. The original data is available on dryad at <https://doi.org/10.5061/dryad.xksn02vkn>. 10.1371/journal.pone.0281130.r001 Decision Letter 0 Chaline Nicolas Academic Editor 2023 Nicolas Chaline This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 29 May 2022 PONE-D-22-05221A survey of UK Beekeeper’s Varroa treatment habitsPLOS ONE Dear Dr. Martin, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Dear Steve,Although all referees thought your paper and its objectives interesting, they felt that some important informations were missing with regards to the methods and analyses used and whether this would influence the results and conclusions of the paper. One main issue is how colony loss was assessed (if it was) and how numbers were extracted from the survey. Please attend to the comments and explicit your answer in a rebuttal lettercheers Please submit your revised manuscript by Jul 13 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Nicolas Chaline Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <h ttps://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_b ody.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. If you are reporting a retrospective study of medical records or archived samples, please ensure that you have discussed whether all data were fully anonymized before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent. If patients provided informed written consent to have data from their medical records used in research, please include this information \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: I Don't Know Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: No Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The submitted survey of UK beekeeper´s varroa treatment habits presents interesting new data and fills a gap in our knowledge on recent beekeeping practice. The study is well presented with a clear description of the data sampling and evaluation procedure. The discussion picks up relevant links to related surveys and the establishment of mite resistance in bee populations depending on the treatment strategies. The length of the different chapters is appropriate and well balanced. Some confusion may arise about the number of respondents for the data in table 3. The table reports on a total of 2,868 responses from 2,897 responses in general (L116). If 25 responses from the Isle of Man have been excluded, some of the 16 participants with zero colonies (L146f) seem to be included here. However, this wouldn´t make any sense for my understanding. The authors should give a precise description on this. L102 refers to Fig S1 which is missing in the supplied manuscript. In L120 Table S1 is mentioned but it should be Table 5 as far as I understand. In L 264 the second “by” has to be erased. Reviewer \#2: The paper has the potential to be interesting, but is currently missing the access to the data and the supplemental information, so difficult to assess. Also the conclusions are very broad and the data presented don't support the conclusions the authors make, as the survey did not include any assessment of colony losses. Without distinguishing what the actual loss rates of the treatment free beekeepers have, it is not possible to state that the bees are actually resistant to varroa or have a high tendency for survival. The higher colony numbers actual suggest the opposite. Also there are a number of odd sentence structures and grammatical errors. I have made note of the mores specific comments in a separate document. Reviewer \#3: The study aims to estimate the number of beekeepers in the UK who are not treating their colonies for varroa, and to estimate whether the existence of untreated and resistant colonies in the UK could be assumed. Information on beekeeping practices are currently limited in the scientific literature, and this question and the data the study provides are really relevant and valuable. The article is generally clear and well written. The methodological choices are globally sound and with a high number of respondents to the survey, which is important to note. Still some information should be added in the methodological section, and the limits induced by some methodological choices should be identified and discussed as they can have important impacts on the interpretation of the results. Also, the findings should be discussed regarding the scientific literature on the subject. For these reasons I recommend major revisions. Major comments Some methodological limits should be identified and discussed, especially i) the possible influence of the sampling through the beekeeping associations (are these associations involved in prescribing varroa treatments?) and ii) the lack of information that the survey provides about the colony losses and about the beginning year of the beekeeping activity. Without this last information, the share of beekeepers who have “never treated” cannot be interpreted as it is done currently regarding the possible existence of surviving colonies (those beekeepers could have started recently and their colonies may not have faced this absence of treatment for varroa for a long time). It should not be suggested that these colonies had never known any treatment (it could be the case, but the data of the survey do not allow to conclude on this question). The existence of colonies surviving without treatment for varroa and the possible losses that the absence of treatment may induce should also be discussed regarding the scientific literature on these questions. Other comments L. 57: “due to almost universal treatment by beekeepers”: this statement should be either nuanced or supported by a reference about American beekeepers’ practices. E.g. see Thoms et al. (2019 - <https://doi.org/10.1007/s13280-018-1130-z>), which indeed reveals a high percentage of American backyard beekeepers reporting not to treat for varroa. L. 100-105: If I understand correctly, there was no question about the colony survival (cf. L100-105). It would have been a valuable information, as the emergence of resistant colonies is only possible if the non-treated colonies survive (in the case of backyard beekeepers replacing regularly their colonies that died from varroa, these non-treated colonies would not be more resistant than other). L. 106-113. As the beekeeping associations who forward the survey to their adherents play a major role in the sampling, it would be useful to provide more information about these associations (as such beekeeper associations and their role can differ from one association or one country to the other). Especially, can these associations interfere with their adherents’ treatment choices (e.g. by advising or not advising to treat for varroa)? L. 123. One of the main analysis choices is the investigation for a possible spatial pattern. As many other factors could possibly influence beekeepers’ choice about their treatment for varroa (beekeeping experience, age, etc), this hypothesis of a spatial pattern could be explained and justified. L. 130. Some biomechanicals and natural methods that were reported by beekeepers (cf. Table 5) are not really efficient against varroa. As the study focuses on the possible existence of untreat / resistant colonies and not on the choice of beekeepers to treat or not to treat (reasons for such choices, etc), these “low- efficacy” treatments could have been considered jointly with the treatment-free group, or a third variant for the response variable could have been considered. So the choice to gather all the treatments together in a single response variable could be explained. L. 133. As one of the questions of the study is the existence of untreated and possibly resistant colonies, it would be useful to know how many colonies in total these 2,872 beekeepers manage and what percentage of the total number of colonies in the UK it represents. L. 141-144. This passage is not very clear, especially what is supposed to “explained why there was significant increase in treatment-free beekeepers as the numbers of colonies they managed increased”. It would be useful to rephrase or to complete it. L. 146 / Table 1. The third column heading is not very clear: does it represent the percentage of colonies of the group (1-5,…) which are not treated? So 73% of the colonies of the 16-30 group are not treated? An additional column providing the percentage of treatment-free beekeepers in each of the groups would be welcome. Besides, the final percentage of untreated colonies (gathering all groups) should be added. L. 160-162. Grouping the beekeepers who indicated that they have "never treated" with those indicating a specific number of years since the last treatment (here 10 years and more, or 6 years and more L. 201) is questionable as the date of installation of the beekeepers was not in the survey and is not known. Some of them could have started beekeeping recently, and the fact that they have “never treated” for varroa does not presume that their colonies had to face a significant period without treatment. This group of beekeepers with an unknown number of years without treatment should be considered separately and the confusion with the groups where a long treatment-free period is known should be avoided. L. 163-164. The fact than some association gather more than 75% of treatment- free beekeepers raises questions about the role of these associations, and about the possible exchanges related to varroa treatments that its members may have. Even if it was not the objective of the study to understand the determinants of the absence of treatment, it would be interesting to discuss the possible role of associations on this point given their central place in the survey sample. L. 178. “the majority are treated”: the majority of colonies? L. 180-184. It would be useful to add the global percentage of beekeepers not using any treatment and to more clearly distinguish if the percentages given for the types of treatments are exclusive of each other or not. E.g. can the 3% of beekeepers using biomechanical methods be also in the 78% using chemical? Or are they only using biomechanicals methods? L. 201. See comment on L. 160-162 L. 202. The extrapolation of the study results to the 30,000 estimated UK beekeepers should be supported by a discussion about the representativity of the sample. That joins the question of giving information about the beekeeping associations (see comment on L 103-115), and about their representativeness (are they some apicultural practices which may differ between members of these associations and beekeepers who are not members, e.g. if the associations are involved in prescribing treatments for varroa? Are the members of these associations representing a large share of the UK beekeepers?) L. 257-260. As varroa is usually considered in scientific and technical literature as an important factor of colony losses, some scientific references about the place of varroa in colony losses need to be added and the technical references which are the only provided references should be discussed regarding these scientific references. L. 267. It would be useful to precise “in Hawaii” and not only “in the USA” as the context of an island can be specific. Fi. 1. A. A legend with the number of respondents corresponding to the different pie sizes is needed here. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0281130.r002 Author response to Decision Letter 0 1 Jul 2022 Data is said to be publicly available, but no link included. Same goes for the supplemental information. THE DATA WILL BE PLACED ON DRYAD IF ACCEPTED Please apply the use of commas for numbers consistently. For example, compare line 35 and line 38. Also Varroa is not consistently italicized. DONE I don’t understand why the feral population of bees is mentioned in line 43. You surveyed beekeepers with managed colonies. This has no implications for the feral bees and no evidence is presented that the UK has truly feral colonies. DELETED ‘FERAL COLONIES’ AS THE IDEA WAS THAT MORE VARROA-RESISTANT MANAGED COLONIES WILL LEAD TO MORE VARROA-RESISTANT FERAL COLONIES BUT WE AGREE WITH THE COMMENT TO DELETED. Line 56-57, awkward sentence structure. Unclear what remained high. ENTIRE SECTION REWORDED Line 60, based used twice REPLACED ONE WITH ESTABLISHED Line 65-66, not clear why an IPM strategy would be linked to long-term resistance and low mite levels. IPM, as I understand it, is threshold base and involves monitoring and then treating appropriately. How does this encourage low mite levels? And why would this lead to long-term resistance? Needs explanation and evidence. THIS IS JUST A REMOTE POSSIBILTY SO THIS STATEMENT HAS BEEN DELETED Line 102: I have no access to the Figure S1 Line 120: I have no access to Table S1 SORRY THIS HAS OCCURRED Line 201-203, that assumes that the survey reached and was completed by a representative wedge of the beekeeping population. Many treatment-free beekeepers tend to be younger and more active online, so this may have skewed the responses. These caveats should be mentioned. THE COMMENT ABOVE ONLINE ACTIVE HAS NOW BEEN ADDED TO THE RELEVENT PART OF THE DISCUSSION L 239-241. Line 218-219: This is not surprising at all. Most treatment free beekeepers in the United States expect annual losses of 50%. Thus, in order to rebuild the following year, they must keep more colonies in reserve. I am disappointed this survey didn’t ask for self-reporting of annual loss rates or honey production per colony. The high level of treatment-free suggests that beekeepers and the much higher colony numbers suggests that the resistance is probably not enough for low annual losses. WE DID NOT INCLUDE COLONY DEATHS IN OUR SURVEY SINCE THIS IS ALREADY COVERED BY THE LONG RUNNING BBKA SURVEY DATA FROM WHICH WE HAVE NOW INCULDED(see Fig S2) AND WE HAVE EXPANDED THE PARAGRAPHY COMPAIRING COLONY LOSS RATES FROM UK TREATMENT-FREE BEEKEEPERS. MY PERSONAL EXPERINCE WORKING WORK WITH SEVERAL TREATMENT-FREE BEEKEEPERS THEIR LOSSES ARE SIMILAR TO BEEEKEEPERS THAT ARE TREATING. WE DID NOT ASK ABOUT HONEY PRODUCTION AS IN THE UK THIS VARIES GREATLY DUE TO LOCATION AND MICRO-CLIMATE. CLEARLY THERE ARE BIG DIFFERENCES BETWEEN THE SITUATION IN THE UK AND USA IN NUMBER OF TREATMENT FREE BEEKEEPERS THIS IS NOW MENTION IN THE LAST SECTION AS WELL AS THE (Thoms et al) Line 235-240: This makes the assumption that IPM requires multiple treatments per year. IPM requires monitoring and treating accordingly when a pest level rises above a certain point. Also, many do not consider drone trapping a treatment. As I don’t have an actual copy of the survey questions, I cannot assess if there may have been misunderstandings in the responses. POOR PLACEMENT OF SENTENCE SO RE-STRUCTED THIS SECTION AND COMMENTED THAT ONLY VERY BASIC MITE MONTIROING TAKES PLACE. Line 253-255: This needs more clarification. Is the 4% what beekeepers reported was specifically due to varroa? This would then be an underrepresentation of actual winter losses due to varroa. Numerous surveys have shown that when varroa levels are high, winter losses increase, even though beekeepers blame other factors for the loss. Many beekeepers with colony winter losses don’t know what killed their colony, but beekeepers with more experience can often diagnose the contracted brood nest area and varroa frass. WE HAVE GIVEN MORE CLARIFICATION ON THE PRINCIPAL CAUSES OF COLONY LOSSES IN THE UK Line 254-255: Why is the low winter loss attributed to treatment-free beekeeping. There could be other explanations, such as very active beekeeping groups that teach how to monitor and treat effectively. The evidence presented does not warrant this statement, as there is no segregation of losses by those treating and those not treating presented. WE AGREE THAT NOT ENOUGH EVIDENCE IS PRESENT AT THE MOMENT SO THE SENTENCE HAS BEEN DELETED Line 256-260: These statements require the reader to go to the original articles and assess the numbers of colonies surveyed. Important information such as colonies assessed in the survey should be included with the paraphrased results, especially as these are not peer-reviewed studies but association results and so the statistics have not been evaluated. The way they are presented in this paper, the reader assumes they are peer-reviewed papers, unless the references are checked in detail. WE AGREE SO ADDED IN ADDITIONAL DATA/INFO BOTH IN THE TEXT AND SUPP DATA Line 276-279: This is a very broad statement and is not justified by the results presented, as no losses were calculated in the survey. WE HAVE SHOWEN USING BBKA SURVEY BASED ON 2,613 BEEKEEPERS THAT THERE IS NO SIG RELATIONSHIP BETWEEN %TREATMENT AND %COLONY LOSSES AT A REGIONAL LEVEL IN THE UK. WE HAVE ALSO ADDED IN MORE REFRENCES TO SUPPORT OUR BROAD STATEMENT. FINALLY THE PAPER HAS BEEN CAREFULLY PROOFED NOW 5\. Review Comments to the Author Reviewer \#1: The submitted survey of UK beekeeper´s varroa treatment habits presents interesting new data and fills a gap in our knowledge on recent beekeeping practice. The study is well presented with a clear description of the data sampling and evaluation procedure. The discussion picks up relevant links to related surveys and the establishment of mite resistance in bee populations depending on the treatment strategies. The length of the different chapters is appropriate and well balanced. Some confusion may arise about the number of respondents for the data in table 3. The table reports on a total of 2,868 responses from 2,897 responses in general (L116). If 25 responses from the Isle of Man have been excluded, some of the 16 participants with zero colonies (L146f) seem to be included here. However, this wouldn´t make any sense for my understanding. The authors should give a precise description on this. THE 25 RESPONSES FROM THE ISLE OF MAN WERE EXCLUDED FROM THE 2897 (= 2872) WHICH IS NOW STATED IN THE METHODS. IN TABLE 1 THE 16 BEEKEEPERS WITH ZERO COLONIES WHERE EXCLUDED WHICH LEAVES 2856 BEEKEEPERS AS STATED IN TABLE 1. WE HAVE REWRITTEN THE METODS TO INCLUDE ALL SAMPLES EXCULDED L102 refers to Fig S1 which is missing in the supplied manuscript. (Sorry our error, CORRECTED) In L120 Table S1 is mentioned but it should be Table 5 as far as I understand. (deleted) In L 264 the second “by” has to be erased. Deleted Reviewer \#2: The paper has the potential to be interesting but is currently missing the access to the data and the supplemental information, so difficult to assess. Also, the conclusions are very broad, and the data presented don't support the conclusions the authors make, as the survey did not include any assessment of colony losses. Without distinguishing what the actual loss rates of the treatment free beekeepers have, it is not possible to state that the bees are resistant to varroa or have a high tendency for survival. The higher colony numbers actual suggests the opposite. AS MENTIONED ABOVE WE HAVE A PARAGRAPH COMPARING LOSSES IN TREATMENT-FREE POPULATIONS VS THE NATIONAL AVERAGE. ANOTHER SURVEY COLLECTS LOSS DATA BUT NOT AIMED AT SEPARATING OUT THE TREATED AND NOT TREATED. WE HAVE ALSO INCLUDED THE MAIN REASONS IN ENGLAND & WALES FOR COLONY LOSSES AND VARROA ACCOUNTS FOR ONLY 4% ALTOUGH WE AGRREE THIS MAY BEEN UNDERESTIMATED BY THE BEEKEEPERS. Also there are a number of odd sentence structures and grammatical errors. I have made note of the mores specific comments in a separate document. ALL SHOULD BE FIXED NOW Reviewer \#3: The study aims to estimate the number of beekeepers in the UK who are not treating their colonies for varroa, and to estimate whether the existence of untreated and resistant colonies in the UK could be assumed. Information on beekeeping practices is currently limited in the scientific literature, and this question and the data the study provides are relevant and valuable. The article is generally clear and well written. The methodological choices are globally sound and with a high number of respondents to the survey, which is important to note. Still some information should be added in the methodological section, and the limits induced by some methodological choices should be identified and discussed as they can have important impacts on the interpretation of the results. Also, the findings should be discussed regarding the scientific literature on the subject. For these reasons I recommend major revisions. Major comments Some methodological limits should be identified and discussed, especially i) the possible influence of the sampling through the beekeeping associations (are these associations involved in prescribing varroa treatments?) THERE IS NO PRESCRIPTION OF TREATMENTS VIA ASSOCIATIONS SINCE TREATMENT ADVICE COMES CENTRALLY FROM THE NATIONAL BEE UNIT & THE MAP INDICATES WE HAD A VERY WIDE SPREAD OF RESPONSE and ii) the lack of information that the survey provides about the colony losses and about the beginning year of the beekeeping activity. Without this last information, the share of beekeepers who have “never treated” cannot be interpreted as it is done currently regarding the possible existence of surviving colonies (those beekeepers could have started recently, and their colonies may not have faced this absence of treatment for varroa for a long time). It should not be suggested that these colonies had never known any treatment (it could be the case, but the data of the survey do not allow to conclude on this question). THERE ARE A GROWING GROUP OF BEEEKEEPERS IN THE UK THAT NEVER TREAT SUCH AS OXFORDSHIRE NATURAL BEEKEEPING ASSOCIATION AND THEIR LOSSES FOR TWO WINTERS WERE SIMILAR TO THE NATIONAL AVERAGE. THIS INFORMATION IS PROVIDED IN THE MS AS WELL AS A NEW SUPPLEMENTAL FIGURE SHOWING THE NATIONAL AVERAGE COLONY LOSSES FROM 2007 TO 2021. The existence of colonies surviving without treatment for varroa and the possible losses that the absence of treatment may induce should also be discussed regarding the scientific literature on these questions. WE HAVE EXPANDED THE SECTION ON COLONY LOSSESS IN THE UK AS REQUESTED Other comments L. 57: “due to almost universal treatment by beekeepers”: this statement should be either nuanced or supported by a reference about American beekeepers’ practices. E.g. see Thoms et al. (2019 - <https://doi.org/10.1007/s13280-018-1130-z>), which indeed reveals a high percentage of American backyard beekeepers reporting not to treat for varroa. I CHECKED THOMS ET AL AND ALOUGHT MOST TREATMENT SKEPICS DO NOT TREAT THIS GROUP MADE UP LESS THAN 3% OF THE SURVEY EG (63 VS 2276). THUS, IT APPEARS THAT THERE ARE MUCH FEWER TREATMENT-FREE BEEKEEPERS IN THE USA THAN UK. HOWEVER, WE HAVE CHANGED ‘UNIVERSAL’ TO ‘HIGH’ WHICH IS MORE ACCURATE AND CITED THIS PAPER AT THE END OF THE DISCUSSION. L. 100-105: If I understand correctly, there was no question about the colony survival (cf. L100-105). It would have been a valuable information, as the emergence of resistant colonies is only possible if the non-treated colonies survive (in the case of backyard beekeepers replacing regularly their colonies that died from varroa, these non-treated colonies would not be more resistant than other). SEE EALIER COMMENTS ABOUT COLONY LOSSES L. 106-113. As the beekeeping associations who forward the survey to their adherents play a major role in the sampling, it would be useful to provide more information about these associations (as such beekeeper associations and their role can differ from one association or one country to the other). Especially, can these associations interfere with their adherents’ treatment choices (e.g., by advising or not advising to treat for varroa)? THIS IS SOMETHING THAT IS NOT DONE THE ASSOICATIONS ONLY FORWARED THE SURVEY LINK TO ALL ITS MEMBERS USING THERE (CONFEDENTIAL DATABASE). IF THESE TYPE OF BEHAVIOR DID OCCUR, WE WOULD HAVE PICKED IT UP IN THE ANAYLSIS. FURTHERMORE, THE ASSOICATIONS THAT SEND OUT THE LINK HAVE NO WAY OF SEEING WHAT THEIR MEMBERS WORTE. L. 123. One of the main analysis choices is the investigation for a possible spatial pattern. As many other factors could possibly influence beekeepers’ choice about their treatment for varroa (beekeeping experience, age, etc), this hypothesis of a spatial pattern could be explained and justified. EACH ASSOICATION HAS A WIDE RANGE OF AGES, GENDERS, AND EXPERIENCE TO REQUEST THIS TYPE OF DATA AND CONSDIER THEM IN THE ANAYLSIS LIES WAY BEYOND THE SCOPE OF THIS STUDY. L. 130. Some biomechanical and natural methods that were reported by beekeepers (cf. Table 5) are not efficient against varroa. As the study focuses on the possible existence of untreated / resistant colonies and not on the choice of beekeepers to treat or not to treat (reasons for such choices, etc), these “low- efficacy” treatments could have been considered jointly with the treatment-free group, or a third variant for the response variable could have been considered. So, the choice to gather all the treatments together in a single response variable could be explained. THE REVIEWER IS CORRECT IN THAT THE DIFFERENT TREATMENTS RANGE WIDLY IN THEIR EFFICACY’S. TO ERR ON THE SIDE OF CAUTION WE HAD TO MAKE IT VERY CLEAR TO UNDERSTAND WHAT ‘TREATMENT-FREE’ MEANT. OTHERWISE, WE COULD HAVE RUN THE PROBLEM OF BEEKEEPERS DECIDING WHAT THEY CONSIDER IS TREATMENT FREE. WE AGREE THE LOW- EFFICACY TREEATMENTS COULD BEE ALLOWING NATURAL RESISTANT TO DEVELOP AND THIS ONE OPTION FOR BEEKEEPERS TO BECOME TREATMENT FREE. IN THE FUTURE IT WOULD BE INTRESTING TO SEE IF THE USAGE OF THESE LOW-EFFICACY METHODS INCREASED. L. 133. As one of the questions of the study is the existence of untreated and possibly resistant colonies, it would be useful to know how many colonies in total these 2,872 beekeepers manage and what percentage of the total number of colonies in the UK it represents. WE HAVE NOW ADDED IN A SECTION AT THE START OF THE RESULTS SHOWING THAT SURVEY ALSO COVERED ABOUT 10% OF THE UK COLONIES L. 141-144. This passage is not very clear, especially what is supposed to “explained why there was significant increase in treatment-free beekeepers as the numbers of colonies they managed increased”. It would be useful to rephrase or to complete it. THIS SECTION HAS BEEEN REEWORDED L. 146 / Table 1. The third column heading is not very clear: does it represent the percentage of colonies of the group (1-5,…) which are not treated? So, 73% of the colonies of the 16-30 group are not treated? An additional column providing the percentage of treatment-free beekeepers in each of the groups would be welcome. Besides, the final percentage of untreated colonies (gathering all groups) should be added. WE RE-CHECKING THE DATA IN TABLE 1 WE DISCOVER AN ERROR SO THANKS FOR THAT. WE HAVE NOW RE-DONE TABLE 1 AND TABLE 2 SO SHOULD BE MUCH CLEARER AS REQUESTED BY THE REVIEWER SORRY ITS BEEEKEEPERS NOT COLONIES SO WE HAVE REWORDED IT. L. 160-162. Grouping the beekeepers who indicated that they have "never treated" with those indicating a specific number of years since the last treatment (here 10 years and more, or 6 years and more L. 201) is questionable as the date of installation of the beekeepers was not in the survey and is not known. Some of them could have started beekeeping recently, and the fact that they have “never treated” for varroa does not presume that their colonies had to face a significant period without treatment. This group of beekeepers with an unknown number of years without treatment should be considered separately and the confusion with the groups where a long treatment-free period is known should be avoided. IN TABLE 2 WE HAVE SEPRATED OUT THE NEVER ADMINISTRED GROUP SO THEY ARE CONSIDER SEPRATLY. THE MANY OF THESE BEEKEPERS BELONG TO THE OXFORD NATURAL BEEKEEPING. THIS GROUP HAS BEEN AROUND FOR 10 YEARS AND WE PROVIDE COLONY LOSS DATA FROM THIS GROUP AND COMPARE AGAINST THE NATIONAL AVERGAE AND ITS VERY SIMILAR. L. 163-164. The fact than some association gather more than 75% of treatment- free beekeepers raises questions about the role of these associations, and about the possible exchanges related to varroa treatments that its members may have. Even if it was not the objective of the study to understand the determinants of the absence of treatment, it would be interesting to discuss the possible role of associations on this point given their central place in the survey sample. AS PREVIOUSLY MENTIONED VARROA TREATMEENTS ARE RECOMMENDED BY THE GOVERMENTS NATIONAL BEE UNIT. THE TREATMENT-FREE BEEKEEPERS IN MOST ASSOICATIONS DO NOT MENTION THAT THEY DO NOT TREAT AS THEY ARE IN MINORITY HENCE WHY WE DID THE SURVEY TO GET A BETTER ESITATMTION OF THEE THE NUMBER OF TREATMENT FREE BEEKEEPERS. L. 178. “the majority are treated”: the majority of colonies? YES, SO CHANGED L. 180-184. It would be useful to add the global percentage of beekeepers not using any treatment and to more clearly distinguish if the percentages given for the types of treatments are exclusive of each other or not. E.g., can the 3% of beekeepers using biomechanical methods be also in the 78% using chemical? Or are they only using biomechanical methods? THIS DATA DOES NOT CURRENTLY EXIST THE Thoms et al survey and this study are some of the first attempts to collect such data. L. 201. See comment on L. 160-162 L. 202. The extrapolation of the study results to the 30,000 estimated UK beekeepers should be supported by a discussion about the representativity of the sample. That joins the question of giving information about the beekeeping associations (see comment on L 103-115), and about their representativeness (are they some apicultural practices which may differ between members of these associations and beekeepers who are not members, e.g., if the associations are involved in prescribing treatments for varroa? Are the members of these associations representing a large share of the UK beekeepers?) IN THE UK THERE ARE AROUND 30,000 AMATEUR BEEKEEPERS BUT ONLY 400 PROFESSIONALS THAT BELONG TO THE BEE FAMERS SOCIETY. THIS IS NOW MENTIONED IN THE METHODS. L. 257-260. As varroa is usually considered in scientific and technical literature as an important factor of colony losses, some scientific references about the place of varroa in colony losses need to be added and the technical references which are the only provided references should be discussed regarding these scientific references. THIS IS VERY TRUE PARTICULAR IN THE NORTHERN HEMISPHERE BUT IN MANY COUNTRIES (E.G., AFRICA, BRAZIL-MEXICO, CUBA) VARROA IS NOW NOT CONSIDERED A PROBLEM SINCE THE BEES ARE ABLE TO KEEP THE INFESTATION LEVEL DOWN BELOW 10-4% DEPENDING ON THE POPULATION. HENCE IN THE EYES OF WELL-ESTABLISHED (6YRS+) TREATMENT-FREE BEEKEEPERS VARROA IS NOT CONSIDERED AS A SERIOUS PROBLEM. THE SAME CAN BE SAID BY BEEKEEPERS THAT REGULAR TREAT THEIR VARROA POPULATIONS. WE HAVE NOW ADDED IN A SHORT SECTION SHOWING IN ONE SURVEY THE BEEKEEPERS REPORTED THAT ONLY 4% OF THEIR DEATHS WAS DIRECTLY CAUSED BY VARROA. L. 267. It would be useful to precise “in Hawaii” and not only “in the USA” as the context of an island can be specific. CHANGED AS SUGGESTED Fi. 1. A. A legend with the number of respondents corresponding to the different pie sizes is needed here. THIS HAS NOW BEEN DONE 10.1371/journal.pone.0281130.r003 Decision Letter 1 Corby-Harris Vanessa Academic Editor 2023 Vanessa Corby-Harris This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Dec 2022 PONE-D-22-05221R1A survey of UK Beekeeper’s Varroa treatment habitsPLOS ONE Dear Dr. Martin, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 21 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Vanessa Corby-Harris, Ph.D. Academic Editor PLOS ONE Additional Editor Comments: Dear authors, (First, I apologize for my lengthy response. This manuscript was re-assigned to me and so I was reading it for the first time.) This manuscript summarizes a survey of whether UK beekeepers treat for mites or not. Overall, I think that this manuscript could be a good fit for PLoS One based on the journal’s publication criteria, with one exception noted below. I also think that the general idea of treatment-free beekeeping is interesting and that it would be a useful addition to the literature. However, I do agree with reviewer 2 that the manuscript is somewhat narrow in scope and rather descriptive, and would be better suited for a more specialized or beekeeper- friendly journal. However, many of those journals are pay walled and so this could be a better way to more widely disseminate the authors’ results. This manuscript has been through one round of review and was much improved after the reviewer comments were addressed. You will see that the reviewers have some lingering concerns that still need to be addressed. Please address each comment in your revision. Specifically, reviewers 2 and 3 have multiple concerns, such as the rationale behind several hypotheses, statements that need supporting citations, the level of detail in the methods, and concerns about the survey itself. There are also several instances where the conclusions are not supported by the data (ex. treatment-free beekeeping is becoming more common). Please revise those statements so that the manuscript meets the PLoS One publication criteria, specifically criteria \#4 that “Conclusions are presented in an appropriate fashion and are supported by the data”. I also saw several small grammatical errors and had a few minor questions. I apologize that these are coming up now, but they need to be addressed. 1\) There are several sentences that start with “Although”, “Thus”, “Whereas”. Please revise these sentences because they are often grammatically incorrect. L56 – decades L59 – I wouldn’t say they are highly effective. For example, most or all can’t get into the brood cell, where mite reproduction happens. Revise?  L61-62 – watch what you capitalize (ex. oxalic?) L65 – take out the comma after reference 9 L83 – is this resistance or tolerance? L103 – six questions: L134 – isn’t “treatment status” the response variable? L140 – widely or wildly? L141-143 – this should be in the methods and needs more detail (see reviewer 2 comment) Figure 1 – it could be helpful to remind the reader that the 158 area associations in the spatial study had ≥5 respondents, so associations with \<5 respondents were not colored in black. At least that was my interpretation of what you said in the text. L160-161 – Is this an important or real trend if you don’t see a difference between these groups and the larger groups? Can you explain or speculate why you didn’t see differences between these small and medium colony number groups and the larger group? Table 2 – had no effect (see legend text); also should you have a range for the colony number group estimate since you had four groups? I like Tables 4 and 5, although I agree with reviewer 3 that there are issues with the “treatment-free” group. Some of those natural methods were interesting (rhubarb??). And how does queen trapping differ from a brood break? Just wondering, really. L218-219 – This is repeating part of my point above, but I agree that this group could also include beekeepers that have kept bees for only a few (\<6) years and so combining them with the beekeepers that have been treatment-free for 6 years or more could be misleading. It seems that there was no criteria that respondents have kept bees for 6 or more years. L275-276 – revise the sentence where it says weather-related and Varroa – should this be split up somehow? L281- that’s not a lot of data (N=6) to base a conclusion on. I am also of the opinion that the discussion should not include too many results/statistics, so this is something to potentially omit. It is appropriate to talk about it, just not as something that you tested using data and stats. L307-310 – revise this sentence, it is too long. I like the general idea of the sentence though! \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed Reviewer \#3: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes Reviewer \#3: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: please correct the following typos in the revised document: L168:...had no effect... L270: overwinter losses. Over the last... L307:...than found currently in the UK. Regarding your statement in L249-251I suggest a more careful wording as the BBKA report is restricted on the treatment period between August and September while OA as the most common treatment in UK should be mainly used on broodless colonies during winter. Reviewer \#2: The paper is well written, but the survey lacks depth. The authors wished to investigate how wide spread treatment free beekeeping is within the UK. They did this by conducting an online survey with limited questions that could be completed within five minutes. It would have been far more interesting if they then did a follow up with the beekeepers who qualified as treatment-free to learn more about their beekeeping practices. I feel the current results presented are more appropriate for a bee journal than a rigorous scientific publication. For specific comments regarding the manuscript, see attached PDF. Reviewer \#3: SPECIFIC COMMENTS L. 72 References are still missing to support this assertion L 127-132 The choice to explore a potential spatial pattern (rather than the other possible explanatory factors for the varroa management) should still be explained: why is this hypothesis made? L 233. References or studies on the number of treatment free beekeepers a few years ago would be appropriate here to support the assertion that the number of treatment-free beekeepers is growing. L 240-244 The BBKA studies the authors refer to are about a specific time of the year: even if some of the beekeepers who do not treat at this time of the year could be treatment-free beekeepers as the authors underline, some of them could also treat at another time of the year. Thus, the results of these BBKA studies should not be directly used to compare with the percentage of treatment-free beekeepers which was estimated in the study, as they do not estimate the same thing. L.279 It is unclear if the percentage of colonies not treated here refers to the “august to September non treated” (cf. L 241) or to year-long non treated (and the BBKA study the authors refer to does not seem to be available online to find this information). It would be appropriate to precise this information. GENERAL COMMENTS This study provides useful information about the number of treatment-free beekeepers in the UK. Some relevant changes have been made to the paper, considering the reviewers’ feedback on the first manuscript. Still, the authors did not respond to some of the feedback, in particular regarding their methodological choices, which were not always explained (e.g. why explore a spatial pattern?) and regarding the limits of these methodological choices and therefore of the study. One of the major methodological issues is the choice to gather beekeepers who indicated that they have “never treated against varroa” with beekeepers who have not treated for six years or more. This choice relies on the hypothesis that beekeepers who have “never treated” manage colonies which have not been treated for a significant period of time (6 years or more, or a least several years). This cannot be stated, as the survey lacks a question on the duration of the colony management by beekeepers: the beekeepers who have “never treated” could also be beekeepers who just started beekeeping in the year the survey was made. The number of years in which their colonies would have been treatment-free is then totally unknown, and could also be less than a year. Considering this issue, the number of treatment-free beekeepers have to be estimated without this “never treated” category. Also, one of the main conclusions of the authors seems to be that treatment-free beekeeping would be increasing and that it would be a reliable solution as the colony losses would not be greater: these conclusions are not supported by the study as the survey was not designed properly to study these questions. It would have required questions about the colony losses and about the age of the beekeeping activity. These questions were not asked in the survey and the references that the authors provide to support these conclusions are clearly not sufficient to conclude about this. Indeed, the compared category is not the same for the BBKA study, and the given references e.g. the Oxfordshire Beekeeping Group are particular cases and not scientific literature. This does not seem to be strong enough arguments to support such an important conclusion as the supposed absence of additional colony losses in the case of treatment-free beekeeping. There are of course relevant questions and perspectives that this study opens up. Still, it should be seen as such and not considered as actual and current results. To conclude on that new version of the manuscript, some efforts were made by the authors since the first draft and these efforts must be acknowledged. Still, it seems that this manuscript still requires some modifications to address properly the methodological issues and to ensure a better and clear distinction between what is an actual conclusion of the study and what remains a perspective without sufficient evidence to conclude here. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No Reviewer \#3: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0281130.r004 Author response to Decision Letter 1 20 Dec 2022 1\) There are several sentences that start with “Although”, “Thus”, “Whereas”. Please revise these sentences because they are often grammatically incorrect. L56 – decades CORRECTED L59 – I wouldn’t say they are highly effective. For example, most or all can’t get into the brood cell, where mite reproduction happens. Revise? REVISED L61-62 – watch what you capitalize (ex. oxalic?) DONE L65 – take out the comma after reference 9 DONE L83 – is this resistance or tolerance? SEVERAL YEARS A GO I USED TOLERANCE BUT WAS TOLD TO USE RESISTANCE WHEN TALKING ABOUT THIS MECHANSIM. IN FACT BOTH TERMS CAN APPLY DEPENDING WHAT VIEW YOU TAKE. THEREFORE, I NOW USE RESISTANCE AS THIS IS THE DOMINANT TERM USED IN THE FIELD AND DEFINE IT AS I HAVE DONE HERE. L103 – six questions: DONE L134 – isn’t “treatment status” the response variable? YOU ARE CORRECT SO CHANGED L140 – widely or wildly? WIDELY IS THE CORRECT TERM L141-143 – this should be in the methods and needs more detail (see reviewer 2 comment) DONE AND ADDED NEW DETAIL REQUESTED Figure 1 – it could be helpful to remind the reader that the 158 area associations in the spatial study had ≥5 respondents, so associations with \<5 respondents were not colored in black. At least that was my interpretation of what you said in the text. WE HAVE NOW ADDED THIS INFO INTO THE LEGEND L160-161 – Is this an important or real trend if you don’t see a difference between these groups and the larger groups? Can you explain or speculate why you didn’t see differences between these small and medium colony number groups and the larger group? FURTHER ANAYLSIS (by calculating the 95% CI) INDICATED THAT DUE TO A HIGH VARIABILITY IN THE DATA IN THE DATA WE CAN CONCLUDE THERE IS NO SIGNIFICANT RELATIONSHIP BETWEEN COLONY NUMBER AND TREATMENT STATUS. THIS IS NOW MENTIONED. Table 2 – had no effect (see legend text); also should you have a range for the colony number group estimate since you had four groups? ALEX I like Tables 4 and 5, although I agree with reviewer 3 that there are issues with the “treatment-free” group. Some of those natural methods were interesting (rhubarb??). And how does queen trapping differ from a brood break? Just wondering, really. TABLE 4 WE HAVE CHANGED TO ‘NOT TREATING’ AS THIS IS WHAT THEY WERE DOING AT THE TIME OF THE SURVEY, BUT WE DO AGREE WITH YOU AND REVIEWER 3 THAT ELSEWHERE WE BREAK THIS GROUP UP INTO ITS VARIOUS GROUPS AS WE HAVE DONE IN TABLE 3. AS FOR TABLE 5 WE COULD COMBINE BROOD BREAKS WITH QUEEN TRAPPING BUT THESE ARE THE TERMS THE BEEKEEPERS USED IN THE SURVEY. RHUBARD LEAVES CONTAIN OXALIC ACID BUT THE LEVEL WILL BE LOW L218-219 – This is repeating part of my point above, but I agree that this group could also include beekeepers that have kept bees for only a few (\<6) years and so combining them with the beekeepers that have been treatment-free for 6 years or more could be misleading. It seems that there was no criteria that respondents have kept bees for 6 or more years. IN HINDSIGHT THIS WAS A OVERSIGHT OF THE QUESTIONS ASKED AND RESPONSE OPTIONS AVAILABLE. BUT NOW WE HAVE SEPARATED OUT THE GROUP OF NEVER TREATED AND 1-5 YEARS GROUP. THE READER SHOULD NOW BE ABLE TO DISTINGUSH BETWEEN THOSE NOT- TREATING AND THOSE THAT ARE TREATMENT FREE (I.E. 6+ YEARS) L275-276 – revise the sentence where it says weather-related and Varroa – should this be split up somehow? DONE L281- that’s not a lot of data (N=6) to base a conclusion on. I am also of the opinion that the discussion should not include too many results/statistics, so this is something to potentially omit. It is appropriate to talk about it, just not as something that you tested using data and stats. SECTION REMOVED L307-310 – revise this sentence, it is too long. I like the general idea of the sentence though! SPLIT INTO 2 SENTENCES Reviewer \#1: please correct the following typos in the revised document: L168:...had no effect... CORRECTED L270: overwinter losses. Over the last...CORRECTED L307:...than found currently in the UK…CORRECTLY Regarding your statement in L249-251I suggest a more careful wording as the BBKA report is restricted on the treatment period between August and September while OA as the most common treatment in UK should be mainly used on broodless colonies during winter. THE BBKA SURVEYS ALSO COVERS THE OCT TO APRIL PERIOD AS WELL SO THIS DATA IS NOW GIVEN IN ADDITION TO THE AUG-SEPT DATA. Reviewer \#2: The paper is well written, but the survey lacks depth. The authors wished to investigate how wide spread treatment free beekeeping is within the UK. They did this by conducting an online survey with limited questions that could be completed within five minutes. It would have been far more interesting if they then did a follow up with the beekeepers who qualified as treatment-free to learn more about their beekeeping practices. I feel the current results presented are more appropriate for a bee journal than a rigorous scientific publication. For specific comments regarding the manuscript, see attached PDF. ALL COMMENTS ADRESSED ON ATTACHED PDF (see attached PDF) Reviewer \#3: SPECIFIC COMMENTS L. 72 References are still missing to support this assertion IT IS NOW MADE CLEAR THAT THERE IS NO EVIDENCE TO SUPPORT OR CONTRADICT THIS IDEA. L 127-132 The choice to explore a potential spatial pattern (rather than the other possible explanatory factors for the varroa management) should still be explained: why is this hypothesis made? WE HAVE INSERTED THE FOLLOWING TEXT ‘WHILE CONDUCTING PREVIOUS RESEARCH INTO VARROA-RESISTANCE COLONIES \[14\] IT APPEAR THEY EXISTED IN MANY PARTS OF THE UK. THEREFORE, MAPS WERE CREATED……’ L 233. References or studies on the number of treatment free beekeepers a few years ago would be appropriate here to support the assertion that the number of treatment-free beekeepers is growing. THIS SECTION HAS BEEN DELETED AS THIS STUDY IS MEANT TO PROVIDE BASE-LINE DATA FROM WHICH WE CAN SEE IF IN THE FUTURE THINGS INCREASE OR DECREASE. L 240-244 The BBKA studies the authors refer to are about a specific time of the year: even if some of the beekeepers who do not treat at this time of the year could be treatment-free beekeepers as the authors underline, some of them could also treat at another time of the year. Thus, the results of these BBKA studies should not be directly used to compare with the percentage of treatment-free beekeepers which was estimated in the study, as they do not estimate the same thing. WE HAVE INCLUDED ADDITIONAL ‘NON-TREATMENT’ DATA (OCT-APRIL) FROM THE BBKA SURVEY L.279 It is unclear if the percentage of colonies not treated here refers to the “august to September non treated” (cf. L 241) or to year-long non treated (and the BBKA study the authors refer to does not seem to be available online to find this information). It would be appropriate to precise this information. TO ACCESS THE DATA YOU MUST FIRST JOIN THE BBKA IN ORDER TO ACCESS THERE NEWSLETTER INCLUDING THE BRITISH BEE JOURNAL, SINCE THE JOURNAL/NEWSLETTER LIKE MANY JOURNALS IS BEHIND A PAYWALL. GENERAL COMMENTS This study provides useful information about the number of treatment-free beekeepers in the UK. Some relevant changes have been made to the paper, considering the reviewers’ feedback on the first manuscript. Still, the authors did not respond to some of the feedback, in particular regarding their methodological choices, which were not always explained (e.g. why explore a spatial pattern?) and regarding the limits of these methodological choices and therefore of the study. THE REASON HAS NOW BEEN GIVEN One of the major methodological issues is the choice to gather beekeepers who indicated that they have “never treated against varroa” with beekeepers who have not treated for six years or more. This choice relies on the hypothesis that beekeepers who have “never treated” manage colonies which have not been treated for a significant period of time (6 years or more, or a least several years). This cannot be stated, as the survey lacks a question on the duration of the colony management by beekeepers: the beekeepers who have “never treated” could also be beekeepers who just started beekeeping in the year the survey was made. The number of years in which their colonies would have been treatment-free is then totally unknown, and could also be less than a year. THE ‘NEVER TREATED’ GROUP IS NOW TREATED SEPARATLY Considering this issue, the number of treatment-free beekeepers have to be estimated without this “never treated” category. THIS HAS NOW BEEN DONE Also, one of the main conclusions of the authors seems to be that treatment-free beekeeping would be increasing and that it would be a reliable solution as the colony losses would not be greater: these conclusions are not supported by the study as the survey was not designed properly to study these questions. It would have required questions about the colony losses and about the age of the beekeeping activity. These questions were not asked in the survey and the references that the authors provide to support these conclusions are clearly not sufficient to conclude about this. Indeed, the compared category is not the same for the BBKA study, and the given references e.g. the Oxfordshire Beekeeping Group are particular cases and not scientific literature. This does not seem to be strong enough arguments to support such an important conclusion as the supposed absence of additional colony losses in the case of treatment-free beekeeping. There are of course relevant questions and perspectives that this study opens up. Still, it should be seen as such and not considered as actual and current results. WE HAVE REDUCED AND RE-WRITTEN THAT SECTION AND ADDED THAT AN ‘independent scientific studies are required to see if Varroa-resistance effects the rate of colony losses.’ 10.1371/journal.pone.0281130.r005 Decision Letter 2 Corby-Harris Vanessa Academic Editor 2023 Vanessa Corby-Harris This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 16 Jan 2023 A survey of UK Beekeeper’s Varroa treatment habits PONE-D-22-05221R2 Dear Dr. Martin, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Vanessa Corby-Harris, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): After carefully reading through the reviewers' comments on the first revision and the authors' response to these comments, I feel this article is acceptable. 10.1371/journal.pone.0281130.r006 Acceptance letter Corby-Harris Vanessa Academic Editor 2023 Vanessa Corby-Harris This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Jan 2023 PONE-D-22-05221R2 A survey of UK Beekeeper’s *Varroa* treatment habits Dear Dr. Martin: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Vanessa Corby-Harris Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist. [^2]: Current address: School of Natural Sciences, National University of Ireland Galway, Galway, Ireland

# Introduction Plant pathogenic *Xanthomonas* bacteria cause severe losses of crop production worldwide. Their virulence mainly relies on a type-III-secretion system that translocates effector proteins into plant cells. Such proteins interfere with cellular processes and manipulate the plant to the benefit of the pathogen. Transcription activator-like effectors (TALEs) constitute an important group of effectors that manipulate the transcriptome of the host plant. After entering the nucleus TALEs bind to promoter sequences and initiate transcription. TALE- induced plant genes that support pathogen virulence encode transporters for sugar or sulfate, factors that stabilize small RNAs, or transcription factors. DNA association of TALEs is initiated by their N-terminal region whereas the central part of TALEs provides specific binding to matching sequences. This central DNA binding domain is composed of tandem repetitions of a nearly identical 34 amino acid repeat. Each repeat mainly differs in two-amino acids, the so-called RVD (repeat-variable diresidue), which encodes the specificity for one or several DNA bases. The TALE repeats form a right handed superhelix that wraps around the DNA double strand. Although these repeats can be rearranged to specify practically any given target sequence (called TALE-box or EBE/effector- binding element), a well-balanced presence of so called "strong" and "weak" RVDs is required for efficient activation. A nuclear localization signal (NLS) and an acidic activation domain in the C-terminal part ensure import into the plant nucleus and gene activation, respectively. This modular structure and designable DNA-binding specificity has turned TALEs into versatile tools for targeted genome modification and gene regulation in many organisms. Computational analysis of genomic target sites of natural TALEs showed a preferential occurrence in apparent core promoter regions of -300 to +200 bp around the transcriptional start site (TSS;). Previous studies based on the TALEs AvrBs3, AvrXa7, and AvrXa27 showed that they shift the natural TSS of target genes around 40–60 bp downstream of the position at which the TALE is binding the DNA. Moving the AvrBs3-box in the *Bs3* promoter to a position further upstream resulted in a concomitant upstream shift of the TSS. These observations led to the impression that TALEs control the onset and the place of transcription functionally analogous to the TATA-binding protein. One host component that is involved in TALE-mediated gene induction has been identified. A rice line that is homozygous for the *xa5* allele exhibits an increased resistance to *Xoo* infection due to a loss of efficiency of TALE-mediated transcriptional initiation. *xa5* represents a recessive point mutation of the transcription factor IIA small γ subunit. Apparently, this mutation interferes with the interaction of TALEs with TFIIA and likely accompanying basal transcription-complex components. Furthermore, interactions with parts of RNA stabilizing and modifying enzymes have been suggested for TALEs of the PthA family, but so far a functional link is missing. In contrast to the already intensively studied DNA-binding mode, the circumstances for successful TALE-mediated transcriptional initiation *in planta* are poorly understood. If the presence of a target sequence alone is sufficient to initiate transcription at any genomic locus or if the target locus has to satisfy additional requirements is still puzzling. Here, we systematically analyze TALE-mediated gene activation potential in different promoter contexts, from different positions or orientations within a promoter, and in a possible cooperative fashion. To dissect whether an effect is dependent on the TALE activation domain or the TALE DNA-binding domain, we replaced the TALE activation domain with a heterologous activation domain and the TALE DNA- binding domain with the Cas9-DNA-binding domain. Our findings suggest that TALEs can not dictate transcriptional activation on their own, but rather rely on further promoter elements. Furthermore, TALEs can support transcription by binding in either forward or reverse orientation, an observation which has not been reported, before. We show that this mode of activation is biologically relevant, because reverse-binding TALEs are able to support virulence of *Xoo* during an infection. This study is significant for understanding TALE-mediated gene activation, and will have implications for target site prediction *in planta*. # Material and methods ## Bacterial strains and growth conditions The bacterial strains used in this study were *Escherichia coli* Top10 (New England Biolabs Frankfurt am Main, Germany), *Agrobacterium tumefaciens* GV3101 pMP90 and *Xanthomonas oryzae* pv. *oryzae* (*Xoo*) strains BAI3 and BAI3Δ*talC*. *E*. *coli* cells were cultivated at 37°C in LB, *Xoo* strains at 28°C in PSA and *A*. *tumefaciens* GV3101 at 28°C in YEB medium. ## Plant material and plant inoculations *Oryza sativa* ssp. *japonica* cv. Nipponbare was grown under glasshouse conditions at 28°C (day) and 25°C (night) at 70% relative humidity (RH). Leaves of 4-week-old plants were infiltrated with a needleless syringe and a bacterial suspension of an optical density (OD) at 600 nm of 0.5 as previously described. Symptoms (watersoaked lesions) were scored 5 days post inoculation (dpi). *Nicotiana benthamiana* plants were grown under 16 h of light, 40–60% RH, at 23°C (day) 19°C (night) in the growth chamber. Leaves of 4- to 6-week-old plants were inoculated with *A*. *tumefaciens* strains using a needleless syringe. ## Construction of artificial TALEs and the dCas9 activator TALEs were constructed by using the Golden TAL technology as described by Geißler *et al*. (2011). Up to six individual repeats were subcloned in an assembly vector, resulting in a hexa-repeat module. Three of these hexa-repeat- modules were then fused to the N- and C-terminal parts of Hax3 and inserted in an expression vector for *Agrobacterium* or *Xanthomonas*, respectively. This results in a N-terminally fused GFP for the *Agrobacterium* vector and in a C-terminally fused FLAG Epitope in the *Xanthomonas* vector. The dCas9 variant fused to the C-terminal region of Hax3 was used as dCas9 activator and generated as described before. The activity score of the used sgRNAs was analysed by using deskgen. ## β-Glucuronidase (GUS) reporter constructs and GUS activity analysis β-Glucuronidase assays from plant samples were performed as described by Boch *et al*., 2009. Briefly, the PCR-amplified fragments of the promoters (*Bs4*, *OsSWEET14*) were cloned into pENTR/D-TOPO (Life Technologies GmbH, Darmstadt, Germany) and fused to the *uidA* reporter gene by LR recombination into pGWB3. To analyze reporter activity, *A*. *tumefaciens* strains delivering TALE constructs and GUS reporter constructs were resuspended in infiltration medium, resulting in an OD₆₀₀ of 0.8, mixed in equal amounts and inoculated into *N*. *benthamiana* leaves. Two dpi, leaf discs were sampled and GUS activities were quantified using 4-methylumbelliferyl-β-D-glucuronide (MUG). Total protein concentrations were determined using Bradford assays. Data were compiled from triplicate samples originating from different plants. Error bars represent the standard deviation. ## RNA isolation and qRT- PCR Leaves of 4-week-old Nipponbare plants were infiltrated with 10 mM MgCl₂ or with the different *Xoo* strains using an OD₆₀₀ of 0.5. At 1 dpi, 5 cm segments were harvested and rice total RNA was isolated using the Qiagen RNeasy kit. cDNA was generated from 2 μg RNA using the Fermentas first-strand cDNA synthesis kit (Thermo Fisher Scientific Inc.,Waltham, MA, USA) and real-time PCR was performed using the iCycler (Bio- Rad, München, Germany) as described before. The amplification efficiency for each primer pair was analyzed using a standard curve plot of a dilution series. cDNA amounts were normalized using actin as a reference gene. The fold change induction was calculated in comparison to leaves treated with the BAI3Δ*talC* mutant by using the ΔΔCt method. ## RNA isolation and 5´RACE TALE-dependent transcriptional start sites at the *OsSWEET14* promoter were determined by using 5´RACE. Leaf discs were harvested from the same plant material that was used for determining GUS-activity in *N*. *benthamiana* and total RNA was extracted by using the Qiagen RNeasy kit. 1 μg total RNA was used to produce cDNA for 5´RACE by using the SMARTer RACE cDNA amplification kit (Takara Bio, Inc., Shiga, Japan). The 5´ends were amplified by using the universal forward primer and the gene specific primer GUS 5\`RACE, cloned, and sequenced. # Results ## TALE-mediated gene activation depends on flanking DNA sequences So far it is unknown if TALEs that bind to a certain genomic target site (TALE- box) are sufficient to initiate transcription at that location or if surrounding DNA sequences influence gene induction. To address this, we designed reporter constructs in which the Hax3-box, that is bound by Hax3, a natural TALE from *Xanthomonas campestris* pv. *campestris*, is placed in front of varying downstream sequences. A 300 bp fragment of the tomato *Bs4* promoter was used. This promoter fragment has a very low basal activity which makes it well suited for transcriptional activation assays. We subdivided this fragment into portions of equal length (75 bp) and placed them downstream of the Hax3-box and upstream of a promoterless *uidA* reporter gene. This setup ensured a fixed distance between the TALE-box and the reporter gene open reading frame (ORF;). Additionally, we questioned whether non-promoter regions can trigger TALE- mediated gene expression. Hence, we included two random 75 bp fragments from the *Bs4* ORF in our analysis. The reporter constructs and the constructs containing *hax3* were co-transformed in *N*. *benthamiana* using *Agrobacterium*-mediated T-DNA transfer. Quantification of the reporter gene activity showed that the three *pBs4* derived fragments *pBs4-278*, *pBs4-126* and *pBs4-50* mediated the strongest Hax3-dependent gene activation. In contrast, the two *Bs4* ORF sequences *oBs4+550* and *oBs4+2362* and the *pBs4-202* promoter fragment facilitate only a slight or no gene induction by Hax3. These findings indicate that TALEs cannot initiate transcription at any genomic location but rely on specific surrounding DNA sequences for efficient gene induction. To get a general idea of other putatively involved DNA-binding factors we predicted potential cis-regulatory motifs by using the program PlantCARE. To further dissect which sequences and potential promoter elements support TALE activity, we chose the fragment *pBs4-50*, that was highly responsive to Hax3, and *pBs4-202*, that was not, and exchanged parts of them. We divided the promoter fragments into four parts (A1 to A4 and B1 to B4) and generated promoter-swap constructs fused to the Hax3-box and the *uidA* reporter gene (75bp). In comparison to the reference *pBs4-50* only the constructs 1 and 10 showed a comparable or even higher activation by Hax3. This already indicates two findings (i) the reason why *pBs4-50*, but not *pBs4-202* allows TALE- dependent expression is not due to sequences directly downstream of the Hax3-box that might interact with C-terminal parts of Hax3 and (ii) the presence of the 5´UTR in *pBs4-50* is not crucial for robust activity by e.g. influencing mRNA stability. In contrast, all other constructs showed a decreased or no activity (construct 2–9). Apparently, the predicted CAAT-box, the TATA-box, and the sequence between them are relevant (, construct 6 vs. 10). In summary, these results demonstrate that TALE-boxes which are targeted by TALEs are not always sufficient to initiate transcription. This implies that TALEs, binding to these sequences, do not initiate gene expression on their own. They likely need to bind in proximity to supporting sequences—probably core promoter elements—to ensure efficient gene activation. ## Systematic analysis of TALE-dependent reporter gene activation at the *OsSWEET14* promoter To systematically investigate from which promoter positions TALEs can initiate transcription we used a complementary approach in which we kept the promoter sequence constant, but used designer TALEs targeted to different positions. The well characterized rice *OsSWEET14* promoter was used, that is activated by a number of natural and artificial TALEs. *OsSWEET14* encodes a sugar exporter and represents an important rice virulence target whose TALE-dependent activation supports bacterial virulence. An *OsSWEET14* promoter fragment 1000 bp upstream of the ATG was chosen that comprises possible core and distal promoter elements. The fragment was inserted upstream of a promoterless *uidA* reporter gene which allows a systematic comparison of TALE activities in *N*. *benthamiana*. 14 artificial TALEs with an equal number of 17,5 repeats were built to target various positions within this fragment (; ;. By using the TALgetter prediction program, all TALEs used in this study have been carefully examined to bind exclusively their desired target sequence and not at other possible locations in the *OsSWEET14* promoter. As control to compare the impact of the varying RVD sequences on TALE activity and to exclude an impact of variable flanking sequences, we first analyzed the corresponding TALE-boxes in the *pBs4-*based reporter construct and verified protein integrity *in planta*. Next, the TALE constructs were co-delivered together with the *pOsSWEET14* based reporter construct in *N*. *benthamiana* via *Agrobacterium*. The two natural *Xoo* TALEs AvrXa7 and TalC which induce the *OsSWEET14* promoter in the natural *Xoo*-rice interaction were used as positive controls and the *OsSWEET14* promoter in combination with a non-matching TALE (Hax3) was used as negative control, respectively. The majority of the analyzed TALEs activated the reporter gene irrespective of their position in the *OsSWEET14* promoter. This demonstrated that TALEs can drive gene expression from highly different positions in a promoter. TAL7, TAL14, and TAL18 show a particularly low activity (\<35% compared to TalC) at the *OsSWEET14* promoter although they efficiently induce expression of the *pBs4*- driven reporter gene. This suggests that their binding sites are not favorable or not accessible in this particular genomic context. ## Artificial TALEs can activate the *OsSWEET14* promoter in reverse orientation All known TALE targets are based on TALEs binding a promoter in "forward" orientation, with the C-terminal activation domain facing the downstream ORF. In contrast, classical enhancer elements are often orientation-independent. Therefore, we investigated whether TALEs also function in "reverse" orientation *in planta*. Nine TALEs were designed to bind in reverse orientation in the *OsSWEET14* promoter and TALgetter target site predictions verified that these are the only possible binding-sites within the promoter. The two natural *Xoo* TALEs AvrXa7 and TalC were used as positive controls. TALE constructs and the reporter construct were co-delivered in *N*. *benthamiana* via *Agrobacterium*. Remarkably, the three reversely placed TALEs, TAL-11, TAL-1, and TAL-2, activated the reporter highly efficient with an activity of 71%, 104%, and 179%, respectively, compared to TalC. We used the highly active TAL-2 to analyze experimentally whether its activation potential was indeed due to its binding in reverse orientation. We designed two reporter constructs based on the *pBs4* reporter system to exclude any *pOsSWEET14*-based effects. One reporter covers the TAL2 binding site on the forward and the TAL-2 binding site on the reverse strand fused to the minimal *pBs4* promoter and the *uidA* reporter gene. Both TALEs activated this reporter albeit TAL-2 to a significantly lesser extent compared to the *pOsSWEET14* context. As control, we mutated all five Guanine nucleotides in the upper strand to Adenine (construct TAL2/-2mut). TAL2 should still bind this sequence because its RVD NN recognizes G and A, whereas TAL-2 has HD RVDs matching only to C on the reverse strand. Accordingly, TAL-2 lost the ability to activate the reporter verifying that TAL-2 in fact binds to the reverse DNA strand. The other reverse TALEs resulted in weak activities of 10 to 17% compared to TalC although some of these (i.e., TAL-AvrXa7, TAL-81, TAL-82, TAL-9) target the same region as highly active forward TALEs. All TALEs were active in combination with forward-oriented TALE-boxes upstream of the *pBs4* minimal promoter, and protein levels were comparable, indicating that the RVD composition is not causing the observed variability. In summary, the systematic analysis of differentially positioned forward and reverse binding TALEs resulted in three important findings: (i) Some TALEs (TAL11/-11, TAL2/-2 and TAL1/-1) activate transcription in both orientations, suggesting that their DNA target sequences are accessible and allow a flexible interaction with the transcriptional machinery. (ii) The activity of some TALEs (AvrXa7/TAL-AvrXa7, TAL8/-81/-82 and TAL9/-9) is orientation-dependent, suggesting that the reverse orientation in these cases prevented an efficient recruitment of the transcriptional machinery. (iii) Some TALEs do not efficiently activate gene expression irrespective of their orientation, suggesting that either the TALE-box is not accessible within the promoter context or the relative position is not suitable to recruit the transcriptional machinery. The observed activity of reverse binding TALEs breaks the long- standing assumption that TALEs only activate genes in one-direction. ## TALEs can synergize to induce transcription at the *OsSWEET14* promoter Reverse-binding TALEs seem to be particularly dependent on further promoter elements and we wondered whether they can enhance the capacity of other TALEs to induce transcription. The phenomenon of synergistic gene activation has been described in human cells using TALEs fused to the heterologous VP64 activation domain, but it is unknown whether the native TALE has a similar potential in plant cells. For this, we used low-activity artificial TALEs from our collection. Individual TALEs or combinations of two or three TALEs and the *OsSWEET14*- reporter were co-transformed into *N*. *benthamiana* and GUS activity was measured. In most cases, combinations of TALEs at the promoter increased transcriptional induction. Even the presence of low activity reverse- binding TALEs can stimulate other TALEs (e.g., TAL3 and -AvrXa7), but this requires that their relative position does not hinder each other e.g., TAL3 and TAL-7 whose activation domains face each other and might spatially block the simultaneous assembly of the transcriptional machinery. Together, this demonstrates that multiple TALEs, including reverse TALEs and low-active TALEs can act synergistically to efficiently induce gene expression in plants. ## The activity of TALEs targeted to reverse oriented sequences is dependent on other promoter elements Based on our finding that the presence of core promoter regions can influence TALE-mediated gene activation, we aimed to analyze the influence of the TATA-box and surrounding DNA sequences on the activity of TALEs. For this, we used an *OsSWEET14* promoter variant in which 33 bp encompassing the TATA-box were deleted in rice via genome editing. This *ossweet14-11* variant lacks the AvrXa7-box and is thus resistant to *Xoo* strains carrying the TALE AvrXa7. The edited promoter was amplified from the rice mutant and inserted into the GUS reporter. TALE and reporter constructs were co-transformed into *N*. *benthamiana* and GUS activity was measured. The deletion of the TATA-box and the surrounding sequences did not significantly influence the activity of the majority of the forward binding TALEs, indicating that an interaction with the transcriptional machinery in these cases is TATA- box independent. The binding sites of TAL4, TAL5, and Avrxa7 are at least partially deleted resulting in an expected loss of activity. Surprisingly, the activities of the TALEs that bind directly upstream of the TATA-box, including TAL1, TAL2 and TalC, were reduced by 60, 54, and 70%, respectively, if compared to the activity measured at the wild type promoter. This indicates that the activity of TALEs binding to a distinct region directly upstream of the TATA-box is positively supported by this core element and surrounding sequences whereas other TALEs are not. This effect is even more pronounced if the activity of reverse binding TALEs is analyzed. The highly active reverse TALEs TAL-11, TAL-1 and TAL-2 show a strong decrease in their activity of 61, 81 and 91% compared to the WT promoter. This shows that the activity of reverse binding TALEs in particular is highly connected to the presence of supporting DNA-elements like the TATA-box. ## The use of alternative activation and DNA-binding domains, respectively, do not change the relative activation pattern at the *OsSWEET14* promoter TALE derivatives that were fused to the synthetic VP64 transactivation domain and analyzed in human cell lines showed a highly position and orientation independent activity and activated genes over long distances. TALE-VP16 fusions are in principle known to be functional *in planta*, but with a much decreased activity in comparison to natural TALEs. To analyze if the VP64 AD changes the activation pattern relative to the binding position of TALEs *in planta* a subset of our artificial TALEs targeting the *OsSWEET14* promoter was fused to VP64 instead of the TALE AD. Their gene activation capacity was compared at the *pOsSWEET14*::*uidA* reporter construct following transformation into *N*. *benthamiana*. The overall activity of the VP64-fused TALEs was lower than the activity of the TALEs containing their native AD, likely because VP64 is not completely compatible with the plant transcriptional machinery. Importantly, the use of VP64 instead of the TALE AD did not profoundly change the relative activation level of the analyzed TALEs at their respective positions. This indicates that the TALE AD functions in a manner that is analogous to the VP64 AD in plant cells. Although TALEs employ a highly flexible DNA binding domain their access to DNA can be inhibited by methylated cytosines or nucleosomes occupying the sequences. The *Streptococcus pyogenes* Cas9 is a non-related DNA-binding protein that is guided to target sequences by a mechanism that is unrelated to TALEs. DNA- binding of SpCas9 is directed by a guide RNA that forms base pairing to 20 nucleotides in the target sequence and is therefore not affected by methylation. To test whether the TALE DNA-binding domain or its activation domain is responsible for the orientation-independent gene activation we analyzed the TALE-activation domain in the context of the Cas9-DNA binding platform. We fused the catalytically inactive "dead Cas9" (dCas9, D10A; N863A; lacking nuclease activity to the C-terminus of Hax3 to generate a dCas9 activator that is functional in plants. To compare the activity of TALEs and the dCas9 activators we designed sgRNAs (single guide RNAs) that bind in proximity to existing TALE- boxes. T-DNAs producing TALE and Cas9-activator, respectively, were co- transformed with the reporter construct containing the *OsSWEET14* promoter into *N*. *benthamiana* and reporter gene activation was determined. Those sgRNAs that hybridized to the region neighboring the highly active TAL1 and TalC mediated strongest reporter gene activation. Taken together, the use of an alternative DNA binding domain identified the same promoter region upstream of the TATA-box as highly suitable for activation by the TALE-AD. ## TALEs control the transcriptional start site dependent on their position Previous studies indicated that TALEs can shift the transcriptional start site (TSS) of a plant gene. Since some of them bound to sequences overlapping the TATA-box, it is not clear whether the TSS shift in these previous examples is based on the action of the TALE or by spatially blocking the TATA-box. Hence, we aimed to systematically analyze TALE dependent TSS shifts in the *OsSWEET14* promoter by using our collection of differentially positioned artificial TALEs. The TALE constructs were co-transformed together with the reporter construct into *N*. *benthamiana* followed by RNA extraction and 5´RACE. The AvrXa7-dependent TSS at the *OsSWEET14* reporter construct in *N*. *benthamiana* resembled the one following *Xanthomonas*-mediated delivery in rice indicating that both systems are comparable. The TALEs that were located downstream of the TATA-box (TAL3/-3, TAL-9, TAL-10; ) shifted the *OsSWEET14* TSS to positions around 40–60 bp downstream of the respective TALE-box which is in concordance to previous observations (;. Interestingly, the TSS for TAL3 and TAL-3 are comparable although the activation domain which is supposed to be the platform to assemble the initiation complex is located in a flipped orientation. TALEs binding upstream of the TATA-box (TAL2/-2, TAL1/-1 and TAL11) on the other hand show a different response. TAL2 and TAL-2 initiate transcription around 50–60 bp downstream of their target site which mainly resembles the natural *OsSWEET14* TSS. Although the binding site of TAL1 and TAL-1 is located further upstream it primarily triggers the same TSS as TAL2 and TAL-2. This suggests that the TSS in presence of TALEs that bind in a certain region may not primarily be dictated by the position or orientation of the TALE itself but by the proximity to other active promoter-elements e.g. the TATA-box. In contrast, TAL11, which binds even further upstream of the TATA-box, again resulted in a TSS 40-60bp downstream of its binding site and additionally in the natural TSS. These results show that TALEs can trigger transcription in two ways. Either they directly influence the TSS, suggesting a direct role in the assembly of the pre-initiation complex (PIC) or they support other promoter elements e.g. the TATA-box which recruit the PIC. ## Artificial reverse binding TALEs restore the virulence of the BAI3Δ*talC* mutant in rice To analyze whether reverse binding TALEs can activate expression of target genes in a natural infection, a subset of artificial TALEs was introduced into the *Xoo* mutant BAI3Δ*talC* (. This mutant is deficient in the major virulence factor *talc* which targets *OsSWEET14*. Consequently, this strain fails to induce *OsSWEET14* and does not cause disease symptoms. Complementation of this *Xoo* mutant strain with *talC* or artificial TALEs that target *OsSWEET14* restored *OsSWEET14* induction and virulence *Xoo* BAI3Δ*talC* was complemented with artificial TALEs from our collection, and *talC*, respectively. The strains were inoculated into the rice variety Nipponbare, and virulence symptoms were documented in comparison to the mutant strain without any *OsSWEET14*-targeting TALE. To test TALE-dependent activation of *OsSWEET14*, qRT-PCR was performed in parallel. The forward-binding TALEs TAL2, TAL8, and TAL10 efficiently induced *OsSWEET14* and supported the development of disease symptoms. Moreover, also the highly active reverse binding TALE TAL-2 resulted in a gain of virulence and significant increase in *OsSWEET14* expression. However none of the forward or the reverse binding TALEs TAL14, TAL3/-3, TAL7/-7, and TAL-82, which had a low activity in the *N*. *benthamiana* transient system, restored the virulence of BAI3Δ*talC* or activated the expression of *OsSWEET14* significantly. One of the TALEs, TAL10, did induce *OsSWEET14* expression, but did not lead to a gain of virulence. Instead, it led to the formation of dark brown lesions around the inoculation spot, suggesting that this TALE triggers a collateral defence reaction in parallel which blocks the development of virulence symptoms. In summary, TALEs can activate a target gene from different promoter positions and via binding in the reverse orientation also in the biologically relevant natural infection system. # Discussion TALEs are versatile virulence factors that have been evolved by the pathogen to target different plant promoter sequences and induce a variety of target genes. The unique modular TALE DNA-binding domain has been well studied, but how the host transcription machinery is recruited and whether further promoter elements are needed is still largely unknown. Protein interaction screens to identify host components have yielded only few candidates, and those shed little light on the gene induction requirements. Instead, we here used a set of functional studies to gain first insights into the conditions under which TALEs induce gene expression. Our reporter gene induction experiments demonstrate that the presence of a TALE binding site alone can not trigger transcription, instead, a specific DNA context is required. This indicates that TALEs function analogous to enhancer- binding proteins that require the basal transcription machinery for their action and function as control agents for gene expression. Previous studies have revealed a binding preference for natural TALEs at -300 to +200 bp around the TSS of a target gene, a region that typically includes core promoter elements. One of such elements apparently supporting TALE-mediated gene activation in some of our experiments, is the TATA-box. Interestingly, several natural TALE-boxes overlap with the TATA-box of their respective target genes. This finding led to the assumption that TALEs functionally replace the TBP. According to our present data, this is presumably not the case. Instead TALEs can benefit from the presence of a TATA-box. In *Oryza sativa* ssp. *japonica* only around 19% of the promoters contain a TATA-box. Indeed, some TALE-activated promoters (*Xa23*, *OsSWEET2b*, *OsSWEET4*, *pAGT582-1*, and the fragments *pBs4-278*, *pBs4-126* and *oBs4+550* used in this study) lack a prototype TATA-box. This shows that the presence of a TATA-box is not absolutely required for TALE-dependent induction, but can be substituted for by certain other sequences. Using the prominent TALE virulence target *OsSWEET14* from rice, we could further show that TALEs function from many different positions within a given promoter, but not equally well. Although individual artificial TALEs have been used before to induce promoters a quantitative comparison of multiple TALE positions within a native promoter context has not been done, so far. The variable efficiency of TALEs could be caused by several reasons: sequences are blocked by other proteins or nucleosomes, TALEs do not bind equally well because of their RVD composition, methylation of target sequences interferes with RVD- base recognition, or by the relative position of TALEs to other transcription factors. We surveyed available data to identify occupied and accessible DNA regions in the *OsSWEET14* promoter. Although these data do not give a consistent picture, the region upstream of the TATA-box should in principle be accessible for TALEs. To exclude factors like RVD composition, DNA methylation, and other effects specific for the TALE-DNA interaction, we employed designer activators as fusions between dCas9 and the C-terminal domain of TALEs. Comparison of a collection of sgRNAs to target the dCas9-activator to different positions within the *OsSWEET14* promoter revealed that they are also position- dependent and that it is not the DNA-binding domain, but rather the position of the TALE activation domain within the promoter context that dictates the efficiency of the activator. Although classical enhancers are typically regarded as being highly independent in their positioning relative to a given promoter, this is not unambiguously the case. In fact, re-positioning of such enhancer elements within promoters demonstrated that they also show a different activity at different positions. We postulate that TALEs can in principle function from different positions within a promoter, thus functionally resemble classical enhancer-binding proteins. It has been reported earlier that TALEs shift the transcription start site (TSS) to a position 40–60 bp downstream of their binding position upon gene activation, but do not do so in a synthetic promoter with multiple inserted TALE-boxes (and TATA-boxes). The analysis of our collection of TALEs targeting the *OsSWEET14* promoter now suggests, that the TALE-dependent TSS shift does not depend on the TALE alone, but depends on the presence of additional promoter elements. TALEs positioned in a region closely upstream of the natural TATA-box supported the natural TSS, whereas TALEs positioned further upstream or downstream triggered novel TSS. We envision that TALEs are enhancer-binding proteins that define their own binding position according to their modular DNA- binding domain, but at the same time, they directly cooperate with and recruit the transcription machinery. In a breakthrough study one of these interactions could recently be shown. TALEs interact with the gamma subunit of the basal transcription factor TFIIA, which is well known to interact with activation domains of several activators similar to the one from TALEs, belonging to the family of acidic activation domains (e.g. VP16, Zta and Gal4). Indeed, when we replaced the native acidic activation domain of TALEs with the VP64 activation domain, these activators contained a similar relative activity profile, with their highest activity at comparable positions to native TALEs at the *OsSWEET14* promoter. This shows again that the position within the promoter dictates the overall efficiency of the activator and that the activation domain can functionally be replaced by related ones. Our TALE-VP64 fusions still contained most of the C-terminal domain of native TALEs and the portion that interacts with TFIIAy. This domain might therefore have facilitated the interaction to the basal transcription machinery, and caused the comparable activity profiles along the different promoter positions. Because the region upstream of the TATA-box in the *OsSWEET14* promoter is so well suited for activators, we propose that our TALE-positioning screen actually identified a region spanning a potential natural enhancer element that regulates the expression of *OsSWEET14* in rice. We were highly surprised to realize that TALEs also function when bound in a reverse orientation relative to the open reading frame. This has, so far, not been considered for the identification of TALE targets and the reverse binding mode will be an important implementation to identify novel TALE virulence targets. It further points out that the orientation, relative to the TSS, of the TALE activation domain, which is interacting with host components, does not play a role under certain, but not all, conditions. This observation further suggests that TALEs may initiate transcription bidirectionally. A bidirectional transcription describes a situation in which—originating from one promoter—two divergent transcripts are initiated. This phenomenon has frequently been observed for many promoters of human and animal origin and increasingly also for plants. During the review process of our study another manuscript was published which supported our findings that TALEs function when bound in reverse orientation and initiate transcription bidirectionally. In this work, the authors suggest that forward and reverse binding TALEs function equally well. In contrast, in our experimental setup most reversely placed TALEs did not induce transcription to the same level as forward placed ones ( an). This suggests that although the bidirectional transcription is a general feature of TALEs it is possibly restricted by surrounding promoter sequences. Therefore, we propose that TALEs only function in reverse orientation if placed in defined promoter regions that allow bidirectional transcription, e.g., the region upstream of the TATA-box in the *OsSWEET14* promoter. This indicates that both orientations are not equally well suited to induce gene expression and that TALEs prefer the forward orientation. Possibly, in reverse orientation, the TALE protein itself is blocking the path of the polymerase and an efficient establishment of a novel TSS can only occur downstream of the TALE. This is supported by our observation that both, forward- and reverse-binding TALEs upstream of the TATA-box trigger the same TSS downstream. This further suggests that TALEs bound in either one of the orientations can activate already present, but paused polymerase II complexes. Comparing the activity of our TALE collection at the wild-type and a promoter- derivative lacking the TATA-box and surrounding sequences revealed that those TALEs that are located closely upstream of the TATA-box function less well, whereas further upstream positioned TALEs and TALEs downstream of the TATA-box are not influenced. This indicates that TALEs can either use existing elements or use novel downstream promoter sequences if they are suitable. In the latter case, the TSS is changed. This observation is consistent with our promoter-swap experiments which revealed that the region downstream of the TALE is crucial for TALE-dependent gene activation. Intriguingly, the activity of the reverse- oriented TALEs in particular is negatively impacted by the removal of the TATA- box. This indicates that transcription initiated by reverse TALEs is especially dependent on the presence of potent other promoter elements. This restriction is likely the reason why so far most natural TALEs bind in forward orientation at their respective target promoters. Our combination of several weakly active TALEs targeted to either forward or reverse binding sites in the *OsSWEET14* promoter revealed that TALEs can act synergistically in plants. This feature was only shown in human cells before, where TALE derivatives with a truncated C-terminus were fused to the VP64 activation domain. Interestingly, the synergistic effect was only observed if TALEs do not spatially hinder each other. Our results further suggest that TALEs *in planta* may not only act synergistically with each other but are further supported by promoter elements which are possibly occupied by host transcription factors (e.g. the TATA-box). Whether the synergistic activity of TALEs is based on increased chromatin remodeling or other effects remains to be elucidated. Our systematic analysis of variably positioned TALEs shows that they act in principle as enhancer-binding proteins. When placing designer TALE activators in promoters to trigger efficient target gene activation one should consider the presence of existing promoter elements, e.g. the TATA-box, and the expected possible shift of the TSS. The novel reverse binding mode of TALEs now allows for the detection of formerly overlooked virulence targets in host plants. Therefore, we have added the option to search for reverse-binding TALEs in the TALE-target prediction programs TALgetter, and the feature is also implemented in the alternative programs TALVEZ and TALE-NT 2.0, respectively. Forward and reverse binding TALEs will furthermore be excellent tools to tackle fundamental questions of gene induction in plants. # Supporting information We thank Ulla Bonas for support, and the 2013 MSc course *"Plant Genetics"* at the Martin Luther University for demonstrating that reverse binding TALEs can activate gene expression. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** J. Streubel HB JB. **Formal analysis:** JG. **Funding acquisition:** JB JG. **Investigation:** J. Streubel HB JB. **Resources:** J. Stuttman. **Writing – original draft:** J. Streubel JB. **Writing – review & editing:** J. Streubel JB.

# Introduction Intermittent hypoxia (IH) is a key pathological feature of obstructive sleep apnea (OSA), the most common sleep related breathing disorder. Epidemiological studies show that patients with OSA suffer an accelerated decline in kidney function. However, it is unclear whether this is due to OSA *per se* or to confounding factors such as obesity, hypertension, diabetes or other concomitant disorders. A rodent model of OSA was developed in 2001 that simulates moderate to severe OSA in clinical settings, and has been used to investigate the systemic effects of OSA such as insulin resistance, endothelial vascular dysfunction, and alterations in tumor-associated macrophages function. Importantly, the pathological pattern of IH is distinct from the pathological feature of sustained hypoxia, where IH promotes excessive production of reactive oxygen species and inflammatory mediators, and increases sympathetic activity. With respect to kidney disease, OSA enhances the activity of the renin angiotensin-aldosterone system, increases the activity of the sympathetic nervous system and generates systematic and local reactive oxygen species; these alterations are known to induce functional and structural kidney damage. For instance, focal segmental glomerulosclerosis (FSGS) is an important histopathological component of impaired renal function that can result from several metabolic and haemodynamic factors including a diabetic milieu, increased blood pressure and mitochondrial oxidative stress. Mechanistically, mesangial matrix expansion (MME) is the cornerstone of FSGS. Moreover, excessive accumulation of glomerular extracellular basement membrane (mesangium) is driven by an over- production of a number of growth factors including transforming growth factor-β1 (TGF-β1), connective tissue growth factor (CTGF) and vascular endothelial growth factor-A (VEGF-A). However, the histopathological alterations of the kidney in response to IH have not been reported. We tested the hypothesis that IH alters glomerular structure and modulates the expression of glomerular growth factors. # Materials and methods ## Animals After approval from the University of British Columbia Animal Care Committee, ten 8-week old wild-type male CB57BL/6 mice were obtained from JAX Laboratories and allowed to acclimatize for one week with free access to water and regular chow before initiating the IH procedure. ## IH protocol Animals were randomly assigned to receive either IH or intermittent air (IA) for 60 days as we described previously. Briefly, mice assigned to receive IH were placed in specially designed cages containing oxygen sensors for measuring the fraction of oxygen inspired (FIO₂) in the cages. The cages were connected to a gas regulator that allowed the flow of sufficient amounts of nitrogen gas to reduce FIO₂ to 8% within 30 seconds, after which the gas regulator allowed for a rapid replacement of the nitrogen by oxygen and causing the FIO₂ in the cage to be restored to 21% (room air) within 30 seconds. This 1-minute IH cycle was repeated 60 times per hour for a total of 12 day-light hours per day. ## Animal monitoring and euthanasia Animals were monitored twice a week for the following parameters: body weight, activity, appearance, breathing and posture/gait. A score of 1 or 2 in any category resulted in increased monitoring, including daily weighing, supportive care (e.g. supplemental heat, SQ fluid replacement, gel food or food treats/moistened pellets). A score of 3 in any category or a cumulative score of \>5 after appropriate supportive care resulted in immediate euthanasia. Animals were euthanized by inhaling an anesthetic followed by carbon dioxide. ## Kidney sectioning After 60 days of either IH or IA, mice were euthanized and the right kidney from each animal was fixed in 4% paraformaldehyde (Sigma-Aldrich, Germany) for 24 hours and then sectioned (5 μm) for histopathological and immunohistochemical studies. ## Histopathological assessment Two paraffin-embedded kidney samples from each group were sectioned (4 kidney sections per slide) and stained with hematoxylin and eosin, periodic acid–Schiff (PAS) and Masson's trichrome stains for pathological evaluation. Sections were evaluated by a certified pathologist who was blinded to the study protocol (i.e. blinded to identity of IA or IH samples) for evaluation of glomerular congestion, matrix expansion and pelvic inflammation. ## Glomerular and mesangial area calculation Following deparaffinization, kidney sections were stained with PAS for assessment of the total glomerular tuft area and glomerular basement membrane (mesangium) area. The mean glomerular tuft area was determined by counting all available glomeruli (where the glomerular vascular pole is evident) within each kidney section, where one kidney section was used from each animal. The mesangial matrix area was calculated as the PAS-positive area compared to the total glomerular tuft area using 50 randomly selected glomeruli per kidney section (see below). Images were taken at 400x magnification using an Aperio ScanScope**®** CS microscope slide scanner, and measurements of glomerular tuft area were made using Aperio ImageScope (V 12.1.0.5029) software. ## Glomerular selection for area measurements and protein expression We created a virtual grid with 25 different slots to enable sampling of all areas of each kidney section. A minimum of two different glomeruli were randomly selected from each slot for mesangial area measurements and protein semi- quantification. As some slots did not include any glomeruli (example: slot E1), the total number of missing glomeruli from such slots was calculated, allowing for a similar number of glomeruli to be reinvestigated so that we were able to sample 50 glomeruli per kidney section. ## Immunohistochemistry Sections of kidney tissues were deparaffinised and rehydrated using xylene and downgraded concentrations of alcohol, after which slides were immersed in 10 mM sodium citrate (pH 6) for 20 minutes in a steam bath (95 °C) to allow for antigen retrieval. Slides were then incubated in 3% hydrogen peroxidase (Sigma- Aldrich–Germany) for quenching of endogenous peroxidase activity, and non- specific proteins binding was blocked using 1.5% normal blocking serum for 90 minutes. Kidney sections slides were then incubated in goat anti-mouse TGF-β1 (1:50 sc-146-G), CTGF (1:50 sc-14939) and VEGF-A (1:50 sc-152-G) primary antibodies (Santa Cruz Biotechnology–USA) for 24 hours at 4°C. After several washes, slides were incubated with avidin biotinylated horseradish peroxidase- labeled secondary antibody as per manufacturer instructions (ImmunoCruz™ goat ABC Staining System: sc-2023). Finally, slides were stained with 3,3′-diaminobenzidine (DAB) chromogen for protein detection and quantification, and counterstained with hematoxylin for detection of nuclei. Images were taken at 400x magnification using an Olympus BX61 electron microscope. Only 50 images of randomly selected glomeruli were considered per kidney section, and images were analyzed using IHC Profiler plugin within ImageJ free software, which measures the intensity of the brown (DAB) color caused by the antigen-antibody reaction. The percentage of only “highly-positive” brown staining in each glomerulus was measured, so reducing the possibility of other colors influencing the brown stain and thus enhancing the accuracy of protein quantification. The % of highly-positive staining was averaged from 50 different glomeruli per kidney section (to give n = 1). ## Western blotting Upon euthanasia, 20 mg of the left kidney cortex of each animal was harvested and homogenized in RIPA lysis buffer (sc-24948), sonicated thoroughly and then kept frozen (-80 degrees) for protein analysis. A total of 50 μg total protein from each tissue homogenate was used for protein electrophoresis separation on 12% precast polyacrylamide gel (Bio-Rad: 4568044) as per manufacturer instructions. Proteins were then transferred to nitrocellulose membranes (Bio- Rad: 1620115) and incubated with rabbit anti-mouse 1:1000 β-actin (cell signaling: 4967), 1:1000 GAPDH (Cell signaling: 2118), 1:100 TGF-β1 (sc-130348), 1:100 CTGF (sc-365970), 1:100 VEGF (sc-7269), 1:100 hypoxia inducible factor-1α (HIF-1α) (sc-13515), 1:100 Bax (sc-7480) and 1:100 Bcl-2 (sc-7382) primary antibodies overnight at 4 degrees. After several washes, the membrane was incubated for 1 hour in horseradish peroxidase-conjugated anti-rabbit secondary IgG (cell signaling: 7074S) for β-actin and GAPDH proteins detection, and with horseradish peroxidase-conjugated mouse IgG kappa binding protein (sc-516102) for detection of other mouse monoclonal antibodies. Finally, 1 ml of Clarity™ Western ECL Blotting Substrates (Bio-Rad: 1705060) was added to the membrane for chemiluminescent signaling detection, and images were taken by ChemiDoc™ XRS+ System device (Bio-Rad: 1708265). Images were analyzed by Imagej software and reported as the pixels’ intensity of a target protein relative to either β-actin or GAPDH housekeeping proteins. ## Fluorometric *In situ* cell death detection Paraffin-embedded kidney sections were deparaffinized and rehydrated as described earlier. Slides were then incubated in phosphate-buffered saline with tween-20 (PBST) solution for 10 minutes for the purpose of permeabilization. After washing with PBS, 50 μl of TUNEL reaction mixture (Roche Applied Science– 12156792910, USA) was added to each kidney section, and sections were then incubated at 37 °C for 1 hour in a humidified chamber. After several washes, slides were counterstained with 4',6-diamidino-2-phenylindole (DAPI) stain for detection of nuclei and finally mounted. Fluorescence images were taken with Olympus BX61 electron microscope at 200x. Four random images were taken from each kidney section to give n = 1, with one kidney section per animal. The number of TUNEL staining-positive cells was normalized to the number of available cells in each image, and averaged from 4 images per kidney section. Cell counting was performed using cell counter plugin within ImageJ free software. ## Renal function index At the end of study protocol, each mouse was kept separately in a metabolic cage for 24-hour urine collection. Urine samples were kept at -20 degrees for total urinary albumin excretion analysis. Total urinary albumin was quantified using a mouse albumin ELISA kit (41-ALBMS-E01) as per manufacturer instructions. Serum creatinine was measured by a creatinine enzymatic assay (Crystal Chem: 80350) as per manufacturer instructions. ## Statistical analysis Data are presented as mean ± SD. The normal distribution for each set of the results was tested using the Shapiro-Wilk test of normality; an unpaired t-test was used to calculate differences between groups using GraphPad Prism (Version 6) statistical software, and p values less than 0.05 were considered significant. # Results ## Animal characteristics Some animals showed signs of minor distress during the initial phase of exposure to IH, but no signs of illness were observed in any animals during the study period. There were no differences in kidney weight, body weight or fasting blood sugar for IH and IA groups. Details of animal characteristics at the end of the study are shown in. ## Histopathological assessment A subjective histological assessment by an independent pathologist indicated no obvious differences between groups in PAS and trichrome staining, except for a variation in the amount of mesangial matrix in the glomeruli of the IH group that required further non-descriptive analysis. On the other hand, differences were evident in hematoxylin and eosin stained slides. Kidney sections from the IH group had a mild degree of glomerular congestion, mesangial expansion and pelvic inflammation. shows examples of kidney sections stained with hematoxylin and eosin, PAS and Masson's trichrome stains. ## Glomerular area and MME A minimum of 145 different glomeruli per kidney section was included in the analysis of glomerular tuft area. The glomerular area in mice exposed to IH was increased by 13.4%; the average glomerular tuft area in the IH group was 2782.84 ± 72.68 μm² (95% CI: 2692–2872 μm²) compared to an average glomerular tuft area of 2454.98 ± 73.59 μm² (95% CI: 2362–2546 μm²) (p\< 0.001) in the IA control group. Further examination of PAS stained kidney sections indicated an expansion of mesangial matrix in the IH group by 1.8-fold compared to controls. The mean fraction of mesangial matrix to glomerular tuft area in IH exposed mice was 0.087 ± 0.015 (95% CI: 0.032–0.065), compared to 0.049 ± 0.012 (95% CI: 0.064–0.108) in controls (p\< 0.01). ## Immunohistochemistry We examined a number of profibrotic proteins to explore their role in MME secondary to IH. The glomerular expression of TGF-β1 and one of its essential downstream proteins, CTGF, were significantly increased by 2.7 (p\< 0.05) and 2.2 (p\< 0.01) fold respectively in response to IH. Moreover, the glomerular expression of VEGF-A was significantly increased by 3.7-fold (p\< 0.05) in the IH group compared to control IA group. summarizes the differences in glomerular protein expression between IH and IA groups. ## Western blotting Cortical TGF-β1, CTGF and VEGF proteins were semi quantified by western blotting analysis. Compared to the IA group, the expression of TGF-β1, CTGF and VEGF-A proteins in IH-exposed mice was greater by 1.9, 4.0 and 2.6-fold respectively (p\< 0.05 for all). The cortical expression of HIF-1α protein was also higher in the IH group by 2.9 times (p\< 0.05). Finally, the ratio of apoptotic (Bax)/antiapoptotic (Bcl-2) proteins was 2.4 fold higher in IH-exposed mice (p\< 0.01). summarizes cortical protein expression in IH and IA groups. ## Fluorometric *in situ* cell death detection Five paraffin-embedded kidney sections from each group underwent TUNEL-staining for detection of cells with DNA damage, and counterstained with DAPI for detection of nuclei. The percentage of TUNEL-positive nuclei in the IH group was 3.2 ± 0.67% (95% CI: 2.7–3.8%) compared to 0.38 ± 0.27% (95% CI: 0.14–0.61%) in the IA control group (p\< 0.001). ## Renal function index An average of 0.9 ± 0.13 ml and 0.86 ± 0.22 ml of urine was collected from IH- exposed and control mice respectively (p = 0.45). The average 24-hour urinary albumin excretion in the IH-exposed mice was 43.4 ± 16 μg (95% CI: 29.3–57.5 μg), while in control mice urinary albumin excretion averaged 9.7 ± 4.6 μg (95% CI: 5.7–13.8 mg) in 24-hours (p\< 0.01). In contrast, the average serum creatinine level in IH-exposed mice was 108.02 ± 56.1μmol/L, compared to 86.1 ± 46.4 μmol/L in the IA control mice (p = 0.51). # Discussion Our study shows for the first time that IH causes glomerular hypertrophy and expansion of the glomerular mesangial matrix in a mouse model of OSA. We also show that mice exposed to IH have increased expression of glomerular TGF-β1, CTGF and VEGF-A proteins. In addition, there was a significant increase in renal cellular apoptosis after 2-months of IH. Our data also shows that IH-exposed mice excrete higher amounts of urinary albumin compared to control mice, suggesting that there was mild disruption of glomerular filtration in these mice, possibly related to glomerular hypertrophy. There was no evidence of severe renal functional damage secondary to IH, at least based on serum creatinine data. Glomerular hypertrophy is associated with glomerulosclerosis, where the interplay of hemodynamic and various growth factors determines the balance between glomerular matrix accumulation and degradation. Although MME occurs in diabetes, other diseases such as hypertension also cause glomerular hypertrophy in humans. Nevertheless, factors such as obesity and hyperuricemia are also associated with glomerular hypertrophy. Increases in glomerular profibrotic cytokines is suggested to be the principal pathway in the pathogenesis of MME or glomerular hypertrophy. With respect to renal pathophysiology, TGF-β1 is a pleiotropic cytokine secreted by glomerular mesangial cells and podocytes in response to common fibrogenic stimuli, and is considered the primary fibrogenic growth factor in the pathogenesis of renal fibrosis. In addition, TGF-β1 is also a major inducer of CTGF in renal fibrosis, where excessive production of glomerular CTGF is implicated in glomerulosclerosis and thickening of glomerular basement membrane. Several epidemiological studies suggest a possible relationship between sleep apnea and declining kidney function. A retrospective cohort study reported that about 30% of patients with sleep-related breathing disorders were also diagnosed with chronic kidney disease (CKD), which was significantly greater than the prevalence of CKD in the healthy population. However, the apnea-hypopnea index was not a significant determinant of CKD in sleep-related breathing disorder patients. Another retrospective cross-sectional study detected significant reductions in glomerular filtration rates as the severity of OSA increased. The prevalence of diabetes and hypertension is also significantly higher in patients with severe OSA. On the other hand, sleep apnea is a significant comorbidity in patients with CKD as shown in a prospective study where the prevalence of sleep apnea (predominantly obstructive) increased significantly with declining kidney function. This study reported that 57% of patients with end-stage renal disease had sleep apnea, compared to 41% of patients with CKD not on dialysis and 27% of patients with glomerular filtration rates of ≥60 ml/min. Suggested mechanisms linking OSA and CKD include, but is not limited to, activation of the renin-angiotensin-aldosterone system (RAAS), increased sympathetic regulation and elevations in systemic and local reactive oxygen species. The effect of nocturnal hypoxia is common in OSA patients and cannot be overlooked in the context of OSA-related CKD. A recent cohort study reported that OSA patients with nocturnal hypoxia (defined by oxygen saturation\< 90% for\> 12% during night-time monitoring) were at significant risk for accelerated loss of kidney function, with an adjusted odds ratio of 2.89 (95% CI: 1.25–6.67). In addition, another study reports that RAAS can be influenced by nocturnal hypoxemia in OSA patients, where patients with severe hypoxemia (mean SaO₂≤ 90% during overnight) have significantly greater renal RAAS activity compared to moderate hypoxemia patients (SaO₂≥ 90%) and control subjects. Interestingly, continuous positive airway pressure (CPAP) therapy reverses the increased activity of RAS in patients with OSA, as shown by a recent clinical trial where CPAP therapy increased renal plasma flow and significantly reduced plasma aldosterone and urinary protein excretion in non- diabetic normotensive OSA patients. Although our study focusses on glomerular signs of injury secondary to IH, we have not examined specific markers of tubular structural or functional injury. The chronic hypoxia hypothesis states that hypoxia is a key player in inducing primary glomerular injury and that it creates a hypoxic tissue environment that eventually triggers tubular injury. In fact, renal tissue hypoxia is a major pathological mechanism in triggering several renal pathologies such as hypertensive and diabetic nephropathies. Renal tissue hypoxia induced in rats by administering dinitrophenol (a mitochondrial uncoupler that increases oxygen consumption) increases markers of renal injury similar to those occurring in hypertensive and diabetic nephropathy, e.g. increased renal tissue oxygen consumption and urinary protein excretion. Despite growing clinical evidence of OSA-related CKD, the direct effect of OSA on kidney structure and function is largely unexplored, leaving the pathological association between OSA and CKD unclear. A possible explanation for this is that the majority of OSA patients frequently present with multiple comorbidities, making it difficult to establish the direct effects of OSA on CKD. Studies on animal models may therefore be a useful approach to examine the direct effects of OSA on kidney pathology. For example, a study in mice using similar study design as in our study reported that 8 weeks of IH upregulated a number of inflammatory and profibrotic proteins (including CTGF) in kidney tissues, suggesting a direct destructive effect of IH on kidneys. However, unlike the urinary protein analysis data reported in our study, Sun et al reported no differences in 24-hour excretion of urinary proteins. Others report that metallothionein (potent antioxidant) knock-out mice were more prone to renal damage when exposed to 8 weeks of IH. Another recent study reported that 2 weeks of IH significantly increased plasma VEGF in mice; interestingly, mouse macrophages exposed to IH *in vitro* also have significant increases in VEGF expression. Such studies not only strengthen a clear pathological linkage between OSA and the renal system, but also illustrate potential treatment strategies to avoid OSA-related renal consequences. The animal model of IH we use has also been used to investigate other sleep apnea-related comorbidities such as cancer. It is important to note variations in the severity and duration of IH in this model; for instance, Almendros and colleagues reported a significant increase in lung epithelial TC1 cell tumors in mice exposed to levels of IH that are close in its severity to the hypoxic levels in our study (FIO₂ = 6%), but with a duration of only 28 days. Another study by the same group reported increased melanoma lung metastasis in mice after being exposed to IH for 30 days using FIO₂ of 5%, but for only 6 hours per day. Our study has a number of strengths; the use of an animal model excludes the effects confounding variables such as pre-existing hypertension, diabetes, and obesity that are potential confounders in human studies. However, our study also has a number of limitations. First, the number of mice studied was relatively small. Second, our mouse model is not a perfect representation of human OSA. For example, patients with OSA tend not to have such severe desaturation, and suffer from hypercapnia during events as opposed to hypocapnia. However, IH is one of the key components of OSA, and this model has been used extensively in the literature especially in terms of elucidating cardiometabolic complications associated with OSA. Another important limitation of this model is that it does not allow for easy recording of factors usually accompany sleep apnea, such as sleep fragmentation. The recurrent arousals combined with hypoxemia/reoxygenation can activate the sympathetic nervous system, oxidative stress, and inflammation; these may represent important components in the increased risk of cardiovascular diseases including myocardial infarction and stroke in OSA patients. Although mice exposed to IH in our study did not show any signs of diabetes and obesity, we were unable to reliably monitor blood pressure during the wake and sleep cycles of mice. Others have reported increases in the mean arterial pressure (by \~14 mmHg) in a rat model of IH. Therefore, we are unable to rule out the effects of secondary hypertension due to IH on glomerular alterations. In summary, we show that kidneys are an important target for the harmful effects of IH. We provide evidence for glomerular hypertrophy and MME in response to IH, with simultaneous increases in glomerular TGF-β1, CTGF and VEGF-A proteins. Taken together, these findings allow for a better understanding of the mechanisms by which IH induces renal damage, and suggest potential targets for mitigating the potential kidney damage related to OSA. # Supporting information The support of the Saudi Arabian Cultural Bureau in Canada (to BA) is gratefully acknowledged. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work.

# Introduction Secondary metabolites are often bioactive and are thus an attractive source of lead compounds in drug discovery efforts. In many cases, secondary metabolites isolated from higher organisms such as marine invertebrates are thought to ultimately derive from symbiotic bacteria residing in these hosts. In the majority of cases, little is known about the symbionts that produce secondary metabolites due to difficulties in culturing and/or sequencing their genomes directly from complex microbiomes. However, understanding these symbionts' lifestyle is of critical importance, both to natural products discovery and chemical ecology. In our own efforts, we have used the tunicate *Lissoclinum patella* as a model system to understand the interactions between microbial symbionts, host animals and secondary metabolite chemistry. *L. patella* is a colonial tunicate in the family Didemnidae, with a wide distribution across much of the Western Pacific. Like many didemnid species, *L. patella* harbors a photosynthetic symbiont, the cyanobacterium *Prochloron didemni*. *P. didemni* carries out several metabolic functions for the host, and has been shown to synthesize a series of highly modified cyclic ribosomal peptides, termed cyanobactins. The biosynthetic pathways that make cyanobactins are all related to the prototypical patellamide pathway. One or several precursor peptides are expressed, containing the residues that are incorporated into the finished compounds, flanked by recognition sequences and a leader peptide on the N-terminus. A heterocyclase may act on the immature precursor peptide, to produce methyloxazoline, thiazoline and oxazoline from threonine, cysteine and serine, respectively. Optionally, these heterocycles can be oxidized by an oxidase in the pathway (i.e. to methyloxazole, thiazole, oxazole). The leader peptide and the 5′ recognition sequence are then cleaved by a protease homologous to PatA. The last step in the process is the cleavage of the 3′ recognition sequence and macrocylization of the precursor peptide by a PatG homolog. In some pathways, the macrocycle can be further prenylated, if suitable side chains remain. Remarkably, the patellamide pathway and relatives are extremely tolerant to altered precursor cassettes, and are capable of processing precursor sequences quite unlike the naturally occuring compounds. It is clear that the currently known cyanobactins account for a miniscule portion of the chemical diversity that is biosynthetically possible, suggesting strong evolutionary or other influences on natural systems. Beyond *P. didemni*, *L. patella* has a complex microbiome of resident microflora, which contribute to secondary metabolite production and which vary according to microhabitat within the animal. We recently described another symbiont, *Candidatus* Endolissoclinum faulkneri, which is found only in a subset of *L. patella* animals and is associated with the presence of the highly cytotoxic patellazoles, that may serve a protective function for the host animal. Our analysis of *Ca.* E. faulkneri genomes indicates that the bacterium is a long-term symbiont that is exclusively vertically transmitted. Strains of *Ca.* E. faulkneri from phylogenetically distant hosts are correspondingly divergent in genomic sequence, indicating genetic isolation. The patellazoles biosynthetic pathway is a large polyketide synthase (PKS) system, which has been maintained for the ∼6–31 million years that *Ca.* E. faulkneri has been associated with *L. patella*. In contrast to *Ca.* E. faulkneri, several lines of evidence suggest that *P. didemni* can be transmitted between hosts both horizontally and vertically. *P. didemni* genomes obtained from geographically distant animals are remarkably similar (above 97% nucleotide sequence identity across the whole genome), indicating that these strains are not genetically isolated. This strongly suggests there is at least a transient free-living fraction of the *P. didemni* population that can move between hosts. Consistent with this notion, although *P. didemni* has never been cultured outside of its host, genome analysis suggests that independent life may be possible. In fact, it has been found that the microenvironment inhabited by *P. didemni* varies significantly in terms of O₂ saturation and pH during dark/light cycles, indicating that in contrast to reduced-genome intracellular symbionts, *P. didemni* must maintain the ability to adapt to different environmental conditions. Stable *P. didemni* symbiosis is limited to the didemnids, but unstable associations have been reported in other host groups, such as holothurians and sponges. Amongst the didemnids, *P. didemni* phylogeny has been found to be independent of host species, and the distribution of photobionts amongst this group suggests multiple symbiosis establishment events. The presence of a specialized vertical transmission apparatus only in *Diplosoma* likewise suggests parallel evolution of such mechanisms corresponding to multiple origins for this symbiosis. Consistent with findings in *P. didemni*–didemnid relationships, we previously found a random distribution of *P. didemni* secondary metabolite pathways across and within host species. However, we observed that in some cases, different animal species collected in the same vicinity contained different secondary metabolites, suggesting a degree of symbiont selection. We also observed that chemistry is not consistent across all *L. patella* samples. Many ascidian species have been found to encompass cryptic species, due to their similar or identical morphologies. We hypothesized that similarly, the designation *L. patella* may contain several divergent populations. In the present work, we show that *L. patella* encompasses at least three distinct phylogenetic groups, and that these groups contain different secondary metabolites. In the case of symbionts that can be horizontally-acquired, such as *P. didemni*, this suggests the host is involved in secondary metabolite selection from a free-living pool. # Results and Discussion As a result of our long-standing interest in didemnid tunicates, *P. didemni* and the cyanobactins, we have collected specimens of *L. patella* spanning a large geographic area from Fiji to Palau between 2002 and 2011. We were able to amplify host 18S rRNA genes from a number of these samples, and all show \>98% nucleotide identity to an 18S rRNA sequence in the NCBI database identified as *L. patella* (accession no. AB211085, see). We constructed a phylogenetic tree based on these nucleotide sequences plus other members of Didemnidae from the NCBI database and the *Ciona intestinalis* 18S rRNA gene to act as an outgroup. All of our sequences formed a clade along with the type *L. patella* sequence. We then turned our attention to the mitochondrial cytochrome *c* oxidase I (COXI) gene, because this marker has been used for fine phylogenetic distinctions and identifying cryptic species, including many types of tunicate. Using a variety of primer sets, we were able to amplify COXI genes from 15 animals. The resulting sequences were pooled with other Didemnidae COXI sequences from NCBI along with a *Ciona savigyni* COXI sequence to act as an outgroup in a tree based on the translated protein sequences. Unfortunately, we found that amongst these NCBI sequences, two non-overlapping sections of the COXI gene were present, and thus we could not make one complete COXI tree with all sequences. In particular, the only reported *L. patella* COXI sequence could not be included in the tree containing our sequences, although we were able to compare it to full length COXI sequences assembled from shotgun Illumina data obtained from three animals (vide infra). Our collections of *L. patella* fall into at least three separate clades. We already observed that animals containing *Ca.* E. faulkneri and patellazoles were restricted to a divergent clade we termed group ‘B’. Detailed analysis of the COXI nucleotide identities of this clade reveal it contains three highly divergent cryptic populations which could be different species. We now see that the animals of group ‘A’ are also somewhat divergent. When we examined in detail the pairwise nucleotide identities of COXI sequence, we saw that animals collected in 2005 from Southeastern Papua New Guinea have significantly diverged from other group ‘A’ animals, with identities ranging from 91–95%. This divergence is on the order of other cases of cryptic speciations claimed in tunicates based on COXI divergences between 2 and 16.5%. In a survey assessing the use of COXI for phylogenetic distinctions, it was found that conspecific samples rarely diverged more than 2% in nucleotide identity. Therefore it is possible that animals from the ′05 collection are a distinct cryptic species from other group ‘A’ animals. We found evidence that there may be further cryptic populations in the remaining group ‘A’ animals, as animal E11-097 exhibited an intermediate divergence between both the 2005 collection and the remainder of group ‘A’ (90–95% identity). We therefore term these groups as A1 (05–033, 05–039, 05–027, 05–044), A2 (E11-097) and A3 (L4, 07-103, 03-005, 07-002B, 07-005, and 07-110). Interestingly, we were able to compare the *L. patella* COXI sequence from NCBI (AB602781.1) with full length COXI sequences assembled from Illumina shotgun sequencing data of L2, L5 and L6. This revealed that AB602781.1 was as closely related to L5 and L6 as it was to L2. The pairwise sequence identities between all three groups (L5/L6, L2 and AB602781.1) were similarly and significantly different. This suggests that AB602781.1, collected in Sanur, Bali, may be another member of group B and could contain *Ca.* E. faulkneri and the patellazoles. The divergent group B contains individuals collected over a wide swath of the Pacific, from the Eastern Fields region south of Papua New Guinea to Fiji. Conversely, group A1 contains only individuals collected off Southeastern shores of Papua New Guinea and is minimally divergent. These results alone would suggest that a primary influence on *L. patella* phylogeny is geography along with founder effects, similar to some populations of sponges. However, group A3 is phylogenetically quite uniform and yet covers a large area encompassing the Bismarck Sea and Palau. Further sampling is required to determine whether A1 and A2 are truly more geographically restricted than A3, and this will likely reveal the full extent of these groups' ranges and also determine whether any coexist in the same locale. Coexistence might suggest that there is little genetic exchange between the groups; for instance there are several color morphotypes of the didemnid tunicate *Didemnum molle* that are phylogenetically distinct. Sometimes several morphotypes occupy the same area and thus there may be mechanisms to maintain reproductive isolation of these forms. From analysis of the aligned COXI sequences in our tree set, the different *D. molle* morphotypes shared between 89 and 97% nucleotide identity. Another didemnid tunicate, *D. vexillum*, was recently found to be composed of two phylogenetic clades. The two clades share ∼96% COXI nucleotide identity, and the authors concluded that these were not separate species. Nevertheless, colony fusion experiments showed significantly higher success rates amongst the genetically similar invasive form from New Zealand versus the genetically diverse population from Japan. We then examined the secondary metabolite chemistry of *L. patella* animals by LCMS, identifying known cyanobactins and patellazoles previously found in *L. patella* based on their mass. We used skiff, a Perl script used in the Clovr-16S pipeline, to analyze the tabulated peak areas. This script's intended purpose is to take tables of 16S abundances, binned at an arbitrary taxonomic level, and create a heatmap and dendrograms that reflect the Euclidean distance between normalized samples and bins. Because skiff is agnostic as to the type of data it receives, we were able to use it to visualize the Euclidean distance between samples based on *L. patella* chemistry as shown by a dendrogram. With this method, we observe that the secondary metabolites present mirror the clades we determined in our COXI phylogenetic tree, except that E11-097 is closer chemically to the A1 clade than phylogeny would suggest. The A1 members are classified by the presence of lissoclinamides 5–8, ulicyclamide and isomers, as well as ulithiacyclamide. The A3 group lack lissoclinamides 5–8 and in addition can contain lissoclinamide 1, 9 and patellamides. The B group are quite chemically distinct. As well as containing the patellazoles, they also contained different cyanobactins compared to group A animals, including the prenylated patellins and trunkamide A. It is not surprising that the patellazoles are limited to group B animals, because these compounds are produced by an exclusively vertically transmitted symbiont that is not found in other clades (vide supra). However, it is notable that even though *P. didemni* populations are highly uniform and do not correlate with host phylogeny, there is a strong host phylogenetic signal for their secondary metabolites. We previously found a high sequence identity across three whole *P. didemni* genomes (\>97%), in both patellazole containing and patellazole-negative animals. This suggests that *P. didemni* strains are not genetically isolated, and that there must be significant horizontal exchange between strains in different hosts, in addition to well-established vertical transmission mechanisms. The P2 and P3 *P. didemni* genomes come from animals we now know are significantly divergent (L2 and L3, respectively). The P1 genome comes from animal L1. The majority of the extracted DNA from this animal was used in the extensive sequencing of the P1 genome, and therefore we were not able to obtain a COXI sequence, but chemically it aligns to members of group A3 and was collected from the same geographic area as other members of this group. We show that although P1 and P2/P3 have highly similar genomes, they produce different secondary metabolites. Together, our results suggest that there is some degree of selection for *P. didemni* strains based on secondary metabolism, and that this correlates with host phylogeny. The patellazoles are highly toxic and therefore likely to be defensive in function. This may also be the case for the cyanobactins, although they are not generally as cytotoxic as the patellazoles. Some have suggested that cyanobactins may have metal binding capabilities, and some have moderate cytotoxicity, but their true ecological function remains obscure. We have found, however, that their distribution is not random, and this may be used as a basis for further exploration of their function. Previous reports of natural product isolation may indicate the potential ranges for the different *L. patella* clades. For instance, patellazoles were previously isolated from an animal collected in Guam, and compounds related to both trunkamides and patellamides have been isolated from animals collected in the Great Barrier Reef. The mitochondrion is thought to be the result of an ancient endosymbiosis event in the early evolution of eukaryotes, and its tiny genome is therefore likely the end result of the process of genome degradation and erosion observed in endosymbiotic bacteria. Like more recent endosymbionts, mitochondia exhibit accelerated evolution because of their population structure and lack of DNA repair pathways, and their genome sequences can be used to infer a great deal about the hosts' evolutionary history. Additionally, in tunicates it has been shown that gene order in mitochondria is hypervariable, potentially providing an additional phylogenetic signal. We had previously obtained shotgun metagenomic sequence in Illumina HiSeq runs for three group B animals (L2, L5 and L6, see Materials and Methods), and set out to assemble mitochondrial genomes from them. In all cases, contigs that appeared mitochondrial were high coverage (several hundred × or more), and could be separated from other genomes on this basis. In all cases, mitochondrial assemblies were resolved to single contigs, ranging in size from 12,562 bp (L6) to 14,403 bp (L2), and all were ∼21% GC. Annotated ORFs all correspond to genes previously found in other tunicate genomes, but some genes commonly found were missing from the assemblies (NADH dehydrogenase subunit 4L was missing from L2 and L6, while NADH dehydrogenase subunit 6 and ATPase F0 subunit 8 were not found in any of the assemblies). This may be due to general difficulties in assembling such high-AT sequence. The L2 mitochondrion assembly contained an unannotated section, roughly equivalent in size to the small and large subunit rRNA genes in other tunicate mitochondria. This section may include the L2 mitochondrial rRNA genes, but only small parts of the sequence showed any homology to tunicate mitochondrial rRNA genes in BLASTN searches against the NCBI database. Independent assembly efforts in L5 and L6 were syntenic with the L2 assembly, but they appear to lack the putative rRNA region in L2. The consistent synteny across three samples suggests that there are no misassembled portions of the L2 assembly. As with previously reported tunicate mitochondrial genomes, all genes are on the same strand, but the gene order exhibited is unique. We previously determined that the members of group B may represent several cryptic species, with COXI identities suggesting that animals L2 and L5 diverged somewhere between 6 and 31 million years ago. As expected, L5 and L6 share greater gene identities with each other than either one has with L2. Interestingly, our analysis shows that L5 and L6 are not clones. These animals were collected in the same vicinity (within ∼100 m); while L5 contains *Ca.* E. faulkneri and patellazoles, L6 has lost *Ca.* E. faulkneri and contains a potentially pathogenic bacterium in its place that we termed *Ca.* Xenolissoclinum pacificiensis. Because L5 and L6 have highly similar mitochondrial genomes, colony fusion may be possible, and therefore we cannot exclude the possibility that the loss of *Ca.* E. faulkneri in L6 is reversible through this mechanism. Further studies will be needed in order to investigate the structure of this population, and whether the loss of *Ca.* E. faulkneri in L6 is recent and/or stable. Such studies might be a unique opportunity to investigate the influence of symbionts and natural products on host population structure and speciation. Variation in the distribution and abundance of natural products is a significant problem that affects the utility of natural compounds in drug discovery efforts. Often vanishingly small amounts of a compound are isolated in an initial collection. Although only small amounts are required for the characterization of structure and *in vitro* activity, recollection for further development is often challenging. Our efforts here show that one of the factors that can control chemical variation in symbiotic systems is unappreciated cryptic speciation of the host, even when symbionts may be obtained from the environment. Similar issues are beginning to be recognized in other secondary metabolite producers. For instance cyanobacteria designated as *Lyngbya majuscula* have been credited with several hundred natural products in the literature. Recent genomic sequencing of a strain that fell under this classification (now *Moorea producta*) suggests *Lyngbya majuscula* could be a morphologically identical species complex whose natural products may be a marker of phylogeny. Our mitochondrial genome sequences hint at the complexity of symbiotic interactions in the environment, where individual hosts can be found that have lost even stable symbionts. Loss and gain of strictly vertical symbionts may play a major role in host speciation if these events affect fitness and reproductive compatibility, consistent with previous models of symbiosis. We observed one such local extinction in animal L6, which lacks *Ca.* E. faulkneri and is likely unable to regain this symbiont, potentially driving host speciation and adaptation in L6's descendents. This study shows that the previously supposed random distribution of *P. didemni*–produced cyanobactins is in fact based on host phylogeny. Genomic data indicate that this symbiont is highly uniform, and therefore populations within individual hosts are *not* genetically isolated and must undergo frequent horizontal exchange between didemnid hosts. Thus, our results implicate host- dependent recruitment of *P. didemni* based on secondary metabolite production. Because cyanobactin pathways are highly tolerant to precursor peptide mutations, the host may play a major role in maintaining ecologically important precursor sequences. These results have implications for biodiversity and drug discovery. In tandem with previous results showing new compounds can be isolated by surveying individual ascidian colonies, it is now clear that cryptic populations of ascidians are an untapped source of new potential pharmaceuticals. In turn, if a local, cryptic population goes extinct due to habitat loss, there is a likelihood that potential pharmaceuticals will be permanently lost to science. While extinction of cryptic species is often proposed to negatively impact drug discovery, there is a prevalent contrary view that, at least among bacteria, ‘everything is everywhere’. Here we provide a concrete example of how important bacterial compounds might be lost through destruction of local habitats. # Materials and Methods ## Extraction of DNA and sequencing of phylogenetic markers Permission to perform field research was granted by the Papua New Guinea Department of Environment and Conservation, the governments of Palau, Fiji and Solomon Islands. These efforts were facilitated by the University of Papua New Guinea, the Coral Reef Research Foundation and the University of South Pacific, respectively. Samples of *L. patella* were collected from the sites indicated in and preserved in RNAlater. Portions of each sample were set aside as frozen specimens for chemical analysis. DNA was extracted from tunicate samples preserved in RNAlater either using an established tunicate method, the Qiagen DNeasy kit, or by direct pulverizing of tissue in DMSO. Tunicate 18S rRNA and mitochondrial COXI genes were amplified using primers shown in, with Platinum Taq High Fidelity (Invitrogen). In all cases, PCR reactions were 10 µL volume, containing 0.1 µL Taq, 1× of the supplied buffer, 2 µM each primer, 0.2 mM each dNTP (Invitrogen) and 2.0 mM MgSO₄. Reactions consisted of hot start (94°C, 2 min), followed by 35 cycles of \[94°C/30 s, variable annealing temperature/30 s, 68°C/1 min per kb extension (minimum 45 s)\], then a final extension step of 68°C/10 min. PCR products were either Sanger sequenced directly with the relevant primers, or else cloned using the Topo-TA cloning kit (Invitrogen) before sequencing. ## Construction of phylogenetic trees and identity matrices Marker sequences (both 18S rRNA and COXI nucleotide sequences) for members of the family Didemnidae and *Ciona intestinalis* or *Ciona savignyi* were downloaded from the NCBI database. The *Ciona* sequences acted as outgroups for rooting the trees. Database 18S rRNA sequences, along with experimental sequences from *L. patella* samples were aligned with Clustal Omega. The alignment was inspected manually in ClustalX, and sequences that were extremely short or unilaterally introduced large inserts were discarded. The alignment was trimmed using a Perl script (tim_aligned_fasta.pl), and used to construct a phylogenetic tree with FastTreeMP with the parameters -slow -spr 5 -mlacc 3 -gamma -gtr -nt. All trees were viewed and manipulated using the Interactive Tree of Life server. COXI sequences were translated into protein sequences using the ascidian mitochondrial translation table (NCBI translation table 13), then aligned with Clustal Omega. It was found that COXI sequences from NCBI aligned to two distinct, non-overlapping regions of the *Ciona savignyi* sequence (accession no. BAC57000.1), and so two distinct trees were constructed, one of which included sections of COXI sequence obtained from the assembled mitochondrial genomes of L2, L5 and L6 (vide infra). In each case the alignments were manually inspected and trimmed as with the 18S rRNA alignment, before the trees were constructed with FastTreeMP using the parameters -slow -spr 10 -mlacc 3 -bionj -gamma. In order to determine the nucleotide identities of the sequences in the trees, the original nucleotide sequences were aligned to the protein alignments with a Perl script (nucleotide_translation_alignment_2.pl). Pairwise identities were calculated from this alignment with another Perl script (identity_matrix.pl ). ## UHPLC/HRMS analysis Frozen chemistry voucher samples of each animal were freeze-dried, then sequentially extracted with chloroform and methanol. The combined extracts were dried down and passed over a small C₁₈ plug, eluting with methanol. LC/MS data were acquired using a Bruker MaXis ESI-Q-TOF mass spectrometer coupled with a Waters Acquity UPLC system operated by Bruker Hystar software. A gradient of MeOH and H₂O (containing 0.1% formic acid) was used with a flow rate of 0.3 mL/min on a RP C18 column (Phenomenex Kinetex 2.6 µm, 2.1 mm × 100 mm). The gradient went from 10% MeOH/90% H₂O to 97% MeOH/3% H₂O in 12 mins, followed by 97% MeOH/3% H₂O held for 3.5 mins. Full scan mass spectra (*m*/*z* 150-1550) were measured in positive ESI mode. The mass spectrometer was operated using previously published parameters. Tune mix (Agilent, ESI-L low concentration) was introduced through a divert valve at the end of each chromatographic run for automatic internal calibration. ## Construction of secondary metabolite heatmap and dendrogram Raw data files from LCMS runs were converted to mzXML format and processed in MZMine according to the following procedure: 1. peak detection in centroid mode with a noise level cutoff of 5.0 × 103; 2. chromatogram building with a minimum time span of 0.05 min, minimum peak height of 2.5, and 5.0 ppm *m*/*z* tolerance; 3. chromatogram deconvolution using the local minimum search algorithm, a chromatogram threshold of 65.0%, a search minimum retention time range of 0.05 min, a minimum relative height of 5.0%, a minimum absolute height of 5.0 × 103, and a minimum ratio of peak top/edge of 2; 5. isotopic peak grouping with a *m*/*z* tolerance of 5.0 ppm, retention time tolerance of 0.01 min, maximum charge of +2, assuming monoisotopic shape with the lowest *m*/*z* being representative; 6. peaks list row filtering with a minimum of 1 peak in a row, a minimum of 1 peak in an isotopic pattern, a peak duration range of 0.0–2.0 min and auto *m*/*z* and retention time. Compounds were identified in MZMine using a custom database containing compounds previously found in *L. patella*, with a tolerance of 5 ppm error. The peak areas reported in MZMine were tabulated, and compounds arising from the same biosynthetic precursor peptides or pathways were added together (including both \[M + H\]+ and \[M + Na\]+ ions). The table was used as an input for skiff (<http://clovr.org/docs/skiff/>), a Perl script that is a component of the CloVR-16S pipeline. This script normally takes tables of 16S abundances for groups of samples, binned according to a certain level of phylogenetic classification. In this case, the peak areas were expressed as a fraction of the sum of assigned peak areas for each sample, and the log₁₀ of each fraction was used to plot the heatmap. Clustering of samples (i.e. the dendrogram portion of) was achieved by calculating the Euclidean distance between samples based on these transformed values. ## Sequencing, assembly, annotation and comparison of draft ascidian mitochondrial genomes Assemblies of the mitochondrial genomes of *L. patella* animals L2, L5 and L6 were constructed from Illumina HiSeq 2000 datasets previously reported. In each case, 10% of the full dataset was used (8.0 M paired reads for L2, 14.8 M paired reads for L5 and 20.0 M paired reads for L6). A script was used to screen out PCR duplicates (<https://github.com/ibest/GRC_Scripts/blob/master/screen_duplicates_PE.py>), then the reads were filtered for length \> 40 bp and quality \> 30 with Seqyclean (<https://bitbucket.org/izhbannikov/seqyclean>). Only the first 4.8 M filtered reads were used for subsequent processing in L6, due to high mitochondial genome coverage. Overlapping paired reads were then joined with FLASH, then both paired and overlapping reads were subjected to assembly with SPAdes in -careful mode. All three mitochondrial genomes were resolved into a single contig, using K-mer values of 67,73,77,83,87 (L2); 41,45,51,55,61 (L5); and 77 (L6). The genomes were annotated manually in Artemis, and found to all be syntenic (although with different break points in the respective contigs). Sequence comparisons were carried out by first aligning gene protein sequences with ClustalX and then constructing nucleotide alignments from these as described above. The hive plots that form part of were created using the D3JS javascript framework (<http://d3js.org>), by adaptation of an example plot (<http://bl.ocks.org/mbostock/2066415>). ## Accession Numbers The ascidian marker sequences and mitochondrial assemblies have been submitted to the National Center for Biotechnology Information (NCBI) (<http://www.ncbi.nlm.nih.gov>). The accession numbers are as follows. **18S rRNA genes**: L3, KJ009375; L2, KJ009376; L5, KJ009377; L6, KJ009378; E11-097, KJ009379; 05-027, KJ009380; 05-033, KJ009381; 05-039, KJ009382; 05-044, KJ009383; 07-110, KJ009384; 07-103, KJ009385; 03-005, KJ009386; L1, KJ009387; L4, KJ009388; 07-005, KJ009389; 07-002B, KJ009390. **COXI genes**: L3, KJ009363; 05-033, KJ009364; 05-039, KJ009365; 05-027, KJ009366; 05-044, KJ009367; E11-097, KJ009368; L4, KJ009369; 07-103, KJ009370; 03-005, KJ009371; 07-002B, KJ009372; 07-005, KJ009373; 07-110, KJ009374. **Mitochondrial genome assemblies**: L2, KJ596321; L5, KJ596322; L6, KJ596323. # Supporting Information Permission to perform field research was granted by the Papua New Guinea Department of Environment and Conservation and facilitated by the University of Papua New Guinea. Additional permission for field collections was obtained from the governments of Solomon Islands, Fiji and Palau. We thank C. M. Ireland for help with collections, M. K. Harper for many helpful discussions on this paper (both University of Utah), Sam Hunter (Institute of Bioinformatics and Evolutionary Biology, University of Idaho) for helpful discussions on mitochondrial assembly, Anton Korobeynikov (St. Petersberg Academic University) for discussions on the use of the SPAdes assembler, and James Robert White for advice on the use of the skiff algorithm. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JCK TSB EWS. Performed the experiments: JCK MSD TPW. Analyzed the data: JCK MDBT TPW. Wrote the paper: JCK EWS. [^3]: Current address: Pharmaceutical Sciences Division, University of Wisconsin–Madison, Madison, Wisconsin, United States of America [^4]: Current address: Department of Bioengineering and Therapeutic Sciences and California Institute for Quantitative Biosciences, University of California, San Francisco, California, United States of America

# Introduction Organ and tissue donation is essential for fundamental and applied research. A state as close as in-vivo is equally important to ensure reliability of the results and clinical relevance. Needs are steadily increasing while availability is decreasing. Legal and cultural differences between nations regarding donation in general and those for research complicate comparisons. However, quantitative data are lacking as well as concrete solutions to correct imbalance. In ophthalmology, cornea is no exception. A growing shortage of tissue for research is well documented in the USA. Worldwide, shortage also concerns donation for transplantation. Cornea is by far the most transplanted tissue, but in 2015 only 1/70 patients waiting worldwide could benefit from a corneal graft each year. In France, corneal procurement is governed by the Bioethics Laws revised in 2018, based on the European standards. Cornea is procured from a deceased donor, then necessarily stored in an authorized eyebank which carries out tissue quality and microbiological safety tests. Donation chain is an opt-out system with an online national refusal register (NRR). In absence of refusal expressed during his/her lifetime, any deceased person may legally be procured. Routinely, the hospital coordination team systematically seeks to ask relatives (without precise guidelines on degree of kinship) about the intentions of their deceased if no instruction (written or oral) is left by deceased and the relatives themselves refuse, the coordinators trust them and the procurement is never imposed. The scientific human corneas used in France have two different origins: they come mainly from waste products of transplantation donation and a minority from body’s donation to science. The first are corneas discarded by quality controls in eyebanks. They represent about 50% of 11,000 corneas procured, but infected corneas must be withdrawn and only those that don’t have refusal for scientific use can be used. They are mostly poorer quality corneas to those required for transplantation. Besides these corneas are only available after several days of storage and are no longer fresh tissue. The second, resulting from donation in anatomy laboratories of medicine faculties, represents only a few hundred per year in France. They are the only fresh scientific corneas currently available, provided that death-to-procurement time is \<24hours when body arrives in laboratory. According to legislative framework, distribution of these corneas for research are free of charge in France. Among our research axes aims we developed an active storage machine for corneal grafts, which restores equivalent of intraocular pressure and renews storage medium. After testing it on porcine corneas, preclinical validation of this device was performed on a large series of human fresh corneas. To avoid all biases related to discarded corneas from eyebanks, we requested from the Agence de la Biomédecine (ABM, whose mission is to supervise, evaluate, promote procurement and transplantation) opportunity to procure fresh scientific human corneas, for this preclinical study and all other protocols using our device, and for protocols aiming at make evolve eyebanking, corneal imaging and bioengineering. We were concerned that donation for research could raise concerns because of its novelty and specificity. We therefore conducted a-year prospective study comparing characteristics of targeted donation for research and usual donation for transplantation. # Material and methods ## Ethical considerations All procedures conformed to the tenets of the Declaration of Helsinki for biomedical research involving human subjects. The ABM specifically authorized corneas procurement for research (PFS15-008 & PFS16-010) and study was approved by local Institutional Review Board “Ethics Committee of the CHU de Saint- Etienne, Research Commission of Terre d’Ethique” (IORG0007394, N°IRBN272016/CHUSTE). The ABM asked us to select for research only persons with medical contraindication to transplantation, based on the European standards, so as not to reduce corneas number for waiting recipients. All other inclusion criteria didn’t differ from those usually used for corneal transplantation donation. ## Study design We collected prospectively during a-year, all data concerning corneal donation for research and transplantation in the Saint-Etienne university hospital. Main objective was to compare acceptance rates between both groups. Secondary objectives were to analyze whether differences exist between both groups in donor profiles or corneas procured. We compared: 1/ consent seeking process data: obtaining consent methods (face-to-face or telephone), answer delay (immediate or delayed to think about); 2/ reasons for refusal; 3/ methods of expressing opposition to donation (NRR, instructions left to relatives); 4/ donors and corneas characteristics: endothelial cell density (ECD) measured 48hours after procurement (main quantitative quality criterion in eyebanks: done by the Saint-Etienne eyebank technicians for transplantation, and a unique skilled operator (TG) for research) and cataract operated eye (potentially with lower corneal quality); 5/ acceptance rate according to hospital coordinators experience to potentially optimize interview protocol. Finally, to verify if this new research task impacted transplantation activity, we compared this year’s study to previous 13 years' activity carried out in the same hospital environment with the same eyebank connection. ## Hospital coordination team missions Three nurses of the hospital coordination team were involved. One was very experienced (20 years of experience and about 2000 interviews), one was experienced (15 years of experience and about 1500 interviews), the third (devoting 50% of her time to administrative-regulatory tasks) had 5 years of experience and about 500 interviews. Regular mission consisted in obtaining consent to organ and tissue donation for transplantation. In France donation chain follows an opt-out system. More specifically for the corneas, they daily screened all intra-hospital deaths. Following French guidelines health authorities, maximum death-to-procurement time was 24 hours. Were recommended but not mandatory: death-to-body- refrigeration time \<4hours, if not death-to-procurement time should be \<12hours. In parallel, coordinators screened deceased persons with medical contraindication to donation for transplantation. Interviews were conducted according to standard written procedures, either face-to-face or by telephone, as we reported in 2002. For the research group, coordinators explained to the relatives of the eligible deceased that corneal donation wasn’t possible for transplantation due to medical contraindication, but for local medical research: so the coordinators asked the relatives to consent to cornea donation directly for research. General purpose of the corneal grafts research axes conducted by same team of surgeons and researchers was explained, if the relatives so wished. When clarifications were required, a detailed explanatory letter from department’s head (PG) could be provided to the relatives, just as follow-up support if needed. The type of the consent depended of each eligible deceased: written or oral refusal during lifetime for deceased, or oral consent or refusal (by telephone or face-to-face) from relatives contacted and informed. ## Corneas procurement All corneas were procured by 3 trained ophthalmology residents. Fresh human corneas were procured by in situ excision with the same settings whatever the target, transferred to our laboratory within 20 minutes (for research) or to eyebank as usual (for transplantation). Costs of sterile single-use instruments and organoculture medium for the research group were covered by our research laboratory funding. The ABM authorized us to procure only fresh scientific corneas: the remaining eye tissue was left on the deceased as procurements for transplantation, with tegumentary reconstruction ad integrum. ## Statistics Normality of continuous data distribution was analyzed with Shapiro-Wilk test with non-normality threshold of 5%. Normal distribution data were described by mean±SD, \[min-max\]. When the variable followed normal distribution, an unpaired Student t-test was used to compare donors’ characteristics of both groups. Chi² test was used to compare percentages. Statistical significance was set at P\<0.05, with two-tailed tests (unless specified one- tailed) and adjusted with Tukey technique when multiple tests were performed. Analyses were performed with SPSS 25.0 (IBM Corp, Armonk, NY). # Results ## Medical wards of pre-selected deceased In 12 months, on 1442 intra-hospital deaths, coordinators sought absence of opposition to donation for 371 deceased eligible to potential corneal donation: 127 for research, 244 (232 in circulatory arrest, 12 from multi-organ donor) for transplantation. For each group, deceased were mainly from 3 services: emergency (21%), geriatrics (20%) and intensive care (16%) for research, and intensive care (28%), emergency (12%) and pneumology (9%) for transplantation. ## Methods of interviewing the relatives and answer delay Seventy-one percent of interviews were conducted by telephone, without difference between both groups (72% for research, 70% for transplantation P = 0.389). Answer was immediate in 69% (257/371) of cases and didn’t differ between both groups (64% for research, 72% for transplantation P = 0.540). Relatives answered more often immediately during face-to-face interview (79% of cases) than by telephone (66% of cases) (P = 0.016), without difference between both groups (immediate answer was done: a/during face-to-face interview 74% for research, 81% for transplantation P = 0.393; b/ by telephone 61% for research, 68% for transplantation P = 0.218). ## Acceptance rates Consent was obtained in 62% (79/127) of cases for research and 54% (132/244) for transplantation (P = 0.135), providing 158 fresh corneas for research and 264 for eyebank respectively. Acceptance rates weren’t influenced by gender of eligible donors, with 57% (123/216) of acceptance among males and 57% (88/155) among females (P = 0.974) respectively for overall series, without difference between both groups: 62% of females gave their consent for research versus 52% for transplantation (P = 0.212); respectively 62% of males for research versus 55% for transplantation (P = 0.357). details acceptance rates within and between both groups, according to interview type and answer delay by relatives: no significant differences were found. Telephone made it possible to procure 110 corneas for research and 172 for transplantation (i.e. 67% of all corneas). Acceptance rates varied significantly with coordinator's experience: the most experienced obtained 71% and 62% of consent, respectively for research and transplantation, the one with intermediate experience 55% and 52%, the 3rd less experienced 43% and 39% (P = 0.003 for overall comparison between coordinators, both groups combined). None of them had more success for research than for transplantation (P = 0.253, P = 0.730, P = 0.824 respectively for the 3 coordinators). Coordinator's experience was directly correlated to acceptance rates whatever target: r = 0.9966 for research (P = 0.026), r = 0.9972 for transplantation (P = 0.024) (unilateral tests). ## Analysis of opposition to donation No opposition was found on NRR. Clear (oral or written) opposition from subject prior to death, left to the relatives, was found in 19% (9/48) of cases for research versus 37.5% (42/112) for transplantation (P = 0.045). In other cases, refusal was a relatives’ decision (from one person if alone, or from a collegial decision). There wasn’t any influence of the gender or relationship of the person(s) providing consent (P = 0.342 and P = 0.254 respectively). Regardless target (79% for research, 80% for transplantation), the relatives mainly didn’t provide any justification (P = 0.036). When refusal was mentioned, it was: "Donation isn’t priority" in 10% of cases for research and 11% for transplantation (P = 0.999); "Our relative suffered too much, was too sick or too old" in 8% of cases for research and 6% for transplantation (P = 0.999); relatives didn’t call coordinators back in 3% of cases for research and 3% for transplantation (P = 0.999). ## Donors characteristics and corneal cells quality Medical contraindications that allowed selection in the research group were mainly the cognitive disorders in 66% (52/79) of cases, followed by the blood cancers (8%). The details were provided in. Donors in the research group were significantly more likely to be females (P = 0.004) and older by an average of 10 years (P\<0.001); their bodies were also refrigerated quicker (P\<0.001). However, there wasn’t significant difference in death-to-procurement time (P = 0.911), percentage of eyes with cataract surgery (P = 0.120) and ECD (P = 0.071). Twelve pairs of corneas (15%) in the research group were available for experimentation within 6 hours or less after death, and 24 (30%) within 12 hours or less, with an average death-to-experimentation in laboratory time 15h15±6hours \[0h55;24h08\]. Further details of these characteristics were noted in. ## Comparison of transplantation activity with previous years With 244 families contacted for transplantation during this study year, activity wasn’t different from that of previous 13 years, which had concerned an average of 227±24 \[181 to 264\] families (P = 0.994). Similarly, number of corneas procured for transplantation (264) didn’t differ from the average of previous 13 years (284±24 \[from 223 to 330\] (P = 0.998)). # Discussion Corneal donation for research is still too little-known to public and must be encouraged as much as that for transplantation, both of which are essential to fight against blindness. We showed that it is possible to specifically target deceased to procure fresh corneas for research, in parallel with transplantation activity, without putting them in competition. Acceptance rate didn’t differ between the research and the transplantation groups, whereas we feared that donation for research might have raised more reluctance. Several mutual and non-exclusive explanations can be formulated: 1/ people who accept idea of corneal donation don’t oppose research and transplantation; 2/ some may think that they help even more people via research than transplantation, which treats traditionally only two patients; 3/ older age (10 years) in the research group can make higher consent rate; 4/ great experience of the coordinators in presenting legislation and research issues with empathy plays a key role; 5/ longstanding involvement of the coordinators in research projects, directly related here to improve corneal grafts storage keeps motivation over time; 6/ general public’s knowledge of our health region and team's research works (dedicated mainly but not only to corneal eyebanking and transplantation) through regular follow-up support for the relatives may help to make meaning of donation. But in practice relatives didn’t ask the coordinators for so much details about research aims. Our high acceptance rate for transplantation was significantly higher than those reported for transplantation in literature where rates exceeding two-thirds acceptance are exceptional. Comparison between the teams is highly complex because it questions donation regulation, overall organization of the teams, donors screening and experience of the coordinators evolving over time. Our findings result certainly from combination of an opt-out system and a small well-structure, highly motivated experienced team. Key role of "human factor" in success of this deeply personal approach was reported before. Our study further confirms crucial role of telephone interview, reported regularly. In 2002, we showed that for transplantation, telephone accounted for 58% of interviews and 48% of the corneas procured. Here for transplantation, its role is even more important with 71% of interviews and 67% of corneas procured. While in 2002 telephone was less efficient than face-to-face interviews, this difference wasn’t significant in 2017: communication resources evolution with the smartphones pervasive in daily life, is probably involved. Refusals analysis shows that NRR is probably widely underused. The donation opponents tell probably rarely their relatives why, with a potential memory bias, as the justifications collected by our coordinators seem to indicate. Whatever target, donation legislative framework like presence of NRR aren’t well known to general public. Generally speaking, question of donation isn’t still sufficiently addressed in families, even for favorable people with only 50% of them having informed their relatives. At time of death, when relatives have to decide in place of their deceased, fear of doing wrong or assuming responsibility can lead to refusal. The medical contraindications to corneal donation for transplantation are numerous, based mainly on precautionary principle solely and deprive recipients of potentially great intrinsic quality corneas: only rabies (constantly fatal), herpes simplex virus, bacteria, fungi and retinoblastoma have been formally demonstrated to be transmissible via the cornea. So, targeting these contraindications allows researchers to benefit from them. Some sporadic cases of prion disease were reported in recipients, and circumstantial evidence has implicated corneal transplantation as a mechanism of transmission of iatrogenic prion disease: transplantation is presumed (but not formal) to be the source of prion disease in recipients. Although we could be concern about disease transmission or research bias induced by these contraindications, it clearly depends on the research aim(s). In this study, every cornea was procured by in situ excision with single-use sterile instrument, without any contact with retina or optic nerve, which are be considered as specific risk factors to possible iatrogenic spread of sporadic and variant prion disease. We did risk analysis before the use of those corneas to develop our active storage machine to make evolve eyebanking, with assessment of their baseline intrinsic quality (endothelial cell density, transparency, presence of scar or not, presence of previous refractive surgery or not) and safety (no infection), following tests used in daily routine in eyebanks. Furthermore, with these common medical contraindications people may wish to give but feel excluded from donation: donation for research can be a way to give them back this opportunity. However many people are interested in donating their corneas for research but aren’t aware that it was possible to do so. Thus, they leave more often no instruction to their relatives concerning procurement for research. Targeted donation for research has another advantage: it allows the researchers to dispose the fresh tissues immediately after procurement, without passing through the eyebanks. Several positive consequences can be stated: it provides short circuit, without extra-work for the eyebanks and less charges for the researchers at the end of the chain (variations occur in different countries, such as the USA where procurements are done by paid technicians); it removes the damaging processes on cornea quality induced by storage itself. Our specific approach imposed by the ABM made it possible to procure a large quantity of fresh scientific corneas over a short period of time, without difference quality from those grafted, ensuring optimal clinical relevance for the research protocols in eyebanking. Through people giving their bodies to science, we found that we could just procure 41±17 fresh scientific corneas per year over last 3 years in our faculty of medicine: it would have taken almost 4 times longer to obtain same number of fresh corneas versus our study with 158 corneas a-year. Without these fresh tissues, preclinical validation of our active storage machine couldn’t have been performed under same conditions as its future use, like the other following protocols to improve corneal storage using our device with the extensions of authorizations by the ABM. Death-to- refrigeration time difference between groups was explained by difference in the services from which deceased came, not involving same post-death logistics. Donors age was logically higher, with more females in the research group, since we were asked to select potential donors for research on the diseases that were more frequent with aging and females live classically longer. It didn’t impact research protocols, since donor age isn’t a significant factor in survival of corneal grafts. Concerning coordinators, despite their additional workload to obtain targeted donation for research, there wasn’t negative impact on already important transplantation activity, neither on number of the families interviewed nor on number of corneas procured. That underlines possibility to have clinical and research processes coordinated without penalizing each other. The coordinators increased their efficiency without additional working time, didn’t receive bonus, made and have pursued it with belief: whatever research and clinical endeavors, they did both tasks without any conflict. Donation chain may vary in different countries. Regarding cost effectiveness of targeted corneal donation, we believe it may be applied whatever medico-economic system. In France, our laboratory had to pay only materials. But if needed, extra-cost from people in charge of obtaining consent or/and procurements, must be implemented at chain’s end for researchers. Anyway, without passing through the eyebanks, researchers will have less fees and better-quality fresh corneas, immediately available for experimentations, with this targeted corneal donation. Despite being prospective and designed to minimize potential bias, our study presents some limitations: 1/ center-effect with highly motivated coordinators that have strong and long-lasting links with medical-research team for two decades. However, we believe that extrapolation to the other research teams motivated to obtain exceptional quality tissues, could have similar results. 2/ net inter-coordinator effect. Despite using standard protocol, some variations were intrinsically linked to age-related experience: more experienced coordinator probably adapts better to each family profile whatever target donation. Several proposals have recently been made to improve acceptance of corneal donation for research. In the USA, Williams *et al*. advocates for scientific donation, by proposing creation of an eye donation registry for research; collaborations between the eyebanks and the research institutes to recover corneas unsuitable for transplantation, as we have implemented since almost 20 years to procure discarded grafts from our eyebank. Besides, proposed the creation of an online portal, under the aegis of the Association for Research in Vision and Ophthalmology, to specifically link the eyebanks to researchers in need of eye tissue. To make this work useful to other teams, we may suggest some ways to increase number of donation for research: 1/ to develop privileged relationships between the research laboratories and the team in charge of donation in local hospital. Exposing her regularly aims, advances and results of researches, is probably crucial to induce lasting links; 2/ to encourage companionship between the novice and the experienced coordinators; 3/ to integrate in donation promotion campaigns, awareness raising to donation legislative framework and possibility to donate for research; 4/ to increase the resources allocated to the coordinators and ensure that all eligible relatives of deceased can be contacted in a timely manner: pressure comes from legislative framework and bodies transportation from hospital mortuary to private funeral homes, explaining mainly gap between number of intrahospital deaths and deceased eligible contacted in this study. Respectively 413 families were not contacted due to legislative framework with death-to-procurement time \>24h (which discards deceased from eligibility to for corneal donation for research or transplantation); 411 families were not contacted due to fast transport of the deceased body from hospital mortuary to private funeral homes or conservative care (which prevent from interviewing families for research or transplantation). The third cause is that sometimes the coordinators didn’t have enough time to treat every dossier of eligible deceased (n = 243 families) for research or for transplantation in their daily multi-tasks: they share time between corneas, other tissues (vessels, bone), organs (mainly kidney). In summary, in parallel with transplantation activity, by targeting donors with medical contraindication to corneal donation for transplantation, we can obtain many fresh scientific corneas of similar quality to those grafted, immediately available for the researchers, without increasing shortage for waiting recipients or incurring extra-fees by passing through the eyebanks. This targeted corneal donation could be a potential solution to make research advance better and faster. # Supporting information We are grateful to those who donated their corneas to science, and to their families. We also thank the Agence de la Biomédecine for its institutional support and the authorizations. 10.1371/journal.pone.0233392.r001 Decision Letter 0 Liu Yu-Chi Academic Editor 2020 Yu-Chi Liu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 13 Feb 2020 PONE-D-19-35417 Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital PLOS ONE Dear Dr. GARCIN, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by Mar 29 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Yu-Chi Liu, M.D Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements: 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.plosone.org/attachments/PLOSOne_formatting_sample_main_body.pdf> and <http://www.plosone.org/attachments/PLOSOne_formatting_sample_title_authors_affi liations.pdf> 2\. Please provide additional details regarding participant consent. In the Methods section, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. 3\. Thank you for including your ethics statement: The Agence de la Biomedecine (ABM) specifically authorized corneas procurement for research (PFS15-008 & PFS16-010) and study was approved by local Institutional Review Board (IORG0007394, N°IRBN272016/CHUSTE). Please amend your current ethics statement to include the full name of the ethics committee/institutional review board(s) that approved your specific study. Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”). For additional information about PLOS ONE ethical requirements for human subjects research, please refer to <http://journals.plos.org/plosone/s/submission-guidelines#loc-human-subjects- research>. 4\. We note that you have a patent relating to material pertinent to this article. Please provide an amended statement of Competing Interests to declare this patent (with details including name and number), along with any other relevant declarations relating to employment, consultancy, patents, products in development or modified products etc. Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing-interests> by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. This information should be included in your cover letter; we will change the online submission form on your behalf. 5\. Thank you for stating the following in the Competing Interests section: "P GAIN, S ACQUART and G THURET are inventors on “patent US 20160029618A1” submitted by University Jean Monnet that covers “Medical device intended for long-term storage of a cornea, or for ex vivo experimentation on a human or animal cornea”. P GAIN and G THURET are consultant for Thea laboratories and Quantel Medical." We note that one or more of the authors are employed by a commercial company: Thea laboratories and Quantel Medical. a.     Please provide an amended Funding Statement declaring this commercial affiliation, as well as a statement regarding the Role of Funders in your study. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. You can update author roles in the Author Contributions section of the online submission form. Please also include the following statement within your amended Funding Statement. “The funder provided support in the form of salaries for authors \[insert relevant initials\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.” If your commercial affiliation did play a role in your study, please state and explain this role within your updated Funding Statement. b\. Please also provide an updated Competing Interests Statement declaring this commercial affiliation along with any other relevant declarations relating to employment, consultancy, patents, products in development, or marketed products, etc. Within your Competing Interests Statement, please confirm that this commercial affiliation does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to  PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing- interests>). If this adherence statement is not accurate and  there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. Please include both an updated Funding Statement and Competing Interests Statement in your cover letter. We will change the online submission form on your behalf. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: <http://journals.plos.org/plosone/s/competing-interests> 6\. We note that you have included the phrase “data not shown” in your manuscript. Unfortunately, this does not meet our data sharing requirements. PLOS does not permit references to inaccessible data. We require that authors provide all relevant data within the paper, Supporting Information files, or in an acceptable, public repository. Please add a citation to support this phrase or upload the data that corresponds with these findings to a stable repository (such as Figshare or Dryad) and provide and URLs, DOIs, or accession numbers that may be used to access these data. Or, if the data are not a core part of the research being presented in your study, we ask that you remove the phrase that refers to these data. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The study is technically sound with a valid scientific question. The results were interesting and surprising. We note that the paper did not mention whether the research cornea donors were tested for relevant communicable diseases and what the disposition of the corneas would be should the results come back positive. Was there any concern that some of the cognitive disorders were due to prion disease? Our other query would be the consent taking process for biomedical research, perhaps this can be mentioned in the paper? Will this be different from the consent for clinical transplant purposes? Or can we assume that ABM has already provided the necessary authorization and therefore no further consent is required? The manuscript is easily understood but there is room for improvement with regard to overall sentence construction, grammatical flow and word choices. Agree with the author’s view that the high consent rates may be a single centre effect and may not represent the situation in most other institutions around the world. However, it echoes the need for a consistent and sustained messaging to the general public about cornea donation, whether it be for therapy, research or education. Reviewer \#2: In this prospective study, the authors compared the donation acceptance rate for cornea transplantation (deceased patients with no medical contraindication for corneal transplantation) and for research (deceased patients with contraindication for corneal transplantation) and evaluated if the acceptance rate for transplantation donation remained stable over a year. General comments Obtaining fresh corneas for research is highly challenging and a main concern for worldwide researchers. While it is highly interesting to see that acceptance rates for research and for transplant were comparable, this paper opens several questions regarding research regulations of corneas. Indeed, in France, historically, ineligible corneas for transplant were being used for research, in consequence, research corneas came from patients without medical contraindication for transplant. In the present study, corneas from deceased patients with medical contraindication for transplant (including cognitive disorders, blood cancers, neurologic disorder, uncontrolled infection, tumor of CNS, severe influenza and hemodilution) were used for research to increase the number of fresh corneas needed for research purpose. The main question here might be “should we use tissues from cognitive disorders, blood cancers, neurologic disorder, tumor of CNS, uncontrolled infections… for our research?” Indeed, the authors explained that one of the main focus of their lab was to develop an active storage machine for corneal graft. We could be concern about both disease transmission and induced research bias which is not at all discussed in the paper. Thus, the discussion should address those points which can’t be only resumed in one sentence (Line 279-281) “The medical contraindications to corneal donation for transplantation are numerous, based mainly on precautionary principle solely and deprive recipients of potentially great intrinsic quality corneas”. Was any risk analysis performed before the use of those corneas? Point by point comments Some additional points need to be clarified and are listed below. Results Line 153 the authors reported 1442 intra-hospital deaths but only 371 deceased eligible. Can the authors explain the reasons of the gap since cornea for research are based on medical contraindications meaning that families of all deceased patients should be interviewed. What were the limitations? Table 1 is nonreadable, please correct. In addition, in the table, the total of research interviews is 126 but number written in results line 154 is 127. Please review and correct. Line 212 “medical contraindications that allowed selection in the research group” can you please clarify the French rules about research on tissues with medical contraindications for transplant? End of comments \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: Yes: Howard Cajucom -Uy Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0233392.r002 Author response to Decision Letter 0 9 Mar 2020 PONE-D-19-35417 Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'. • A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'. • An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. First, we would like to thank the Academic Editor, reviewer \#1 and \#2 for her/his reviewing and advises. We have chosen to respond to the comments and to resubmit this manuscript. We deeply apologize for grammar or language mistakes. Journal Requirements: When submitting your revision, we need you to address these additional requirements: 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.plosone.org/attachments/PLOSOne_formatting_sample_main_body.pdf> and <http://www.plosone.org/attachments/PLOSOne_formatting_sample_title_authors_affi liations.pdf> No problem, we followed guidelines 2\. Please provide additional details regarding participant consent. In the Methods section, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. Consent was informed. In France we have opt out system. The type of the consent depended of each eligible deceased : written or oral refusal during lifetime for deceased, or oral consent or refusal (by phone or face to face) from relatives contacted. We precised it in Methods section. 3\. Thank you for including your ethics statement: The Agence de la Biomedecine (ABM) specifically authorized corneas procurement for research (PFS15-008 & PFS16-010) and study was approved by local Institutional Review Board (IORG0007394, N°IRBN272016/CHUSTE). Please amend your current ethics statement to include the full name of the ethics committee/institutional review board(s) that approved your specific study. Once you have amended this/these statement(s) in the Methods section of the manuscript, please add the same text to the “Ethics Statement” field of the submission form (via “Edit Submission”). For additional information about PLOS ONE ethical requirements for human subjects research, please refer to <http://journals.plos.org/plosone/s/submission-guidelines#loc-human-subjects- research>. We have amended the full name of our institutional review board that approved our specific study Ethics Committee of the CHU de Saint-Etienne, Research Commission of Terre d’Ethique We precised it in the Methods Section and in the “Ethics statement” field of the submission form. 4\. We note that you have a patent relating to material pertinent to this article. Please provide an amended statement of Competing Interests to declare this patent (with details including name and number), along with any other relevant declarations relating to employment, consultancy, patents, products in development or modified products etc. Please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials, as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing-interests> by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. The patent does not alter our adherence to PLOS ONE policies on sharing data and materials. This information should be included in your cover letter; we will change the online submission form on your behalf. OK no problem. 5\. Thank you for stating the following in the Competing Interests section: "P GAIN, S ACQUART and G THURET are inventors on “patent US 20160029618A1” submitted by University Jean Monnet that covers “Medical device intended for long-term storage of a cornea, or for ex vivo experimentation on a human or animal cornea”. P GAIN and G THURET are consultant for Thea laboratories and Quantel Medical." We note that one or more of the authors are employed by a commercial company: Thea laboratories and Quantel Medical. a\. Please provide an amended Funding Statement declaring this commercial affiliation, as well as a statement regarding the Role of Funders in your study. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. You can update author roles in the Author Contributions section of the online submission form. Please also include the following statement within your amended Funding Statement. OK, no problem. “The funder provided support in the form of salaries for authors \[insert relevant initials\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.” “The funder provided support in the form of salaries for authors \[PG, GT\], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.” If your commercial affiliation did play a role in your study, please state and explain this role within your updated Funding Statement. No commercial affiliation played role in our study. b\. Please also provide an updated Competing Interests Statement declaring this commercial affiliation along with any other relevant declarations relating to employment, consultancy, patents, products in development, or marketed products, etc. Within your Competing Interests Statement, please confirm that this commercial affiliation does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing- interests>). If this adherence statement is not accurate and there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared. Commercial affiliations do not alter our adherence to PLOS ONE policies on sharing data and materials. Please include both an updated Funding Statement and Competing Interests Statement in your cover letter. We will change the online submission form on your behalf. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: <http://journals.plos.org/plosone/s/competing-interests> 6\. We note that you have included the phrase “data not shown” in your manuscript. Unfortunately, this does not meet our data sharing requirements. PLOS does not permit references to inaccessible data. We require that authors provide all relevant data within the paper, Supporting Information files, or in an acceptable, public repository. Please add a citation to support this phrase or upload the data that corresponds with these findings to a stable repository (such as Figshare or Dryad) and provide and URLs, DOIs, or accession numbers that may be used to access these data. Or, if the data are not a core part of the research being presented in your study, we ask that you remove the phrase that refers to these data. We apologize for that, we precised this point in reviewed manuscript. Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes 3\. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The study is technically sound with a valid scientific question. The results were interesting and surprising. We note that the paper did not mention whether the research cornea donors were tested for relevant communicable diseases and what the disposition of the corneas would be should the results come back positive. Was there any concern that some of the cognitive disorders were due to prion disease? Thank you for your comment, we have added the following sentences in the discussion section (lines 291 to 307) “only rabies (constantly fatal), herpes simplex virus, bacteria, fungi and retinoblastoma have been formally demonstrated to be transmissible via the cornea. So, targeting these contraindications allows researchers to benefit from them. Some sporadic cases of prion disease were reported in recipients, and circumstantial evidence has implicated corneal transplantation as a mechanism of transmission of iatrogenic prion disease : transplantation is presumed (but not formal) to be the source of prion disease in recipients \[31-34\]. Although we could be concern about disease transmission or research bias induced by these contraindications, it clearly depends on the research aim(s). In this study, every cornea was procured by in situ excision with single-use sterile instrument, without any contact with retina or optic nerve, which are be considered as specific risk factors to possible iatrogenic spread of sporadic and variant prion disease \[35, 36\].” Our other query would be the consent taking process for biomedical research, perhaps this can be mentioned in the paper? Will this be different from the consent for clinical transplant purposes? Or can we assume that ABM has already provided the necessary authorization and therefore no further consent is required? Thank you for your comment. We clarified this point lines 122 to 139. The manuscript is easily understood but there is room for improvement with regard to overall sentence construction, grammatical flow and word choices. We deeply apologize for grammar or language mistakes. We updated this point. Agree with the author’s view that the high consent rates may be a single centre effect and may not represent the situation in most other institutions around the world. However, it echoes the need for a consistent and sustained messaging to the general public about cornea donation, whether it be for therapy, research or education. Reviewer \#2: In this prospective study, the authors compared the donation acceptance rate for cornea transplantation (deceased patients with no medical contraindication for corneal transplantation) and for research (deceased patients with contraindication for corneal transplantation) and evaluated if the acceptance rate for transplantation donation remained stable over a year. General comments Obtaining fresh corneas for research is highly challenging and a main concern for worldwide researchers. While it is highly interesting to see that acceptance rates for research and for transplant were comparable, this paper opens several questions regarding research regulations of corneas. Indeed, in France, historically, ineligible corneas for transplant were being used for research, in consequence, research corneas came from patients without medical contraindication for transplant. In the present study, corneas from deceased patients with medical contraindication for transplant (including cognitive disorders, blood cancers, neurologic disorder, uncontrolled infection, tumor of CNS, severe influenza and hemodilution) were used for research to increase the number of fresh corneas needed for research purpose. The main question here might be “should we use tissues from cognitive disorders, blood cancers, neurologic disorder, tumor of CNS, uncontrolled infections… for our research?” Indeed, the authors explained that one of the main focus of their lab was to develop an active storage machine for corneal graft. We could be concern about both disease transmission and induced research bias which is not at all discussed in the paper. Thanks for this comment. We added new comment concerning this point. (cf. response to comment of reviewer \#1 about prion disease =\> line 291 to 307) Thus, the discussion should address those points which can’t be only resumed in one sentence (Line 279-281) “The medical contraindications to corneal donation for transplantation are numerous, based mainly on precautionary principle solely and deprive recipients of potentially great intrinsic quality corneas”. We added new comment concerning this point (line 291 to 307). Was any risk analysis performed before the use of those corneas? Yes, We added new comment concerning this point (line 303 to 307). “We did risk analysis before the use of those corneas to develop our active storage machine to make evolve eyebanking, with assessment of their baseline intrinsic quality (endothelial cell density, transparency, presence of scar or not, presence of previous refractive surgery or not) and safety (no infection), following tests used in daily routine in eyebanks.” Point by point comments Some additional points need to be clarified and are listed below. Results Line 153 the authors reported 1442 intra-hospital deaths but only 371 deceased eligible. Can the authors explain the reasons of the gap since cornea for research are based on medical contraindications meaning that families of all deceased patients should be interviewed. What were the limitations?, We added new comment concerning this point (line 371 to 380). Gap can be explained by several causes : 1/ legislative framework with death-to-procurement time \>24h, which discards deceased from eligibility to for corneal donation for research or transplantation (n= 413) 2/ fast transport of the deceased body from hospital mortuary to private funeral homes or conservative care, which prevent from interviewing families for research or transplantation (n=411) 3/ Only 3 nurses of the coordination team are present to do all the work about corneas but also for other tissues (vessels, bone) and organs (mainly kidney). So, sometimes they do not have enough time to treat every dossier of eligible deceased for research or for transplantation in their daily tasks (between corneas, other tissues, organs) (n=243) ; so increase the resources allocated to the coordinators is crucial. 4/ medico-legal impediment with prosecutor objection (n=4). Table 1 is nonreadable, please correct. In addition, in the table, the total of research interviews is 126 but number written in results line 154 is 127. Please review and correct. We apologize for that ; we reviewed and corrected on revised manuscript. Line 212 “medical contraindications that allowed selection in the research group” can you please clarify the French rules about research on tissues with medical contraindications for transplant? The Rules are edited and updated regularly by the Health Authorities, and this is The Agence de la biomédecine provides the rules : for each tissue medical contraindications exist for transplantation. Ineligibility for transplantation makes tissue discarded and so destructed. But some of this discarded tissue which does not represent infectious risks, may be used by specific, certified and authorized laboratories. For ocular tissue, distribution of tissue are free of charge in France. End of comments 6\. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy. Reviewer \#1: Yes: Howard Cajucom -Uy Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0233392.r003 Decision Letter 1 Liu Yu-Chi Academic Editor 2020 Yu-Chi Liu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 30 Apr 2020 PONE-D-19-35417R1 Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital PLOS ONE Dear Dr. GARCIN, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that the manuscript has significantly improved but the reviewers still have some minor concerns. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by Jun 14 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Yu-Chi Liu, M.D Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) Reviewer \#2: Discussion Line 268 misspelling of “iin” please replace by “in” The authors may want to add some additional information regarding the CoVid current situation which will probably change their future practice. In their study, all deceased patients were eligible for research donation, will they now consider testing all deceased patients before harvesting corneas? Are French rules currently changing regarding cornea donation for research? \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: Yes: Howard Cajucom-Uy Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0233392.r004 Author response to Decision Letter 1 1 May 2020 PONE-D-19-35417R1 Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital PLOS ONE Saint-Etienne, May 1st 2020 Dear Professor Yu-Chi Liu, Please find attached a revised version of our article titled “Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital”. We have carefully considered and responded to all the points addressed by the reviewers. Per your instructions, all substantive amendments in the revised version are stated in our point-by-point response, and are marked in red in the article. We greatly hope that this new version will meet the reviewers' expectations and comply with your editorial policy. Yours sincerely, Dr. Thibaud GARCIN, M.D., Ph.D., FEBO Saint-Etienne University Hospital "Corneal Graft Biology, Engineering, and Imaging" Laboratory EA 2521 Faculty of Medicine Saint-Etienne France Dear Dr. GARCIN, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that the manuscript has significantly improved but the reviewers still have some minor concerns. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by Jun 14 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: • A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'. • A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'. • An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Yu-Chi Liu, M.D Academic Editor PLOS ONE Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: (No Response) 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes 4\. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) Reviewer \#2: Discussion Line 268 misspelling of “iin” please replace by “in” This has been corrected The authors may want to add some additional information regarding the CoVid current situation which will probably change their future practice. In their study, all deceased patients were eligible for research donation, will they now consider testing all deceased patients before harvesting corneas? Are French rules currently changing regarding cornea donation for research? Thank you for the suggestion. The continuation of corneal procurement as early as possible is indeed crucial both for patients waiting for transplants and for laboratories like ours that work on these irreplaceable human tissues. The SARS- Cov-2 epidemic has effectively stopped all but multi-organ donation. It appears that we may soon be able to resume therapeutic procurement from symptomatic, non-at-risk individuals, but there are many unanswered questions. In order to be able to answer the question of the risk of donor-recipient transmission, we have already started a new research work and obtained the authorization from our health authority (Biomedicine Agency, PFS-2020-011) to resume retrieval for scientific purposes for the next 150 donors. Thus, we will systematically collect and test all donors (nasopharynx, conjunctiva and cornea) and carry out serologies. This will provide us with robust data to make good decisions about what new tests to perform or not to perform on donors, for transplantation and for research. We hope that the new recommendations will be based on evidence and not on an unfounded precautionary principle. Of course, during this new study we are recording the acceptance rate for corneal donation among all potential donors, and possible reasons for refusal. We will know whether the current epidemic is changing behavior and we will submit a letter to report the results, if you wish. For this article, as suggested, we would like to add the following paragraph as the end of discussion: “By the time this article is accepted, the global SARS-Cov-2 epidemic has stopped corneal procurement for therapeutic and scientific purposes altogether. Scientific knowledge is sorely lacking to establish the risk of transmission of this virus via ocular tissues and therefore to make new recommendations on which virological tests should be added or not to ensure total safety. Our French health authority has authorized us in emergency (PFS-2020-011) to perform a large series of scientific procurements from any potential donor (COVID+ or -) in order to objectively analyze the risks. During this particular study we will also analyze whether the epidemic has modified the acceptance rate and the reasons for a possible refusal.” Respectfully Dr T Garcin, MD, PhD, FEBO 7\. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy. Reviewer \#1: Yes: Howard Cajucom-Uy Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] 10.1371/journal.pone.0233392.r005 Decision Letter 2 Liu Yu-Chi Academic Editor 2020 Yu-Chi Liu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 May 2020 Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital PONE-D-19-35417R2 Dear Dr. GARCIN, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Yu-Chi Liu, M.D Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0233392.r006 Acceptance letter Liu Yu-Chi Academic Editor 2020 Yu-Chi Liu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 May 2020 PONE-D-19-35417R2 Corneal donation for research versus for transplantation: a-year prospective study of acceptance rates in a French University Hospital Dear Dr. Garcin: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Yu-Chi Liu Academic Editor PLOS ONE [^1]: GAIN, S ACQUART and G THURET are inventors on “patent US 20160029618A1” submitted by University Jean Monnet that covers “Medical device intended for long-term storage of a cornea, or for ex vivo experimentation on a human or animal cornea.” The patent does not alter our adherence to Plos One policies on sharing data and materials.

# 1. Introduction At present, water injection into a coal seam is the most effective way to reduce dust but is not an ideal solution in the field, mainly because the pore-fracture structure of coal seams is very complex, and its morphological and structural characteristics determine the physical and chemical properties of the coal. The occurrence and flow characteristics of fluid in different coal seam structures vary. How to reconstruct the pore-fracture structure in coal seams and determine the relationship between the connectivities of the pores and fractures is an important means to determine the seepage flow capacity of the coal. Recently, CT technology has been used to study the pore-fracture structure of coal seams. This method adopts fault imaging technology, which has many advantages, such as nondestructive 3D digitization and refinement. Nondestructive visual measurement of the pore-fracture structure in a coal seam can be realized, and a 3D network model of real pores can be obtained. Many scholars have carried out a considerable amount of research on the process of CT scanning restoration and reconstruction. Researchers R.SH.MIKHALL and others have calculated the volume surface and hydraulic radius of a micropore group by analysing a kind of silica gel pore. J. T. Fredrich et al revealed the geometric complexity of pore spaces by imaging the pore structures of 3D geological materials using laser scanning confocal microscopy and 3D reconstruction. C. R. Clarkson carried out low-pressure nitrogen adsorption and high-pressure mercury intrusion measurements on shale reservoir coal samples. The results showed that the specific surface area and pore volume results were quite different among the coal samples. Hiroshi Okabe et al used two-dimensional rock flakes to describe the three-dimensional pore space of a rock, assuming isotropy in the reconstruction process, and obtained three-dimensional stereograms. Li et al used CT scanning technology and electron microscope scanning technology, the fracture pore structure at different scales is studied, and the influence of pore characteristics on coal seam gas storage capacity is analyzed. Alexandra Roslin et al used CT scanning technology, the influence of confining pressure on coal seam fracture is analyzed. The results show that confining pressure can inhibit coal structure to some extent. Zhang et al results showed that the effect of fracturing on the horizontal and vertical direction of coal body is obtained by the method of fractal dimension counting. After analyzing the pore volume and roar length before and after fracturing, it is concluded that liquefaction fracturing can improve the permeability of coal body. S.M.Shah et al. used lattice Boltzmann and pore network models to simulate single- and two-phase flow, analysed porosity results, predicted porosity, unidirectional permeability and multiple physical properties at different length scales, and improved the flow law from the pore scale to the core scale. At present, most scholars mainly focus on CT scanning to restore the distribution of pore fissures in coal rock. Few studies have been performed on the multiscale characterization and modelling of coal, the seepage flow law of water in real pores and the dynamic seepage evolution mechanism. Based on this work, the dynamic seepage process of coal seam water injection has been accurately described. This paper uses the high-precision CT scanner by Dandong Aolong Ray Co., Ltd., to scan the coal body, filter the interference noise in the original image by a median filter, and reconstruct the three-dimensional pore network model of the coal body by using Avizo numerical simulation software. The pore and fissure structures at different scales are extracted. The maximum sphere algorithm and coordination number are used to analyse the relationship between the connectivities of pore and fissure groups at different scales, and the effects of different quantitative relationships between the pore size and pore throat channel permeability are studied. Avizo software is used to simulate the flow path of a fluid in the seepage channel, and the trend of the variation in fluid velocity between different seepage channels is discussed to explore the mechanism of the microseepage of fluid in a coal body and provide guidance for coal seam water injection and dust control measures. # 2. Material and methods ## 2.1. Coal sample collection The coal was taken from the No. 4 coal seam in Leijia District of Fuxin City, and this coal is long-flame coal. According to the GB/T 482–2008 Chinese national standards for sampling, the coal specimen was cut into 5 cm×5 cm×10 cm cuboids by a standard coal sample cutting machine in the laboratory, and three standard coal samples were obtained. An SDLA618 proximate analysis of the coal was carried out by using an industrial analyser. The results are shown in. Experimental coal samples are shown in. ## 2.2. Experimental equipment and methods The basic principle of CT detection technology is that the cross-section of the penetrating object absorbs the X-ray, and the detector receives the information of the X-ray passing through the section of the layer. When the uniform object is irradiated by X-ray penetration, the attenuation coefficient x of the object is exponential. The black and white colourscale in the grey image from the CT scanning process clearly characterize the density distribution of the object under detection, including the fissures, texture and other defects inside the object. The absorption ability of the object to the X-ray is also different, which is reflected by its attenuation coefficient. Thus, the pore structure of the coal body can be revealed without damage, and the coal sample can be further tested after CT scanning. As shown in, using a Dandong Oron Ray Instrument Co., Ltd., AL-CT-225 industrial X-ray CT detection system, the coal sample was placed in the centre of the platform. A variety of detection methods, such as cone beam scanning and DR real-time imaging, were used. During 360° coal sample rotation, scan information about the defect locations and porosity was obtained. A total of 720 projection images were collected. The coal sample scan was performed at a radiation source voltage of 60 kV at a high temperature and constant pressure and a current of 240 μA. The minimum CT spatial resolution of the tomography was 6 μm, and the density resolution was 0.5, a maximum of 1024×1024 pixels. ## 2.3. Analysis of scan results An initial image of the coal sample is obtained by CT scanning technology, as shown in. A group of slices in the yz, xz, and xy planes are selected. The black linear feature is a fissure in the coal sample, and the yz plane has a large number of pore-fracture groups. There are clearly fracture channels and pores visible in the coal matrix in the xz and xy planes. However, there is a considerable amount of noise in the original image, so it is necessary to select the appropriate filtering function to further process the image. ## 2.4. Median filtering of CT image noise There is system noise in the original image obtained by X-ray CT scanning. The system noise is mainly due to the noise that inevitably arises in the process of data transmission. The noise is randomly distributed in the image. These noise points will increase or decrease the real pixel value of the image. Noise is often represented as an additional pixel position or block on the image. Therefore, the quality of the image is reduced, and the restoration, feature extraction and interpretation of the image are affected. In this paper, a median filter is selected to process the noise in the image. Processing the noise in the image can not only eliminate the noise in the image but also maintain the data of the original image. The grey value of the original image is not changed. is the original image of a slice of the CT scan result, and is the image filtered by the median filter. The median filter is used to filter the noise points in the image, to protect the data of the original pore structure and clarify the edge position of the image clearly so that it is not blurred. The processed image can be used as the basis image for reconstructing the three- dimensional coal body structure. # 3. Results and discussion ## 3.1. Comprehensive characterization of pore-fissure spaces at different scales The coal images from the CT scanning reflect the pore-fissure characteristics of each layer. To analyse the three-dimensional pore-fissure distribution characteristics of the coal samples, a three-dimensional reconstruction of the coal samples is carried out. The internal structure at different scales is extracted and characterized. The two-dimensional CT slice after median filtering is used to construct the three-dimensional model of the coal sample by using the volume rending module in Avizo, as shown in. The pore-fracture characteristics in the coal samples are extracted with the threshold segmentation module, The data parameters of pore volume 3d, pore radius and roar length of coal sample can be analyzed by using Label Ansysaily function in Avizo, and the pore model after segmentation of the coal matrix is accurate undamaged, as shown in. shows the comprehensive spatial characterization pattern of the pore fissures, in which the pore fissures of the coal sample result in massive, isolated pore blocks with a zonal vertical distribution. These blocks in the coal are distinguished by different colours, and the connected fissures are displayed in the same colour. The results of the quantitative analysis of the coal samples show that the total pore volume imaged is 3.6735×10¹³ μm³, the total pore area is 9.098×10¹⁰ μm², there are a large number of micron-sized pores in the coal samples, and the coal samples show good three-dimensional uniformity in terms of the pore distribution. As shown in the, the three-dimensional spatial distribution of the coal pores at different scales is extracted. shows that there are a large number of very small pores; larger pores are added in, and some connected pores are added in. As the scale increases, the pore connectivity is gradually revealed. By extracting coal bodies at different scales, the distribution of different components at each scale can be clearly seen. The diagram shows that the pores of a sample will disform pore-fracture groups with increasing scale. This part of the connected pores will provide the main seepage channels for the coal body. In the process of seepage flow, hydraulic fractures communicate with connected pore groups. ## 3.2. Quantitative analysis of pores and pore throats The diversity of the pore system of coal and rock depends on the pore structure distribution, which is reflected in the quantitative relationship among the pore size, pore volume area, pore throat channel radius and pore throat channel length. The distribution law of the pores and pore throat channels in coal samples is the key to determining the permeability of coal seams. The pore radius determines the connectivity between pore throat channels, and the radius and length of the pore throat channels directly determine the permeability of the coal seam. As shown in, the distributions of pore radius, volume contribution rate and pore throat radius and length of coal samples are given. Analysis of the data shows that the pore radii below 500 μm account for 85% of all the pores and that the pore radii greater than 500 μm account for 15% of all the pores. The coal samples have pores with sizes of 200–300 μm, and the pore throats in coal are well developed. The pores in a coal sample are well connected by the pore throats. The micron-sized pores in the coal samples are divided into micropores, transition pores and mesopores, defined in pore radius ranges of 0–500 μm, 500–2000 μm and 2000 μm, respectively. The three-dimensional spatial distribution constants of the micropores, transition pores and mesopores in the coal samples are shown in. The pore size analysis of the coal sample data shows that the number of micropores, transition pores and mesopores is 2763, 598 and 96, respectively, the contribution of micropores and transition pores to the total pore volume is 0.78% and 11.21%, and mesopores account for 88.01% of the total pore volume. Despite the number of micropores and transition pores, due to their small sizes, micropores and transition pores contribute little to the connectivity of the fluid channels. The mesopores, due to their larger pore size, mainly provide the seepage passages. also show good connectivity in 3d visualization. Based on quantitative pore space analysis of coal samples containing micropores (79.92%) and transition pores (17.29%), the pore volume ratio is 11.99%, and the pore volume ratio also considering the moderately developed mesopores (2.79%) is 88.01%. shows the scatter plot of the pore throat radius and pore throat length results of the coal samples and the corresponding law determined by exponential function fitting. The pore throat radii range from 46.29 to 4286.89 μm, and the lengths of the pore throats range from 4206.28 to 48073.5 μm. The volumes of the pore throats range from 0.4 to 121 mm³. From the diagram, most of the pore throat radii in the coal sample are in the range of 100–2000 μm, and the pore throat lengths are concentrated in the range of 5000–25000 μm. The results show that the distribution of the pore throat radius and length parameters in the coal samples are fairly uniform, and the pores are connected through the pore throat channels. Compared with the results of the complex 3D pore morphology of coal-like porous media, micron-scale pores are more developed in these coal samples. The three-dimensional pores are intricate, indicating that the macroscopic pores and microscopic cracks in coal samples have a complex spatial distribution and interconnectivity. ## 3.3. Connectivity analysis and network characteristics of a porous rock To further explore the permeability of the coal samples, it is necessary to further extract the pore-fracture structure of the coal body, identify the pore position and form an equivalent pore network. The maximum sphere algorithm is generated by uniformly distributing spheres at the pore voxels in the pore space, one for every two voxels in each direction. The maximum ball algorithm is as follows: $$EqDiameter = \sqrt[3]{\frac{6 \times Volume3d}{\pi}}$$ EqDdiameter is the equivalent pore size, μm, and Volume3d is a single pore volume, μm³. In, the spheres represent the radii of the pore throat channels. The coloured cylindrical connection represents the connectivity between the pore throat channels, that is, the permeability of the coal sample. The pore throat channel connectivity, pore throat channel length and pore throat channel radius directly affect the coal permeability. The coal-like pores and pore throats are extracted using the maximum sphere algorithm. The extracted pore network model has a total of 3458 pores, 151 pore throats and 93 connected pore channels (representing the coordination number). The branches and endpoints of the network in the figure are called pores, and the lines connecting pores are called pore throats. The maximum pore volume is 3.51×10¹² μm³, and the corresponding pore area is 5.85×10⁹ μm². The maximum capacity of the pore throat channel is 5.77×107 μm², and the maximum length is 4286.89 μm. As shown in the pore network model in, the coal samples contain a large number of very small pores, which are basically not connected with other pores. With increasing pore radius, the connected channels increase, and the permeability increases gradually. That is, the small cracks in the coal often appear as isolated points. Because they are not closely connected with the internal pores, the very small pores cannot be connected with the pore throat channels or contribute to the seepage. These isolated pore structures often do not have seepage characteristics. Because pores with larger pore sizes are fully developed and connected in the coal body, several complete seepage channels are formed with other pore channel tributaries, which plays a decisive role in the seepage flow capacity of the coal seam. ## 3.4. Matching analysis The connectivity between pores directly determines the permeability of the coal seam. The coordination number reflects the connectivity between pores and pore throats, that is, if a pore is interlinked with pore throats. The pore structure parameters represented by the coordination numbers can reflect the spatial characteristics of the whole pore system and be used to analyse the pore system more accurately. shows the frequency histogram of the coordination number distribution. Pores with a coordination number of zero can be regarded as isolated pores. This part of the pore characteristics often shows that the pore volume area is small, it is basically not connected with other fissures, and the contribution to the mass flow rate of the seepage channel is very small. The flow trajectory of the fluid in the coal body runs along the seepage channel, that is, the connected pores form the seepage zone region. In reality, this part of the pore needs other seepage channels to extend the pressure-driven seepage to the isolated pore area to form a complete seepage zone. For pores with more than one coordination number, a complete seepage channel is formed because of the connectivity with other pores. The seepage water moves along the seepage path, and the larger the coordination number is, the closer the connection and the greater the number of seepage channels in the coal. thus, the larger the coordination number is, the more seepage occurs in the coal, and the better the wettability of the coal samples. ## 3.5. Permeability simulation principle The ability of porous media to allow single-phase fluid passage under a certain pressure gradient is defined in square metres, but square micrometres are more widely used in practice because 1 square micrometre is closer to 1 Darcy (d) and 1 d = 0.9869233 is the constant coefficient of the absolute permeability of a material according to Darcy’s law of fluid flow. <img src="info:doi/10.1371/journal.pone.0252277.e002" id="pone.0252277.e002g" /> Q S = − k μ △ P L Q is the flow (m³/s) of fluid through the porous medium in a unit of time; S is the cross-sectional area of the porous medium (m²); is the absolute permeability (Pa.s); k is the dynamic viscosity coefficient of the fluid flow; △P is the pressure differential of the fluid in the medium (Pa); L is the length of the fluid passing through a porous medium (m). $\frac{Q}{S}$ usually represents the flow velocity of fluid flowing through the surface of a porous medium and is also known as the Darcy velocity. When a porous medium is saturated with single-phase fluid, the permeability is called the absolute permeability, and when a porous medium is filled with a multiphase fluid, the permeability is called the relative permeability. ## 3.6. Permeability simulation results For a further analysis of the percolation ability of the pore-connected channel, seepage simulation of the real pore network model is carried out, and seepage simulation of the coal sample is carried out in the x and z directions. The flow characteristics and transportation capacity of the fluid can be obtained by analysing the flow trajectory. The simulation software adopts AVIZO software, the model size is 5 cm×5cm×10cm, the inlet parameter is set to 1 MPa pressure, the outlet parameter is set to 0 MPa, the dynamic viscosity parameter of the fluid is set to 1.01×10⁻³ Pa·s. is a seepage trajectory simulation diagram for the z direction; the fluid direction is set to flow from left to right, a purple streamline represents the medium-velocity region of the fluid, and a yellow streamline represents the high-speed region of the fluid. The diagram shows that the high-speed region is the connected region among the pore throat channels, that is, the region in which each tributary connects to the others. The convergence of each branch leads to momentum superposition, which makes the velocity increase, and the fluid velocity is in the high-speed region. The medium-velocity region of fluid is mainly due to the poor development of pore fractures and the decrease in the pore throat radius in the coal sample, which leads to weak connectivity between the pore throat channels and the loss of some momentum when the fluid passes through the seepage channel. The fluid velocity is a medium velocity. To better observe the influence of pore connectivity and the length of the pore throat channel on the percolation ability, represents the pore throat channel connectivity and the percolation trajectory superposition diagram. The fluid velocity is faster in the superposition region of the large pores and seepage channels. In the narrow aperture area, the velocity is relatively slow. shows the trajectory map of the seepage in the x direction, and the grey streamlines correspond to the low-velocity zone of seepage. It can be clearly seen that the high-speed zone in the diagram is characterized by tight and large pores and that the region of smaller pores corresponds to the medium velocity region. The main reason for the low-speed region is that the pore throat channels are close to the outer wall of coal, and the pore connectivity is not good there. shows that when several seepage channels converge, the seepage channels are relatively flat, and the velocity is high. In the flow of microscale pore structure, the whole structure of pore system is connected, but there are some isolated pore channels, which are basically zero. In the process of fluid migration in connected channels, the velocity change of fluid trajectory is more complex, which does not follow the linear and nonlinear relationship. First of all, the size of the roar volume is the most important factor to determine the seepage capacity, that is, the degree of pore connectivity. Secondly, the size of pore roar radius is a secondary factor to reflect the seepage capacity of coal seam. Finally, the distribution of pore roar structure is the final factor. # 4. Conclusions \(1\) The pore-fissure structures in coal samples are analysed qualitatively at different scales. The reconstructed 3D pore network models of the coal samples clearly divides the pores, fissures and matrix parts of the coal samples. The pores in the coal samples are nonuniformly distributed and vertically connected, and very small pores exist in isolation. An accurate pore model is of great significance for the analysis of fluid seepage in coal samples. \(2\) The pore radius, pore throat length and pore connectivity parameters of the coal samples are analysed by using the maximum sphere method. The coal samples contain 3458 pore points, 151 pore throat channels, a volume of 3.6735×1013 μm³, and a pore capacity of 9.098×1010 μm². A pore space analysis indicates that the coal samples contain micropores (79.92%) and transition pores (17.29%), the pore volume ratio of which is 11.99%; when including the moderately developed mesopores (2.79%), the pore volume ratio is 88.01%. The coal samples exhibit favourable three-dimensional space constants. The mesopores that connect the pores provide the main seepage channels, and the contribution of isolated pores to the percolation capacity is negligible. \(3\) The seepage simulation results based on the real pore-fracture network model show that the seepage region of fluid in the coal seam is divided into low-, medium- and high-speed regions. The morphological characteristics and connectivity of the seepage channels determine the permeability. The more connectivity there is, the larger the pore radii and the more uniform the morphological properties; thus, the more connectivity there is, the higher the permeability. The fluid seepage in the coal seam is driven by the following factors: pore connectivity compactness \> pore throat channel dimensions \> pore throat channel structure distribution characteristics. \(4\) The pore distribution characteristics of the studied coal seam are analysed qualitatively, and the three-dimensional pore-fracture network structure is characterized quantitatively. From the microscopic point of view, the migration trajectory of fluid in the microscale seepage channels is analysed. The results suggest that this proposed method for evaluating the physical properties and flow capacity of coal seams is accurate. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Background India accounted for 19% of the globally estimated 287 000 maternal deaths in 2010. Although the level of maternal mortality in India has shown a definite decline over the last decade nationally, the MMR declined by 35% from 327 deaths per 100 000 births in 1999–2001 to 212 in 2007–2009 ; the current number of maternal deaths is still unacceptably high. The national MMR in India is an aggregate that conceals wide regional variations. Three large states viz. Kerala, Tamil Nadu, and Maharashtra with MMRs of 81, 97, and 104 per 100 000 births respectively, have already achieved the Millennium Development Goal 5 (MDG 5) target. However, in nine other large states, MMR estimates still range between 258 and 390. These nine states account for 62% of maternal deaths in India, and 12% of the global burden of maternal mortality. The world’s progress towards the achievement of MDG 5 is largely dependent on maternal mortality reductions in India, more specifically, in these nine Indian states. Skilled attendance at all births is considered to be the most critical intervention for ensuring safe motherhood, and has been accepted as one of the indicators for measuring progress to achieving MDG 5. Institutional delivery is expected to improve maternal and neonatal outcomes through timely intervention by skilled birth attendants backed by essential infrastructure and strong referral services when needed. To reduce maternal mortality, the Government of India commissioned nationwide programmes, including the Child Survival and Safe Motherhood Program (1992–1997) that was followed by Phase-1 of the Reproductive and Child Health Programme (RCH-1) (1997–2004). These programmes aimed at increasing the availability of emergency obstetric care (EmOC) services by enhancing institutional capacities. However during this period (1992–2004) of focus on strengthening institutions and investing in the supply side, maternal health care indicators in India were slow to improve, despite it being a period of substantial economic growth in the country. A major bottleneck identified was the low demand for and uptake of institutional deliveries – the proportion of institutional deliveries during this period showed increase from 26% to 41%; skilled birth attendance increased from 33% to 47%, yet more than half of women continued to deliver at home. Demand–side financing programmes, particularly cash transfer programmes, have emerged recently as newer ways of addressing the chronic problem of underutilisation of health and social services, particularly among vulnerable groups. The PROGRESA programme in Mexico provided cash to families in return for accessing children’s education, health, and nutrition services. Evaluation of the PROGRESA programme showed a significant positive impact on school enrolment and health outcomes. Similarly, the conditional cash transfer (CCT) in Honduras showed that conditional payments to households increased the use and coverage of preventive health care interventions. Evaluation of cash transfer programmes in Nicaragua, Colombia, and Brazil also demonstrated a similar positive effect. The most common mechanisms that have been employed in these regions to stimulate demand have been CCTs and voucher schemes. The CCT provides monetary incentive to households/individuals on the condition that they utilise specific services. Given the limited success with supply–side interventions under RCH-1 (1999–2004) in raising the proportions of skilled attendance at births and the growing evidence of the effectiveness of demand–side financing schemes on the utilisation of health services, the Indian government, in 2005, launched a nationwide CCT programme known as Janani Suraksha Yojana (JSY) focussed on maternal health. The JSY aims to reduce maternal and neonatal mortality through the promotion of institutional births by providing cash incentives to mothers on giving birth in a health institution. While the outline of the JSY scheme is the same across the country, it has different eligibility criteria and differential cash transfer size in different states, based on provincial proportions of institutional birth at the time the scheme was designed. In the states with high levels of maternal mortality and low levels of institutional delivery(low performing states), the JSY scheme provides a cash incentive of \$31 and \$22 to rural and urban women respectively, irrespective of socioeconomic status, age, or parity if they give birth in a public or accredited private health facility. In other socioeconomically better-developed states (high-performing states), the cash incentive is about half that paid out in the low performing states and is restricted to the first two live births of women from below the poverty line (BPL) and from scheduled castes (SC) and tribes (ST). All health facilities pay incentives into the mother’s bank account at the time of discharge from health facility after delivery. To provide mothers more options to choose the place for delivery, in some districts, private health facilities are accredited by district-level authorities based on broad guidelines issued by the health ministry. These guidelines include criteria such as infrastructure and human resources required. Women delivering at private facilities receive JSY benefits only on producing official certification of belonging to a vulnerable group. The Indian CCT scheme is the largest CCT in the world, with 52 million beneficiaries since inception. The CCT is underpinned by two major assumptions: (1) Financial barriers exist to access institutional care for childbirth. The cash incentive will enable women to overcome these financial barriers to access institutional care for delivery, and (2) increasing institutional births will provide more women access to skilled birth attendance and, therefore, will reduce maternal and neonatal deaths. Thus, the CCT is envisaged to result in (a) increased institutional births and (b) reduced MMR and neonatal mortality in regions with high proportions of institutional births. Previous evaluations of the Indian CCT, i.e., the JSY, were limited to small geographic areas focussed on processes and/or based on data from early years of the JSY and have limitations owing to unavailability of maternal mortality data at district levels. The Government of India recently set up the Annual Health Survey (AHS) (2010–2011) to capture population-based data on health indicators in these nine states with poor health indicators including high levels of maternal mortality. In this paper, we (i) report on trends in uptake of institutional births after the initiation of the CCT, and (ii) study the association between institutional birth proportions and the MMR in these nine states using AHS data. There have been calls in the literature to investigate the success of such CCT programmes in low-income settings, with more limited health system capacity. This paper contributes to this body of knowledge; - lessons from this study of the Indian CCT will be valuable in informing policies around demand-side financing in other similar settings. # Method ## Study Setting India is federal union of 35 states with distinctly different levels of socioeconomic development. States are further subdivided into smaller administrative units called districts, each with a population of approximately 1.5 million, which is divided into five to ten units called blocks. This study includes nine large states - Bihar, Uttar Pradesh(UP), Uttarakhand, Madhya Pradesh(MP), Orissa, Rajasthan, Jharkhand, Chhattisgarh, and Assam – that constitute about half of India’s population and account for 62% of her maternal deaths. These nine states are subdivided into 284 districts. They have relatively poor socioeconomic indicators; 34%–57% of their populations live below poverty line (based on a defined degree of deprivation) as per national surveys carried out by the Indian government. These nine states have relatively higher MMRs, infant mortality rates (IMR), and birth rates than the national averages of 212/100,000 live births, 50/1000 births, and 22.5/1000 population, respectively. The Government of India has classified these states as ‘high focus states’, implying more focussed attention to and greater allocation of resources towards strengthening the health system in these states. ## Study Design This study is an analysis of secondary data from two large population-based national surveys viz. Sample Registration Survey and Annual Health Survey conducted by the Government of India. ## Data Sources: The Sample Registration System (SRS) The SRS is a large demographic survey carried out periodically in India to generate reliable annual estimates of birth rate, death rate and other fertility and mortality indicators at the national and state levels. At present, SRS is operational in all states of India and covers about 7.27 million people in 1.5 million households. The sample unit in rural areas is a village and in urban areas; the sampling unit is a census enumeration block (population from 750 to 1000). The SRS comprises continuous enumeration of births and deaths in selected sample units by resident part-time enumerators, and an independent survey every six months by SRS supervisors. The data obtained by these two independent functionaries are matched. While recording details of every outcome of pregnancy, the enumerators and supervisors are required to enquire about the type of medical attention received by the mother at the time of delivery, including place of delivery. Proportions of institutional births reported by the SRS between 2005 and 2010 (years) have been used for the analysis in this paper. ### Annual Health Survey (AHS) : The AHS is the Government of India’s recent initiative at recording district level health outcomes in the nine ‘high focus’ states. The rationale for the survey was to identify districts requiring special attention as these often miss detection when studying average statistics at the state level. A special feature of the AHS is that it is the first survey in India to provide estimates of the district level mortality. The first AHS reported in 2010–2011 covered all the districts in the nine states. The AHS adopted a uni-stage stratified simple random sample without replacement except in case of larger villages and surveyed 18 million people in 3.6 million households. The survey collected background information of selected households and information from ever married women aged 15–49 years from these households regarding pregnancy outcome, place of delivery, child immunisation, and breastfeeding, to mention a few, that took place during the reference period (Jan. 2007 to Dec. 2009). Further details of data collection and management procedures are available on the survey website. Data reported by AHS on district-level MMR, proportion of institutional births, caesarean rate, total fertility rate, and proportion of literate and poor population were used in this paper. ### Census of india The Indian national census is conducted every ten years across all (35) states in the country. Each household is visited to collect information on a wide range of demographic and socioeconomic indicators of the household and the individuals. The district level information on level of urbanisation, vulnerable population, and total population from Census 2001 were used as covariates in the analysis presented here. ## Variables For state-level analysis, the proportion of institutional births, i.e., number of births that took place in government or private health institutions out of the total births, before and during the implementation of the JSY (2005–2010) was used. These data were sourced from the SRS during the period. For district-level analysis, the main outcome variable was district MMR. Given that maternal mortality is a rare event, the AHS estimated the MMR for a group of three to five geographically contiguous districts. In this analysis, we attributed the estimated MMR to each district in the group. The main explanatory variable of interest was district level proportion of all institutional births. Other socio-demographic covariates that influence maternal mortality included district-level caesarean rate, total fertility rate, literacy, proportion of households in the lowest 20% quintile of wealth index, proportion of vulnerable (Scheduled cast/tribes) and urban population. A brief description of these variables is given below: ### Caesarean rate Percentage of caesarean deliveries out of total deliveries that occurred at government and private institutions in the given period. Caesarean rate is the proxy indicator for availability of comprehensive EmOC services. ### Total fertility rate The total fertility rate (TFR) in a specific year is the number of children that would be born to each woman if she were to live to the end of her childbearing years and if the likelihood of her giving birth to children at each age was the currently prevailing age-specific fertility rates. ### Literacy The proportion of population in a district with the ability to read and write in any language, expressed as a percentage. ### Proportion of poor households Household wealth index was constructed by the AHS at the state level for each of the nine study states using the assets possessed (such as ownership and status of the house) and the facilities availed (such as electricity, toilet) by the households to determine a household’s relative economic status. Thereafter, the households were ranked according to their individual household asset score and then divided into five quintiles with the same number of households in each. In this paper, we used proportion of households in the lowest-income quintile in each district based on assets possessed as an indicator of level of deprivation (poverty) of the respective districts. ### Proportion of vulnerable (Scheduled cast/tribes) population The proportion of scheduled caste and tribe persons in the population of each district. Scheduled castes and tribes are those communities that were historically subject to social disadvantage and exclusion. They are accorded special status by the Constitution of India and are recipients of special social benefits as part of a programme of positive affirmation. ### Urban population The proportion of the total population in urban areas for every district. Data for district-level MMR, institutional births, caesarean rate, TFR, literacy, and poor population were sourced from the AHS and while that for vulnerable, urban and total population of district was sourced from the national census 2001. ## Ethics Statement The study is based on the data available in the public domain for use. ## Analysis State-level data on proportion of institutional births between the years 2005 and 2010 from the SRS were analysed to describe trends in proportions institutional births before (2005) and during (2006–2010) the implementation of the JSY. A statistical comparison of mean institutional delivery proportions before (2005) and during (2010) the JSY programme was done. Socio-demographic characteristics of the 284 districts in the nine states are presented. The association of district characteristics and institutional births was first examined separately by simple correlation analysis. To study the relationship between the proportion of institutional births and MMR, TFR, literacy, and the proportions of poor, urban and vulnerable populations we first used a simple correlation analysis. Subsequently, multiple regression models were developed to assess the effect of change in the district- level proportion of institutional births on the district MMR when other relevant socio- demographic variables were kept constant. STATA 10 was used for statistical analysis. # Results 1. Change in the proportion of institutional births in the nine states since the inception of the JSY programme: The proportion of institutional births increased in the nine states from a pre-programme average of 20% to 49% in the five years (p\<0.05). While institutional birth proportions increased across all nine states, the magnitude of the increases varied across states. 2. Association between institutional birth proportions and MMR in the district: - 2.1 *Characteristics of 284 study districts*: District characteristics for the 284 districts are presented in. On average, each district had a population of 1.7 million with varying proportions of poverty, literacy, and urbanisation. The proportion of institutional births ranged from 16.8% to 92.5% (mean 56.2%), demonstrating wide variations in utilisation of this service. The MMR ranges from a minimum of 183 to a maximum of 451. - *2.2 Correlation of district characteristics with proportion of institutional births and with MMR:* shows the estimated correlation of district characteristics with institutional birth proportions and MMR. Districts with higher fertility rates and higher levels of deprivation had lower institutional births proportions (r = −0.37 and −0.28, respectively); conversely, higher literacy and urbanisation in a district correlated positively with institutional births proportions (r = 0.38 and 0.32, respectively). The proportion of vulnerable population in a district did not show much influence on the uptake of institutional births (r = 0.07). There was no correlation between the proportion of SC/ST populations in the district and institutional birth proportions when these groups were analysed separately (data not shown). - Simple correlation between district characteristics and MMR showed that the fertility rate and the proportion of the poor in the population were positively correlated with MMR (r = 0.40 and 0.25, respectively). On the contrary, higher literacy and urbanisation were negatively correlated to MMR (r = −0.34 and −0.18, respectively). The proportion of births in an institution and births by caesarean section (CS), each had a small negative correlation with district MMR (r = −0.11 and −0.19, respectively). A scatter plot of institutional birth proportion and MMR does not show any strong relationship between institutional birth proportions and MMR in the districts. - *2.3 Regression analysis:* We built a regression model to explore the association between the proportion of institutional births and MMR. Covariates included are shown in. This model was unable to detect a significant association between institutional birth proportion and MMR \[CI: −0.10, 0.68\] adjusting for other confounders as shown in. Districts with higher fertility rates or higher proportions of poor population were significantly associated with higher MMR. Conversely districts with high literacy and high urbanisation were associated with lower MMR. Districts with high C-section rates were associated with higher MMR. # Discussion This report is based on district level estimates of MMR from the AHS from India. The results are important given that the nine states included in the study contribute to 12% of global maternal deaths. Efforts made to reduce maternal mortality in this area will impact global achievement of MDG 5. These results are also relevant for policy makers planning to initiate or expand cash transfers for promotion institutional births in other similar settings. Our analysis of the nine states indicates a steep rise in institutional birth proportions since the inception of the JSY programme. Although available data do not allow segregation of institutional births into JSY and non-JSY births, a large part of this increase in institutional births is fuelled by the JSY. Despite the steep rise in institutional births, our analysis was unable to detect a significant association between institutional birth proportions and MMR in the districts. ## ### Increase in institutional births When compared with pre-programme levels, the proportion of institutional births at state level increased two to three times over a period of five years since the programme began. Reports from other large-scale household surveys such as the District Level Household Survey and periodic reports from the health system also show increases in the institutional birth proportions after the implementation of the CCT scheme. A slow rise in institutional births during the RCH-I (1997–2004) and a sharp increase after initiation of the JSY scheme indicates its success in converting a significant proportion of home births into institutional births. While some of these institutional births have definitely occurred outside the JSY i.e., in non- accredited private institutions however this proportion remains marginal. Calculations made from the AHS reports show that an average of 13% of all births and 25% of all institutional births occur in private institutions in the study districts. As women delivering at private institutions accredited for the JSY receive the cash benefit, a proportion of institutional births in the private sector (25%) become JSY beneficiaries, leaving only a small proportion of non-JSY institutional births as a whole. Another recent survey report also indicated that in the ‘high focus’ states of MP, Rajasthan, UP, Orissa, and Bihar, the vast majority of all institutional births do occur under the JSY programme: on average, only 12.9% of all institutional births do not occur within the JSY programme. Experience from Nepal has been similar; a cash incentive to women on delivering in a health facility, increased utilisation in maternity services. A review of evidence on demand-side financing for sexual and reproductive health services in low and middle-income countries reports increased utilisation of services as an effect of demand-side financing strategy. Despite the overall increase in institutional births at the state level, within each state there were wide district-level variations that were associated with background socioeconomic characteristics of the districts. Districts with higher literacy and larger urban populations tended to have higher institutional birth proportions, whereas poverty and high fertility rates adversely affected the utilisation of institutional delivery services. As it is known that poor women bear the highest burden of maternal death, the programme still needs to develop mechanisms to reach this most vulnerable group. ### Association between institutional birth and maternal mortality Our analysis was not able to detect a significant association between district institutional birth proportion and MMR. While it is possible that we were unable to detect a significant association given the wide confidence intervals around our estimates, it is also possible that there is a limited influence of institutional birth proportions on MMR. Lim et al, in their evaluation of the JSY scheme in India, also reported an inability to detect its effect on maternal mortality at the district level, possibly because of a lack of programme effect or an inadequate sample size to detect the effect. This finding of a lack of association between institutional birth proportions and MMR could suggest the possibility that it is likely that the CCT disproportionately attracts pregnant women without complications to institutions, i.e., women most vulnerable to maternal death are not entering the programme. The mortality among women with complications is likely to be higher if they deliver at home, than if they deliver at institutions, as the latter should have access to a skilled person and EmOC services. There are currently no estimates of the proportion of mothers with complications among home or institutional births. The lack of association could also be related to the poor quality of care offered at institutions to the mothers with complications. Despite the focus on supply-side strengthening in the earlier RCH programmes, there have been recent reports documenting inadequacies in skilled human resources, infrastructure and supplies, which are critical for provision of good-quality care.Moreover, as reported by a programme evaluation report of the JSY, although all public sector facilities are designated as programme facilities, it is only a rather small number of higher-level facilities that actually have the ability to handle complications. Therefore, women in rural areas reach lower level facilities which are ill-equipped to handle complications. These inadequacies present a challenge to a mother having a safe delivery, even if she reached a facility. Paxton et al found that correlation between proportion of skilled attendance at birth and MMR becomes weaker for developing countries alone than when both developed and developing countries are included together. The correlation further drops for countries having MMR of more than 200. From this analysis the authors conclude that skilled attendance alone is not accountable for higher correlation of skilled attendance and MMR. Countries with low MMR have high proportion of skilled birth attendance and they have high proportions of maternal complications managed with high quality EmOC services. ### The quality of care issues While our analysis does not deal directly with quality of care under the programme, it is possibly an important explanation for the lack of association between institutional birth proportions and MMR. The available literature is summarised below. ### Need to ensure skilled attendance in an enabling environment In promoting institutional births, it was hoped that pregnant women would get skilled attendance at births, and access to appropriate EmOC in the event of complications. The Safe Motherhood Inter-Agency Group has defined *skilled attendance* as a process through which a woman is provided with adequate care during labour, delivery, and the postpartum period. Studies exploring links between skilled attendance at births and maternal mortality suggest a need of 1) a partnership of skilled attendants (health professionals with the skills to provide care for normal and/or complicated deliveries), and 2) an enabling environment of equipment, supplies, drugs and transport for referral. Investigation by Sri B et al on high number of maternal deaths in 2010 in the Barwani district of Madhya Pradesh found a lack of skilled birth attendance, failure to carry out emergency obstetric care in obvious cases of need, and referrals that never resulted in treatment. This report questions the policy of giving cash to pregnant women to deliver in poor quality facilities without first ensuring quality of care and strengthening the facilities to cope with the increased patient loads. Programme evaluation of the JSY reports that the programme has increased access to delivery by an Auxiliary Nurse Midwife (ANM), nurse, or doctor, but not necessarily to skilled birth attendant (SBA), because most nurses and ANMs who are actually providing services were not trained in the SBA training. The conceptualisation of the JSY programme in India has led to the substitution of the critical component of skilled attendance with a notion of ‘institutional births’ as being equivalent to skilled birth attendance, and, therefore, this as a condition to be met in order to receive the cash benefit from the JSY. Evidence seems to now indicate that the assumption that institutional birth is the same as skilled birth attendance in an enabling environment does not hold. This has resulted in pregnant women arriving in institutions, but this in itself is not necessarily giving them access to skilled attendance. ### Need to address other non-financial access barriers The JSY has raised the uptake of institutional births, yet, in some districts, more than half of women continue to deliver at home. This suggests that even if a cash incentive is able to attract more women to facilities, there are still many for whom; other non-financial barriers operate to reduce the likelihood of an institutional birth. Some of these barriers include the inability to access transportation or the costs involved in doing so, non-perception of birth as a risk event, the social status accorded to women, and poor levels of trust in the public health facilities. Some states initiated emergency transport arrangement, which reduce the transportation barrier; however, the other barriers are more complex and require more structural and social change. An analytical framework by Bart Jacob et al provides guidance on interventions to support structural changes in the health system that can reduce some barriers. Significant positive association of literacy, urbanisation and inverse association of fertility rates and poverty in our regression model seem to suggest that overall socioeconomic development contributes significantly in maternal mortality reduction. Although socioeconomic development is merited, reduction in maternal mortality can be achieved even in countries with poor development indicators. For instance, low income-countries such as Sri Lanka, Thailand, and Maldives have shown that universal access to skilled attendance at birth and EmOC services could reduce maternal mortality drastically. ### Negligence of antenatal and postnatal care The JSY scheme incentivises pregnant women for utilisation of health facility for intra-partum care, whereas antenatal (ANC) and postnatal care (PNC) is not a prerequisite for cash benefit. While splitting the cash benefit can imply large administrative burdens on the programme, a narrow focus on institutionalising intra-partum care restricts opportunities of averting deaths by early detection of risk pregnancies and treatment of common postnatal complications like puerperal sepsis. ## Limitations AHS estimates MMR for groups of three to five geographically contiguous districts. These pooled estimates were attributed to individual districts during our analysis. Institutional birth proportions are assumed to be JSY deliveries in this analysis, although a small proportion as reported above are non-JSY institutional births. MMR estimates are based on community-level surveys; no information is available on what proportion of deaths occurred within the programme or outside of it. ### Residual confounding The results of the study should be interpreted with caution because the association between institutional birth proportions and maternal mortality can be confounded by other known and unknown confounders. Examples of these could be road network in the district, a measure of overall quality of care in the district, emergency transport services available in the district, etc. Some confounders such as cultural practices, awareness about need of and availability of health care services, etc. will influence maternal mortality via access to institutional birth. Variables used in the analysis (e.g. poverty, literacy, urbanisation) serve as proxies for these factors. Factors that determine the quality of care provided in the facilities could have contributed to varying levels of maternal mortality, but in the absence of precise data on quality of care provided in study districts, this study is unable to control the confounding effect of district-level variation in quality of care. We acknowledge that residual confounding is likely to be present. However, despite the limitation of this ecological study, it explores the association between district-level variation in uptake of institutional births and maternal mortality using available data at the smallest unit of analysis (district) during the implementation of the JSY; it makes an important contribution. The counter intuitive finding of an association between caesarean rates and MMR could possibly be because confounding factors were not taken into account in the model. It also suggests further exploration of appropriateness and quality of caesarean and post-operative care provided. It could be possible that the sudden rise in institutional deliveries resulted into overcrowding in facilities and quality of care in operation theatres or in postnatal wards was compromised resulting in more deaths. Another possibility could be the higher rates of caesarean in private hospitals, exposing more women to higher risk. In the 284 study districts, the median caesarean rates for public and private sectors reported by the AHS were 5% and 28%, respectively. When we conducted a regression analysis using a stratified caesarean rate in public and private facilities it showed that caesarean rate in private facilities, but not in public facilities, were significantly associated with higher mortality. This needs to be explored more. Bertan et al, in their analysis of global and regional estimates of caesarean rate show that although caesarean rates below 15% are associated with lower maternal mortality; higher rates are predominantly correlated with higher maternal mortality. ## Conclusions We were unable to detect a significant association between the proportion of institutional births and the MMR at the district level, though other indicators of overall development such as literacy showed a significant association with reduction in the MMR. Although the JSY succeeded in raising institutional birth proportions significantly; the same has not translated into significant reduction in the MMR. It is likely that a weak supply side has led to a situation in which increased access to institutional birth has not resulted in reduction in maternal deaths, as mothers are not receiving appropriate or adequate care. It is also possible that the JSY failed to draw mothers with life-threatening complications into institutions, resulting in most of such women continuing to deliver at home, contributing to persistent maternal mortality. Further studies are required to examine the extent to which the JSY increased access to institutional care among mothers with complications. Moreover, to translate the JSY gains in institutional delivery coverage into reduced mortality outcomes, it is important to ensure that all women accessing an institution for delivery receive good quality obstetric care. We acknowledge the peer reviewers – Dr Abhijit Das and two anonymous reviewers, for critical comments on an earlier version of this paper. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BR. Performed the experiments: BR VD ADC. Analyzed the data: BR ADC. Contributed reagents/materials/analysis tools: BR VD. Wrote the paper: BR ADC.

# Introduction Epithelial-mesenchymal transition (EMT), a developmental process by which epithelial cells reduce cell-cell adhesion and lose apical-basal cell polarity, plays a critical role in the embryogenesis and conversion of early stage tumors into aggressive malignancies. EMT promotes multiple physiological processes that increase the invasiveness and metastasis potentials of human tumors. EMT is typically characterized by loss of E-cadherin, gain of N-cadherin and vimentin, and translocation of β-catenin from membrane to the nuclear compartment. The impairment of E-cadherin is a hallmark of EMT, and E-cadherin expression is often inversely correlated with tumor malignancy and patient survival. The E-cadherin expression level is down-regulated by gene silencing with CpG methylation on promoter in hepatocellular carcinoma (HCC), and may be associated with tumor grade and poor prognosis of HCC. Several transcriptional regulators that act as E-cadherin repressors are mediated by recognizing the E-box motif on the E-cadherin promoter region. Factors of Snail zinc finger, Zeb and bHLH families are known to suppress E-cadherin, thereby promoting the EMT process and tumor metastasis. In addition, increased Zeb-1, Snail, SIP1, and Twist expressions are reportedly associated with the clinicopathological significances of HCC malignant progression, including cancer invasion and poor patient survival. 14-3-3 proteins are a family of regulatory molecules with highly conserved homology among all eukaryotic cells. 14-3-3 modulates physiological functions *via* binding intracellular proteins with Ser/Thr phosphorylation-dependent domains, thereby influencing conformation, activity, subcellular localization, and protein complex stability. 14-3-3 proteins comprise seven isoforms (β, ε, γ, η, σ, τ/θ and ζ), which play crucial roles in regulating multiple cellular processes, including cell cycle regulation, DNA repair, apoptosis, cell adhesion, and motility. 14-3-3 proteins have been implicated in various types of human malignancies. Among the variety of 14-3-3 isoforms, increased expression of 14-3-3ε has been demonstrated in breast cancer, lung cancer, vulvar squamous cell carcinoma, follicular and papillary thyroid tumors, meningioma, and HCC,, although reduced expression of 14-3-3ε in gastric cancer has been reported. In addition, previous studies have demonstrated that up-regulation of 14-3-3ε expression protects colorectal cancer cells and endothelial cells from oxidative-stress induced apoptosis, while suppression of 14-3-3ε by nonsteroidal anti-inflammatory drugs induces cancer and endothelial cell death. Furthermore, elevated expression of 14-3-3ε is significantly associated with increased metastatic risk, shortened overall survival, and progression-free survival of HCC. Enhanced expression of 14-3-3ε was suggested to induce and be associated with focal adhesion kinase (FAK) expression *via* activation of NFκB signaling. These results implied that 14-3-3ε was involved in the modulation of cell polarization and migration, which may potentially regulate HCC tumor development and metastasis. In this study, we show for the first time that 14-3-3ε induces HCC cell migration and EMT *via* regulation of Zeb-1/E-cadherin expression. Our results reveal that E-cadherin is a downstream modulator for 14-3-3ε during HCC tumor progression. # Materials and Methods ## Cell Culture and Stable Cells Huh-7 (Japanese Collection of Research Bioresources, JCRB-0403), HepG2 (American Type Culture Collection, ATCC-HB-8065), Hep3B (ATCC-HB-8064), PLC-5 (ATCC- CRL-8024) and SK-Hep1 (ATCC-HTB-52) human hepatocellular carcinoma cells were maintained in DMEM (Gibco, Gaithersburg, MD) supplemented with 10% fetal bovine serum (FBS; Hyclone Thermo Fisher Scientific, Waltham, MA), 100 units/ml penicillin, and 100 units/ml streptomycin, in a humidified incubator (Forma) with 5% CO₂ at 37°C. For stable transfection, 14-3-3ε cDNA was amplified by PCR and then subcloned into the p3XFlag-CMV vector. Huh-7 cells were transfected with p3XFlag-CMV (Control) or p3XFlag-14-3-3ε (14-3-3ε) by use of Polyjet™ transfection reagent (SignaGen Laboratories, Rockville, MD) according to the manufacturer’s instructions. The transfected cells were selected with G418 (500 µg/ml) for 4 weeks. Single colonies of stable clones (at least 3 in each cell line) were maintained in DMEM with 10% FBS and 200 µg/ml of G418. ## Transient Transfection Huh-7 and HepG2 cells were transiently transfected with control and 14-3-3ε by use of Polyjet™ transfection reagent (Signa-Gen Laboratories, Ijamsville, MD). Cells were transfected with control or 0.5 to 1.5 µg of 14-3-3ε vectors per 6-well plate followed by incubation with Polyjet™/DNA complex-containing medium and replaced with complete medium for 24 hours. Transfected cells were incubated for additional 24 hours before performing cell migration assays or protein expression analysis. ## Knockdown Studies Gene silencing was performed using 14-3-3ε, Snail (Stealth RNAi, Invitrogen, Carlsbad, CA), Zeb-1 siRNAs (Santa Cruz, Heidelberg, Germany) and Stealth RNA Negative Control (Invitrogen, Carlsbad, CA) with reported sequences. Transient transfection of siRNA was carried out using Lipofectamine™ RNAiMAX (Invitrogen, Grand Island, NY) according to the manufacturer’s guidelines. ## Western Blot Analysis Cells were lysed in ice-cold RIPA buffer (0.5 M Tris-HCl, pH 7.4, 1.5 M NaCl, 2.5% deoxycholic acid, 10% NP-40, 10 mM EDTA, Millipore, Temecula, CA) containing cocktail protease inhibitors (Roche, IN, USA). Cell lysates were centrifuged at 15,000 rpm for 20 minutes at 4°C, and protein concentrations were determined by a Bio-Rad protein assay kit (Bio-Rad Laboratories, Hercules, CA). Each sample of 20 µg protein was applied to a gradient SDS-PAGE gel and immunoblotted onto PVDF membranes. The membranes were blocked and probed with indicated primary antibodies of Flag and actin (Sigma-Aldrich, St. Louis, MO), E-cadherin and N-cadherin (BD Biosciences, San Jose, CA), 14-3-3ε, Zeb-1, Twist and Slug (Santa Cruz Biotechnologies, Heidelberg, Germany), Snail (Cell Signaling Technology, Beverly, MA), Zeb-2 (Abcam PLC, Cambridge, UK), and vimentin (Millipore, Temecula, CA). The membranes were immersed in PBST containing horseradish peroxidase-conjugated secondary antibody, and protein levels were determined by use of enhanced chemiluminescence reagents. ## Immunofluorescence Staining Immunofluorescence staining was performed as described previously. Briefly, 14-3-3ε and control cells were fixed with 2% paraformaldehyde for 15 minutes at 4°C. After washing, cells were permeabilized with 0.1% Triton X-100 in PBS for 5 minutes and blocked with PBS containing 10% FBS at room temperature for 1 hour. For the immunofluorescence staining, cells were incubated with the primary antibodies of anti-E-cadherin and anti-N-cadherin (BD Biosciences, San Jose, CA), and anti-vimentin (Millipore, Temecula, CA) in PBS containing 1% FBS at 4°C overnight, followed by incubation with Alexa Fluor® 488 secondary antibody (Invitrogen, Grand Island, NY) in PBS containing 5% bovine serum albumin at room temperature for 2 hours. Samples were mounted and images were analyzed by use of the Leica TCS SP5 Confocal Imaging System (Leica, Germany). ## Migration Assay Bio-coat cell migration Boyden chambers were used for cell migration assay (Becton Dickinson, Pont-de-Claix, France). Briefly, cells were trypsinized and suspended in 0.1% BSA-DMEM and cells (1×10⁴ for SK-Hep1, 6×10⁴ for Huh-7 and 2×10⁵ for HepG2) were added to the upper wells with 8-µm pores. Cells were allowed to migrate toward the bottom wells containing 100 µg/ml fibronectin (Becton Dickinson, Pont-de-Claix, France), epithelial growth factor (EGF, 20 ng/ml, Sigma-Aldrich, St. Louis, MO) and 10% BSA-DMEM for 20 hours. Cells remaining on the upper side were removed, and migrated cells on the bottom side were fixed and stained with 0.1% crystal violet containing 20% ethanol and 1% formadehyde for 20 minutes. Cell migration was quantified by counting the total number of migrated cells. ## Quantitative Real-time PCR As described previously, total RNA was extracted by use of the RNAspin Mini Kit (GE Healthcare, Freiburg, Germany). cDNA was synthesized from 2–5 µg RNA by use of the oligo(dT)₁₈ primers and RevertAid™ First Strand cDNA Synthesis Kit (Fermentas, Thermo Fisher Scientific, Waltham, MA). Quantitative real-time PCR involved use of SYBR Green (Kapa biosystem, Woburn, MA) with specific oligonucleotide primers from the AB 7900HT system (Applied Biosystems, USA). Applied Biosystems Relative Quantification (RQ) Manager Software version 1.2 was used to analyze the relative gene expression in each sample by the comparative Ct method. Gene expression was normalized to that of glyceraldehyde-3-phosphate dehydrogenase (GAPDH). ## Clinical Specimens Tissue samples were obtained from 113 HCC patients who had undergone surgery for tumor resection or biopsy at Taichung Veterans General Hospital from January 1999 to December 2001. The mean follow-up time was 51.5±28.7 months. Thirty- three patients (29.2%) developed tissue-proved metastasis in 3 to 87 months after the resection of primary HCC. Slides from paraffin-embedded surgical specimens of primary tumors with surrounding non-cancerous liver parenchyma were subjected to immunohistochemical (IHC) staining. The pathological features, IHC staining results, clinical parameters, including Barcelona-Clinic Liver Cancer (BCLC) staging, and disease outcomes were collected for analysis. This study was approved by the Institutional Review Board of Taichung Veterans General Hospital. The policy that no informed consents are required for using these de- linked samples for retrospective analysis was also approved by the Institutional Review Board. ## Immunohistochemical Analysis For immunohistochemistry analysis, an automatic immunostaining device and ultraView detection kit (Ventana XT Medical System, Tucson, AZ) were used to detect 14-3-3ε expression in paraffin-embedded tissues by use of a primary antibody against 14-3-3ε (1∶800; Santa Cruz Biotechnology, Santa Cruz, CA) and E-cadherin (1∶800; BD Bioscience). A negative control was prepared by the same staining procedure without primary antibodies. The intensity of 14-3-3ε and E-cadherin protein staining was semiquantitatively scored by a Quick-score (Q-score) method based on intensity and heterogeneity,. Staining intensity was scored as 0 (negative), 1 (weak), 2 (moderate), or 3 (strong). For heterogeneity, the proportions of tumor cells positively stained with 14-3-3ε and E-cadherin were scored as 0 (0%); 1 (1–25%); 2 (26–50%); 3 (51–75%) or 4 (76–100%). The Q-score of a given tissue sample was the sum of the intensity and heterogeneity scores and ranged from 0 to 7. A Q-score ≥2 was considered overexpressed, or positive expression, and a Q-score \<2 was considered normal, or negative expression. Cases with \<5% weakly stained specimens were considered as negative expression. ## Statistical Analysis The Student’s *t*-test was used to analyze differences between 2 groups. Kaplan- Meier curves were plotted and the log rank test was used to analyze time-related variables of probabilities for metastasis and overall survival. A *P* value \<0.05 was considered statistically significant. # Results ## 14-3-3ε Promotes HCC Cell Migration To explore the potential role of 14-3-3ε in HCC tumor metastasis, we examined the expression of 14-3-3ε in distinct HCC cell lines. 14-3-3ε was detected in all tested HCC cells. The well-differentiated HCC cells, including Huh-7, HepG2 and PLC-5, expressed lower levels of 14-3-3ε, while the poorly differentiated SK-Hep1 cells expressed higher levels. We next established a stable cell line with 14-3-3ε overexpression. Huh-7 cells were transfected with p3XFlag-CMV (Control) or p3XFlag-14-3-3ε (14-3-3ε) vectors and selected by G418 for 4 weeks. Individual colonies were picked and 14-3-3ε expression was confirmed by Western blot analysis. At least 3 clones were selected, and the representative clone was used for further experiments. To investigate whether 14-3-3ε regulates cell migration, we performed the migration assay with Boyden chamber experiments. We found that 14-3-3ε (stable clones 1–4) significantly induced cell migration. In addition, the induction of cell migration accessed by 14-3-3ε overexpression was confirmed by transient transfection in both of Huh-7 and HepG2 cells. Transiently 14-3-3ε overexpression dose-dependently increased Huh-7 and HepG2 cell migration. To further confirm the effect of 14-3-3ε on modulating cell migration, control or 14-3-3ε stable cells (clone 1) were transfected with scramble or 14-3-3ε siRNAs and the efficiency of 14-3-3ε knockdown was determined by Western blotting analysis (upper panel). Knockdown with siRNA significantly abolished 14-3-3ε-induced cell migration (lower panel). Additionally, knockdown of 14-3-3ε with siRNA significantly suppressed SK-Hep1 cell migration with a dose-dependent manner. These results suggest that 14-3-3ε plays an important role in promoting HCC cell migration. ## 14-3-3ε Promotes Epithelial-mesenchymal Transition of HCC To investigate whether 14-3-3ε expression regulates EMT of HCC cells, we determined the expression of EMT markers, E-cadherin, N-cadherin and vimentin, by Western blotting analysis. We found that 14-3-3ε overexpression significantly reduced E-cadherin expression, but it induced N-cadherin and vimentin expression. The expression levels and subcellular localizations of E-cadherin, N-cadherin, and vimentin were further examined by immunofluorescent confocal microscopy. E-cadherin expression was detected at the cell-cell contacts in control cells, while it was dramatically reduced in 14-3-3ε overexpression cells. Slight expression of N-cadherin and vimentin was detected in control cells, but such expression was significantly induced by 14-3-3ε overexpression (middle and right panels). Furthermore, reduction of E-cadherin expression and the induction of N-cadherin and vimentin expression in 14-3-3ε overexpression cells were abrogated by transfection with 14-3-3ε siRNA, as determined by Western blotting analysis. The regulation of these expressions of E-cadherin, N-cadherin and vimentin by 14-3-3ε knockdown were further confirmed by confocal microscopy. 14-3-3ε siRNA restored E-cadherin expression, which localized at the cell junctions. In addition, knockdown with siRNA suppressed the N-cadherin and vimentin expression induced by 14-3-3ε. These results indicate that 14-3-3ε overexpression promotes EMT of HCC. ## 14-3-3ε Promotes HCC Cell Migration via Upregulation of Zeb-1 and Snail To understand the molecular regulation of how 14-3-3ε induces EMT and reduces E-cadherin expression in HCC, we examined the expression levels of distinct E-box transcriptional suppressors. We found that 14-3-3ε overexpression selectively induced Zeb-1 and Snail expression but had no significant effect on Zeb-2, Twist or Slug. The induced expression of Zeb-1 and Snail by 14-3-3ε was further confirmed by quantitative real-time PCR analysis. In addition, results of transient transfection indicated that overexpression of 14-3-3ε dose- dependently induced Zeb-1/Snail and reduced E-cadherin expression in Huh-7 and HepG2 cells. Furthermore, 14-3-3ε-induced expression of Zeb-1 and Snail was abrogated by knockdown of 14-3-3ε with siRNA. These findings suggest that Zeb-1 and Snail may be involved in 14-3-3ε-induced HCC cell migration and EMT. We next determined the role of 14-3-3ε-induced Zeb-1 and Snail on cell migration. We found that knockdown of either Zeb-1 or Snail expression by siRNA significantly abolished 14-3-3ε induced cell migration. These results indicate that Zeb-1 and Snail play important roles in regulating 14-3-3ε-induced HCC cell migration. ## 14-3-3ε Suppresses E-cadherin Expression Selectively Mediated by Induction of Zeb-1 To further explore the role of Zeb-1 and Snail on 14-3-3ε-induced cell migration or EMT, we knockdown Zeb-1 and Snail with siRNAs and examined E-cadherin expression by Western blotting analysis. Interestingly, 14-3-3ε-reduced E-cadherin expression was specifically restored by Zeb-1 siRNA, but not by Snail siRNA. This specific effect of Zeb-1 on regulating 14-3-3ε-reduced E-cadherin was validated by quantitative real-time PCR. In addition, the selective effect of Zeb-1 knockdown restoring E-cadherin expression in 14-3-3ε overexpression cells was further confirmed by confocal microscopy. SK-Hep1 cells expressed higher levels of 14-3-3ε and lower levels of E-cadherin than other HCC cell lines. We next transfected SK-Hep1 cells with 14-3-3ε siRNA and determined the expression of Zeb-1, Snail and E-cadherin by Western blotting analysis. Knockdown of 14-3-3ε reduced the expression of Zeb-1/Snail and induced that of E-cadherin in SK-Hep1 cells with a concentration-dependent manner. These results demonstrate that the reduction of E-cadherin expression by 14-3-3ε is selectively mediated by regulation of Zeb-1. Thus, Zeb-1/E-cadherin expression is a downstream factor of 14-3-3ε for promoting EMT in HCC. ## Correlation and Impact of Positive 14-3-3ε with Negative E-cadherin Expression in HCC To further support the likelihood that 14-3-3ε suppresses E-cadherin and regulates EMT as well as tumor progression, we examined the expression of 14-3-3ε and E-cadherin by immunohistochemical analysis in HCC tumors. Expression of 14-3-3ε was higher in HCC primary tumors than in the surrounding non- cancerous liver tissues. We next determined the expression of E-cadherin and found that E-cadherin expression was reduced in HCC tumors. Positive 14-3-3ε expression was significantly correlated with negative E-cadherin in HCC tumors (*p* = 0.043). In addition, expression of 14-3-3ε was correlated with Zeb-1 in HCC tumors. ## Association of 14-3-3ε/E-cadherin Expression with Extrahepatic Metastasis and Patient Survival of HCC We have previously shown that 14-3-3ε overexpression in HCC primary tumors was significantly associated with subsequent extrahepatic metastasis and reduced 5-year overall survival. To evaluate whether E-cadherin plays an important role as a downstream effector of 14-3-3ε in promoting tumor progression, the associations of E-cadherin with clinicopathological characteristics and with 14-3-3ε expression were compared. In addition to 14-3-3ε positivity, expression of E-cadherin is significantly correlated with gender (*p* = 0.011), histology grade (*p* = 0.001), BCLC staging (*p* = 0.030), tumor size (*p* = 0.003), and subsequent extrahepatic metastasis (*p* = 0.004). Patients with positive E-cadherin expression exhibit a lower risk of metastasis (, *p* = 0.013) and better overall survival rate (*p* = 0.047) than do those with negative E-cadherin expression in 14-3-3ε positive HCC tumors. These results provide clinical evidence to support the hypothesis that E-cadherin is one of the crucial downstream regulators of 14-3-3ε that modulate HCC tumor progression. # Discussion We previously demonstrated that 14-3-3ε expression is increased in primary and metastatic HCC. Elevated 14-3-3ε expression is correlated with higher risk of extrahepatic metastasis and lower survival rates of HCC patients. In this study, we investigated the molecular mechanism to determine how 14-3-3ε regulates tumor progression. Attenuated expression of E-cadherin has been recognized as an important determinant and biomarker of tumor progression, one especially indicative of EMT in various tumors. In addition, gene silencing and loss of E-cadherin expression in the malignant progression of HCC have been demonstrated, , and it is suggested that E-cadherin is associated with reduced survival of HCC patients. Our current investigation indicates that 14-3-3ε promotes HCC EMT and cell migration and also suppresses E-cadherin expression *via* upregulation of Zeb-1. We found that the expression of Zeb-1 was increased (14 of 113) in HCC primary tissues, although the increase is not as significant as in a previous report. This difference may due to the sensitivity of reagents, sample size or differences in the cohort. It has been suggested that TGF-β and downstream signals of Smad2/3 activation regulate Zeb expression and EMT. Although we performed the experiments of Smad2 knockdown in 14-3-3ε overexpression cells, we did not observe significant restoration of E-cadherin. In addition, increased 14-3-3ζ expression has been shown to promote EMT *via* associating with TGF-β receptor signaling and PI3-K subunit p85 in breast cancer cells. However, treating cells with TGF-β receptor or PI3-K inhibitors (SB-431542 or LY-294002) did not abolish E-cadherin-suppression induced by 14-3-3ε. These results suggest the effect of 14-3-3ε-suppressed E-cadherin expression may not be regulated through TGF-β/Smad2/3 or PI3-K signal pathways. Thus, 14-3-3ε contributes to EMT via induction of Zeb-1 may be mediated by a novel mechanism. Further work is currently ongoing to investigate how 14-3-3ε regulates Zeb-1 expression. A functional motif for 14-3-3 binding in Snail was demonstrated, and the ternary protein complex comprised of 14-3-3, Ajuba and Snail plays an important role in transcriptional repression and EMT. Snail contains two potential phosphorylated residues at Ser11 and Thr177 in putative motifs for 14-3-3 protein binding. Phosphorylated Snail selectively interacts with 14-3-3γ, 14-3-3ε, 14-3-3θ/τ, 14-3-3η and 14-3-3β, but not with isoforms of 14-3-3σ and 14-3-3ζ. This interaction has been demonstrated to be required for E-cadherin suppression in MCF breast cancer cells and 293 cells. To find out whether 14-3-3ε suppresses E-cadherin *via* a similar mechanism, we co-immunoprecipitated 14-3-3ε stable cells with anti-Flag antibody, followed by Western blotting analysis of Zeb-1. However, no significant interaction of 14-3-3ε with Zeb-1 was detected. Further studies using proteomic approaches are currently underway to investigate the potential 14-3-3ε interaction partners in regulating Zeb-1 and EMT of HCC. Expression of transcriptional repressors for E-cadherin, including Zeb-1, Snail, SIP1 and Twist, is associated with cell invasion, EMT, metastasis and poor patient survival of HCC. However, no previous studies have shown that 14-3-3ε promotes HCC tumor progression *via* modulating E-cadherin transcriptional repressors. Our study shows for the first time that 14-3-3ε induces Zeb-1 expression, thereby repressing E-cadherin expression and promoting EMT. The 14-3-3ε regulation of E-cadherin reduction occurs through Zeb-1, and not through Snail or other E-cadherin repressors, as supported by. To further clarify the regulation of 14-3-3ε-reduced E-cadherin expression by Zeb-1, 14-3-3ε overexpression cells were transfected with Zeb-1 siRNA or control scramble siRNA, and the gene expression profile was analyzed by use of microarray analysis. Altered gene expression (fold change \>2) was identified of 557 transcripts in 14-3-3ε overexpression *vs.* the control cells and 160 transcripts in Zeb-1 siRNA *vs.* scramble siRNA cells. Among them, *CDH1* (E-cadherin), *SMAD2*, and *PLA2G2A* were regulated in 14-3-3ε overexpression cells but had a reversed expression pattern in Zeb-1 knockdown cells (Data not shown). These results provide additional evidence to support our findings. In addition to Zeb-1, our results indicated that 14-3-3ε induces Snail expression and promotes HCC cell migration. However, knockdown of Snail did not restore 14-3-3ε-reduced E-cadherin expression. Interestingly, partially increased of Snail expression was found by treatment with Zeb-1 siRNA. As Snail and Zeb-1 regulate EMT of HCC may be mediated by separate and complicated pathways, a compensative effect is possibly involved. Further investigation is needed to elucidate this finding. Additionally, our results indicated that 14-3-3ε overexpression-induced EMT (increase of N-cadherin, Vimentin, Zeb-1 and Snail as well as decrease of E-cadherin expression) was impaired by 14-3-3ε siRNA. However, knockdown of 14-3-3ε has no significant effect on affecting EMT markers in control cells. We therefore postulate that other endogenous house- keeping regulators may be involved in maintaining basal level of Snail/Zeb-1 expression. Endogenous level of Snail/Zeb-1 modulates expression of EMT markers which is independent of 14-3-3ε expression in HCC. Moreover, 14-3-3ε may upregulate FAK expression *via* activation of NFκB to enhance HCC cell migration. These results reveal the complicated signal mechanisms that are involved in 14-3-3ε induced HCC cell migration, EMT, and metastasis. Uncovering the complex role of 14-3-3ε in tumor progression could contribute to the development of therapeutic strategies for treatment of aggressive and advanced HCC. # Supporting Information We thank the Comprehensive Cancer Center of Taichung Veterans General Hospital for providing information concerning the outcomes of patients. We also thank the core laboratory of National Health Research Institutes for the helpful assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JYL YJJ BSK TAL. Performed the experiments: TAL YJJ BSK SML SCC. Analyzed the data: JYL YJJ BSK TAL YMW CH JW. Wrote the paper: JYL.

# Introduction Tuberculosis (TB) continues to be a major global health problem including in Saudi Arabia. High proportion of immigrants living in the country and annual massive influx of pilgrims mostly from TB endemic areas constitute favorable conditions for TB transmission. One of the most important challenges in global TB control is the early detection and treatment of latent tuberculosis infection (LTBI). About 10% of individuals infected with *M*. *tuberculosis* develop pulmonary TB, and the remaining 90% suppress the bacterial invasion through their immune systems and persist with LTBI. Unfortunately there is very limited data on the prevalence of LTBI among health care workers in Saudi Arabia, except reports from few institutions. Health care workers (HCW’s) in general are considered as a high-risk group of LTBI because of the increased risk of exposure. A recent systematic review showed, among HCW’s of low and middle income countries, the LTBI prevalence is ranged between 33–79% with a median of 5.8% annual LTBI incidence. Usually, HCW’s may come into contact with patients with undiagnosed cases or unknown/unsuspected cases of active TB, which elevates the acquisition of LTBI. HCW’s employed in sections like emergency, intensive care, internal medicine, radiology, are at greater risk of acquiring *M*. *tuberculosis*. Due to the epidemiological evidence of TB as a consequential occupational disease, HCW’s with negative TST results must undergo annual LTBI screening. Surveillance study on LTBI among HCW’s vaccinated with BCG has been hampered by the non- specificity of TST. False-positive results secondary to cross-reactions caused by BCG vaccination and/or exposure to non tuberculous mycobacteria, booster phenomenon and technical issues, like interpretation of the result may all lead to unnecessary treatment of presumed LTBI thereby, rendering TST unsuitable as a surveillance tool in TB risk groups. The advent of Interferon Gamma Release Assays (IGRA) and their increasing availability show promise for more accurate LTBI detection in HCW’s. The higher specificity of QuantiFERON TB gold assay (QFT) compared with TST can reduce unnecessary treatment, follow-ups and thereby, treatment cost of LTBI. Most IGRA studies have been done in low TB-endemic regions whereas; data from low to middle-income settings, with high background of TB infection rates have been fairly scarce. We are not aware of any report describing the use of IGRA’s among HCWs in Saudi Arabia or in Gulf Cooperation Council (GCC) countries. In Saudi Arabia, TB screening is recommended for all HCW’s. Annual LTBI screening is mandatory for all employees at King Faisal Specialist Hospital and Research Centre (KFSHRC) as part of their job contract renewals. The TST is the currently following LTBI diagnostic method at KFSHRC. However due to the limitations of the TST and to promote a more sensitive technique, IGRA was considered. The cosmopolitan nature of the employee population (citizens of more than 60 countries) has not been considered for optimizing the LTBI diagnosis. Moreover, the majority of the employee populations are from countries where BCG vaccination is mandatory. In addition, there was an increasing prevalence of NTM diseases also reported in the country recently. Thus a new LTBI diagnostic test was badly needed for the institution. Therefore, a cross sectional study has been designed with two major set objectives. The key objectives were to analyze the prevalence of LTBI among HCW’s of highly diverse origin at KFSHRC and to compare the feasibility of using TST and QFT to screen the LTBI among this diverse population. # Materials and Methods This study was carried out for a period of 36 month (February 2012 to January 2015) in KFSHRC, Riyadh. The protocol has been reviewed and approved by the Research Advisory Council (RAC) of KFSHRC. All the candidates enrolled were subjected to a short interview and completed the consent form before withdrawing the blood samples. ## Study population The inclusion criteria for study subjects was any new (undergoing pre-employment checkup) or existing employee (annual re-contracting checkup), who can consent for the study. The exclusion criteria included candidates with previous history of active TB or undergoing TB medication. The study subjects were recruited from both clinical and non-clinical health care workers designated in various departments at KFSHRC. Clinical health-care workers (those included in any direct patient contact or giving direct care) are doctors, nurses and allied health professionals. Non-clinical health-care workers included those who did not involve in direct patient care, such as administrative staffs, researchers, housekeepers, and hospital technical maintenance staffs. According to the job profile of the candidate, the degree of TB exposure level was defined into low, medium and high. Doctors, nurses and allied health professionals were classified into high risk group. The medium risk group consisted of candidates who are not directly giving patient care but in contact with patient samples (ex; medical lab technicians) or technicians managing respiratory units or equipments. The low risk group included researchers, maintenance staffs and administrative staffs respectively. A standard questionnaire was used to collect the information on key variables, such as age, gender, nationality, previous exposure or treatment details of TB, previous chest x-ray details, BCG vaccination and BCG scar, prior TST (date and result), job category, service and years in the health profession. The questionnaire has been completed by face to face interview and reference to medical records. All the TST positive converted staffs were offered with a follow up of Chest X-ray and a 9 months isoniazid treatment according to the standard guidelines. The decision of treatment was kept optional to the employee after the discussion with the physician. However a continuous follow-up was mandatory. ## Performing TST and QFT All new HCW’s, who did not have a documented TST result, were subjected to TST during the routine examination at the time of employment in the family medicine department of KFSHRC. TST was performed by administrating 2-TU of PPD RT23 (Staten’s Serum Institute, Copenhagen, Denmark). Induration was measured 48–72 h after the application. Well trained nurses performed and interpreted the results in 48–72 hours according to the American Thoracic Society (ATS) and Centers for Disease Control and Prevention guidelines. The positive interpretation followed was a TST with an area of induration ≥10 mm. If HCW’s had a previous positive TST, we took note of the place and the year. If HCW’s had negative TST, it is repeated annually or after an exposure as part of KFSHRC’s infection control practice. The QFT was carried out by using the commercially available QuantiFERON TB gold In-Tube Assay (Cellestis, Australia) according to the manufacturer’s protocol. The raw optical density was measured and interpreted with the software QuantiFERON TB Gold analysis Software v1.51 (Cellestis, Australia). ### Statistical analysis The candidates were classified according to the WHO’s geographical classification into 6 major groups; American, African, European, Eastern Mediterranean, South East Asian, and Western Pacific. The statistical analysis has been carried out by using the SPSS v-20 software package. A chi squared test was utilized for categorical data. All the putative predictive variables were subjected to calculate the odds ratio and 95% confidence interval. # Results Overall 1603 candidates were enrolled during the study period. Of the total, 8 HCWs (1.2%) were excluded from the analysis due to indeterminate QFT results and 1595 were included for final analysis. The study population included candidates from 33 countries. The native Saudi population was only 17%. The geographical classification on enrolled candidates showed, 52.4% were from Western-Pacific followed by 22.9% Eastern Mediterranean and 11.3% South East Asians. The median of age in the study was 35.5 years with a predominance of female candidates (67.9%). Clinical staffs represented the major study group by71.5% and among them nurses were predominant (53.2%). Overall, 185(11.6%) candidates reported with a previous exposure to known pulmonary TB cases. Majority of the candidates (90.6%) received at least one shot of BCG vaccination before enrollment. Candidates holding a healthcare experience of 2–5 years were comparatively more (37.6%), though the mean duration of experience in the study was 8.4 years. Positive TST was noticed in 31.5% of the enrolled candidates. Among the TST positive cases 45.2% were QFT positive while 54.8% were negative. On the other hand, among TST negative cases, 15.7% were QFT positive. The disagreements between the two tests were relatively high (kappa co-efficient-0.312±0.026, p value- \<0.00001). Among 170 TST converted participants only 46 opted for INH prophylaxis and none of these cases progressed to active TB during the study period. The regression analysis of the putative predictable factors showed a significant association of younger age groups (*p* -0.030, 0.049), BCG vaccination (*p*- 0.028), European (*p* -\<0.001) and south East Asian (*p*-0.009) origin with TST positivity. QFT positivity was associated with profession of candidates and occupational TB exposure risk. In addition, the origin of the candidates (*p*-\<0.001) was highly significant in QFT positivity. Analysis of agreement of results of TST and QFT showed 14.23% concordant positive results while 57.7% is concordant negative. However, 10.8% cases were TST negative and QFT positive. On the other hand 275(17.2%) cases were TST positive and QFT negative. Analyzing the association of various putative factors showed the origin of the patient (*p* \<0.001), occupational TB exposure level (*p*-\<0.001), profession (*p-* 0.001) and BCG vaccination history (*p-* 0.001) have a statistical significance on the agreement of both the tests. Among the discordant results, candidates with Western Pacific origin showed a predominant QFT positivity (65.7%) and negativity (46.9%). Interestingly, among the different professions, nurses have the highly discordant results. # Discussion This is the largest Saudi Arabian study, which evaluated the performance of QFT and TST among highly diverse HCW’s population. The study has targeted onto the putative risk factors with an emphasis on the impact of geographic origin of study subjects. It is also the only cross sectional study in the country that looked at the possibilities of developing active TB in recently identified LTBI in HCW’s. This study has been carried out in a tertiary care center which has been listed among the top five medical facilities in the country employing staffs from more than 60 countries. The impact of this huge diversity in the origin of the staffs has been highly considered on the LTBI prevalence, as it influence the agreement between TST and IGRA. The overall agreement between TST and QFT was 72%, that can be considered as fair, when many previous international studies reported moderate or fair agreement only between both the tests. Supportively, recent studies from Saudi Arabia on dialysis patients also showed 75.5% and 90.9% agreement between TST and QFT/TSPOT Assay respectively. The prevalence of LTBI measured in this study by TST (31.5%) and QFT (25.0%) was relatively high. This elevated rate may be typical for a moderate TB burden country like Saudi Arabia. Furthermore, the previous estimates showed that, Saudi Arabia has only an intermediate prevalence of LTBI (2–14%). In Saudi Arabia, only limited information is available on prevalence of LTBI among HCW’s particularly screened with IGRA testing. A multicenter study utilizing TST alone in HCW’s showed 11% LTBI prevalence, while the current study showed 31.5%. However, the elevated prevalence level of LTBI in the current study may largely depend on the origin of the candidates, that particularly from TB endemic regions. The current study population consisted only 17% Saudi nationals. The results of TST and QFT stratified by the Saudi and non-Saudi origin showed a prevalence of 18.1% and 17.3% among Saudis. This finding highlights the massive role of immigrant HCW’s in the institution and mostly their origin from TB endemic countries. After applying the geographical region classification of WHO into the study, showed the Western Pacific origin candidates (mainly Philippines) as the largest group in the study followed by the Eastern Mediterranean and south East Asians. Interestingly, there are 970(60.8%) candidates truly from high TB burden countries namely, Philippines, India, Bangladesh, Pakistan, Sudan, South Africa, Eritrea, Ethiopia and Kenya. Analysis of the TST results against the nationality of the candidates, showed 305(60.6%) of the total 503 TST positive and 234 (58.6%) of the total 399 QFT positive candidates are from this group. The agreement between two results was really narrow (only 2%). Among the candidates from high TB burden countries only 94(9.7%) had TST conversion during their employment in the study center. This speculates the scope of a remote infection among those remaining 876(90.3%) candidates from their previous destination that is mostly their mother country. This finding has a support from molecular studies by Varghese et al., which showed remote TB infection among immigrants is very common Saudi Arabia. In the current study the maximum discordance was observed in TST positive and QFT negative results. Among the TST positive cases 54.7% were negative for QFT. This finding corroborates with published studies which reported positive TST and negative IGRA is the most common discordance. Majority of the study candidates were from a country where BCG vaccination is mandatory, thus 90.6% of the enrolled candidates were BCG vaccinated. This could probably affect the higher level prevalence of TST positivity. Supportively, a statistically significant association (OR 1.039, 95% CI-0.664–1.625, P value 0.028) was noticed between BCG vaccination and TST positivity while no significant association with QFT. This finding is in concordance with previous studies, which showed BCG vaccination significantly elevates the likelihood of TST positivity. When compared with age \>60 years, the younger age has significant association with TST positivity in concordance with various previous studies. The rate of TST positivity among different professionals showed no significant association. In contrast, the TB exposure level based on profession showed a high significance towards QFT positivity, when low risk group was kept as a reference as seen in previous studies. The origin of the candidates also has significant association with both TST and QFT positivity. The maximum discordant results were noticed among candidates from Western Pacific region. Overall 65.7% of the TST negative QFT positive candidates and 46.9% of TST positive and QFT negative candidates were from mainly Philippines and Malaysia, while candidates from Eastern Mediterranean region showed the rates of 17.4% and 30.9% respectively. The massive enrollment of Filipino candidates (45.2%) has an impact on the overall findings. The two main reasons are the mandatory BCG vaccine administration and high TB burden of Philippines. Perhaps, the higher discordant results of TST negative and QFT positive could not be explained in detail. No significant association could be established between the discordant results against age of the candidates, gender, and years of work in the medical field. However, strong associations could be noticed on origin of the candidates (*p* \<0.001), profession (*p* 0.001), TB exposure level (*p* \<0.001) and BCG vaccination (*p* 0.001). Nurses and support service staffs showed majority of the discordant results. This finding was supported by the rate of TB exposure level as the high TB exposure group has highest discordance between results. This study has certain limitations; candidates reported as TST negative and QFT positive were not retested for confirmation and quantitative analysis of QFT testing was not considered. # Conclusion In conclusion, the prevalence of LTBI estimated by QFT is high among Saudi Arabian HCW’s. The disagreements between TST and IGRA results were relatively high and thus QFT alone cannot be recommended to screen LTBI. The origin of the candidates has significant role in both TST and IGRA positivity. However due to the high level of LTBI prevalence, the screening and management of LTBI in HCW’s in the country must be immediately streamlined. # Supporting Information We deeply acknowledge, Maribel Carbonel, Diana Oliveira, Florence Saguid, Lenora Reyes, Tracy Lynn Alsarhani, Rania Omar Albadawi and Maria Minette Valesteros for their immense support for recruiting the candidates. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SAH A. Alrajhi A. Alkhenizan. Performed the experiments: BV AD GF. Analyzed the data: BV MS. Contributed reagents/materials/analysis tools: SA A. AlSaif A. Alzahrani. Wrote the paper: SAH BV MS A. Alzahrani A. Alrajhi GF.

# Introduction Sleep plays an important role in maintaining health. Sleep disorders have been shown to negatively affect lifestyle-related diseases, such as metabolic syndrome, hypertension, diabetes, and cardiovascular disease. The following underlying mechanisms have been suggested for this association: neurobiological and physiological stressors; the inhibition of glycemic control, which increases dietary intake through the secretion of ghrelin; and the promotion of insulin resistance by increasing cortisol, IL-6, and TNFα levels. Therefore, preventing sleep disorders is important for maintaining health. Exercise has been noted to have a positive impact on sleep disorders. Epidemiological studies on sleep disorders and exercise/physical activity have examined non-restorative sleep in middle-aged and elderly individuals and sleep quality in nursing homes. Intervention studies have reported improvements in sleep quality in adults and elderly individuals with depression. Exercise affects sleep via the following mechanisms: its thermoregulatory effects reduce wake times during the night, it facilitates sleep onset by activating a heat dissipation mechanism controlled by the hypothalamus to increase central body temperature, and it improves mood due to its antidepressant/anxiety effects. In animal studies, exercise was shown to increase the levels of adenosine, which activates the sleep center in the hypothalamus, and serotonin, which synthesizes the sleep hormone melatonin. Nutrition has been investigated as another factor related to sleep disorders. Epidemiological studies have reported a relationship between sleep quality and micronutrients such as carotenoids, vitamin B₁₂, calcium, and selenium. However, the relationships between sleep and macronutrients remain unclear. A previous study indicated the presence of a relationship between sleep and protein intake, but another study reported that no such relationship existed. The latter study demonstrated that sleep was associated with lipid and carbohydrate intake. Thus, the findings obtained from previous epidemiological studies are inconsistent. These discrepancies may exist because of the lack of a uniform method for evaluations using questionnaires. Therefore, we conducted an epidemiological study to investigate the factors affecting sleep using the Pittsburgh Sleep Quality Index (PSQI), which is one of the most frequently used indices for evaluating self-rated sleep quality in sleep medicine. The role of environmental factors must be considering when examining the association between nutrient intake and sleep. However, previous studies have not yet investigated this triangular relationship in detail. Two previous studies that reported different findings on the relationship between protein intake and sleep did not perform an analysis adjusted for environmental factors such as exercise. Therefore, the effects of interactions between environmental factors and nutrient intake on sleep remain unclear. The present cross-sectional study examined the effects of regular exercise and nutrient intake on sleep quality. # Methods ## Data collection In this cross-sectional study, comprehensive health survey data were collected from the residents of Shika Town, Ishikawa Prefecture, Japan, between November 2017 and February 2018. As of November 2017, there were 21,007 residents in Shika Town, and 15,012 were older than 40 years. The Shika study epidemiologically investigates the causes of lifestyle-related diseases through interviews, self-administered questionnaires, and comprehensive medical examinations. Previous studies have also examined the relationship between nutrition and health. ## Participants This study was conducted on participants recruited from those who underwent a medical examination in Shika Town. For details, a total of 378 people aged 40 years and older who live in four model districts (Horimatsu, Tsuchida, Higashimatsudo, and Togi) provided their consent to participate in this sleep study. Of these individuals, 193 were excluded because they did not meet the survey criteria \[169 participants did not complete the brief-type self- administered diet history questionnaire (BDHQ), 1 participant did not have energy records within 600 – 4000Kcal/day, and 23 participants did not complete the smoking, drinking, or exercise questionnaire\]. shows the inclusion criteria. In total, 185 participants (95 males and 90 females; mean age ± standard deviation: 60.5 ± 9.7 years, range 41–83 years) who answered all relevant questions in the questionnaires and did not withdraw their consent were included in the analysis. ## Questionnaire and measurements The participants completed a self-administered questionnaire on lifestyle and underlying diseases. Lifestyle items included the number of exercise days per week and the mean exercise time during each session, whether they were current smokers (1. yes, 2. no) and/or current drinkers (1. yes, 2. no), and education (1. junior high school, 2. high school 3. junior college, 4. university or higher). Underlying disease items included metabolic syndrome (1. yes, 2. no), hypertension (1. yes, 2. no), diabetes (1. yes, 2. no), angina (1. yes, 2. no), myocardial infarction (1. yes, 2. no), and depression (1. yes, 2. no). Body mass index (BMI) was measured using health survey data from the Shika study. Nutrient intake was assessed using the BDHQ. The BDHQ is a four-page structured questionnaire that assesses the consumption frequency of 58 foods and beverages that are commonly consumed by the general Japanese population. The BDHQ estimates dietary intake in the last month using an *ad hoc* computer algorithm. The validity of the BDHQ has been demonstrated in previous studies in Japanese populations. To analyze nutrient data, the density method was used to estimate intake per 1000 Kcal. The following formula was used to calculate the energy intake ratio (% energy) of energy-producing nutrients (proteins, lipids, carbohydrates, and alcohol): energy intake from each nutrient/energy intake (EN) × 100, adjusted intake of non-energy-producing nutrients: crude intake of various nutrients/EN × 1000 Kcal. Sleep status was assessed using the PSQI. The PSQI assesses sleep quality and disturbances over a one-month period and consists of the following components: subjective sleep quality, sleep latency, sleep duration, habitual sleep efficiency, sleep disturbance, the use of sleeping medication, and daytime dysfunction. Each component of the PSQI is scored from 0 to 3. The PSQI global score is a sum of these components that ranges between 0 and 21, with higher scores indicating poorer sleep quality. The validity of the PSQI has been demonstrated in previous studies in Japanese populations. ## Statistical analysis Regular exercise was defined as exercise for at least 30 minutes at a time and twice a week. Buysse *et al*. defined a PSQI score \>5 as poor sleep quality; however, their study population included adolescents and differed from that in our Shika study, which included those aged 40 years and older. In the present study, the median PSQI of the participants was 10; thus, they were classified into PSQI ≤10 and ≥11 groups. The age distribution of the participants in the present study did not differ from that of the Shika Town inhabitants. The distribution of variables was checked by the Kolmogorov–Smirnov, and Shapiro–Wilk normality tests, or the normal distribution curve in the histogram was confirmed before using other statistical tests. The Student’s *t*-test was used to compare the means of continuous variables and the Chi-square test was performed to compare the proportions of categorical variables. All participants were stratified into two groups based on their PSQI scores (PSQI ≤10 and ≥11) and whether they participated in regular exercise (regular exercise group and non-regular exercise group). A two-way analysis of covariance (ANCOVA) was used to examine the effects of the interaction between regular exercise and PSQI on nutrient intake. The following confounding factors were adjusted for: age, sex, BMI, current smoker, current drinker, education, hypertension, and diabetes. A multiple logistic regression analysis was conducted to examine the effects of regular exercise and nutrient intake on sleep quality. The dependent variable was the PSQI (≤10 and ≥11). We used three models in the logistic regression analysis. Model 1 included the individual factors of age, gender, and BMI. Model 2 included the environmental factors of current smoker and current drinker together with individual factors. Model 3 included the disease factors of hypertension and diabetes together with individual factors. Additionally, the analyses were stratified by whether the participants performed regular exercise. Pearson’s correlation coefficient was used to confirm multicollinearity. Specifically, there was no value of \| r \| \>0.9 in the correlation matrix table between independent variables. The forced input method was used for variable selection. The significance level was set at 5%. IBM SPSS Statistics version 25 for Windows (IBM, Armonk, NY, USA) was used for the statistical analysis. ## Ethics statement The present study was conducted with the approval of the Ethics Committee of Kanazawa University (No. 1491). Written informed consent was obtained from all participants. # Results ## Participant characteristics The participants’ sleep quality and nutrient intakes are shown in. Among the 185 participants, there were 95 males with a mean age of 60.9 years (*SD* = 9.6) and 90 females with a mean age of 60.3 years (*SD* = 10.1); there was no significant difference between genders. BMI (*p* \< 0.001) was significantly higher in males than in females. Significantly more males were current smokers (*p* \< 0.000), current drinkers (*p* \< 0.001), and had metabolic syndrome (*p* \< 0.001), diabetes (*p* = 0.049), and angina (*p* = 0.019) than females. The mean PSQI was 10.7 (*SD* = 2.7) in males and 10.7 (*SD* = 2.8) in females, with no significant difference between genders. When comparing nutrients, the total energy (*p* \< 0.001) was significantly higher in males. Conversely, the intakes of other nutrients, excluding carbohydrates, sodium, vitamin D, and vitamin B₁₂, were significantly higher in females. ## Comparison with the PSQI The mean age of the 106 participants in the PSQI ≤10 group was 59.2 years, which was significantly younger than that of the 79 participants in the PSQI ≥11group (62.4 years, *p* = 0.026). The PSQI ≤10 group reported significantly more exercise days per week (*p* = 0.004) and a significantly longer mean exercise time per session (*p* = 0.008). Furthermore, these factors were significantly greater in the PSQI ≤10 group even after adjusting for age, sex, BMI, current smoker, current drinker, education, hypertension, and diabetes (exercise days and exercise time: *p* \< 0.001 and *p* = 0.004, respectively). Therefore, regular exercise was beneficial for sleep quality. The proportion of participants with metabolic syndrome (*p* = 0.026) was significantly higher in the PSQI ≤10 group. When comparing each nutrient, the intakes of retinol equivalent (*p* = 0.044) and vitamin B₂ (*p* = 0.024) were significantly higher in the PSQI ≤10 group than in the PSQI ≥11 group. ## Comparisons with regular exercise The mean age of the 59 participants in the regular exercise group (64.2 years) was significantly older than that of the 126 participants in the non-regular exercise group (58.8 years, *p* \< 0.001). The mean PSQI (*p* = 0.003) was significantly higher in the non-regular exercise group. When comparing each nutrient, the intakes of protein (p \<0.001), minerals (p \< 0.001), and 12 kinds of vitamins were significantly higher in the regular exercise group than that in the non-regular exercise group. ## Effects of the interaction between regular exercise and the PSQI on nutrient intake The regular exercise group was divided into two groups based on PSQI scores; there were 43 participants in the PSQI ≤10 group and 16 in the PSQI ≥11 group. The non-regular exercise group was similarly divided into two groups based on the PSQI scores; there were 63 participants in the PSQI ≤10 group and 63 in the PSQI ≥11 group. A two-way ANCOVA adjusting for age, sex, BMI, current smoker, current drinker, education, hypertension, and diabetes was used to examine the effects of interactions between regular exercise and the PSQI on nutrient intake. Interactions were observed for age (*p* = 0.006), education (*p* = 0.002), protein (*p* = 0.002), carbohydrate (*p* = 0.045), phosphorus (*p* = 0.008), zinc (*p* = 0.031), vitamin D (*p* = 0.015), vitamin B₁₂ (*p* = 0.007), and pantothenic acid (*p* = 0.008). A post hoc Bonferroni analysis indicated that there was significantly higher protein intake in the PSQI ≤10 group than in the PSQI ≥11 group with regular exercise (*p* = 0.001); however, there was no difference between the two PSQI groups without regular exercise. ## Effects of regular exercise and protein intake on sleep quality shows the results of a multiple logistic regression analysis with PSQI (≤ 10 and ≥11) stratified by regular exercise. Protein intake was a significant independent variable in any models that were adjusted for individual factors (age, sex, and BMI; OR: 1.260; 95% CI: 1.037, 1.531; *p* = 0.020), individual and environmental factors (current smoker and current drinker; OR: 1.357; 95% CI: 1.081, 1.704; *p* = 0.009), and individual and disease factors (hypertension and diabetes; OR: 1.675; 95% CI: 1.206, 2.326; *p* = 0.002) in the regular exercise group but not in the non-regular exercise group. This result implies that sleep quality is better with a high protein intake, even after adjusting for different confounding factors only in the regular exercise group. # Discussion In the present study, the PSQI was selected as the most frequently used index for sleep evaluation. Epidemiological studies on sleep have been performed using sleep times and questionnaires. However, evaluating sleep by time alone lacks objectivity because sleep times and measurement items differ among studies. For example, one study considered the appropriate sleep time to be 7–8 hours, but another considered it to be 7–9 hours; other studies have evaluated sleep- related time based on sleep latency (difficulty falling asleep) or sleep efficiency (maintaining sleep). Therefore, comprehensively evaluating sleep quality using a questionnaire may provide more objective findings. A previous study that compared the diagnostic screening characteristics of the Insomnia Severity Index, the Athens Insomnia Scale, and the PSQI reported similar sensitivities and specificities. Therefore, the PSQI in the present study was confirmed to be a valid and comparable questionnaire to those used in other studies. Buysse *et al*. defined a PSQI score \>5 as poor sleep quality. Conversely, Das *et al*. demonstrated that the mean PSQI was 8.59 ± 5.35 in a community-based study among a geriatric population, and they described that the difference in the PSQI may be due to the different cultures and lifestyles of people in different countries. The mean PSQI of all the participants in the present study was 10.7 ± 2.7, which seemed to reflect the current average Japanese lifestyle. Comparisons between the PSQI ≤10 and ≥11 groups in the present study revealed that regular exercise was beneficial for sleep quality, which is consistent with previous findings. However, other studies did not observe a relationship between exercise and sleep. Briefly, the lowest or highest levels of exercise were not associated with sleep disorders, and the effects of short-term resistance exercise on sleep were inconsistent. By contrast, previous studies reported a positive relationship between exercise intensity and sleep with 30 minutes or more of exercise each time, moderate to intense physical activity of 150 minutes or more per week, or 500 to 1500 metabolic equivalents of task minutes/week of physical activity. Exercise intensity in the regular exercise group in the present study was considered intermediate because the mean days of exercise per week was 4.3 (*SD* = 1.7) and the mean exercise time per session was 80.3 minutes (*SD* = 64.2). Accordingly, the relationship observed between exercise intensity and sleep in the present study appears to support previous findings showing that intermediate exercise intensity has a positive effect on sleep quality. The multiple logistic regression analysis in the present study revealed a positive correlation between good sleep quality and protein intake only in the regular exercise group. These results seem to indicate that there is a mechanism by which regular exercise promotes protein absorption. In addition to exercise, the ingestion of protein just before sleep has been reported to improve nighttime protein synthesis by enhancing its digestion and absorption. Tryptophan is a constituent amino acid of protein that competes with the other larger neutral amino acids to gain access to the transport system to cross the blood-brain barrier. Dietary carbohydrates pull larger amino acids into the muscle tissue, allowing tryptophan to access the transport system, cross the blood-brain barrier, and contribute to the synthesis of serotonin and melatonin. Tryptophan has been shown to affect the serotonin-melatonin pathway because an intraperitoneal injection in rats increased serotonin levels; likewise, its administration to patients with moderate insomnia significantly reduced sleep latency. The reason for good sleep quality among participants who reported high protein intake in their daily diet and regular exercise was thought to be a result of the pathogenesis in which the tryptophan-serotonin-melatonin pathway was activated due to the enhanced protein absorption. In the present study, the following micronutrients were associated with sleep quality after adjusting for confounding factors (age, sex, BMI, current smoker, current drinker, education, hypertension, and diabetes): phosphorus, zinc, vitamin D, vitamin B₁₂, and pantothenic acid. Many of these micronutrients showed similar results as previous studies. Frank *et al*. reported that lower intakes of phosphorus and zinc were associated with a shorter sleep duration. Komada *et al*. reported that vitamin D and vitamin B₁₂ in adult males were associated with sleep duration. Grandner *et al*. reported that vitamin D was associated with sleep maintenance difficulties. Since the results for many of the micronutrients examined in the present study agreed with those of previous studies, the relationships observed between micronutrient intakes and sleep in the present study were considered reliable. One limitation of the present study is that the number of inputs for the independent variables was restricted in the multivariate analysis because of the small number of participants. This study might include selection bias because the subjects participated voluntarily in this study. Since the PSQI was evaluated only with a questionnaire, using a more objective method such as a polysomnogram to assess sleep quality is necessary. Since we have not examined the effects of obstructive sleep apnea, future studies should be performed that incorporate a design to evaluate obstructive sleep apnea. Moreover, since this was a cross-sectional study, interventions analyzing regular exercise and protein intake could not be conducted. Further multicenter randomized controlled trials with target values for exercise intensity and protein intake are necessary to clarify the effects of regular exercise and nutrient intake on sleep quality. # Conclusions We conducted this cross-sectional study on Japanese participants to investigate the relationship between regular exercise and nutrient intake as factors affecting sleep quality. Protein intake was higher among the participants with a PSQI ≤10 in the regular exercise group (mean, 17.13% of energy consumption) than in those in the non-regular exercise or PSQI ≥11 groups. Furthermore, the results of the multiple logistic regression analysis showed that sleep quality was better in the regular exercise group when protein intake was high; this relationship was not observed in the non-regular exercise group. # Supporting information We would like to thank the officials of Shika Town, Ishikawa prefecture and the staff of the Department of Environmental and Preventive Medicine, Kanazawa University Graduate School of Medical Sciences, Kanazawa University Graduate School of Advanced Preventive Medical Sciences, Department of Bioinformatics and Genomics, University of Tsukuba and Keio University. 10.1371/journal.pone.0247926.r001 Decision Letter 0 Moran Jose M. Academic Editor 2021 Jose M. Moran This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 24 Nov 2020 PONE-D-20-33985 Protein intake in inhabitants with regular exercise is associated with sleep quality: Result of the Shika study PLOS ONE Dear Dr. Suzuki, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. It is necessary that the authors address the requirements expressed by the reviewers. Please submit your revised manuscript by Jan 08 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Jose M. Moran Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. In your Methods section, please provide additional information about the participant recruitment method and the demographic details of your participants. Please ensure you have provided sufficient details to replicate the analyses such as: a) the recruitment date range (month and year), b) a description of any inclusion/exclusion criteria that were applied to participant recruitment, c) a table of relevant demographic details, d) a statement as to whether your sample can be considered representative of a larger population, e) a description of how participants were recruited, and f) descriptions of where participants were recruited and where the research took place. Finally, please report whether sample size calculations were performed. 3\. Please include additional information regarding the survey or questionnaire used in the study and ensure that you have provided sufficient details that others could replicate the analyses. For instance, if you developed a questionnaire as part of this study and it is not under a copyright more restrictive than CC-BY, please include a copy, in both the original language and English, as Supporting Information. Moreover, please include more details on how the questionnaire was pre-tested, and whether it was validated. 4.  We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see <http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data- access-restrictions>. In your revised cover letter, please address the following prompts: a\) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. b\) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see <http://www.bmj.com/content/340/bmj.c181.long> for guidelines on how to de- identify and prepare clinical data for publication. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data- availability#loc-recommended-repositories>. We will update your Data Availability statement on your behalf to reflect the information you provide. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: I have read with great interest the study of Fumihiko Suzuki and collaborators and I want to comment the following: 1\. I am struck by the division of two study groups based on a cut-off point of 10 in the PSQI without a clinical justification. The original PSQI study concluded that a value greater than 5 indicates a negative impact on sleep. This data is mentioned by the authors in the methodology. However, it makes no mention of the Japanese validation study that found similar data. That is, values \>5.5 are indicative of poor sleep quality. By the way, in the study the control group had an average 38 years old and a cut-off point of 7.5, a specificity of 97% was found for those patients with primary insomnia. Then, an average of 10 points seems to be high and difficult to be considered normal. (Doi Y, Minowa M, Uchiyama M, Okawa M, Kim K, Shibui K, et al. Psychometric assessment of subjective sleep quality using the Japanese version of the Pittsburgh Sleep Quality Index (PSQI-J) in psychiatric disordered and control subjects. Psychiatry Res. 2000;97: 165–172.) The authors could use the cutoff points that best fit their purposes, but as a post-frame analysis. In strict adherence to the above, it would be desirable for the researchers to add an analysis considering the cut-off point of\> 5.5 and continue with the groups they consider. 2\. The discussion should mention the possible causes for which the average population has poor sleep quality (mean PSQI of 10) and be careful to consider the group with \<10 as having a good quality of sleep. Reviewer \#2: the manuscript is well written with novel findings comments: methodology: the author needs to define what is PSQI and BDHQ and did you use Japanese version and if so, any validation study done Results: good analysis was done Discussion: PSQI has its own drawback for sleep quality assessment. furthermore, people with metabolic syndrome are at greater risk of developing obstructive sleep apnea and that may hinder their sleep quality please add these comments in the discussion \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0247926.r002 Author response to Decision Letter 0 14 Jan 2021 I want my identity to be published for this peer review. 10.1371/journal.pone.0247926.r003 Decision Letter 1 Moran Jose M. Academic Editor 2021 Jose M. Moran This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Feb 2021 PONE-D-20-33985R1 Protein intake in inhabitants with regular exercise is associated with sleep quality: Result of the Shika study PLOS ONE Dear Dr. Suzuki, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Although authors have appropriately addressed the requirements of the reviewers, before recommending publication of the article, and given that the authors have included in their analysis the t-Student, ANCOVA, multiple linear regression and Pearson correlations, the authors should state in the methodology section that all the variables involved fulfilled the assumptions required by parametric methods and that these were tested by the appropriate tests (indicate which ones the authors have used). Please submit your revised manuscript by Mar 28 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Jose M. Moran Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: I Don't Know \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors' response to the observation of the cut-off points is valid because they are supported by a recent report on the geriatric population and therefore responds favorably to the question about the high value of the Pittsburgh sleep quality scale in the population of this study. Reviewer \#2: minor linguistic errors, i would suggest writing a review linking, sleep, exercise fitness and diet. the focus on sleep not only from OSA prospective but rather descriping the link between sleep duration, timing and physical fitness in addition to healthy diet \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **Mohammed A. Al-Abri \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0247926.r004 Author response to Decision Letter 1 16 Feb 2021 Feb 16th, 2021. Jose M. Moran, Ph. D Academic Editor PLOS ONE Dear Dr. Moran, Thank you for inviting us to submit a revised draft of our manuscript entitled, “Protein intake in inhabitants with regular exercise is associated with sleep quality: Results of the Shika study” to PLOS ONE. We also appreciate the time and effort you and each of the reviewers have dedicated to providing insightful feedback on ways to strengthen our paper. Thus, it is with great pleasure that we resubmit our article for further consideration. We have incorporated changes that reflect the detailed suggestions you have graciously provided. We also hope that our edits and the responses we provide below satisfactorily address all the issues and concerns you and the reviewers have noted. To facilitate your review of our revisions, the following are our point-by-point responses to the questions and comments delivered in your letter dated Feb 11th, 2020. Major points. Q1. Although authors have appropriately addressed the requirements of the reviewers, before recommending publication of the article, and given that the authors have included in their analysis the t-Student, ANCOVA, multiple linear regression and Pearson correlations, the authors should state in the methodology section that all the variables involved fulfilled the assumptions required by parametric methods and that these were tested by the appropriate tests (indicate which ones the authors have used). A1. We have added the following phrase to the revised manuscript: “The distribution of variables was checked by the Kolmogorov–Smirnov, and Shapiro–Wilk normality tests, or the normal distribution curve in the histogram was confirmed before using other statistical tests.” (P9 L180–182). Supplementary explanation: Although some variables had a p-value of 0.05 or less in the Kolmogorov–Smirnov, and Shapiro–Wilk normality test, it was confirmed that a normal distribution curve was drawn in the histogram. Therefore, we believe that the variables used in this study can be expected to have a multivariate normal distribution. Reviewer \#1: There were no additional comments. Reviewer \#2 Q1. Has the statistical analysis been performed appropriately and rigorously? (Reviewer \#2: I Don't Know) A1. We have added the following phrase to the revised manuscript: “The distribution of variables was checked by the Kolmogorov–Smirnov, and Shapiro–Wilk normality tests, or the normal distribution curve in the histogram was confirmed before using other statistical tests.” (P9 L180–182). Supplementary explanation: Although some variables had a p-value of 0.05 or less in the Kolmogorov–Smirnov, and Shapiro–Wilk normality test, it was confirmed that a normal distribution curve was drawn in the histogram. Therefore, we believe that the variables used in this study can be expected to have a multivariate normal distribution. Q2. minor linguistic errors, i would suggest writing a review linking, sleep, exercise fitness and diet. the focus on sleep not only from OSA prospective but rather descriping the link between sleep duration, timing and physical fitness in addition to healthy diet. A2. Since we were conducting a cross-sectional analysis, we cannot examine the sleep duration and timing in detail, but the mechanism by which regular exercise and nutrition on sleep quality were described as follows: "The reason for good sleep quality among participants who reported high protein intake in their daily diet and regular exercise was thought to be a result of the pathogenesis in which the tryptophan-serotonin-melatonin pathway was activated due to the enhanced protein absorption" (P20 L348-351). Again, thank you for giving us the opportunity to strengthen our manuscript with your valuable comments and queries. We have worked hard to incorporate your feedback and hope that these revisions persuade you to accept our submission. Sincerely, Fumihiko Suzuki Department of Environmental and Preventive Medicine, Graduate School of Medical Science, Kanazawa University, 13-1 Takaramachi, Kanazawa, Ishikawa 920-8640, Japan. Tel: +81-76-265-2218 Email address: <f-suzuki@stu.kanazawa-u.ac.jp> 10.1371/journal.pone.0247926.r005 Decision Letter 2 Moran Jose M. Academic Editor 2021 Jose M. Moran This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 17 Feb 2021 Protein intake in inhabitants with regular exercise is associated with sleep quality: Result of the Shika study PONE-D-20-33985R2 Dear Dr. Suzuki, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Jose M. Moran Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0247926.r006 Acceptance letter Moran Jose M. Academic Editor 2021 Jose M. Moran This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Feb 2021 PONE-D-20-33985R2 Protein intake in inhabitants with regular exercise is associated with sleep quality: Results of the Shika study Dear Dr. Suzuki: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Jose M. Moran Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Gene targeting (e.g., targeted gene replacement) is one of the main molecular tools used in yeast science, which helps in understanding gene functions and interactions as well as cellular and molecular processes in yeasts. Model yeasts, such as *Saccharymyces cerevisiae* and *Schizosaccharomyces pombe*, have very efficient gene targeting systems. For instance, disruption cassettes with short flanking regions that range from 30 bp to 50 bp can integrate with high efficiencies via homologous integration at the correct genomic locus (routinely \>70%) in *S. cerevisiae*. In contrast, the gene targeting efficiencies of various “non-conventional” yeasts, such as *Pichia pastoris*, *Hansenula polymorpha, Yarrowia lipolytica, Pichia stipitis* and *Kluyveromyces lactis* can be extremely low (usually \<1%) with the same length of flanking regions. Accordingly, homologous arms varying from 200 bp to approximately 2000 bp are usually required to ensure efficient gene replacement in non-conventional yeasts. Nevertheless, for some “stubborn” genes, the probability of obtaining the desired gene replacement event is so low that transformation and/or screening procedures have to be iteratively performed, which is laborious and time-consuming. Therefore, the low targeting efficiencies of non-conventional yeasts greatly limits the researches and applications of these industrially important microorganisms. To improve the targeting efficiencies of non-conventional yeasts, molecular mechanisms of gene targeting should be understood. Yeasts have the homologous recombination (HR) pathway and the non-homologous end-joining (NHEJ) pathway. The HR pathway, which depends on the Rad52 epistatic group, is responsible for the targeted integration of DNA. Integration can also be mediated randomly via the NHEJ pathway, which depends on the Ku70/Ku80 protein complex. When foreign DNAs are transformed into cells, they are competed by these two recombinant pathways. Therefore, efficient gene targeting is determined by the relative strength of the HR pathway compared with that of the NHEJ pathway. The bias, which favors the NHEJ pathway in non-conventional yeasts, determines their low targeting efficiencies. However, NHEJ pathway bias is not the only factor that contributes to the low targeting efficiency of non-conventional yeasts. The efficiency of homologous integration in strains with the same genetic background can be very locus specific. For example, the disruption of *ARG1, ARG2, ARG3, HIS1, HIS2, HIS5,* and *HIS6* in *P. pastoris* GS115 strain with flanking arms that range from 200 bp to 900 bp is considerably efficient at frequencies of 44% to 90%. However, the deletion of *PIM* and *OCH1* from the same strain that contains both flanking arms that were longer than 1 kb occurred at a frequency of \<1%. To date, this locus specific phenomenon is not well understood. One explanation is the presence of “hotspot” regions for yeast genome, in which the underlying mechanism still remains unclear. Another, which is often overlooked in previous studies, may be the loss of function effect. Several non-conventional yeasts such as *P. pastoris*, *H. polymorpha*, *Y. lipolytica*, and *P. stipitis* are predominantly haploid. Thus, the knockout of genes with important physiological functions often means great loss of cellular fitness, which would lead to a delayed or failed appearance of correct disruptants. Therefore, the calculated frequency of correct gene targeting by colony counting is likely lower than the actual frequency of gene targeting events that happen genetically because of the unformed colonies. To validate this assumption as well as provide an efficient solution to delete “stubborn” genes in haploid non-conventional yeasts, in this study, we proposed a new strategy for enhancement of gene targeting efficiency by improving cellular fitness of mutant cells, specifically by increasing the genetic redundancy of host cells. To achieve this goal, the targeted gene was cloned into an expression vector (helper plasmid), and transformed into yeast cells to generate the transition host. Targeted gene disruption was then applied in the transition host by transformation with the disruption plasmid. After gene deletion in the genome was successfully validated, the helper plasmid can be easily removed. *P. pastoris*, a methylotrophic yeast, was used as a model in this study for two reasons: 1) the efficiency of targeted gene replacement in *P. pastoris* is very low, which is estimated to occur at a frequency of 0.1% when the total length of the target fragments is 500 bp ; and 2) *P. pastoris* is of great industrial importance. It is by far one of the most often used yeast species in the production of recombinant proteins. The *OCH1* gene of *P. pastoris* was chosen as an illustrative example because its deletion procedure by double homologous recombination is notoriously inefficient. To demonstrate the effectiveness of this strategy, the deletion of two other genes, namely, *SGS1* and *KU70*, was also assessed using the same method. # Materials and Methods ## Strains, Plasmids, and Oligonucleotides A complete list of strains and plasmids is presented in. Oligonucleotide synthesis and DNA sequencing were performed at the Shanghai Sangon Biological Engineering Technology and Service Co. (Beijing, China). ## Media and Growth Conditions *Escherichia coli* DH5α strain was grown at 37°C in LLB broth (10 g L^–1 of tryptone, 5 g L^–1 of yeast extract, 5 g L^–1 of NaCl) or LB medium (10 g L^–1 of tryptone, 5 g L^–1 of yeast extract, 10 g L^–1 of NaCl). *P. pastoris* was cultivated aerobically at 30°C in yeast extract peptone dextrose (YPD) medium (10 g L^–1 of yeast extract, 20 g L^–1 of peptone, 20 g L^–1 of glucose). Antibiotics were added at the following concentrations: 100 mg L^–1 of ampicillin, 50 mg L^–1 of kanamycin, and 25 mg L^–1 of Zeocin (Invitrogen) for *E. coli*; 500 mg L^–1 of geneticin (G418, Invitrogen) and 50 mg L^–1 of Zeocin for *P. pastoris*. ## Construction of Disruption Plasmids pZeoloxp, the base vector used to construct all of the other disruption plasmids, was constructed based on the following procedures. Zeocin resistance cassette was amplified from pPICZA with P_zeo1/P_zeo2, digested with *Sac*I and *Bgl*II, and cloned into pUG6 to replace the kanMX cassette between the loxP sites, thereby producing the pUG-zeoloxp plasmid. The origin of replication (ori), which was amplified from pPICZA by polymerase chain reaction (PCR) with P_ori1/P_ori2 primer pair and digested with *Sal*I, was inserted into the *Xho*I site of pUG-zeoloxp. The DNA fragment with ori and ampicillin resistance gene was then excised through *Not*I digestion and religation. To construct the disruption plasmid for *OCH1* deletion, the upstream and downstream homologous regions of *OCH1* were initially amplified through PCR from *P. pastoris* GS115 strain by using two primer pairs, namely, OCH1-N₅/OCH1-N₃ and OCH1-C₅/OCH1-C₃. The two fragments were cut with *Spe*I, *Not*I, and *Pst*I, ligated into pZeoloxp which was cut with *Spe*I and *Pst*I, thereby generating pZeoloxp-OCH1. The disruption plasmids used to delete the *SGS1* and *KU70* genes, namely, pZeoloxp-SGS1 and pZeoloxp-KU70 were constructed following exactly the same procedure with their corresponding primer pairs. ## Construction of the Episomal Plasmids pGAPZB and the kanamycin resistance gene, amplified from pPIC9K with P_kan1/P_kan2, were cut using *Nco*I/*Stu*I and ligated by the T4 DNA ligase to replace the original Zeocin-resistant gene, thereby generating the pGAPKB plasmid. PARS2, an autonomous replication sequence of *P. pastoris*, was amplified from the genomic DNA of GS115 strain with P_pars2-F/P_pars2-R, cut with *Bam*HI and *Bgl*II, and inserted into *Bgl*II site of pGAPKB to generate the episomal expression plasmid, pGKARS. To express *OCH1*, *SGS1*, and *KU70* in *P. pastoris* with pGKARS, their encoding genes were amplified from the GS115 genome by using their respective primer pairs. PCR fragments were cut using *Eco*RI and *Nde*I, and then ligated into the corresponding pGKARS sites to produce the episomal expression plasmids, namely, pGKARS-OCH1, pGKARS-SGS1 and pGKARS-KU70, respectively. An inducible *mazf* cassette was inserted into pGKARS-OCH1 to facilitate the removal of pGKARS-OCH1 as necessary. To construct the *mazf* cassette, the *mazf* gene was amplified from *E. coli* by using the P_mazf1/P_mazf2 primer pair, digested with *Eco*RI and *Not*I, and ligated into the corresponding pPICZA sites to place the *mazf* gene under the control of the alcohol oxidase (AOX1) promoter. The *mazf* cassette was then excised from the resulting plasmid by *Bgl*II and *Bam*HI digestion and inserted into the *Bgl*II site of pGKARS-OCH1, thereby generating the pGKARSmazf-OCH1 plasmid. ## Competent *P. pastoris* Cell Preparation and Transformation The competent cells were prepared based on a revised version of a previously described method to achieve a highly efficient transformation of *P. pastoris*. Briefly, a fresh single clone was inoculated into 100 ml YPD medium and grown overnight at 30°C by shaking at 200 rpm until the cell density reached an OD₆₀₀ of 1∼2. The cells were pelleted and resuspended at room temperature for 30 min in 8 ml of 100 mM LiAc, 10 mM dithiothreitol, 0.6 M sorbitol, and 10 mM Tris-hydrochloride at pH 7.5. The resulting cells were then washed thrice with 2 ml to 3 ml of 1 M ice-cold sorbitol. Finally, the cells were suspended in 1 M ice-cold sorbitol and transferred to 1.5 ml microcentrifuge tubes with an aliquot of 80 µl. *P. pastoris* GS115 strain was transformed by electroporation according to the protocols outlined by Invitrogen. ## Direct Gene Disruption and Verification The disruption plasmids (pZeoloxp-OCH1, pZeoloxp-SGS1, or pZeoloxp-KU70) were linearized by *Not*I and transformed into *P. pastoris* GS115 strain. The cells were then spread on YPD plate that contains 50 mg L^–1 of Zeocin to screen the Zeocin-resistant transformants. The transformants were further analyzed by colony PCR to verify whether they contained the correct chromosomal integrations of the gene disruption cassette. The parental strains were used as the control group. Two primer pairs, P_1f (located upstream of the 5′ homologous region in the genome)/P_1r (located within pZeoloxp) and P_2f (located within pZeoloxp)/P_2r (located downstream of the 3′ homologous region in the genome), were used to verify each gene. The successful amplification of both bands with the expected size indicated that the chromosomal integrations were correct. The amplification of one band with either primer pairs corresponded to single crossover recombination. The P_3f/P_3r primer pair, located within the open reading frame, was used for further verification, and no band should be amplified for the correct disruptants. ## New Disruption Method and Verification A typical process involves three steps. First, the helper plasmid, such as pGKARSmazf-OCH1, was introduced into the GS115 strain to generate the transition host, GS-tranOCH1. The cells were screened on the YPD medium with 500 mg L^–1 of G418. The transformants that harbor the episomal expression plasmids were confirmed by performing colony PCR with the P_5′GAP/P_3′AOX1 and P_5′AOX1/P_3′AOX1 primer pairs. Second, the transition host was subjected to a one-step disruption method. Third, the helper plasmid was removed by streaking on MM agar plates \[13.4 g L^–1 of YNB, 0.5% (v/v) methanol, 0.1 g L^–1 of histidine, 0.04 g L^–1 of Zeocin, and 0.0004 g L^–1 of biotin\]. The P_5′GAP/P_3′AOX1 and P_5′AOX1/P_3′AOX1 primer pairs were used to verify the removal of episomal expression plasmid. # Results ## Direct Disruption of *OCH1* Gene α-1,6-Mannosyltransferase, the product of *OCH1* gene, adds the first α-1,6-mannose to the Man9ClcNAc2 core oligosaccharide and initiates several subsequent high mannose-type N-glycosylations. Accordingly, *OCH1* is often chosen as the deletion target to avoid the hyperglycosylation of the expressed heterologous proteins in yeasts. However, *OCH1* deletion is a very inefficient process according to previous studies. Choi and his colleagues obtained only one *P. pastoris* strain with the inactivated *OCH1* gene from approximately 1000 clones using double homologous strategy with 2878/1011 bp length of homologous flanks. Using homologous flanks as long as 3 kb, Vervecken et al. failed to delete the *OCH1* gene following the same strategy. To test the efficiency of *OCH1* deletion, we first used the direct one-step knockout method, and the results were used as a control group for the succeeding experiment. We initially constructed a disruption vector for the *OCH1* gene inactivation. This vector was generated by inserting two flanking regions, which were amplified from the *P. pastoris* GS115 genome, into the pZeoloxp module plasmid by performing a three-fragment ligation. The two homologous flanks, 5′ and 3′ regions of the disruption target, were 874 and 852 bp, respectively. The disruption plasmid, designated as pZeoloxp-OCH1, was then linearized at the *Not*I site and transformed into *P. pastoris*. More than 1000 clones were initially analyzed by PCR using the P_1f/P_1r and P_2f/P_2r primer pairs. Only one clone, designated as KO24#, showed one expected amplified fragment (1.42 kb). However, the other expected 1.25 kb fragment was not amplified from KO24#, suggesting that KO24# may be the result of a single crossover recombinant event at the 5′ homologous region. ## Construction of Plasmid pGKARSmazf-OCH1 for Modification of *P. pastoris* To apply our proposed strategy, we initially provided a backup *OCH1* gene for *P. pastoris* before deletion to avoid compromising the fitness of the yeast cells because of function loss. This process was performed by cloning *OCH1* into a carefully designed episomal pGKARSmazf-OCH1 vector, which is characterized by three elements: 1) *Pichia* ARS2 (PARS2) fragment, which kept the plasmid inside the cells for a considerable time and allowed the easy removal when needed; 2) the strong constitutive glyceraldehyde-3-phosphate dehydrogenase (GAP) promoter; and 3) the *mazf* expression cassette regulated by promoter AOX1, whose presence is essential for plasmid removal (as explained later). The pGKARSmazf-OCH1 plasmid was introduced into the GS115 strain by electroporation without linearization. The transformants were screened on YPD plate that contains 500 mg L^–1 of G418. Approximately 195 colonies were formed per µg of DNA. Colony PCR was performed using the two primer pairs, P_5′GAP/P_3′AOX1 and P_5′AOX1/P_3′AOX1, to verify the positive clones. Among the 16 clones, 15 showed the two expected bands, particularly 618 and 1837 bp for *mazf* and *OCH1* cassettes, respectively (data not shown). ## Disruption of the Chromosomal *OCH1* Gene and Elimination of the pGKARSmazf-OCH1 Plasmid The transition host that carries the redundant *OCH1* copies (designated as GS- tranOCH1) was used as the host for the *OCH1* gene disruption. GS-tranOCH1 was transformed using the linearized pZeoloxp-OCH1 disruption plasmid based on the same procedure used in the direct one-step method. Approximately 60 transformants per µg of DNA were obtained, randomly picked, and analyzed by colony PCR using the P_1f/P_1r and P_2f/P_2r primer pairs. The results showed that the expected 1.42 kb band was amplified by the P_1f/P_1r primer pair from six clones, which were further analyzed using the P_2f/P_2r primer pair. All of the clones also revealed the expected 1.25 kb band (data not shown), suggesting that the chromosomal *OCH1* was successfully deleted by gene replacement. The results were also confirmed by sequencing the 1.42 and 1.25 kb PCR products. To remove pGKARSmazf-OCH1, *och1* mutant transition strain (designated as GS- tranOCH1-ΔOCH1) was streaked on a MM plate, in which the *mazf* could be expressed with the AOX1 promoter induced by methanol in the medium. The *mazf* gene, which was obtained from *E. coli*, encodes the MazF toxin that functions as an mRNA interferase and inhibits the growth of prokaryotes and eukaryotes. Therefore, *mazf* expression likely causes a strong selection pressure on streaked strains and forces them to lose the obtained plasmids. The colonies, which were much smaller than that of the GS115 strain, appeared after the strain was cultured for 3 d at 30°C. A change in the colony morphology from smooth to rough appearance was also observed, which was consistent with the previous characterizations of *och1* mutant strains. Ten randomly selected colonies on the MM plate were determined by colony PCR with the P_1f/P_1r, P_2f/P_2r and P_3f/P_3r primer pairs to verify *OCH1* deletion and plasmid clearance. All of the 10 clones exhibited the desired amplification pattern, in contrast to control, suggesting the efficient removal of helper pGKARSmazf-OCH1 plasmid from *och1* mutant strains (resultant strains were named GS-ΔOCH1). The growth profiles of GS115, GS-tranOCH1-ΔOCH1, and GS-ΔOCH1 were then compared. The results showed that the duration of the lag phase of GS-ΔOCH1 was approximately doubled compared with that of GS115 strain, indicating that the loss of the *OCH1* gene is severely detrimental to yeast growth. By contrast, the lag phase of GS-tranOCH1-ΔOCH1 was slightly longer than that of GS115, suggesting that the *och1* phenotype can be rescued by the presence of the redundant *OCH1* gene in the episomal expression plasmid to a large extent. ## Disruption of the Chromosomal *SGS1* and *KU70* Genes To assess the effectiveness of our strategy, two *P. pastoris* genes, namely, *SGS1* and *KU70* were selected for gene-targeted deletion. The conventional one-step strategy and the new strategy were both applied. The disruption plasmids for *SGS1* and *KU70*, namely, pZeoloxp-SGS1 and pZeoloxp-KU70, as well as two transition strains with the redundant copies of the corresponding genes, were generated based on the same procedure used in the *OCH1* gene. shows that the *KU70* deletion by the one-step strategy resulted in five positive clones among the 34 selected transformants, whereas ten positive clones were obtained by performing the new strategy, indicating a 1-fold increase in the frequency of *KU70* gene targeting. For *SGS1*, the frequency of obtaining positive clones was increased from 1% to 24%, which is a 23-fold increase, compared with that in one-step strategy. # Discussion For non-conventional yeasts, efficient gene targeting remains a challenge, which largely limits the study and the application of these industrially important strains. To address this issue, two strategies were most commonly employed: 1) increasing the HR efficiency, usually by increasing the homologous arm length. However, longer flanks are not always sufficient for a high percentage of homologous integration, ; 2) suppressing the NHEJ pathway by deleting important functional proteins, such as KU70 and KU80, involved in the NHEJ pathway. For instance, the deletion of the *KU70* gene in *K. marxianus* yields 80% homologous gene targeting efficiency by using homologous sequences with at least 40 bp in length. The integration at *HIS4* and *ADE1* loci results in \>90% targeting efficiencies with only 250 bp of flanking homologous DNA when the *KU70* homolog of *P. pastoris* is knocked out. Nevertheless, the stability and robustness of these strains should be further evaluated. While previous studies have mainly focused on increasing the HR efficiency or decreasing the competition from the NHEJ pathway, one fact that is overlooked is that the strains after gene deletion on molecular and cellular basis, require a recovery and proliferation process to form colonies. Unfortunately, several non- conventional yeasts, such as *P. pastoris*, *H. polymorpha*, and *K. lactis*, are predominantly haploid. The deletion of functionally important genes, particularly when these genes do not have paralogs in the genome, often results in the loss of fitness, which inhibits the proliferation into sizeable colonies. This condition may be more evident under unfavorable conditions, such as cellular recovery from electroporation and growth on solid media with poor nutrition. For instance, the *acs2* mutant of *K. lactis* requires three weeks to form large colonies. Therefore, the possibility that the actual gene targeting efficiency can be reduced is high because of the unformed colonies. Based on this rationale, we aimed to improve the fitness of haploid yeast cells by providing backup genes presented in a well-designed episomal vector. The fitness of mutant cells was improved effectively by this strategy. The growth defect of *och1* mutant cells was restored to a large extent. As a result, the improvement of gene targeting efficiencies with this strategy is significant. The *och1* disruptants, which cannot otherwise be acquired in practice by direct one-step deletion, can be obtained at a frequency of approximately 10% with mediate length of homologous flanks. The efficiencies of the other two genes, *KU70* and *SGS1*, were also increased by 1- and 23-fold, respectively. These results validated our assumption that cellular fitness is an important factor that limits the efficiency of gene targeting in non-conventional yeasts. In summary, we provided an efficient gene targeting strategy for non- conventional yeasts. The targeted gene was initially amplified by cloning into a helper expression plasmid. After the gene was successfully deleted from the genome, the helper plasmid was removed. Thus, the *och1* disruptants were obtained at a frequency of 10%. The gene targeting efficiencies of *SGS1* and *KU70* were increased by 1- and 23-fold, respectively. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CZ TZ. Performed the experiments: CZ HS PL. Analyzed the data: CZ TZ YL. Wrote the paper: CZ TZ NH YL.

# Introduction In the Near East, cradle of the domestication process, goats were among the first to be domesticated ca. 10,500 years ago. Several thousands years later, domestic goats (*Capra hircus*) were dispersed beyond the natural distribution of its wild ancestor (*Capra aegagrus*). They spread in Anatolia and Europe (starting from 8,800 calBP) throughout the Neolithic dispersal, along with pigs, cattle and sheep. Today, goats are present all over the world with more than 867 million of individuals. In order to better assess the historical processes of domestication, goat mitochondrial genetic diversity has been largely studied across the old world (Europe, Asia, and more recently Africa). It is structured in six different haplogroups A, B, C, D, F and G,, with more than 90% of goats solely from the A haplogroup. Moreover, a very weak phylogeographic structure is observed at the continent scale contrary to other domestic species, such as sheep. Thus, A and C haplogroups have a worldwide distribution although B is mostly present in Asia. Some genetic structure is suspected however at a more restricted geographical scale and some haplogroups, such as G and F, are now restricted to small regions (Middle-East and Sicily respectively). If many studies have been dedicated to the characterization of the mainland diversity few were pursued on islands. However, large Mediterranean islands are of particular importance in the description of the genetic diversity of domestic species since they are considered as biodiversity hot spots, have a high degree of endemism and present a reservoir of cultural practices that have disappeared from the Mainland. In the case of goats, the study of genetic diversity on large Mediterranean islands is highly relevant for several reasons. First, goats are found on most of these islands from the beginning of the Neolithic diffusion and can serve as testimony of this spread. Second, imported domestic goats, are physically isolated from their Mainland relatives. Third, breeding and husbandry practices on the Mediterranean islands are usually different to those on the continent because islands present large but restricted areas and preserve traditional practices mentioned previously. From these observations, we expect that mitochondrial diversity observed on islands can present evidence of historical events or ancient diffusion that would be lost elsewhere, such as the presence of the F haplogroup in Sicily that is unique outside the wild ancestor's area of distribution. Here we characterize goat mitochondrial DNA (mtDNA) diversity through time by studying modern, but above all historical, goat populations from Corsica. We compared them with current continental or island breeds of the Mediterranean Basin to document the microevolution and the influence of insularity and husbandry practices on their genetic diversity. Corsica is an 183 km long and 83 km wide island located in the Northwestern part of the Mediterranean Basin. It formed a unique block with another island, Sardinia, during most of the Pleistocene until the land masses split approximately around 11,000 years ago. The presence of domestic Caprinae is attested in Corsica from the beginning of the Neolithic, ca. 7,600 calBP. The herding system in Corsica, although close to other free-ranging and seasonal transhumance characteristic of other Mediterranean islands, displays a very interesting peculiarity namely the “wandering” of flocks in the mountains for weeks, between the end of the lactation period and the beginning of the births, under very loose surveillance from the herder. These free-ranging practices in Corsica are not recent as Polybius had already mentioned them in his book XII during the II^nd century Before Christ. Nonetheless, being able to ascertain that transhumance and free ranging were common herding practices in Corsica since the Neolithic is a difficult task for zooarchaeologists. Goats are known to be hardy, tough and able to adapt to very difficult habitats compared to other livestock. More than 200,000 goats were still present in Corsica less than 80 years ago but this number has decreased to 30,000 individuals during the last decades. Nevertheless, goats have retained a particular status in Corsica where pastoralism is still strongly established. Besides, a Corsican breed has been recently recognized by the French CNAG (Commission Nationale d'Amélioration Génétique depending on Ministry of Agriculture) in 2003 and by decree of the French Ministry of Agriculture in 2007. It is a dairy breed, with relatively long hairs, either of uniform colour or multicoloured, characterized by its rustic character and ability to adapt to difficult grounds. This was possible because efforts were made to protect and promote local breeds by avoiding mixings with commercial breeds. However, the traditional Corsican breed is now endangered because the traditional husbandry system seems difficult to maintain. The strains on this traditional system are not only economic but also social. Traditional goat husbandry practices implicate daily mobility from the herder, which most of the new generation does not wish to pursue. More recently, the occurrence of Johne's disease (i.e., paratuberculosis) presents another very serious threat and has decimated flocks. This study aims to (i) better characterize the current and past (medieval) mitochondrial genetic diversity of Corsican goats using a Control Region (CR) fragment; (ii) bring information about goat dispersal in the Neolithic by testing the congruence between scenario proposed and data observed on this island; (iii) explain the maintenance of endemic variability in Corsica; (iv) discuss implications for conservation of the Corsican breed. For these purposes, we gathered samples of 28 present-day individuals and 29 bones dated from the Middle Ages (XII^th and XIV^th centuries). # Methods ## The archeological site of Rostino The castrum (strong hold) of Rostino is situated in the North East of Corsica. Occupied between the XII^th and XIV^th centuries AD, this late medieval site has yielded the largest assemblage of Caprinae in Corsica in a good state of preservation, which is rare in the acidic soil of Corsica. Among the domestic species of the XIV^th century deposit, Caprinae represent more than 70% of the identified mammal remains. The economy of the castrum relies on specialized caprine exploitation where the production of sheep and goat complement each other: sheep for meat production and goat for milk and hair production. During the XIV^th century, the caprine exploitation specialized in goats with *Capra hircus* representing more than 70% of the total caprine remains. This large amount of late medieval Corsican *Capra hircus* represents a great opportunity to investigate the consequence of the secular herding practices and the selective choices made by herders to renew their flock given the genetic diversity of goats in large Mediterranean islands. ## Archeological samples We analysed 29 bone fragments from Rostino: 17 were excavated from a XII^th century deposit and 12 from a deposit dated to the XIV^th century. These bones have been identified as *Capra hircus* using morphoscopic criteria on both dental, and appendicular – characters. There is no ambiguity about the origin of bones from domestic animals as the wild ancestor (*Capra aegagrus*) has never been present in Europe. According to the type of bone (mandible, radius, humerus …), the laterality (left, right) and detailed information from the excavation, we were able to clearly identify 25 different individuals from the 29 fragments. The four other fragments very probably came from previously identified individuals. Two samples were subsequently identified as sheep by the molecular analyses. This is not surprising as the inter-specific distinction between sheep and goat on fragmented bones cannot be ascertained with a 100% reliability. Molecular analyses have precisely proven to be useful in this case as has, more recently, the analysis of collagen by mass spectrometry. ## Ancient DNA extracts and PCR amplifications Retrieval of the DNA preserved in bones was performed in ultra clean rooms dedicated to ancient DNA experiments (French National Platform of Paleogenetics PALGENE, CNRS, ENS Lyon). No more than 4 *capra* samples were treated in the same session of DNA extraction along with a bone from another species (cervids, ursids) and a blank control. 100 to 500 milligrams of each bone fragments were reduced in powder and suspended in 5 or 10 ml of EDTA buffer as described in. We extracted the DNA using one, or both, of the two following protocols: a classical phenol-chloroform approach, or direct purification using Qiaquick column (Qiagen kit). 25 out of the 29 samples were independently re-extracted in a second laboratory dedicated to ancient DNA in another city, Grenoble, using the Qiaquick protocol. A 130 bp fragment of the CR (HVI) was amplified with the CapFII and CapRII primers with conditions identical to those described in. At least 2 or 3 independent positive amplifications per sample were obtained, cloned and sequenced following protocols described in. The final sequence of one individual was obtained by making the majority-rule consensus of all consensus of all different clones obtained from each of the independent amplification. More than 80% of the differences observed between clones were G to A or C to T punctual substitutions. This result is consistent with ancient DNA degradation profiles where deamination of cytosines is known as the major factor of artifactual substitutions. ## Present-day Corsican goat sequences We sampled 28 goats in 5 different localities in Corsica (7 from Moltifao, 6 from Tralonca, 11 from Corte, 3 from Altiani, 1 from Quenza;). 14 sequences were already published and 14 were produced for this study. To amplify the CR fragment we used the same primers (CapFII and CapRII) than for medieval sequences with slightly modified conditions: 40 cycles were performed instead of 50–60 for ancient DNA and products were directly sequenced on both DNA strands. ## Sequence analyses All the sequences obtained were aligned (Seaview v4) and the primers removed leading to a fragment of 130 base pairs. Four data sets were constituted: (i) all medieval sequences, (ii) the XII^th century sequences, (iii) the XIV^th century sequences and (iv) the present-day sequences. Firstly, the mitochondrial haplogroup of each Corsican sequence found was determined by performing phylogenetic analyses. The different Corsican haplotypes were analyzed together with 20 haplotypes of known haplogroups. These 20 haplotypes corresponded to the 20 reference sequences that were different for the 130 bp fragment under study, in the dataset selected by Naderi et al. to represent the worldwide variability of the whole HVI-control region (558 bp). Identical sequences, or haplotypes, were identified in the Corsican dataset by using Fabox. After estimating the better model of evolution using jModeltest program and the Akaike Information Criterion (AIC), we performed Bayesian analyses (BA) with MrBayes v3.1.2, (independently confirmed by Maximum Likelihood analyses, not shown). The parameters used for BA were the following: GTR+I+G (nst = 6 and rates = invgamma), 5,000,000 generations sampled every 1000^th generation, 4 chains, a burn-in period of 500 trees (i.e. 10% of generations) visually confirmed using Tracer v1.4.1 (developed by Rambaut A and Drummond A; available from <http://beast.bio.ed.ac.uk/Tracer>), allcompat option. Two independent runs were performed with an average standard deviation of split frequencies at completion of 0.006598. The average values obtained for alpha and proportion of invariable sites parameters were 0.287 and 0.323 respectively. Secondly, we assessed the relationships between only medieval or all Corsican sequences, by performing median-joining networks, using the Network software (available at fluxus-engineering.com) with default parameters. Network Publisher was used to manipulate the networks. To compare the Corsican mitochondrial genetic diversity with the Mediterranean or worldwide diversity, we defined supplementary datasets, one by haplogroup observed, gathering all the sequences published and covering the 130 bp fragment. These datasets combined with the Corsican sequences were used to draw median-joining networks (references are given in the legends of). Thirdly, to assess the past demographic history of the Corsican goats we performed a Bayesian Skyline Plot (BSP) using BEAST v.1.5.4 software. All the Corsican sequences were used and average tip dates were given for all medieval sequences: 850 years BP for the XII^th century and 650 years BP for those of the XIV^th century. BEAUti v1.5.3. was used to build the xml file by using the following parameters: HKY+I+G4 (best model for this dataset assessed by jModeltest and AIC criteria); uncorrelated lognormal relaxed clock; 5 groups and 100 millions of iterations with parameters saved every 10 000 iterations; Burn-in: the first 10% were discarded. The results of 4 independent runs were analysed and the Bayesian Skyline Plot reconstructed with Tracer v.1.4.1. Finally, to describe the genetic diversity observed for the Corsican goats, we computed different classical parameters using either DnaSP v5 or Arlequin v3.5.1.2 including haplotype diversity and frequencies, sequence diversity, pairwise comparisons, and population comparisons (*F_ST*, Fu tests). We compared the different Corsican datasets with each other but also with other datasets corresponding to mainland or island populations (see for accession numbers and geographical origin). We focused in particular on Sardinia's island (75 sequences, accession numbers FJ571522 to FJ571596, ; Corsica's closest island) and Portugal (the biggest dataset generated for mainland goats, 288 sequences, accession numbers AY961629 to AY961916). We also considered 4 different datasets (see Supporting Information for details and references) corresponding to the southern or the northern area of the Mediterranean Sea, the Mediterranean islands (Sicily and Sardinia), and finally 8 Neolithic goats of Baume d'Oullens. To reduce possible biases due to large differences in the size of the datasets (low number of Corsican sequences), we randomly sampled 49 sequences of the non-Corsican sequences and repeated this operation at least three times. The genetic parameters estimated by DnaSP v5 or Arlequin v3.5.1.2 were then computed on these resampled datasets of equal size. # Results ## Medieval goat samples We analyzed 29 bones and obtained reproducible and congruent sequences of *Capra hircus* for 25 of them. We are confident that these results are authentic as we obtained the same results in both laboratories. We also took the ancient DNA precautions recommended by the community as we are used to do. When different bones were supposed to be from a single individual (Ro-5, Ro-10 and Ro-22), systematically we obtained the same sequence confirming the first assessments. Finally, we determined 21 sequences coming from different individuals: 10 dated to the XII^th century and 11 dated to the first half of the XIV^th century. All medieval sequences were from the A haplogroup as shown by the phylogenetic analysis. The six haplogroups appeared monophyletic and were supported by posterior probabilities (pp) higher than 0.9 for 3 of them B, C, G (F is not concerned as only one sequence is used). The A haplogroup that had the highest number of sequences and that was the more diverse received the lowest support (pp\<0.5). Nevertheless, the clustering of the newly determined sequences inside the A haplogroup raises no doubt as confirmed by subsequent network analyses (see also mismatch distributions). Substantial diversity is observed among medieval haplotypes as seen on the network. Among the 21 sequences, we detected 14 unique haplotypes with haplotype Ha 04 being the most frequent (5 individuals). Six out of the 14 haplotypes have never been described before. On average, the mean number of pairwise differences observed between sequences reaches 4.65±2.37. The diversity appeared not significantly different between the XII^th century with 9 haplotypes for 10 sequences (13 polymorphic sites) and the XIV^th century with 8 haplotypes for 11 sequences (15 polymorphic sites). According to the network performed on all medieval sequences (14 haplotypes, 19 polymorphic sites), only 3 haplotypes were shared between both periods. Two of them had a central position in the network (Ha 04 and 09;). ## Present-day goat samples 28 individuals from 5 different localities were studied ( for details). Two different mitochondrial haplogroups were observed (A and C;) with a higher proportion of A sequences (92.8% i.e. 26 out of 28 sequences). Considering all A and C sequences, the mean number of pairwise differences reaches 7.01±3.39 (29 polymorphic sites;). However, when only sequences of the A haplogroup were considered, this value drops to 4.90±2.47 (19 polymorphic sites) which is close to the one observed for A medieval goats. The two sequences from the C haplogroup came from the same locality, Tralonca, and shared the same haplotype Ha 26 that has not been described elsewhere. 15 haplotypes were obtained for the 26 sequences of the A haplogroup among which, 10 were only observed in Corsica. The most frequent haplotypes were, like for the late medieval goats, Ha 04 and Ha 09, with 4 individuals from 3 different localities in both cases. ## Comparison of Corsican goat diversity through time No significant difference was observed between the goats of the XII^th and the XIV^th centuries (non significant *F_ST-*value;). The difference became significant when medieval and present goats were compared (0.036, p-value 0.027). This was due to the presence of the C haplogroup in the present-day sequences since the test was no more significant when only A sequences were taken into account (0.027, p-value 0.099). Finally, there were no significant differences between the medieval and present-day sequences, with or without considering the C sequences. Medieval and present-day goats shared 4 haplotypes (Ha 04, 06, 07 and 09) but only one was specific to Corsica (Ha 06). The 3 remaining haplotypes were generally frequent in other populations. No significant changes were observed in the demographic history of the Corsican population using the Bayesian Skyline Plot. Neither a sign of expansion nor of a crash were observed from the medieval times to date, as rather a constant population size pattern was obtained. This is congruent with the Tajima's D values obtained but not with all the Fu's Fs values computed. However, this latter parameter can reflect other factors than population growth (selection, bottleneck, …). ## Comparison of Corsican goat diversity with other geographical places We compared the diversity observed in Corsica with other Mediterranean island populations and mainland breeds. The 75 sequences of Sardinia were clustered in 46 haplotypes that were all from the A haplogroup. Similarly, the 288 Portuguese sequences were represented by 104 haplotypes, from which only one was from the C haplogroup and all the others from the A haplogroup. By expanding the comparison to larger or other areas, we observed that the C haplogroup is described only in Europe (in present-day Northern Mediterranean area and already at the Neolithic time, and not shown). Both the medieval and present-day Corsican groups appeared significantly different from all other groups whatever the resampled datasets taken into account. Similarly, all the non-Corsican groups of goats also appeared significantly different from each other (data not shown). However, four different haplotypes (Ha 05, 08, 09 and 10) were shared between Corsica and Sardinia islands and six (Ha 07, 08, 09, 10, 12 and 21) between Corsican and Portuguese goats. The median-joining network performed on 39 worldwide sequences of the C haplogroup revealed a classical expansion structure with European sequences on one side and Asian sequences on the other (with the single exception of one Swiss haplotype;). As expected, the sequences obtained from Neolithic goats of the archeological site of Baume d'Oullens in France, among the first goats to have been diffused in Europe, appeared in the central part of the European cluster. The Corsican haplotype showed the highest number of substitutions with this central node (3 substitutions). The same analysis performed on the sequences from the A haplogroup restricted to the Mediterranean area, showed a more complex history with no clear emerging pattern. # Discussion ## DNA preservation in medieval samples From the 29 archeological bones dated back to the XII^th and XIV^th centuries, we obtained 21 sequences from different ancient goat individuals: 10 from the XII^th century deposit and 11 from the XIV^th century deposit, which represents a surprisingly high DNA preservation for remains in Corsica. Such a good preservation of the DNA, which is here correlated to the good preservation of the bones themselves like in most late Medieval Corsican sites, is probably due to the recent age of the site. ## Characterization through time of mitochondrial genetic diversity in Corsican goats The comparison of Middle Ages and present mitochondrial diversity was carried out using 28 present-day goats from five different localities. This comparison may be slightly biased by differences existing in the time span and geographic distribution for either ancient or modern samples. Indeed, the sampled geographic area is larger for the modern goats whereas the time span is longer for the ancient samples. Our results tend to prove that the mitochondrial diversity of Corsican goats has remained relatively constant since the Middle Ages. Moreover, we detect no significant demographic changes (*F_ST*, BSP and) or decrease of genetic diversity. The only difference between both periods is the occurrence of two C haplotypes in the present-day samples, all other goats belonging to the A haplogroup. However, given a dataset of 21 medieval individuals and assuming a constant frequency of the C haplogroup in Corsica (2/28 = 0.0714), we have a 21% of chance of having missed this haplogroup in the Middle Ages sampling. ## The occurrence of the C haplogroup in Corsican goats in the context of the Neolithic diffusion The presence of the A and C haplogroups in Corsica is in agreement with the goat mitochondrial variability observed in European countries. Most of European goats are from the A haplogroup, with C haplotypes found at a rare frequency in Switzerland, Portugal, Spain and Slovenia. This is also consistent with previous paleogenetic studies that already detected both haplogroups in goats from Southern France in the early Neolithic period. The median-joining network performed on all the C haplotypes found worldwide gave results congruent with the diffusion of goats in the Neolithic. Indeed we observed: i) a clear separation of the European haplotypes from the Asian ones with only one exception, a shared haplotype between China and Switzerland; ii) a star-like pattern for European haplotypes suggesting a population expansion with a central position for the Neolithic haplotypes ; iii) a divergent haplotype for Corsica compatible with a subsequent isolation. Without a doubt, the origin of the C haplogroup in Europe can be traced back to the Neolithic spread where its frequency was probably higher than the one observed now. Today, the C haplotypes are relatively scarce in goats and appear more like reminiscent testimonies of this first diffusion. Interestingly, not one has been detected yet in the Southern Mediterranean area or in other Mediterranean islands than Corsica (see Supporting Information). According to the few data we have, it is difficult to conclude on the ancient origin of the Corsican C haplotype or a more recent origin linked to subsequent exchanges with the Northern Mediterranean mainland. However, its position in the network is compatible with the first explanation. Further ancient DNA studies, for a larger area and older period, would be very interesting to highlight this question. ## Variability in Corsica vs other Mediterranean areas The relative stability through time of goat diversity in Corsica could be explained by regular importation of goats from other continental areas or islands, e.g. for commercial trade, as many contacts by sea have been reported during the last centuries in the Mediterranean area. However, a striking point is that the diversity observed in Corsica is substantial for both periods (Middle Ages and present-day, and respectively) and differs from that of other places (data not shown); only about half of the Corsican haplotypes (14/26) are shared with goats from other geographical regions despite our study focused on a short CR fragment (130 bp). This result is not unexpected however. Previous studies - usually targeting the 480 bp fragment first described in and covering our shorter fragment - have already shown that goats were more polymorphic than other livestock (cattle, sheep, pig) on the CR. Moreover, the specific analysis of the A haplogroup (more than 90% of the modern goats) confirmed high haplotype diversities for 20 populations/countries with values close to one. Pereira et al. showed a strong correspondence between mitochondrial genetic and geographic distances suggesting that after the initial expansion, differentiation among regions has been established and maintained. A similar conclusion was obtained with large-scale nuclear SNP analyses obtained for 16 breeds of goats and from microsatellites analyses including Corsican and other European goat populations. This seems in agreement with what we observed in Corsica when compared with other Mediterranean populations (*F_ST* comparisons, data not shown). ## Implication of the traditional Corsican husbandry in the maintenance of the variability Except for some haplotypes that are common in many countries and probably constitute traces of the initial diffusion, 46% of the Corsican haplotypes found have not been previously described elsewhere. The preservation through time of this endemic genetic diversity and its constant level since medieval time could suggest that relatively large effective population sizes have been maintained in domestic goats through exchanges of animals. But ethnographic insights into the herding practices carried out in the Niolu provide another possible or complementary explanation for the preserved genetic diversity of the Corsican goat breed. Typically in the past for the Mediterranean area, goats were usually moved according to the seasonal changes (transhumance) to gain access to more reliable food. Because fodder resources fluctuate in the wild according to different factors (e.g. annual climatic conditions), the system developed in Corsica has been extensive with goats left free-ranging most of the times although under the careful control of the herders. This particular system led to the characteristics observed in the Corsican goats, : i) the high diversity of coat colours is encouraged as individuals can be more easily identified by sight; ii) only the strongest and toughest goats can generally survive in this relatively hostile environment, explaining why the introduction of goats from industrial breeds usually failed; iii) large herds are usually managed to maximize the herd's productivity overall instead of individual productivity. Indeed, selection is not performed to optimize for instance, either milk productivity or the fecundity, but instead to obtain a constant productivity of the herd by year whatever the weather or difficulties encountered. All these points lead to a strategy where the phenotypic diversity, and the underlying genetic diversity, is maximized in order to obtain herds that are highly adaptable, rustic and robust. In order to do that, herders exert strong selection pressure while forming their herds to gain in productivity but also to favour behavioural traits. Indeed, along with the search for constant productivity, one of the main objectives of herders is to design flocks that will maintain “families” from the same maternal lineage. A young descendant female is usually chosen according to its mother's and grandmother's family, taking into account its desirable productive traits as well as its abilities to endure the tough conditions. Animals coming from the same family/pool will live together and move together more easily and naturally. Such “familial” behaviour is extremely beneficial to the cohesion of herds under free-ranging exploitation and especially during the movements on the pasturelands (transhumance). On the contrary, introduction of new animals from other herds can induce significant disturbance in the herd movements by breaking the cohesion of the group. This practice of herders of course has the effect of reducing the genetic diversity. Hence, to insure the persistence of the genetic diversity and to “change the blood”, “cambià u sangui” in the language of traditional Corsican herders, selected males of these different maternal lineages or “families” will be exchanged between flocks to avoid inbreeding and, without control, renew the pool of the mating males within a seasonal time span. As mitochondria are inherited maternally, this system will naturally lead to maintaining high mitochondrial genetic diversity between herds or “families”. So, the diversity we observed would be not promoted by large herds with many exchanges of females between them but on the contrary by traditional practices. These ones rely on very few introductions of females coming from other herds to keep the kinship within the herd as tight as possible to reinforce cohesive behaviour during the ranging but also on a mixing of the genetic diversity through the exchanges of males only. Such entangled practices in the Corsican husbandry system can explain why we do not observe change of the mitochondrial diversity in Corsican goats since the late Medieval period, and probably earlier if we had been able to investigate the genetic diversity of goats from earlier periods. It's thus highly probable that sustainable husbandry practices in today's Corsica that are so well adapted to their environment are, at least partly, the result of practices over millenaries contributing to the maintenance of a relatively high level of genetic diversity. Finally, analyses of additional samples and genetic markers coupled with population simulations using varying population genetic parameters should help to test between both hypotheses, regular importation of goats or millenaries husbandry practices. ## Towards a protection of the Corsican goat breed In this study, we observed that the mitochondrial diversity of goats in Corsica Island has been maintained since the Middle Ages to date. In a time where rustic breeds are endangered and industrial breeds tend to reduce the genetic resources, Corsican goats, by the model of husbandry used and the high diversity observed constitute an interesting pool to preserve for the future management of domestic genetic resources. All the actors concerned by the Corsican goat (regional association in charge of its management, public authorities, researchers and extension services) should so pay attention to make their preservation successful all the more because paratuberculosis has started to decimate flocks. ## Ethic statement The medieval bones of goats were excavated from the Rostino archeological site. Daniel Istria, in charge of the excavation, authorized their analyses. Tissue samples were collected in Corsica in the framework of the ECONOGEN project (European Union contract QLK5-CT-2001-02461), following the European ethical rules implemented in all European projects. # Supporting Information We thank Hamid Reza Rezaei and Saeid Naderi for having given us details about samples of Corsican goats they have published. We also thank members of the paleogenetics lab and the Palgene Platform (CNRS, ENS Lyon) for help during the experiments. We are also grateful to Javier Oliver and Benjamin Gillet for critical reading of the manuscript. Finally, we express gratitude to Joanne Burden who kindly proofread the English. [^1]: Conceived and designed the experiments: PT HF TC SH CH. Performed the experiments: MD HF. Analyzed the data: SH TC. Contributed reagents/materials/analysis tools: DI J-DV PT CH. Wrote the paper: SH FP TC. Provided expertise for Corsican husbandry, ancient DNA and archeozoological data: FC CH J-DV. [^2]: The authors have declared that no competing interests exist.

# Introduction HCV is a worldwide health concern with severe consequences. Globally, HCV is estimated to affect around 3% of the world's population, counting to approximately 170 million people. While it may remain asymptomatic for years, it can lead to serious liver diseases, which include cirrhosis or hepatocellular carcinoma. As with all viruses, HCV is prone to genetic mutations that lead to multiple reproducible variants. Seven genotypes of HCV with various subtypes have been discovered around the world. The genotype HCV-1 is common in America, Europe, and Japan. The subtype HCV-1a is predominant in North American and Northern Europe whereas HCV-1b is the most common subtype in Japan and Eastern Europe. Additional countries where HCV infection rates are very high are Egypt (15% of population, 18 million people) and Pakistan (4.8%, 8.5 million). Approximately 90% of those infected in Egypt carry the genotype 4, with subtype 4a (HCV-4a) predominating. In Pakistan, around 67% of the HCV infections are due to genotype 3, with subtype 3a (HCV-3a) being the most common. Genotype 1 has been the focus of intensive investigations over decades and a variety of effective antiviral drugs and/or inhibitors have been developed. Conversely, variants that are predominant in developing countries have not received much attention. As a result of the crucial role of the nonstructural protein 3 (NS3) in the replication cycle of HCV, the protease domain of NS3 has been an attractive target for direct-acting antiviral agents. The NS3 protease cleaves four downstream sites in the HCV polyprotein and is characterized as a serine protease with a chymotrypsin-like fold, which is activated by the NS4A cofactor. Similar to chymotrypsin, the catalytic triad of the HCV NS3 protease is made of three essential residues, histidine-57, aspartic acid-81, and serine-139. These three residues are collectively known as the catalytic triad and will perform general acid-base catalysis on target peptides. In summary, a charge relay system is formed in which the carboxylic group of D81 forms a hydrogen bond with N*δ*1 of H57. This event increases the p*K*a of the histidine side chain from 7 to about 12. Consequently, H57 deprotonates the hydroxyl group of the S139 side chain and a proton shuttles to Nε2 of H57. The Oγ of S139 then nucleophilically attacks the carbonyl carbon of a substrate's scissile bond resulting in the formation of an oxyanion-containing tetrahedral intermediate,. At this point, the protonated H57 acts as a general acid assisting in the collapse of the tetrahedral intermediate and the cleavage of the substrate. Zinc, which is a part of NS3 protease, plays an important role in the structural stability of the protease by enthalpically disfavoring protein denaturation. Additionally, a bound peptide cofactor (NS4A) increases the protease activity by nearly 1000-fold. It is noteworthy to mention that HCV-3a and HCV-4a NS3 proteases exhibit a several fold decrease in catalytic efficiency relative to that of HCV-1b. This implies a possible correlation between the catalytic efficiency of the NS3 protease and its responsiveness to inhibition, at least by the linear inhibitor Telaprevir. The differential susceptibilities of different drugs to protease variants in relation to their enzymatic activities have been investigated. The commercially available linear NS3 protease inhibitors Telaprevir and Boceprevir are shown to be effective against HCV-1. Furthermore, HCV-3a and HCV-4a NS3 proteases show several fold resistivity to inhibition to Telaprevir compared to HCV-1b, with HCV-4a being the most resistant. Additionally, the newly developed macrocyclic NS3 protease inhibitor Danoprevir has been shown to be effective against genotypes 1b and 4a, but not as effective against genotype 3a. Another macrocylic drug, Simeprevir, has been shown to be somewhat effective in all HCV genotypes, but is most effective against HCV 1a and b. Very recently, the drug Sofosbuvir, which inhibits a different HCV target protein (NS5B polymerase), has been approved by the FDA to treat genotypes 1 through 4. Although Simeprevir and Sofosbuvir provide a way to treat multiple genotypes of HCV, the cost of the typical 3 month treatment (\$66,000 and \$84,000 respectively), is too expensive for use in developing countries. In addition, due to the high mutability of the virus, a sub-type can emerge during the course of treatment, resisting the administered antiviral drug and complicating the treatment regimen. In our earlier work investigating the drug resistivity in HCV-4a, we developed a computational methodology to analyze, in 4D, the active site geometry in HCV NS3 protease. The results showed that both proteases share very similar rigid and overall dynamics features. Conversely, both exhibit significantly different local dynamics and distance distribution profiles, in peak values and broadness, at the catalytic triad. Here we have improved the methodology further and extended our investigation to include the drug resistant genotype HCV-3a. Our data, consistent with our previous report, suggest that genotype-dependent structural dynamics could play a significant role in the stability of the catalytic triad, and possibly, in drug response among HCV genotypes. The results show that the divergent dynamics behavior of the catalytic triad in the NS3 protease of genotype 3a, represent an intermediate state between that of genotype 1b (most stable) and genotype 4a (most unstable). This correlates well with their reported catalytic activities and drug susceptibilities to the linear inhibitor Telaprevir. Therefore, our comparative investigation reported here, illuminates a possible variant- dependent pathway from an active/drug responsive protease to a weakly active/drug resistive one, an understanding with implications in catalysis and drug design. # Results and Discussion HCV-3a and HCV-4a NS3 proteases share 80% and 83% sequence identities respectively, with HCV-1b protease. The HCV-3a and HCV-4a protease structure models superpose very well on the threading template structure (HCV-1b, PDB- ID:1dy8), along with nearly identical structural features. When the three structures superpose, the RMSD in back-bone positions is about 0.3 Å and none of the 174-threaded amino acids fall within the disallowed Ramachandran area and no steric clashes or stereochemical outliers were detected. The three catalytic residues H57, D81, and S139 are located in a crevice between the two protease β-barrels as shown in. The active site is nonpolar and shallow. The central region of NS4A is buried almost completely inside the NS3 protease and serves as a cofactor for proper folding of the protease. The rigid structures indicate that access to the active site is nearly identical in the structural models (HCV-3a and HCV-4a) and template (HCV-1b). Molecular dynamics simulations predict that both HCV-4a and HCV-1b proteases share more or less similar average RMSD in the Cα positions, around ∼1.2 Å at equilibrium. However, the average RMSD in the Cα positions for HCV-3a protease is higher (∼1.6 Å). This implies that the main chain of HCV-3a protease, as a whole, experiences more fluctuations compared to that of the two other genotypes (1b and 4a). Interestingly, as will be shown later, this malleability in HCV-3a protease, which is even greater than the barely functioning genotype 4a, does not propagate fully to the catalytic triad region. In this regard, HCV-3a represents an interesting case where the conformational stability of the catalytic region of the enzyme is somehow “shielded” against an overall positional instability of the protein. Locally, molecular dynamics simulations revealed a strain-dependent, gradually divergent dynamics behavior within the catalytic triad region, with HCV-1b being the most stable, the HCV-4a the most divergent and HCV-3a representing an intermediate state. These dynamic differences seem to correlate well with the differences in catalytic activities and drug susceptibilities to Telaprevir seen in the three genotypes. This result strongly suggests that the local dynamics within the triad region in the NS3 protease could be used as a direct predictive measure for HCV pan-genotype drug susceptibilities. ## 4D simulation of the interactions between the catalytic residues Following the same methodology we reported on previously, we have investigated the local positional dynamics of the catalytic triad residues during the course of simulation and used the distance distribution profiles of catalytically relevant distances as indicators of the 4D variations. The alpha carbons (Cα) of the catalytic residues D81 and S139 exhibit somewhat similar dynamics throughout the simulations for the three genotypes (1b, 3a and 4a). However, the Cα of the catalytic residue H57 of the HCV-4a and HCV-3a shows a slight to moderate increase in RMSD of ∼0.2 Å and 0.7 Å respectively, relative to that of HCV-1b. As entire residues, H57 and D81 in both HCV-4a and HCV-3a models, demonstrate dynamics behavior divergent from that of the template HCV-1b. The RMSD of H57 in the HCV-3a model varies from that predicted in the HCV-1b template by up to 1.5 Å at certain points during the simulation. A very similar trend is observed for HCV-4a. A slight increase in the Cα dynamics of H57 in HCV-4a is associated with a significant increase in the dynamics of the entire residue. Therefore, one should expect a much higher RMSD for H57 (as an entire residue) in HCV-3a, if only the corresponding Cα instability is considered. The observation that H57 (as an entire residue) shows similar RMSD pattern in both HCV-3a and HCV-4a (in spite of the relative stability of the Cα in HCV-4a) points to additional stabilizing interactions acting on H57 in HCV-3a. This is evident in HCV-3a from the intermediate RMSD dynamics of D81 residue, which is known to fulfill a stabilizing role for H57. The RMSD of D81 in the HCV-3a and HCV-4a models gradually diverge from that of HCV-1b template by nearly 0.5 Å and 1 Å respectively. S139 exhibits similar dynamics behavior in the three genotypes, which is expected for a relatively small side chain whose movement is sterically encumbered by the nearby residues. The distance distribution profiles between Nε2 of H57 and Oγ of S139, as well as between Nδ1 of H57 and Oδ2 of D81, of the template (HCV-1b) and models (HCV-3a and HCV-4a) vary widely in both peak value and breadth. In the template structure (HCV-1b), the distance between Oδ2 of D81 and Nδ1 of H57 exhibits a sharp distribution with a peak value around 3 Å. In the model HCV-4a, the corresponding distance distribution is bimodal, much broader, and distributed around 4.5 and 7.5 Å. It is noteworthy to mention that in our previous report, the 4.5 Å peak was not conspicuous. With the improved methodology in the current analysis , this peak becomes clearer. In the model HCV-3a a broader bimodality (∼70% of the distribution) still dominates, with the recovery of a sharp peak around 3 Å, overlapping with that of HCV-1b peak. It is interesting that the distribution for the HCV-3a model is almost a mixture of that of HCV-1b and HCV-4a. Furthermore, the distance distribution between Oγ of S139 and Nε2 of H57 in the template HCV-1b shows a peak value at around 4 Å, while the corresponding distribution in the model HCV-4a is broader and bimodal. Again, the distribution in HCV-3a is a mixture of HCV-1b and HCV-4a shifted by about 0.5 Å. The collective dynamic behavior of the three catalytic residues as vertices of a triangle is analyzed. This is to gain an insight into the respective relative positions of the three residues simultaneously. The atoms chosen were Oδ2 of D81, Nδ1 of H57 and Oγ of S139. The choice of Nδ1 over Nε2 of H57 was to include conformations in which the H57 has rotated in such a way that Oδ2 of D81 and Nδ1 of H57 can no longer hydrogen bond. While more than three atoms are involved in the hydrogen bonding network, the vertices of the triangle collectively probe the hydrogen bonding distances. Thus, any other choice of vertices would only shift the values, not the distribution. The distribution profiles of the area of the triangle during the course of simulation indicate a uni-modal sharp peak in HCV-1b, a broad bimodal distribution in HCV-4a and a hybrid behavior in HCV-3a. The area of the triangle somehow represents a “catalytic plane” whose distribution profile could be predictive of distortions of optimal catalytic geometries. In this sense, HCV-1b is predicted to be the most stable (most active) and HCV-4a is the least stable (least active) while HCV-3a represents an intermediate state. This is consistent with the observation that the catalytic activity of HCV-4a NS3 protease is several orders of magnitude less than that of HCV-1b, while the catalytic activity of HCV-3a is still less than that of HCV-1b, but not as hampered as that of HCV-4a. The same trend was observed in the genotype-dependent drug susceptibility seen in HCV against the linear inhibitor Telaprevir; HCV-1b is the most responsive (least resistive), HCV-4a the least responsive (most resistive) and HCV-3a representing an intermediate state. Telaprevir is a linear inhibitor that fits within the natural substrate' binding site “envelope”. Therefore, it is expected that variations (whether rigid or dynamic) altering inhibitor binding will simultaneously interfere with the binding of substrate; thus, impact the enzymatic activity. However, other inhibitors protrude from the substrate binding envelope, interacting with sites remote from the substrate binding site. Variations occurring at these sites incur drug resistivity, with little effect on the catalytic activity,. In general, drug susceptibilities to protease variants depend on both the 3D location of the variation (mutation) sites, and the stereochemical structure and conformation of the inhibitor. However, dissecting the molecular and structural basis of the differential susceptibilities of different drugs to protease variants in relation to their enzymatic activities is rather extensive, beyond the scope of this work. Together, these data indicate that in the model HCV-3a, H57 spends less time positioned within a probable hydrogen bonding distance to both S139 and D81 compared to HCV-1b, but more time compared to HCV-4a. Thus, H57 is less likely to act as an efficient general acid–base in case of HCV-3a. The effect is more severe in case of HCV-4a where H57 is not positioned within hydrogen bonding distance with D81 and barely with S139. As mentioned before, it seems that the mode of Telaprevir binding somehow allows the protease drug responsiveness to follow the enzymatic activity trend. It is important to note that the predicted divergent dynamics behavior in HCV-3a and HCV-4a is completely hidden by the apparent similarity seen in the catalytic site in the rigid structures. Further, the global instability of the protein's backbone fails to accurately account for the stability of the triad as does the stability of the Cα of the H57. These results highlight the importance of utilizing molecular dynamics as a method of future investigations into protease activity. Furthermore, the correlation between the divergent conformational stability of the catalytic triad region with both the catalytic activity and drug resistivity seen in HCV-proteases cross genotypes, opens an interesting avenue for inquiry with potential predictive applications. # Methods ## DNA sequencing For HCV-4a (strain ED43), we used the amino acids sequence, as described previously. To obtain nucleotide/amino acid sequence of NS3 protease from HCV 3a, RNA was extracted from blood sample of a Pakistani patient and cDNA was synthesized using first strand cDNA synthesis kit (Fermentas, cat no. K1612). Freshly synthesized cDNA was used to amplify the full length NS3 gene using forward primer 5' TATAGGATCCATGCACCATCACCATCATCACGCCCCGATCACAGCATAC3' and reverse primer 5' GAGCAAGCTTTTAGGTGGTTACTTCCAGATCAG 3' containing the *Bam* H1 and *Hind* III sites, respectively, through a gradient PCR reaction. The amplified product was cloned in pET 11a vector and sequenced. The sequence was submitted to NCBI GenBank under the accession number JQ676838. The Research Ethics Review Committee of National Institute for Biotechnology and Genetic Engineering (NIBGE), Faisalabad, Pakistan has approved the protocols and procedures used to collect the blood samples from HCV patients. A written informed consent (as outlined in PLOS consent form) to participate in this study and publish the case details was taken from every donor. ## 3D structure prediction and validation The 3D structure of HCV-3a and HCV-4a NS3 proteases were predicted by threading its amino acid sequence through the X-ray crystal structure of HCV-1b NS3 protease (1dy8) via the threading program LOOPP. LOOPP is a fold recognition program that generates atomic coordinates of a sample molecule based on an alignment with a homologous template structure. By integrating the results from direct sequence alignment, sequence profile, threading, secondary structure, and exposed surface area prediction, the LOOPP builds main-chain and all-atom models. Nearly identical models were also obtained via homology modeling using the SWISS-MODEL Workspace. The RMSD values between models obtained using LOOP and SWISS-MODEL were about 0.2 Å and 0.16 Å for HCV-3a and 4a respectively. To build the NS4A cofactor, we superposed the model structures onto the template structure (1dy8, RMSD 0.3 Å) and built the sequence of the NS4A cofactor for the model based on the corresponding coordinates found in the template crystal structure. Similarly, a single zinc ion was manually docked at the cysteine triad C97, C99, and C145 into the model guided by the corresponding position in another structure of the template protein (1dxp) in which zinc is present. With the cofactor and zinc bound, the model was energy minimized using the CCP4 program suite – and the GROMOS96 program, an implementation of the Swiss-pdb viewer. The final models were validated using the NIH MBI Laboratory for Structural Genomics and Proteomics Structural Analysis and Verification Server. This server utilizes five programs (Procheck, What_Check, ERRAT, Verify_3D, and Prove) to analyze the stereochemical parameters and the quality of the model. Additionally, CCP4 programs suite 6.0 was used for the calculation of a Ramachandran plot, structure superposition, and RMSD value calculation in addition to the evaluation of the stereochemistry. ## Molecular dynamics simulation The molecular dynamics simulation (MD) was performed using NAMD 2.9 under the CHARMM27 force field for proteins. Initially, the 3D structures were solvated using the solvation tool in VMD. The TIP3P model was used for the water molecules. Lengevin dynamics for all nonhydrogen atoms with a damping coefficient of 1 ps⁻¹ was used in maintaining a constant temperature of 310 K throughout the system. A constant pressure of 1 atm was maintained using a Nosé–Hoover Langevin piston with a period of 100 fs and damping timescale of 50 fs. Periodic boundary conditions were used on a 61 Å cubic box with the long-range electrostatics calculated using the particle-mesh Ewald method with a grid point density of 0.92 Å⁻¹. This process ensured that adjacent copies of the protease were never close enough for short-range interaction. A cut-off of 10 Å for van der Waals interactions and a switching distance of 8 Å were found to give convergent results, thus used for production runs. The solvation box was neutralized, using VMD's Autoionize plugin version 1.3, with sodium chloride placed at distances greater than 5 Å from the protease. A time step of 1 fs was used in order to resolve the hydrogen motion of water. The initial structure was first subjected to three rounds of an 800 cycle conjugate gradient energy minimization flanked by 100 ps of MD simulation at 278 K. The system was then heated up in increments of 5 K with 100 ps of MD simulation at each temperature increment until the desired temperature of 310 K was established (the last heating increment was 2 K). This is an overly cautious stochastic heating scheme to ensure that the models explore wider space around a local minimum, given that HCV-3a and 4a proteases are predicted models, not crystal structures like HCV-1b. The system was then simulated for 25 ns. Time frames used for the measurements within the protease were only done for frames where the protease had equilibrated: 15–25 ns. The equilibrium state of the protease was determined by the RMSD of the entire protein's backbone. For HCV-3a, longer runs (40 ns) were performed to ensure equilibration beyond 25 ns. Multiple copies of each protease, which included the cofactor and a zinc ion (nonbonded), were run with different initial conditions to ensure that the results were well converged. All data presented are averaged over six distinct runs in order to ensure a representative sample of the parameter space the protease explores. In order to quantify the relative positions of the three catalytic residues (H57, D81 and S139) simultaneously, three atoms were used as vertices of a triangle. The atoms chosen were Oδ2 of D81, Nδ1 of H57 and Oγ of S139. The area is calculated by where is the vector from Oγ of S139 to Nδ1 of H57 and is the vector from Oγ of S139 to Oδ2 of D81. The distribution of the area of the triangle was monitored during the course of the simulation. We thank Tom Teague for his technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MY EA. Performed the experiments: MY EA SS MR MI MIA MK DH. Analyzed the data: MY EA MK DH. Contributed reagents/materials/analysis tools: MY EA. Contributed to the writing of the manuscript: MY EA SS MR MI MIA MK DH.

# Introduction In 1983 Gouras et al. reported an unusual type of retinal dystrophy, which was associated with characteristic alterations in the rod electroretinogram (ERG). This rare, autosomal recessive condition has been reported in several further studies, and was named “cone dystrophy with supernormal rod responses (CDSRR)”. CDSRR is characterized by an early markedly reduced central visual acuity with central scotoma, photophobia, severe color disturbances, and occasionally nystagmus. In contrast to other cone dystrophies, a disease-typical alteration of the rod system could be observed: while rod sensitivity to weak flashes was reduced, an augmented responsiveness to higher levels of flash stimuli could be detected, and implicit times were considerably prolonged. These characteristics were unique for CDSRR, however, the underlying disease mechanism could not be elucidated at that time. In 2006 Wu et al successfully linked the disorder to chromosome 9p24 and the *KCNV2* gene, which is predominantly expressed in retinal rod and cone photoreceptors. It encodes a member of voltage gated potassium channels (Kv channels), representing a silent subunit (Kv8.2) that is able to assemble with Kv2.1 to form functional heteromeric channels. This results in a shift in the steady-state activation curve of the Kv2.1 channel towards more negative potentials due to a permanent outward K⁺ current, a lower threshold potential for activation, a shortened activation time and slower inactivation kinetics. A mutation in *KCNV2* may thus alter important characteristics of the I_kx current that influences the photoreceptor membrane potential. However, the dysfunction and mechanisms that link *KCNV2* mutations with the clinical picture still remain to be elucidated. Over 50 different mutations in *KCNV2* have been reported so far, mainly small indel mutations or point mutations that constitute protein truncation mutations and amino acid substitutions. Recently, several large deletions within or of the *KCNV2* gene of up to 237 kb in size have been described. Although the genetically detected patients did show altered rod responsiveness, the term “supernormal rod response” was in many cases deceptive, as previously shown. The term “supernormal rod ERG” is a misnomer and most recently, the disorder has been referred to as “KCNV2 retinopathy”. This study employs detailed psychophysical and electrophysiological testing as well as spectral domain optical coherence tomography (OCT) and fundus autofluorescence (FAF) to reveal novel insights into disease-specific functional changes in KCNV2 retinopathy. Additionally, we explore differences of disease specific functional aspects in the phenotype that correlate with the underlying *KCNV2* gene alterations. The genotype of three patients has already been published elsewhere, the remaining three patients' genetic findings are presented here for the first time. # Methods ## Patients Six otherwise healthy patients of German origin (3 female and 3 male; 2 simplex cases and 2 sibling pairs; mean age: 39 years, range 28–60 years) with previously diagnosed stationary retinal disorder and known mutations in the *KCNV2* gene were examined. All examinations were carried out after written informed consent and in accordance with the Declaration of Helsinki. The study was approved by the Ethics Committee of the Medical Faculty of University of Tübingen. ## Molecular Genetics Genomic DNA was extracted from venous EDTA-blood samples according to standard procedures. Genetic testing for point mutations was performed by PCR amplification and subsequent Sanger sequencing of both coding exons and flanking intronic sequences of the *KCNV2* gene, as described previously. Analysis for genomic deletion was investigated by quantitative copy number analyses of the *KCNV2* gene, with realtime PCR employing TaqMan technology or SYBR Green detection assays, as reported earlier. Comparative genome hybridizations (CGH) using a predesigned chromosome 9 specific 385k oligonucleotide array (HG18 CHR9 FT; Roche NimbleGen Inc., Madison, WI) was performed for subject CHRO8.I who had suspected deletions at the *KCNV2* locus (Roche NimbleGen). The deletion junctions in patients CHRO8.I and RCD307 were determined by long distance PCR amplifications and subsequent Sanger sequencing to define the precise breakpoints. Independent segregation of the mutations within the families were conducted by Sanger sequencing of PCR amplified genomic DNA for point mutations, and by qPCR in the two families segregating the *KCNV2* gene deletion. ## Clinical Examination A complete ophthalmological examination was performed including psychophysical tests (Snellen visual acuity, Lanthony Panel D-15 and Nagel anomaloscope color vision tests, visual field and dark adaptation) and an extended electrophysiological protocol (Ganzfeld and multifocal ERG). ## Psychophysical testing Kinetic 90° and static 30° visual field tests were carried out with an Octopus 900 perimeter (Haag-Streit International, Germany). Dark adaptation curves were measured with a dark adaptometer (Roland Consult GmbH, Brandenburg, Germany) after pupil dilation with Tropicamid. After 3 minutes of bleaching with bright white light (intensity 5.5 log photopic trolands), a staircase procedure was used to estimate detection thresholds over a period of 40 minutes. Thresholds were alternately measured for red (635 nm) and green (565 nm) circular targets, presented 20° nasal of the fovea. Cone and rod thresholds were then determined by a model fit using the equation:with *t_k* describing the time to the rod-cone break and *I*1, *R*1 the exponential decay of the cone, *I*2, *R*2 of the rod thresholds. Cone and rod parameters are derived from red and green target functions respectively. For group analysis and for comparison to normals the model was fitted to the raw data of each group of subjects. ## Electrophysiological testing Ganzfeld and multifocal electroretinograms ERGs were recorded according to the standards of the International Society for Clinical Electrophysiology of Vision (ISCEV). All tests were performed using DTL electrodes with an Espion E² (Diagnosys LLC) recording device coupled with a ColorDome (Diagnosys LLC) as light source. After 30 minutes of dark adaptation a series of responses to increasing flash intensities (4 ms–0.0001 cd.s/m² to 10 cd.s/m² in 0.5 log unit steps) were recorded and the stimulus- response (S-R) functions modelled using the equation:with the saturated b-wave amplitude *V_max*, the flash intensity *K* required for semi- saturation as a measure of retinal sensitivity and the slope related exponent *n*. Rod response characteristics were estimated from the a-wave by the Hood and Birch (1994) formulation of the Lamb and Pugh model of the biochemical processes involved in the activation of rod phototransduction. The a-wave ensemble was fitted with a computational model describing the response (*P_III*) as a function of time (*t*) and intensity (*I*):where *Rm_pIII* is the maximum amplitude, *S* is a sensitivity variable and *t_d* is a brief delay before the response onset. *P_III* was then subtracted from the original ERG waveform to give the *P_II* response, which is thought to represent mainly the ON-bipolar cell response, but also the postreceptoral activity in other second- and third- order retinal neurons. The relation between flash intensity and the delay between stimulus onset and reaching a given arbitrary criterion voltage of the *P_II* component was then plotted on a log-log coordinate and the slope of this function was calculated. The voltage criterion chosen in this study was 50 µV. Finally, dark-adapted responses to a series of blue flicker (LED 470 nm) with an intensity of 0.03 cd.s/m² and frequencies between 5 and 30 Hz were recorded to isolate temporal retinal characteristics of the rod system. The light-adapted protocol (10 min of light adaptation to a background luminance of 30 cd/m²) included a single flash cone stimulus and a 30 Hz flicker (both: 4 ms, 3.0 cd.s/m²). In addition, responses to a series of flicker white stimuli of 3.0 cd.s/m² with increasing frequency from 5 to 45 Hz were included to investigate possible alterations in the temporal resolution of the cone retinal pathway. Multifocal ERG (mfERG) was performed with a VERIS System (Version 5.1) using a Grass amplifier (model 12, Quincy, USA). The stimulus, consisting of 61 scaled hexagonal elements covering a central visual field of 60×55°, was presented on a 19″ monitor at a frame rate of 75 Hz at a distance of 32 cm from the subject's eyes. The same DTL electrodes as those for the Ganzfeld recordings were used. Responses were amplified (200 000×), bandpass-filtered (10–100 Hz), and analysed according to ring averages. ## Morphological testing Color and infrared fundus photography, autofluorescence (FAF) and spectral domain OCT recordings (Heidelberg Engineering GmbH, Germany) were performed. # Results ## Molecular Genetic Findings Mutation screening and segregation analysis led to the identification of mutations in the *KCNV2* gene in our patients. The genotypes of the six patients are listed in. and the mutation localization is shown in. We observed compound heterozygous mutations in the two sib pairs (CHRO8.I and CHRO8.II, and BD27.I and BD27.II): both patients from family CHRO8 carried two compound heterozygous nonsense mutations p.Cys113stop and p.Glu148stop, while both patients of family BD27 were compound heterozygous for a complete deletion of the *KCNV2* gene and a missense mutation p.Leu404Pro located in the linker between transmembrane domains S4 and S5. The simplex subject BCM5 harboured two compound heterozygous small deletions: c.8_11del and c.447_449del. The c.8_11del mutation created a frame-shift at the very beginning of the *KCNV2* polypeptide, resulting in a premature stop codon and a severely altered and truncated protein (p.Lys3ArgfsX95). The other deletion only resulted in the loss of a single phenylalanine at position 150 (p.Phe150del) within the NAB domain. The last patient RCD307 was homozygous for another large deletion spanning from exon 1 into the 3′UTR. All mutations, except for the missense mutation p.Leu404Pro and the single amino acid deletion p.Phe150del (see.), are expected to result in the complete loss of Kv8.2. Consequently the siblings BD27.I and BD27.II, and patient RCD307 most likely did not express any KV8.2 gene product, causing an altered subunit composition of the respective Kv-channel and an altered or lost potassium channel function. Previous clinical data have indicated that both the complete absence of Kv8.2 (where Kv2.1 was unaffected) and its altered forms result in CDSRR, which suggests that the special constellation of Kv2.1/Kv8.2 heteromeric channels are essential for functionality in the photoreceptor cells. To examine whether these genotype differences are also evident in the phenotype, we divided the six patients into two groups: group 1, NOP (no protein), included the three patients with a complete absence of Kv8.2, and group 2, ALP (altered protein) with three patients with mutant Kv8.2 subunits. ## Clinical Findings Clinical findings are summarized in. All patients reported an early onset of their visual symptoms without any progression or change over the years. Only the oldest patient (RCD307) reported a slight reduction of his visual acuity in the last three years. Marked photophobia and a prolonged light adaptation time were evident in every case. Only the oldest patient complained of nyctalopia, other subjects denied having difficulty with night vision. All patients were myopic with variable degrees of astigmatism; patients of the ALP group showed a slight tendency for higher myopia and astigmatism. Four of six patients had undergone strabological surgery in childhood and two of them had suffered additionally from infantile nystagmus. ## Psychophysics All subjects presented with reduced central visual acuity (mean VA (logMAR): 0.97±0.2 SD). There was a slight tendency to poorer VA in the NOP group (1.06±0.23 SD) compared the ALP group (0.88±0.13 SD). Color vision testing using the Lanthony D-15 Panel desaturated and saturated tests and with fixation with the preferred retinal locus (PRL) showed severe color confusions in all patients predominantly along the scotopic or red-green axis with relative sparing of the tritan axis. The Rayleigh anomaloscope matches, presented eccentrically at the PRL, were consistent with a rather rod dominated function in five of six patients. Only one patient's (RCD307) results suggested protanopia. Perimetric results showed nearly normal outer boundaries of the visual field in all cases. Static perimetry results revealed relative and absolute defects in the central 30° area, being more pronounced in the NOP than the ALP group. To test cone and rod function loss we measured dark adaptation thresholds for red and green targets.. All patients showed significantly elevated thresholds for red and green stimuli, although the elevation was more pronounced in the NOP group with a final rod threshold of −1.9 log cd/m² and a cone threshold of 0.4 log cd/m² compared to −2.6 log cd/m² and −0.3 log cd/m² for the rod and cone threshold in the ALP group. Thresholds estimated for normals were −1.5 log cd/m² and −3.7 log cd/m² for red and green stimuli, respectively. For green stimuli the rod-cone break was normal (11.0 min for the NOP and ALP group, normal: 10.7 min) and for red stimuli the rod-cone break tended to appear earlier (after 15.6 and 15.9 min for NOP and ALP respectively, normal: 16.9 min). ## Electrophysiology Basic clinical investigation included the Ganzfeld ERG according to the ISCEV standard, for which all patients showed the previously described characteristic responses (see for typical results of a patient from the NOP and ALP group). Most interestingly, oscillatory potentials (OPs) were almost completely absent in the patients' ERG recordings. shows for each subject the amplitudes and implicit times of the a- and b-waves for the scotopic response series recordings. Typical low or undetectable response amplitudes to weak flashes were evident, with markedly delayed implicit times of the a- and b-wave component. There was also an abrupt increase in amplitude with increasing flash intensity accompanied by a normalization of b-wave implicit times. While mean saturation amplitudes *V_max* of the b-wave model fit were similar for both groups and normals (526 µV for normals, 528 and 493 µV for the NOP and ALP groups, respectively) the intensity *K* at semi-saturation was significantly shifted to higher intensities (−2.5, −1.7 and −1.7 log cd.s/m² for normals, NOP and ALP, respectively). Additionally, the peak a-wave amplitude was normal within the entire stimulus intensity range, but the peak implicit times were prolonged for each stimulus step. In this study the rod a-wave showed three interesting features (.): First, in some patients the a-wave of the response to the highest intensity stimulus (4.7 log td\*s) was smaller than that to the 4.2 log td\*s stimulus. This is depicted for one subject in. While R² of the fits in normal subjects was above 0.02 in only two eyes, it was higher in seven of the patient eyes. The difference in goodness of the fit is seen in the normal subject in and in the representative patient BCM5 in. Second, the latency of the negative deflection from baseline appeared with normal delay (average of 4.1±0.92 ms and 4.3±0.35 ms for patients and normals respectively). And third, while the maximum response amplitude Rm_PIII estimated from the model fit was not different from that of normals (on average 241±71 µV and 270±83 µV for patients and normals respectively), the sensitivity parameter S was significantly lower in patients (0.73±0.38 SD) than in normals (1.14±0.36 SD). *P_II* responses were calculated by removal of the fitted *P_III* (.). The latency at which *P_II* reaches 50 µV is plotted as a function of stimulus intensity on log-log coordinates in. In the normal retina we found a slope of −0.18±0.014 SD. In patients with *KCNV2* mutations the mean slopes of regression lines were not significantly different from each other or from the normal mean slope (−0.24±0.065 SD and −0.20±0.038 SD for the NOP and ALP group, respectively), however, a similar and clear shift either representing a response delay or a horizontal shift to higher intensities of approx. 1 log unit was observed in both groups. Latter was consistent with the psychophysically (dark adaptation) estimated rod threshold elevation of approx. 1 log unit. We additionally recorded responses to various flicker frequencies under scotopic conditions. Magnitude and phase of the responses to increasing flicker frequencies are demonstrated. The individual data under scotopic conditions showed reduced magnitudes and a phase difference, which was independent from flicker frequency thus suggesting either a rather small constant prolongation of rod photoreceptor recovery or may result from the sensitivity reduction to the flash strength. Cone responses recorded under photopic conditions showed a marked suppression of response amplitudes, which remained diminished even at the higher flash intensities (.). Additionally, the amplitudes did not show a photopic hill phenomenon. Implicit times were markedly prolonged for each stimulus step. The photopic negative responses (PhNR) were almost undetectable in patients with *KCNV2* mutations, as depicted in. As found for scotopic conditions, the photopic responses to stimuli of increasing frequency (.) showed a similar dependency but also lower amplitudes and a shift in phase. Interestingly, even at the highest frequency of 45 Hz the response waveform was still significant with an almost normal phase. Finally, multifocal ERGs showed reduced amplitudes and delayed implicit times in every ring (.) being more distinct in the central 2 rings. In the outer rings more preserved responses could be obtained. The dysfunction was more sharply limited to the central two rings in the NOP group, which was not observed that clearly in the ALP group. This could be related to the slightly better VA of the ALP group. ## Morphology Fundus photographs, FAF images and OCT scans for each subject can be seen in. There was a range of macular appearances including discrete disturbances of the retinal pigment epithelium (RPE) and bull's eye maculopathy. Patients with a complete absence of Kv8.2 (NOP group) showed more pronounced changes in the macular area, the mean central retinal thickness was 103.5 µm for the NOP, and 135.1 µm for the ALP group, respectively. In FAF imaging mild RPE-alterations were present as small areas with decreased autofluorescence. In contrast, marked RPE-atrophies were seen as sharply demarcated areas of absent autofluorescence surrounded by a ring of increased signal. The oldest patient (RCD307) revealed additional RPE-defects of the posterior pole and epiretinal gliosis. The OCT images also demonstrated the variety of morphological findings. In the milder cases –mainly in the ALP group- OCT revealed a thinner photoreceptor layer (PRL) in the foveal area, there was no disruption in the inner segment/outer segment (IS/OS) border. In the more severe cases (NOP group) the PRL was missing, the IS/OS border was diminished and an increased backscatter from the choroid was observed due to RPE-atrophy. In one case additional granular echoes were present due to deposits on the fundus of patient BD27.I. Based on the SD-OCT volume scans detected from the central 30°×15° retinal area the peripheral outer retinal structure was well preserved in every patient. # Discussion This study describes the genotype and phenotype of six patients with a retinal dystrophy due to changes in the *KCNV2* gene. In three patients the mutations resulted in a complete absence of Kv8.2 encoded by the *KCNV2* gene (NOP group). In the other three patients heterozygous mutations of the first allele resulted in a lack of protein product, mutations of the second allele led to mutant subunits with presumably remaining pore function (ALP group). On the morphological level, macular pathology varied from mild RPE disturbances to large atrophic areas, while the periphery was normal in every case. These findings are consistent with the results described by Robson et al. and Sergouniotis et al., It is interesting to see that the changes are strictly limited to the central macular area in every case, while the specific functional changes of this retinal disorder affect rod-rich mid periphery as well. The question of why cones degenerate and rods keep at least their morphological integrity still remains to be unraveled, but the distribution of Müller cells in the retina (i.e. absence in the fovea) could be one important factor, since their regulatory and buffering effect on the extracellular K⁺ is missing in the cone-rich area, making cones more vulnerable. Although there was a tendency for a more pronounced macular lesion (seen in the OCT and FAF imaging), higher myopia and more elevated dark adaptation thresholds in the NOP group, no clear-cut line can be drawn between the two groups. Interestingly, the only homozygous patient (RCD307) seemed to differ in a few aspects from the heterozygous patients: night blindness, protanomaly and reported progression were present only in his case. There was no family history of color disturbances in this subject and the further genetic analysis of the L-M pigment genes showed an intact OPN1MW/OPN1LW gene cluster, indicating a severe protanomaly rather than protanopia. Furthermore, his morphological results revealed more distinct changes of the fovea resembling a macular hole on the left eye. However, the additional RPE-defects of the posterior pole, the macular hole formation and the accompanying epiretinal gliosis could be due to age related changes, explaining the decreasing visual acuity observed by the 60-year-old patient. However, the investigated cohort is small, further large studies are necessary to better highlight the correlations between genotype and phenotype. Several studies exist on the electrophysiological characteristics of KCNV2 retinopathy.. Nevertheless, there are still several aspects of this special retinal disorder, for which an explanation is needed The lack of Kv8.2 or the presence of mutant subunits eliminates the functional characteristics of Kv2.1/Kv8.2 heteromers leading to a retinal disorder. Only the intact heteromers have the essential specifics to function as a high-pass amplifier and so regulate photoreceptor responses to light flashes. Kv2.1 can form homomeric channels, but without the intact Kv8.2 subunits they activate more slowly and inactivate faster, whereas the voltage dependence of the steady- state inactivation remains unchanged. The absence of intact Kv8.2 subunits therefore leads to a positive shift of the steady-state membrane potential, decreasing the dark current and elevating intracellular K⁺ level. Kv2.1 channels alone do not produce a permanent outward K⁺ current, which also affects the K⁺ homeostasis of the photoreceptors. While restoring the integrity of the K⁺ levels, secondary mechanisms can also lead to a small drop of intracellular Ca²⁺ levels, probably due to altered/enhanced function of the Na⁺/Ca2⁺-K⁺ exchanger. The small drop of cytoplasmic Ca²⁺, however, can result in increasing cGMP levels due to disinhibiting the activated guanylate-cyclase and finally increases the number of open cyclic nucleotide-gated channels. This shift to a more depolarized state in the dark may also have consequences for recovery: for very brief light flashes the membrane potential may remain below the critical limit of −50 mV before the hyperpolarization-activated cyclic nucleotide-gated channels (HCN channels) become activated, thus leading to a prolonged hyperpolarization phase. These mechanisms are well reflected in the electrophysiological findings observed in patients with KCNV2 retinopathy. The responses are characteristically undetectable or markedly reduced with delayed implicit times for dimmer stimuli and there is an abrupt rise in amplitudes and shortening of implicit time with increasing stimulus intensity. The intensity *K* at semi-saturation is significantly shifted to higher intensities, the estimated difference in our cohort is around 1 log cd.s/m². This shift correlates with the threshold elevation during dark adaptometry (for rod thresholds approximately 1 log unit elevation). However, most of our patients do not suffer from nyctalopia, confirming other reports of this inconsistency between subjective and objective light sensation. Our results are in accordance with previous reports and also confirm that “supernormal rod responses” in the ERG - believed to be characteristic for this rare condition - often seem to be missing, as responses, even to high intensity flashes, stay within normal limits in many cases. The dynamics of the b-wave intensity-response function is a more constant feature. In addition, our detailed electrophysiological data show other specific features. The initial phase or leading edge of scotopic response waveforms reflects the activity of photoreceptor cells and arises from light-evoked closure of Na⁺ channels along the plasma membrane of the outer segments. Based on the model fits to our electrophysiological data we conclude that phototransduction activation in this retinal dystrophy is normal, since the onset of the deflection from baseline appears with normal delay. While the maximum amplitude Rm_PIII was within normal limits, the sensitivity parameter S was significantly lower. Similar results were reported for the patient described by Tanimoto et al.. On the contrary, Hood et al. found essentially normal sensitivity S and a slightly lower Rm_PIII, the maximum amplitude, for the rods. However, our patients were included in the study based on confirmed alterations of *KCNV2*, while patients in the study of Hood et al. were chosen on a clinical-electrophysiological basis (i.e. retinal dystrophy with supernormal rod responses). While higher Rm_PIII might be explained by an overshoot due to delayed HCN channel activation, lower sensitivity may be related to higher cation channel sensitivity to cGMP due to lower Ca²⁺ levels. There is also a delayed postreceptoral response, which seems independent from flicker frequency. The delay of the emerging b-wave may have two different origins. Firstly, voltage dependent transmitter release may be delayed due to a small prolongation in reaching hyperpolarization. Secondly, a delayed HCN channel activation and depolarization of the photoreceptor might result in an overshoot of the response of the downstream neuron. Such overshoot would not be associated with changes in the G-protein activation cascade of the bipolar cell, which can be assessed with the *P_II* response analysis. In our *KCNV2* patients the kinetics of the ON-bipolar cell G-protein cascade seems to be normal, however, the cascade is activated with a clear delay. In the study of Robson et al. including 25 patients with the characteristic scotopic b-wave signs, this delay was seen for the ON-responses and particularly for the OFF- responses as well. A further interesting finding was the marked reduction or even absence of the oscillatory potentials, either suggesting an altered function of the inner retina (i.e. amacrine or interplexiform cells) or more likely a significantly reduced cone function. Moreover, we studied the temporal characteristics of the retina for the first time in this specific retinal disorder. Repetitive stimulation is very demanding for the metabolic process in neurons and changes in the temporal dynamics (e.g. in channelopathies) are evident in flicker ERGs. The impairment of temporal response characteristics can occur due to photoreceptor disturbances as well as postsynaptic mechanisms. Kv2.1/Kv8.2 heteromeric channels contribute to the generation of the K⁺ current responsible for the dynamic signal amplification of photoreceptors. The hyperpolarizing overshoot in response to rapid onset illumination has an important role in increasing the sensitivity to fast changes of illumination. Altered Kv2.1/Kv8.2 heteromers lose their ability to function as a high-pass amplifier, which explains the altered temporal characteristics observed in our patients (i.e. there is a constant prolongation of photoreceptor recovery time). Interestingly, even at the highest frequency of 45 Hz the response waveform was still significant with an almost normal phase, which indicates that even though we find clearly reduced amplitudes and a phase shift due to cone dystrophy, the temporal dynamics seem to be almost normal. There is little known about critical flicker fusion in retinal dystrophies in humans, but it has been shown in RCS rats that with progression of the degeneration the amplitude for higher frequency waveforms declines and the critical flicker fusion frequency is shifted to lower frequencies. However, Kv2.1/Kv8.2 heteromeric channels are not the only components of the outward K⁺ current in photoreceptors, although they are essential for their functional properties. These special electrophysiological features are considered to be strictly associated with *KCNV2* mutations, but recently Thompson et al. have shown somewhat similar changes of the dark-adapted electroretinogram in patients with *KCNJ10* mutations. The K⁺ channel expressed by the *KCNJ10* gene (Kir4.1) has previously been recognized as pathogenic in man, causing a constellation of symptoms, including epilepsy, ataxia, sensorineural deafness and a renal tubulopathy (EAST syndrome). Kir4.1 constitutes the primary inward rectifying potassium channel of retinal Müller cells and is responsible for the regulation of extracellular K⁺. Thompson et al. nicely demonstrated similar dynamics of scotopic intensity-response function in two of four patients. ERGs to dimmer flash stimuli showed a delay of up to 20 ms before the onset of the b-wave, and with increasing intensity a sudden elevation of amplitudes and normal implicit times could be detected. These similarities of the scotopic ERG could be explained with an altered sensitivity at the synapse between rod and ON-bipolar cells due to mutations in *KCNJ10*. Photopic ERGs of all patients showed reduced amplitudes of the photopic negative response (PhNR) and showed a delay in b-wave time to peak, but the photopic hill was preserved. However, these patients did not develop a cone dystrophy, since mutations in *KCNJ10* primarily affect the Müller cell functions, while mutations in *KCNV2* lead to disturbed functional integrity of the photoreceptors and probably impair their postreceptoral signaling. KCNV2 retinopathy is considered a very rare retinal disorder associated with high but often normal mixed rod-cone response amplitudes, a marked prolongation of b-wave implicit times and a delayed, almost sudden, steep amplitude-versus- intensity relationship under scotopic conditions. Furthermore, while rod phototransduction is intact, there is a constant delay of the responses, which suggests changes in the synapse or in postreceptoral signaling pathway. Inner retinal involvement is also probable, since oscillatory potentials are almost absent. These findings are diagnostic and are exclusively linked to *KCNV2* mutations. [^1]: The authors have declared that no competing interests exist. [^2]: Performed the experiments: DZ SK BW EZ HJ. Analyzed the data: DZ SK BW EZ HJ. Contributed reagents/materials/analysis tools: DZ SK BW EZ HJ. Wrote the paper: DZ. corrected the manuscript: SK BW EZ HJ.

# Introduction 2-Hydroxypropyl-Beta-Cyclodextrin (HPβCD) is a commonly used excipient to stabilize and solubilize pharmaceuticals. HPβCD reduces cholesterol and lipid accumulation, and has emerged as a possible treatment for Niemann-Pick type C disease, Alzheimer’s disease, and atherosclerosis. But, a negative side effect of HPβCD-based treatments is hearing loss. Recent studies on the chronic effects of HPβCD administered systemically or directly into cerebrospinal fluid found graduated losses of outer hair cells (OHC) along the cochlear spiral, with more severe losses at the cochlear base than the apex. While the origins of HPβCD toxicity within the ear remain uncertain, its effects can modulate cochlear or OHC electromechanics in excised cochleae. In experiments reported here we used a variety of physiological measurements to understand the effects of acute administration of HPβCD, and HPβCD analog methyl-β-cyclodextrin (MβCD), directly into cochlear perilymph. Results were compared to measurements made during treatment with salicylate, which has effects that have been well characterized, and the effects of injecting artificial perilymph alone. # Methods ## Animal preparation We used NIH-strain pigmented guinea pigs of either sex (between 400–600 g). Initially animals were anesthetized with an intraperitoneal injection of sodium thiobutabarbital (100 mg/kg). Cutting shears were used to shave head and neck fur. A tracheotomy was performed, and the animal was artificially ventilated with isofluorane (\~1% in oxygen) with respiratory volume maintained (5% end- tidal CO₂). We monitored heart rate, O₂ saturation, and expired CO₂ level with a pulseoximeter/CO₂ analyzer. The right cochlea was accessed with a ventral surgical approach. Soft tissue of the right ear canal was removed. The animal was mounted with hollow ear bars that allowed delivery of acoustic stimuli. A cannula placed in the left jugular vein was used to administer pancuronium bromide (0.06 mg/kg) to eliminate middle-ear muscle contractions just before the start of making auditory measurements. Body temperature was maintained (38°C) with a dc-powered heating blanket and rectal thermometer system. Experimental protocols for this study were approved by the Animal Studies Committee of Washington University (protocol numbers 20120113 and 20130069). ## Solution administration by injection into the cochlear apex Ototoxic solutions were administered from a pipette sealed into the cochlear apex. Previously we have shown that apical injection drives solutions toward the cochlear aqueduct at the base of scala tympani, allowing the entire scala to be uniformly treated. This overcomes the limitations of classic administration techniques to the cochlear base that do not reach therapeutic levels at the regions tuned to mid to low frequencies that code speech vowels and background noise. Administering into the cochlear base is certainly not the ideal approach to understand how HPβCD affects the entire cochlear length. Solutions injected into the cochlear apex can be administered slowly if the goal of the experiment is to sequentially affect finely spaced cochlear regions contributing to a response. For the experiments reported here, we used a constant, relatively fast (0.5 μL/min) injection rate. Animals received either artificial perilymph alone (controls, n = 3 guinea pigs) or, in artificial perilymph, 20 mM salicylate (n = 3 guinea pigs), 13 mM HPβCD (n = 3 guinea pigs), 27 mM HPβCD (n = 3 guinea pigs), or 13 mM MβCD (n = 3 guinea pigs). The composition (in mM) of artificial perilymph was NaCl (127.5), KCl (3.5), NaHCO₃ (25), CaCl₂ (1.3), MgCl₂ (1.2), NaH2PO₄ (0.75), and Glucose (11). Animals treated with 13 mM HPβCD were in the “low-dose” group and those treated with 27 mM HPβCD were in the “high-dose” group. Salicylate alters surface cisternal system of hair cell bodies, reduces turgor pressure, and effects OHC motility. In short, salicylate attenuates the gain of the cochlear amplifier. Solutions were injected from a pipette sealed into the cochlear apex in the 4^th cochlear turn. The mucosa covering the apex was removed with a damp, cotton wrapped applicator or cellulose wipe. Cyanoacrylate glue was applied to the dry bone of the apex, then a thin layer two-part silicone elastomer was applied. This effectively made the surface hydrophobic. The cochlear apex was fenestrated through the adhesive layers on the bone. The fenestration was made with a 30º, 1/3mm oval window pick by resting the pick on the surface at one location and then lifting the pick off the surface. The fenestra diameter was made to fit a 20–30 μm diameter tip pipette that was pulled from 100 mm x 0.58 mm inner-diameter glass tubing. We made a fluid-tight seal between the glass injection pipette and the hydrophobic surface by wicking fluid (accumulated from either condensation or cochlear fluid accumulation) and applying additional cyanoacrylate glue. Solutions were driven at 500 nL/min for 15 min for a total of 7500 nL through a 50 μL Hamilton gas- tight syringe (1710TLL), glued to a World Precision Instruments plexiglass coupler (MPH6S10), mounted on a computer-controlled World Precision Instruments Ultrapump. ## Acoustic stimuli and physiologic measurements Electrophysiologic measurements were made using procedures that we have previously described. Measurements were made with Tucker-Davis System 3 hardware controlled by custom-written software in Visual Basic (Microsoft) on a personal computer. TD-RP2 modules were used for stimulus generation. Stimuli were passed through TD-PA5 attenuators, and TD-HB7 headphone amplifiers. Acoustic stimuli were delivered in a closed sound system: an Etymotic ER-10C coupled to the hollow ear bar. Calibrations were completed in individual ears by tracking 70 dB SPL tones from 125 Hz to 26 kHz in ¼ octave steps. Cochlear response measurements were made differentially between an Ag/AgCl electrode in the round window niche and a platinum-needle electrode in the vertex. Measurements were made with an optically-coupled TD-HB7amplifier (1000X gain, 0.005–15 kHz bandpass filter), routed to TD-RP2 modules for digitization (48.8 kHz) and averaging. Animals were electrically grounded with an Ag/AgCl pellet electrode coupled to the exposed soft tissue of the neck by a fluid bridge. ## Histological preparation At the conclusion of the apical injection experiments, the experimental cochleae were prepared for fixative injection. The pipette used to inject ototoxins was removed, and the fenestration was occasionally enlarged to a small extent to accept a 20–30 μm diameter tip pipette pulled from 100 mm x 0.58 mm inner- diameter glass tubing used to inject fixative. The pipette for fixative was sealed into the cochlea by wicking away any cochlear or condensation fluid accumulation while applying cyanoacrylate glue to form a fluid-tight junction. The round window membrane was perforated before the start of fixative injection. We injected 2.5% glutaraldehyde and 1.5% paraformaldehyde in a 0.065 M phosphate buffer. This solution was injected at 2000 nL / min for a total of 7500 nL. Cochleae were extracted and placed in fixative solution for at least two days at 4°C. Cochleae were decalcified (0.1 M EDTA with 0.4% glutaradehyde) for 14 days, osmicated (1% OsO₄ in dH₂O) for 60 minutes, dehydrated in ethanols and propylene oxide, embedded in Araldite resins, and sectioned parallel to the spiral axis of the fourth cochlear turn at 40 μM with a carbide steel knife. Sections were mounted in Permount on microscope slides and cover- slipped. Sections were analyzed with light microscopy by an author who was blinded to the treatment each ear received. # Results The time course of cochlear action potential (CAP) threshold shifts differed across treatments and tone-burst frequencies. The ≤10 dB threshold shifts occurring 30 minutes after the start of injecting artificial perilymph alone (control experiments) quantify the extent to which the apical injection procedure itself influenced neural thresholds. The direction and rate (dB / min) of CAP threshold shifts differed between HPβCD 13 and 27 mM treatments, in that threshold shifts essentially raised and then plateaued for 13 mM treatment or, in contrast, steadily increased to a maximum or to the abolition of CAPs for 27 mM treatment. Quick CAP threshold shifts followed by gradual CAP recovery is consistent with the well-known temporary effects of salicylate (e.g.. Maximal effects of salicylate and 27 mM HPβCD on CAP thresholds are consistent with elimination of cochlear amplifier gain, as targeted deletion of prestin in mice raise neural threshold by 40 to 60 dB. Average threshold shifts of 20 to 40 dB from 13 mM HPβCD are far greater than the 6 dB threshold shifts caused by heterozygote prestin knockout mice when electromotility is halved. Total CAP abolition during, or soon after, MβCD treatment is itself an indication of far greater effects than simple cochlear amplifier gain attenuation (see the histological data from these ears presented later). In normal ears CAP to low- level tone bursts originate from the peak of the traveling wave (reviewed in). Across all toxic treatments, CAP thresholds to 2, 4, and 8 kHz started to rise soon after the injection start (i.e., zero minutes re. injection start), but \~3–8 additional minutes was needed to start abolishing CAP thresholds to 16 kHz. This demonstrates that we did not simultaneously treat the length of the cochlea. Comparing maximal effects on CAP threshold shifts in helps to understand the extent to which cochlear amplifier gain was attenuated across treatments. Auditory Nerve Overlapped Waveform (ANOW) measurements show that the apical injection technique can treat the apical half of the cochlear spiral, a region that classic round window administration cannot. The ANOW originates from afferent auditory nerve fibers in the apical cochlear half and can quantify low- frequency auditory thresholds. ANOW amplitudes from supra-threshold, 50 dB SPL sound levels were ablated by salicylate, HPβCD, and MβCD treatments, but not by artificial perilymph alone (control;). Subtle changes to ANOW from control ears during the time of injection suggest that perhaps some ANOW changes in ears treated with toxic solutions resulted from mechanical disturbances from the fast apical injection approach used for these experiments. Nevertheless, unlike CAPs, ANOW amplitude measurements were fully abolished with all four treatments, consistent with the ANOW being more sensitive than CAPs to cochlear manipulations and diseased states. Cochlear microphonic (CM) measurements to 90 dB SPL 500 Hz tone bursts were made from inside the endolymphatic space where responses are not influenced by neural excitation to the sound used to evoke the desired hair-cell-based response. Slight transient changes in CM amplitude from control, 13 mM HPβCD, and salicylate treatment are consistent with slight mechanical disruption caused by the relatively fast injection rate that was used here. The CM remaining after treatment with 27 mM HPβCD originates from current flow through remaining OHC and IHC, perhaps IHCs more than OHCs (as suggested from histological data below). Total abolition of CM measurements occurred for MβCD. Endocochlear potential (EP) measurements were made in the third cochlear turn. Artificial perilymph injection (control) did not markedly affect the EP. Treatment with 13 mM HPβCD and salicylate caused transient and temporary effects: 13 mM HPβCD caused EP to increase during injection and then returned to pre-injection values during the time immediately after injection while, in contrast, salicylate caused EP to decrease before gradually returning toward pre-injection values. Treatment with 27 mM HPβCD moderately affected the EP, causing transient increases during injection that returned to pre-injection values during the time immediately after injection before markedly declining further. MβCD caused a brief increase in EP then total abolition before the injection was complete. EP changes are mirrored by changes in the magnitude of the silent current, e.g., an EP decrease is consistent with a decrease in the standing current through outer hair cells in silence. The time courses of effects on EP during treatment are not identical to the time courses of effects on CM (cf. Figs), which is consistent with the origin of these measurements being different even though they were made with the same electrode in the endolymphatic space of the third cochlear turn. Cubic DPOAE amplitudes were measured at 2*f*₁- *f*₂ with an arbitrarily chosen *f*_*2* ≈ 6 kHz. Individual-ear DPOAE amplitude fine structure was considered by choosing the *f*_*2* primary-tone frequencies closest to 6 kHz that produced a peak in the 2*f*₁- *f*₂ DPOAE amplitude. Injection of artificial perilymph did not affect DPOAE amplitudes in control ears. With 13 mM HPβCD treatment, DPOAE amplitudes dramatically declined during the injection but gradually recovered toward near pre-injection levels. With salicylate and 27 mM HPβCD treatments, effects on DPOAE amplitude were longer, and did not fully recover. MβCD was the only treatment that totally abolished DPOAE amplitudes. Sensory cells within the organ of Corti were evaluated using light microscopy and serial sections of plastic embedded cochleae. All sensory and non-sensory cells were well preserved in the control and 13 mM HPβCD groups. Effects from treatment with 27 mM HPβCD were variable, ranging from well-preserved to severely damaged outer and inner hair cells. Ears with severe IHC and OHC damage had greater effects at the base than apex. While some of this variability in OHC loss may be attributable to the relatively short time-frame of our acute study, interanimal variability is nevertheless consistent with that reported from Crumling et al. and Cronin et al. who studied the chronic effects of HPβCD treatment. Treatment with MβCD consistently caused severe damage directly to the OHCs and IHCs or the regions around these sensory cells. During the time course of these experiments with HPβCD, MβCD, or artificial perilymph (controls), lateral wall structures, including the spiral ligament fibrocytes, stria vascularis, spiral ganglion neurons did not appear to be altered. Specific morphometry was not performed in these structures, but there was no evidence of stria swelling or fluid leakage, the spiral ligament fibrocytes populated the ligament as in control ears, and there were no signs of myelin unwrapping from the spiral ganglion cell soma or evidence of cell swelling. Histological analysis was not performed for treatment with salicylate as the effects are well known or well-characterized. # Discussion Our apical injection procedure overcomes the limitations of classical approaches of administrating drug treatments to the more surgically accessible cochlear base. In particular, apically injected solutions can treat the entire length of the cochlear spiral. We studied the acute effects of injecting HPβCD, and HPβCD analog MβCD, directly into the cochlea. We injected \~2x the volume of guinea pig scala tympani, most of which would have been driven out with our technique through cochlear aqueduct during the injection duration. The experiments ended and the animals were sacrificed 60 minutes after the start of treatment. Results were compared to those from salicylate treatments, as well as artificial perilymph alone (controls). Treatment with 13 and 27 mM HPβCD and salicylate ablated the ANOW amplitude and raised CAP thresholds by approximately 40 to 60 dB, consistent with attenuation of cochlear amplifier gain. Other measurements such as the endocochlear potential, cochlear microphonic recorded from inside the third cochlear turn endolymphatic space, distortion product otoacoustic emissions from the ear canal, and histological measurements generally showed that effects from MβCD were greatest, followed in order of decreasing effect by 27 mM HPβCD, 13 mM HPβCD, salicylate, and artificial perilymph injections alone (controls). ## HPβCD effects on an analysis of the electrical cochlear response Here we address how treatment with 13 mM HPβCD affected harmonic distortion in the electrical cochlear response measured with a round window electrode. Sigmoidal, saturating, nonlinear functions are commonly used in analyses of empirical, gross measurements to study general transfer of acoustic sound into neural excitation (*f*_TR, e.g.,). Even and odd order distortions respectively associate with the asymmetry and saturation of *in vivo f*_TR and analyses of distortions can be used to estimate the operating point of *f*_TR that is associated with the amplitude of distortion products (e.g.. Expressing an even order distortion product as a function of operating point is proportional to the second derivative, and an odd order distortion product is proportional to the third derivative, of a function used to describe the sigmoidal, saturating, nonlinear *f*_TR (e.g.. Cochlear response harmonics to 90 dB SPL 500 Hz tone bursts were measured from animals in the control, salicylate, and 13 mM HPβCD groups. These are the experimental groups where cochlear response was still present after solution administration. Artificial perilymph injections (control) did not affect even order harmonic distortion (*f*_TR asymmetry, gray), but caused a small and transient decrease in odd order harmonic amplitude (*f*_TR saturation) at \~18 min. after the injection start (black). Salicylate caused a slight and brief increase in even order harmonic distortion amplitude immediately after injection (olive), and a decrease in odd order harmonic amplitude that recovered to levels greater than pre-injection levels. (i.e., and “overshoot”;, green). HPβCD caused opposing effects in even and odd order harmonic distortion. The amplitude of even order harmonics maximally decreased at \~21 min after the injection start. In contrast, the amplitude of odd order harmonics increased to a maximum at \~22 min. Both even and odd order harmonics recovered to near pre-injection levels. We suspect that recovery of even and odd order harmonics, as well as slight recovery of DPOAE amplitudes, may have originated from minimal damage that precedes cell death and causes temporary functional deficit. These effects and recoveries suggest that a function which can describe the sigmoidal, saturating, nonlinearities involved with transferring acoustic sound into neural excitation was morphing during our acute experiments. We will now use a Boltzmann analysis of electrical cochlear response measurements from exemplar ears to estimate the operating point of *f*_TR and express harmonic distortions as a function of operating point estimates. The cochlear response is a gross measurement of the cochlear microphonic from inner and outer hair cells, summating potentials, changes to the lateral wall potential from slow or sustained current through hair cells, excitatory postsynaptic potentials, onset or phase-locked compound action potentials from cochlear regions tuned to frequencies of the sound stimulus or those located more toward the base, and the coalescence with spontaneous excitation of single-auditory-nerve-fibers associated with cochlear regions that are not excited by the sound stimulus. We are thus studying cochlear nonlinearity in the general terms of influence on DPOAE amplitude, certainly not mechanoelectric transduction at the apical pole of an OHC as is commonly done with Boltzmann analyses. The Boltzmann analysis of the electrical cochlear response measurements provide a more unique perspective that builds on our qualitative description immediately above. Operating point estimates were obtained by adjusting Boltzmann parameters until the modeled output matched empirical cochlear responses. The Boltzmann function was V_t = V_EP + (− V_sat + 2 V_sat / (1 + exp(-2 S_B / V_sat (P_t + OP)))) where V_EP was a DC potential representing the endocochlear potential magnitude (mV), V_sat as the saturation voltage of the Boltzmann function (mV), S_B represented the slope of the Boltzmann function at its mid-point (mV/Pa), P_t represented the input pressure (Pa) as a function of time, OP represented the operating point of the Boltzmann function (Pa). Harmonic distortions did not markedly vary in control ears when expressed as a function of operating point estimates. Even and odd harmonic distortion variations with operating point estimates during salicylate injections were consistent with results from previous experiments that utilized gel injections into the cochlear apex to cause sustained displacement of the organ of Corti and simulate the effects of endolymphatic hydrops. Variations to harmonic distortions expressed as a function of operating point estimates during 13 mM HPβCD injections were novel in that the trends deviated dramatically from the second and third derivative of the *in vivo f*_TR. HPβCD caused unprecedented effects to physiologic measurements independent of apparent OHC loss that are not consistent with sustained displacement of the organ of Corti, such as what can happen if, for example, alteration to OHC bodies manipulate stereocilia coupling to the tectorial membrane. Above we used empirical measurements in a Boltzmann analysis to understand how 13 mM HPβCD treatment changed operating point estimates of *in vivo f*_TR. We now use the Boltzmann analysis to understand how these operating point estimates can describe the trends in 2*f*₁ and 3*f*₁ harmonic distortion measurements we found during the apical injection procedure. Recall that even (e.g., 2*f*₁) and odd (e.g., 3*f*₁) order harmonics respectively quantify the asymmetry and saturation of *in vivo f*_TR. Using a Boltzmann analysis, we used a sinewave input (P_t) and systematically varied OP with estimates obtained from fitting empirical data (i.e., the x-axis values of). The amplitude of the harmonics in the simulated output (V_t) were measured and expressed as a function of OP. Results show that varying the operating point of the Boltzmann analysis yielded simulated harmonic distortions that were qualitatively similar to empirical harmonics (cf. Figs and, shades of purple). These results add additional support for the interpretation that 13 mM HPβCD caused morphing in the sigmoidal, saturating, nonlinearities involved with transferring acoustic sound into neural excitation during our acute experiments, without causing apparent OHC loss. ## HPβCD effects on electrical cochlear responses help to understand effects on DPOAEs Here we discuss how the effects to the asymmetry, saturation, and sensitivity of *f*_TR related to the changes found to DPOAE amplitude measurements. Administering 13 mM HPβCD did not cause apparent changes to CM amplitude recorded inside the endolymphatic space but caused a large reduction of DPOAE amplitudes (cf. Figs and purple, \~15 minutes after the start of injection). These results suggest intact mechanoelectric transduction but marked attenuation of cochlear amplifier gain. We investigated this further by analyzing CM measurements made before, and 30 minutes after, the start of injection. Before treatment, the CM (red) fit well to a typical Boltzmann function (blue) but not to a Boltzmann function without saturation (V_sat from the Boltzmann analysis, green). These fits to pre-treatment empirical CM can be seen both in the Boltzmann analysis and the time domain. During treatment, the CM (red) fit well to a simple sine wave (light green). But, to achieve a fit to a Boltzmann function, V_sat had to be set to infinity (light blue). These fits to empirical CM measurements made during treatment can be appreciated in both the Boltzmann analysis and the time domain. We suspect that saturation of CM amplitude would have occurred at sound pressure levels much higher than what was used for these experiments because there was no apparent loss of OHC stereocilia and bodies after 13 mM HPβCD treatment. The unexpected result of normal CM amplitude measured inside scala media in the face of maximal decreases in DPOAE amplitude during treatment of 13 mM HPβCD treatment likely originate from lack of saturation to the *in vivo f*_TR. The near normal CM amplitude measured inside the endolymphatic space and transient enhancement of EP measurements made during 13 mM HPβCD treatment could be explained if decreased current flow through OHCs increases the overall resistance between endolymph and perilymph. That is to say, decreased current flow would reduce the OHC-generated potential, but with less shunting to perilymph the amplitude recorded from endolymph may not be reduced to the same degree. ## HPβCD does not affect the auditory nerve or lateral wall of scala media Here we revisit the finding that \~40 min after the start of 13 mM HPβCD treatment DPOAEs returned to near-baseline amplitudes but CAP thresholds shifted to an asymptotic level that is consistent with fully attenuating cochlear amplifier gain (cf. Figs and purple). These findings naturally lead to the question: Does HPβCD cause an auditory neuropathy? Previous work found that intra-cochlear administration of the sodium channel blocker tetrodotoxin (TTX) can be used to study excitatory post-synaptic potentials (EPSPs) recorded with a round window electrode. EPSPs leading to normal spike generation have shorter latencies than CAP latencies. But, post- treatment latencies could decrease from broadened tuning caused by HPβCD affecting the cochlear amplifier, or from the need to increase stimulus level and thus probe a wider region of cochlear tuning curves to achieve a measureable response in damaged ears. Broader filters have shorter delays than sharper filters. TTX effectively causes an auditory neuropathy with raised neural thresholds in the face of normal DPOAE amplitudes and endocochlear potential. It is thus possible that the voltages recorded during the asymptotic threshold shifts in were EPSPs masquerading as CAPs with an amplitude ≥ the typical 10 μV criteria for thresholds while DPOAEs amplitudes approximated pre-injection levels (cf. Figs and \~40 min after injection start). To address this possibility we compared round-window electrode measurements made before and after treatment with 13 mM HPβCD and 250 ng/ml TTX. We used measurements made within the first and last 10 minutes of a 70-minute experiment that used the same 15-minute injection procedure used in the other experiments discussed in this report. CAP measurements before HPβCD and TTX administration are comparable, as expected (red). After HPβCD treatment, the amplitude of the response was reduced but waveform morphology was more like the pre-HPβCD waveform than the post-TTX waveform, consistent with what would be expected by simple reduction of stimulus sound pressure level, or attenuation of cochlear amplifier gain, that drive neural responses (blue). In contrast, after TTX administration the waveform resembles the well-known morphology of gross EPSPs as seen from a round-window ball electrode (blue). *We conclude that HPβCD does not have its primary effect by directly acting on auditory neurons*. CM amplitudes and endocochlear potential measurements made after the apical injection stopped were minimally affected by 13 mM HPβCD. The effects of HPβCD (and salicylate) on distortion products are thus different than those from furosemide which reduces both the DPOAE amplitudes *and* the EP. *Our results do not suggest that HPβCD acts on the lateral wall of scala media*. ## Other theories on the origin of HPβCD ototoxicity β-cyclodextrins can have widespread actions on cochlear function, potentially impacting any process involving cell stiffness or membrane biophysics. In other model systems, cyclodextrins can impact tight junctions, mechanotransduction and synaptic function. Without implicating any one of these, or potentially other interactions, it is clear that the effects of β-cyclodextrins can have multiple origins. While our data suggest the origins extend beyond outer hair cells, it remains unclear whether targets are permanently affected by β-cyclodextrins. ## Do HPβCD effects differ between species and along the length of the cochlear spiral? Histological data from 27 mM HPβCD treatment showed that severe OHC and IHC damage was more pervasive in the basal cochlear half than the apical half, a finding that is consistent with that found by Crumling et al. and Cronin et al. who studied the chronic effects of HPβCD in mice by administering systemically and directly into cerebrospinal fluid. We suspect Cronin et al.’s drug entered the cochleae in cerebral spinal fluid through the cochlear aqueduct in the base of scala tympani, and was therefore at a lower concentration in the apical half of the cochlear length (cf. studies of chronic effects on measurements from mid- frequencies with administration to the cochlear base). In contrast, it is unlikely that our solutions were diluted by mixing with cerebral spinal fluid because our 500 nL / min injection rate is larger than the \~30 nL / min sustained entry of cerebral spinal fluid through the cochlear aqueduct. Methodological differences aside, our histological data showing severe hair cell damage to be more pervasive at the cochlear base than at the apex is consistent with Crumling et al. and Cronin et al.’s findings and suggests that *the graded effects of HPβCD on histological measurements along the cochlear length do not differ between species*. Histological data alone cannot determine if HPβCD has varying effects on *measurements of hearing* along the cochlear length. CAP thresholds to the highest tone burst frequency (16 kHz) were affected less than to the lowest tone burst frequency (2 kHz) by 13 mM HPβCD and salicylate, but were similarly affected by 27 mM HPβCD. Our neural threshold measurements are not consistent with those from Crumling et al. and Cronin et al. who found that auditory brainstem response thresholds to low-frequency tone burst (4 kHz) were shifted by \~40 dB after HPβCD treatment but those to high-frequency tone burst (16 & 32 kHz) shifted \~50–60 dB. The difference between our results and those from Crumling et al. and Cronin et al. might be explained by their low-frequency baseline (control) thresholds that are slightly high in the strain used, a result attributed to greater OHC death. Or, disagreeing data may originate from differences in our methods used to study the acute effects of HPβCD and their methods used to study the chronic effects of HPβCD. Since our sets of histological data agree in that HPβCD has a lesser effect in the cochlear apex, one would ideally want to study the chronic HPβCD effects with an approach that could ensure treatment of the entire cochlear length in many different species and use physiologic measurements that can be obtained throughout the cochlear length. But, as it currently stands, the presently available physiological data suggests that *the graded effects of HPβCD on measurements of hearing from along the cochlear length do differ between species*. Coupling together the available, the histological and physiological findings leads to the hypothesis that apical cells may be less susceptible to death, but are equally susceptible to functional deficit. This hypothesis predicts a continuum of a dose-response gradation from functional deficit to cell death. Addressing the possibility of longitudinal gradients in cell physiology may be a promising area of study, particularly once the perilymph concentrations of clinically applied HPβCD are known. # Conclusions We studied the acute effects of HPβCD injected directly into the perilymph of intact and sealed cochleae. We found that a low-dose of HPβCD raised CAP to an extent that was consistent with attenuating cochlear amplifier gain, had no apparent effect on the EP, altered general nonlinearities involved with transferring acoustic sound into neural excitation without causing apparent OHC loss, and the CM measured from inside the endolymphatic space. In contrast, DPOAEs measured in the ear canal were greatly diminished. A high-dose of HPβCD elevated CAP thresholds, markedly affected the EP, CM and DPOAEs, caused sporadic OHC losses that were consistent with previous studies on the chronic effects of HPβCD. Neither the low- or high-dose of HPβCD caused apparent disruption of the scala media lateral wall or the auditory nerve. But, for the duration of our acute studies, known intra-cochlear concentrations high-dose of HPβCD caused variable effects on OHCs throughout the length of the cochlear spiral. We thank Dr. Uzma S. Wilson and Professor Isabel Varela-Nieto for productively criticizing an early version of this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** JTL KH CAB RKD ANS. **Formal analysis:** JTL ANS. **Funding acquisition:** JTL KH ANS. **Investigation:** JTL. **Methodology:** JTL ANS. **Resources:** JTL KH ANS. **Visualization:** JTL. **Writing – original draft:** JTL KH CAB RKD ANS.

# Introduction In the malignant progression of a solid tumor the gain of an invasive phenotype is the first and a necessary step in the development of metastases. The sequelae of metastasis account for 90% of cancer-related mortality. To date, the biological events underlying invasion and metastasis remain poorly understood. The development of effective anti-invasive/antimetastatic tools and drugs is therefore an unmet clinical need. Our laboratories have been characterizing and validating 4-fluoro-3’,4’,5’-trimethoxychalcone or C16 as a novel invasion inhibitor for the study and potential treatment of metastatic cancer. C16 has nanomolar anti- invasive potency in the chick heart and Matrigel *in vitro* invasion models against several human cancer cell lines (MCF-7/6 breast cancer, BLM melanoma and SK-OV-3 ovarian carcinoma). The molecule shows a defined structure-activity relationship and possesses adequate *in vitro* absorption, distribution, metabolism, and excretion (ADME) properties (*see section ‘Stability and metabolism’ and*). Given its discovery in a phenotypic model, the biological target of C16 is unknown. We have differentiated the molecular mechanism of action of C16 from that of other antimetastatic agents, and are currently characterizing its biochemical interactions in detail. The aim of the present research was to explore the drug metabolism and pharmacokinetics (DMPK) and toxicity profile of C16 in rodents, in preparation of an evaluation of its efficacy as a pharmacological tool in animal models of disease. Here, we have developed appropriate vehicles to accommodate the low bioavailability of C16, including a formulation as a medicated gel suitable for chronic oral dosing. Furthermore, we have analyzed the pharmacokinetic properties and metabolism of C16, as well as its maximum tolerated dose and repeat-dose toxicity profile. Routine pharmacokinetics (PK) and metabolite profiling was conducted in the rat. Additional PK evaluation as well as tolerability and toxicity testing was performed in the mouse, as this species will be used in future proof of efficacy studies. C16 is a chalcone, a chemical class to which various biological activities have been attributed, albeit often at high concentrations. Despite this body of literature, follow-up studies on the DMPK properties of these molecules are rare, and it thus often remains unclear whether sufficiently high plasma levels can be obtained for a sustained period in order to assess *in vivo* actions. The present study may therefore have relevance to research on other members of the chalcone chemical class. # Materials and methods Additional protocols are available in. ## Statistics Data was processed using IBM SPSS Statistics 23 or higher. All data were tested for normality (Shapiro-Wilk test) and homogeneity of variances (homoscedasity, Levene's test). If both conditions were met, a one-way ANOVA with post-hoc Tukey hsd was conducted. If the data was not normally distributed, then a non- parametric Kruskal-Wallis H test or Mann-Whitney U test was performed. If the data was not homoscedastic, Welch’s t-test and a post-hoc Gamess-Howell test were conducted. ## Animal welfare All aspects of this work related to *in vivo* experiments, including housing, experimentation, and animal disposal were performed in general accordance with the *Guide for the Care and Use of Laboratory Animals* (Eighth Edition). Relevant protocols were approved by the Committee on the Ethics of Animal Experiments of Ghent University (Permit Number: ECD 15/59). The persons who carried out the described experiments received appropriate training in animal care and handling. Animal numbers were kept low due to the exploratory nature of the study. Exact numbers were based on relevant literature work using similar or identical animal studies. Animal health was monitored as indicated in the individual protocols, including during every animal manipulation. Throughout all studies, the following criteria were used to remove an animal from the study and humanely euthanize to prevent undue pain or distress: inability to eat and/or drink, inability to thermoregulate, weight loss (\>20%), moribund condition, prolonged bleeding, seizures, paralysis. These endpoints were not reached. No unexpected deaths occurred in any of the described studies. Animals were euthanized by CO₂, except for those that underwent cardiac exsanguination after Avertin anesthesia. Analgesics or anesthesia were administered as indicated in the individual protocols. ## Chemicals C16 was prepared as described earlier. A tribromoethanol injectable solution (Avertin) was prepared as described in Section E in. In-house prepared Milli-Q water was used. Cremophor EL (Sigma, Germany), corn oil (Wako, Japan), dimethyl sulfoxide (DMSO, Sigma-Aldrich, Germany) anhydrous N,N-dimethylacetamide (DMA, Wako, Japan), ethanol (Merck, Germany), Medigel Sucralose (2 oz cups, ClearH2O, USA), 2-methyl-2-butanol (amylene hydrate, Sigma-Aldrich, Germany) 1,2-propanediol (propylene glycol, PG) (Wako, Japan), polyethylene glycol (PEG) 400 (Sigma, U.S.A), PEG 600 (Alfa Aesar, Great Britain or Sigma, USA), sodium heparin (5000 units/mL injectable solution, LEO Pharma, Belgium) Solutol HS-15 (BASF, Germany) 2,2,2-tribromoethanol (Sigma-Aldrich, Germany) were used as obtained from the indicated suppliers. ## Vehicle development ### Nephelometry The solubility of C16 was tested at 10 and 30 mg/mL for the indicated vehicles using laser nephelometry in 96-well plates (BMG LabTech NEPHELOstar microplate reader, BMG LabTech, USA). Solutions were prepared using solubilizing techniques including sonication, heating up to 37 °C and vortex. Solubility data was categorized as soluble (result ≤ +15 of blank), slightly soluble (result between +15 and +25 of blank) or not soluble (result \> +25 of blank). ### Preparation of a C16 solution in 20% DMSO / 10% (DMSO / Cremophor EL 1:1) / 70% H₂O For a 2 mg/mL final concentration, 240 μL of DMSO was added to 2.4 mg of C16 in a glass vial. The mixture was vortexed resulting in a clear solution. Of this stock solution, 210 μL was mixed with 105 μL of a 1:1 DMSO / Cremophor EL mixture and 735 μL of water, yielding a clear solution. ### Preparation of a C16 solution in 10% Solutol HS-15 / 90% PEG 600 Solutol HS15 and PEG 600 were warmed to 37 °C. The pre-calculated volume of Solutol HS15 was added to C16. The mixture was vortexed and then kept at 37 °C with sonication until C16 had completely dissolved. PEG 600 was then slowly added and the mixture was vortexed again, whereupon the solution was visually clear. Solutions up to 30 mg/mL were prepared in this way. ### Stability test of C16 in 10% Solutol HS -15 / 90% PEG 600 Solutions of 30 mg/mL and 10 mg/mL of C16 were prepared and stored at room temperature and 4°C for 24 h. The stability of C16 was evaluated at 0, 1, 2, 4, 8 and 24 h after formulation by high-performance liquid chromatography-mass spectrometry/mass spectrometry (HPLC-MS/MS) (electrospray ionization, ESI+) analysis using oxybutynin as an internal standard (IS). Analyte samples were diluted 500x with acetonitrile (ACN)/H₂O (20:80). This solution was further diluted 100x with an oxybutynin solution (1 ng/μL in ACN/H₂O 20:80), giving a final concentration of 0.6 ng/μL and 0.2 ng/μL, respectively, for HPLC-MS/MS analysis (multiple reaction monitoring scan mode, MRM, *see* Section A of for more details). All analyses were performed in triplicates. ### Preparation of C16-doped medicated gel A pre-weighed amount of C16 was dissolved in 1 mL of DMSO and the resulting solution was delivered into a cup of Medigel sucralose (pre-warmed to room temperature) with a syringe through the foil lid. In order to facilitate even distribution of C16 in the gel, the injection was spread over several points. The resulting suspension of C16 in gel was dispersed evenly by vigorous shaking for at least two minutes. Visual control confirmed a uniform and finely dispersed suspension of C16 particles in the gel. The final DMSO concentration in the gel was 1.64%. ### Tolerability test of vehicles Vehicles of 10% Solutol HS-15 / 90% PEG 600 and 10% DMA / 90% PEG 600 were evaluated for tolerability by oral gavage at 10 mL/kg using the same protocol as for the single dose *per os* (PO) maximum tolerated dose (MTD) study (*see below*). The tolerability of DMSO-doped medicated gel was evaluated during the repeat-dose toxicity test. ## PK studies ### Plasma PK study (IV and PO) in male Sprague-Dawley rats Six male Sprague-Dawley rats (200–350 g) were obtained from Hilltop Labs. Animals were assigned randomly to two groups (IV or PO) of 3 upon arrival. Duration of acclimation was approximately two days. Animals were healthy at the start of the trial. Animals subjected to IV dosing were fitted with a jugular vein cannula (JVC) under isoflurane anesthesia (induction 4%, maintenance 2.0%, oxygen 1 L/min) and under an external heating source. Post-operative analgesia was provided in the form of 2.5–5 mg/kg of ketoprofen subcutaneously. Benzylpenicillin (60 mg) was given intramuscularly to prevent infection. Sterile 0.9% NaCl was injected subcutaneously under the back skin using a 21G needle during recovery to replace fluid loss. Animals were placed under a heat lamp and monitored until they regained full consciousness. Animals were returned to a sterile cage in the animal care room only after full recovery and when exhibiting normal behavior. Surgically modified animals were housed individually, monitored hourly for the first 4 h and then daily over a 2-day recovery period. Animals were maintained in a well-controlled temperature (20–24°C) and humidity (30%-70%) environment with 12 h light/dark cycles. Conventional cages with wood chip bedding were used. Animals were identified by a cage label. The study was not blinded. Food was withheld from all animals for a minimum of twelve hours prior to test article administration and returned at approximately 4 h post- dose. Water was supplied ad libitum. C16 was dosed in cassette with two undisclosed compounds of similar chemical structure. Dosing solutions were prepared in 100% DMSO on the day of dosing. The concentration of C16 was 1 and 5 mg/mL for IV and PO dosing, respectively. C16 was administered at 1 mg/kg IV via JVC or at 10 mg/kg PO. Sampling took place pre-dose, 5, 15, 30 min, 1, 2, 4, and 8 h post-dose. Blood samples (0.3 mL) were collected via the JVC, placed into chilled tubes containing sodium heparin and kept on ice until centrifugation. Plasma preparation involved centrifugation at a temperature of 2 to 8°C at 3,000 *g* for 5 min. Plasma samples were stored frozen at -70°C until analysis. No necropsy was performed. ### Whole-blood PK study (IV, IP and PO) in male CD-1 mice Nine male CD-1 mice (25–35 g) were obtained from Hilltop Labs. Duration of acclimation was at least two days. Animals were randomly assigned to the three dose groups (IV, PO, intraperitoneal (IP), *N* = 3 per group), healthy at the start of the trial, housed one per cage (conventional type with wood chip bedding material), identified by a cage label and maintained in a well- controlled temperature (20–24 C) and humidity (30%-70%) environment with 12 h light/dark cycles. Animals intended for IV dosing were fitted with a jugular vein cannula (JVC, procedure see rat plasma PK study above). The study was not blinded. Food was withheld from the animals for a minimum of 12 h prior to test article administration until 4 h post-dose. Water was offered ad libitum. C16 was dosed in cassette together with two undisclosed compounds of similar chemical structure. The dose of each test article was 1 mg/kg (IV, JVC, 0.5 mg/mL) or 10 mg/kg (IP and PO gavage, 2 mg/mL). Two respective dosing solutions were prepared (protocol *see above*) containing 0.5 and 2 mg/mL of C16 (and equal concentrations of the other two test articles) in 20% DMSO / 10% (DMSO / Cremophor EL 1:1) / 70% H₂O. Formulations were prepared one day prior to dosing and stored at 4 °C. Blood samples were taken pre-dose and 5 (except for PO administration), 15, 30 min, 1, 2, 4, and 8 h after dosing. The samples were collected via the tail vein or by cardiac puncture (8 h time point). Blood (25 *μ*L) was collected and pipetted into a tube with 25 μL heparinized water (1 *μ*L of sodium heparin (1000 unit/ml) + 24 *μ*L water) within 30 seconds of collection. The sample was pipetted up and down five times, immediately frozen on dry ice and stored at -60°C to -80°C until analyzed. No necropsy was performed. ### Data processing Pharmacokinetic parameters were estimated by a non-compartmental model using WinNonlin (v5.2.1 or higher) software. The maximum plasma/blood concentrations (*c*₀) after IV dosing were estimated by extrapolation of the first two time points back to t = 0. The maximum blood concentration (*c*_max) and the time to reach maximum blood drug concentration (*t*_max) after PO dosing were derived from the data. The area under the time-concentration curve (AUC) was calculated using the linear trapezoidal rule with calculation to the last quantifiable data point, and with extrapolation to infinity if applicable. Plasma/blood half-life (*t*_1/2) was calculated from 0.693/slope of the terminal elimination phase. Mean residence time (MRT) was calculated by dividing the area under the moment curve (AUMC) by the AUC. Clearance (CL) was calculated from dose/AUC. Steady-state volume of distribution (*V*_ss) was calculated from CL\*MRT. Bioavailability was determined by dividing the individual dose- normalized PO AUC values by the respective average dose-normalized intravenous (IV) AUC value. Samples that were below the limit of quantification were not used in the calculation of averages, and were treated as zero for pharmacokinetic data analysis. ## Stability and metabolism For other stability and metabolite experiments, *see* Section C in. ### Whole blood stability and metabolite profiling C16 (1 *μ*M) or vehicle (DMSO) were incubated with fresh rat blood (Sprague Dawley, male) at 37°C at 5 time points in duplicate over a 120-min period. The final DMSO concentration in the incubation was 0.25%. Reactions were terminated following 0, 15, 30, 60 and 120 min by acetonitrile containing internal standard. The sampling plate was centrifuged (3000 rpm, 45 min, 4 °C) and the supernatants from each time point analyzed for parent compound by HPLC-MS/MS. The percentage of parent compound remaining at each time point relative to the 0-min sample was then calculated from HPLC-MS/MS peak area ratios (compound peak area/internal standard peak area). Following the blood stability assay the time point at which 30–70% of parent had degraded (15-min sample) was analyzed by HPLC-MS over a mass range of 200–1000 Da, using an extended run time. Potential metabolites were identified by searching against expected biotransformations. Unexpected metabolites were also searched for by comparison to a control sample (0-min time point and/or negative control) run under the same analysis conditions. Metabolites had a ratio of at least two compared to a corresponding peak in the control sample. The metabolite sample was re-injected and data collected for the product ions in the mass window of 50 Da to a point above the metabolite mass, with the precursor ion fixed on the metabolite mass in the time window observed for the metabolite in the original chromatogram. MS/MS spectra of the metabolites were collected and processed into a table representing product ion fragments for the parent and each metabolite. ## Single-dose PO and IP MTD studies Male ICR mice (20–26 g), were obtained from BioLASCO Taiwan. Animals were kept in conventional cages with wood chip bedding. Space allocation for 3–5 animals was 30 x 19 x 13 cm. All animals were healthy at the start of the trial and maintained in a well-controlled temperature (20–24°C) and humidity (30%-70%) environment with 12 h light/dark cycles. Free access to a standard lab diet \[MFG (Oriental Yeast Co., Ltd., Japan)\] and autoclaved tap water were granted. A single PO and IP dose progression test was conducted in which animals were dosed at a minimum of 72- or 24-h intervals, respectively. C16 (vehicle: 10% Solutol HS-15 / 90% PEG 600) was administered by oral gavage at 10 mL/kg to randomly assigned groups of 2 male and 2 female mice, or by IP injection at 5 mL/kg to randomly assigned groups of 5 male mice. Animals received an initial dose of 10 mg/kg. If 50% of the animals survived for 72 h (PO) or 30 min (IP), the dose for the next cohort was increased. If one or more animals died, the dose for the next cohort was decreased. The next dose level was determined by the decision scheme in. The testing stopped when all animals survived at the upper bound, when four (PO) or three (IP) dose levels had been tested or when the upper or lower bound had been reached. At each dose level, animals were observed for the presence of acute toxic symptoms and autonomic effects during the first 60 (PO) or 30 (IP) min, and again at 2, 6 (IP only), 24, 48 (PO only) and 72 h (PO only). Body weights were recorded before dosing and at the 72- (PO) or 24-h (IP) time point. Gross necropsy was performed in all animals without tissue collection. ## 7-Day repeat-dose toxicity study Fifteen 7-week old male CD-1 mice, weighing 27 to 34 g, were obtained from Janvier, France. Animals were randomly allocated to one of four cohorts (*N* = 4 for the 0 mg/kg, 100 mg/kg and 300 mg/kg C16 cohorts, *N* = 3 for the water control group). Animals were healthy at the start of the trial. Each cohort was housed in one conventional cage with wood chip bedding and kept in a room with a well-controlled temperature (20–24°C) and humidity with 12-h light/dark cycles. Duration of acclimation was five days, during which all animals had ad libitum access to feed and tap water. Feed was obtained from Carfil, Belgium (Complete feed for rats and mice, article n° 10783915). Animals were ear-marked under isoflurane anesthesia (induction 5%, maintenance 2.5%, oxygen 1 L/min) on day -3 of the study. The study was not blinded. ### Study design On the morning of day -2 of the study, tap water access was removed from the 0 mg/kg, 100 mg/kg and 300 mg/kg C16 cohorts. These groups were granted access to two non-medicated cups of Medigel sucralose (one in a tailor-made plastic holder with magnetic fixation system to the side of the cage, one in a cardboard holder on top of the bedding material) for an acclimatization period of three days. Note: the magnetic fixation system is preferable, as animals tear and consume the cardboard holders. Medicated gels were prepared in order to provide animals with a PO C16 dose equaling oral gavage dosing at 0, 100 and 300 mg/kg three times per day (0, 78.43 and 235.29 mg C16 per cup, respectively, assuming a consumption of 7 mL of gel per day and an average animal weight of 30 g). On the morning of day 0 of the study, treatment with two medicated gels was started and maintained up to the morning of day 7 (i.e. a 7-day treatment). Fresh cups were prepared and provided every day. Meanwhile, the water control group was continuously granted unlimited access to tap water. Throughout the study, all animals had ad libitum access to feed, except for the water control group, which was fasted on the evening of day 6, i.e. twelve hours prior to the single PO test article administration. Animal weight, feed and water/gel consumption and the presence of toxic symptoms and autonomic effects were monitored on a daily basis for all cohorts. On the evening of day 6, blood was collected via the tail vein for the 0, 100 and 300 mg/kg C16 cohorts for determination of the whole blood C16 level. On the morning of day 7, mice of the 0, 100 and 300 mg/kg C16 cohorts were anesthetized by IP injection with 250 mg/kg of Avertin (20 mL/kg, preparation *see* Section E). Blood for C16 determination and hematological and biochemical analysis was collected via cardiac exsanguination (600 μL to 1 mL). Detailed analysis protocols are described below. Afterward, animals were sacrificed by cervical dislocation. On the morning of day 7, the water group received a single PO administration of 300 mg/kg C16 (10 mL/kg) for high-dose PK evaluation, using a 30 mg/mL solution of C16 in 10% Solutol HS-15 / 90% PEG 600 (prepared as described above). To that extent, blood was collected via the tail vein at the 30 and 60 min time point, and via cardiac exsanguination after 180 min (protocol as for other cohorts). Samples for hematological and biochemical analysis were also prepared. Afterward, animals were sacrificed by cervical dislocation. All animals were necropsied and assessed for gross pathology. Liver, kidneys and spleen weights were recorded, and fragments of these organs were fixed in a 4% formaldehyde solution and processed for microscopic evaluation. ### Whole blood C16 determination Blood samples (25 *μ*L) were collected via the tail vein or cardiac puncture (last time point). Blood was pipetted into a tube with 25 *μ*L heparinized water (1 *μ*L of sodium heparin (1000 units/mL) + 24 *μ*L water) within 30 seconds of collection. The sample was pipetted up and down five times, immediately frozen on dry ice, and stored at -60°C to -80°C until analysis. ### Hematology and biochemical analysis Cardiac puncture as described above afforded 0.6–1 mL of blood. Approximately 500 *μ*L thereof was collected in a Greiner Bio 0.5 mL ethylenediaminetetraacetic acid (EDTA) Eppendorf tube for hematological parameter determination and kept at room temperature. The remaining volume was collected in a regular Eppendorf tube and allowed to clot completely at room temperature over 45 min. The tubes were centrifuged at 1,900 x *g* (3000 rpm) and 4 °C for 10 min using a swinging bucket rotor. The upper serum phase was transferred to a new tube with conical bottom. Care was taken not to disturb the intermediate buffy coat layer. The serum samples were centrifuged at 16,000 x *g* and 4 °C for 10 min in a fixed-angle rotor. The cleared supernatant was transferred to a new tube without disturbing the pellet, affording 200–500 *μ*L of serum. Samples were diluted to 500 *μ*L with normal saline, flash frozen in liquid N₂ and stored at -80 °C prior to analysis. Hematological analysis was conducted on an XE-5000 Automated Hematology System (Sysmex, USA), biochemical analysis on a cobas 8000 series system (Hitachi/Roche, USA). # Results ## Vehicle development ### Vehicles for single-dose studies C16 has low aqueous solubility, which limits the choice of preclinical formulations for *in vivo* studies. In this work, thirteen vehicles were evaluated for their applicability in single-dose studies. For commonly used systems (entries a-e) visual determination of solubility was conducted, while for the more complex systems solubility was assessed by laser nephelometry. As expected, C16 proved highly soluble in DMSO (entry a). A more preferable formulation with 25% DMSO in Cremophor and water (entry e) also proved suitable at low concentrations (up to 2 mg/mL). These vehicles are appropriate for single-dose studies with low amounts of C16, e.g. in PK studies. Dimethylacetamide / 90% PEG 600 and Solutol HS-15 / PEG 600 gave clear solutions at 10 and 30 mg/mL (entries p,q and t,u). These potential high-dose vehicles were further evaluated for tolerability and stability. Solutions of 10 and 30 mg/mL C16 in 10% Solutol HS-15 / 90% PEG 600 proved stable at both room temperature (20–25 °C) and 4°C over 24 h (ANOVA/Tukey HSD or Welch/Gamess- Howell, Table A). No adverse signs were elicited by both vehicles on PO administration (*see section* ‘*Single dose PO and IP MTD studies’*). IP dosing of the more common of these two vehicles, Solutol HS-15 / PEG 600, at 5 mL/kg induced mild adverse effects (decreased exploratory behavior, decreased muscle tone) within 30 min after administration (*see section* ‘*Single dose PO and IP MTD studies’*, 10% DMA / 90% PEG 600 not tested). 10% Solutol HS-15 / 90% PEG 600 thus is a valid vehicle for high-dose experiments, e.g. MTD studies. ### Vehicle for repeat-dose studies For repeated oral dosing over an extended period of time, a formulation of C16 in medicated gel (Medigel Sucralose, ClearH2O) was developed as an alternative to repeated oral gavage. Medigel Sucralose is a non-wetting sucralose-flavored, low calorie water gel with a sweet taste. The chosen formulation as a medicated gel is a stable suspension of C16 particles in water. The stability of this dispersion at room temperature was confirmed over 144 h ( and Tables C-D), and the compound proved uniformly distributed (*see* Table E). The tolerability of Medigel Sucralose had been explored by others. Observations on the tolerability of DMSO in the gel (this co-solvent is used as a solubilizing agent during preparation) are mentioned in the discussion of the repeat-dose toxicity test. ## PK parameters of C16 in the rat and mouse C16 plasma pharmacokinetics were determined in male Sprague-Dawley rats after intravenous and oral administration at 1 and 10 mg/kg, respectively (Tables F-G and Figs B-C). In this exploratory experiment, C16 and two other (undisclosed) compounds were dosed in cassette in 100% DMSO. No adverse effects were observed after the intravenous and oral administration of the materials. Following intravenous dosing of C16 at 1 mg/kg, a *c*₀ of 1257 ng/mL was reached with an average half-life of 1.93 h. The average clearance was 4.95 L/h/kg, the volume of distribution 4.77 L/kg. After oral dosing at 10 mg/kg, a *c*_max of 32.2 ng/mL and oral bioavailability of 6.91% was found. The half-life value of 3.67 h was longer than that observed after IV administration. Next, the whole-blood pharmacokinetics of C16 in male CD-1 mice were evaluated after intravenous (1 mg/kg), intraperitoneal (IP, 10 mg/kg) or oral (PO, 10 mg/kg) administration in cassette dosing with two other (undisclosed) compounds (Tables H-J and Figs D-F). In this experiment, 20% DMSO/10% (DMSO/Cremophor EL 1:1)/70% H₂O was used as a more preferable vehicle with respect to pure DMSO. No adverse reactions were observed. Following intravenous (IV) dosing at 1 mg/kg, C16 had a half-life of 0.746 h, a clearance rate of 45.0 L/h/kg, and a volume of distribution of 32.6 L/kg. Following IP and PO dosing at 10 mg/kg, C16 reached a maximum blood concentration of 107 and 12.5 ng/mL, respectively, both at 15 min post dosing. The average IP and PO bioavailability was 54.7% and 10.5%, respectively. ## Stability and metabolism Interpretation of the above PK data required the generation of additional *in vitro* stability and metabolism data, and revision of known data. As illustrated above, solubility and lipophilicity are acceptable for exploratory work *in vivo* when using suitable vehicles. C16 has a high Caco-2 classification, with a discrepancy in recovery for the A→B and B→A-directions. The compound is stable in simulated intestinal and gastric fluid and in PBS. A much higher intrinsic clearance was obtained for rat liver microsomes in comparison with earlier results for humans. C16 does not significantly inhibit three major drug-drug interaction (DDI)-inducing human cytochromes P450 (CYPs, 1A2, 2D6 and 3A4). Despite the high protein binding of 99%, C16 has a high blood-to-plasma ratio of 9.2. A remarkable difference in half-life in rat plasma and whole blood was obtained. In the whole blood stability assay, comparison of the 15-min sample against the 0-min control revealed one major metabolite (, Tables M-P). HPLC-MS/MS profiling showed that this compound had the mass of a C16-glutathione conjugate. Remarkably, also the *Z*-isomer of C16 was detected in this experiment. Fast partial isomerization apparently occurred under the conditions of this experiment, as the configuration of C16 was initially purely *E* (confirmed by nuclear magnetic resonance (NMR) analysis). This isomerization was also observed after prolonged dissolution in Medigel Sucralose (*see below* and Table C). A similar isomerization was also made by Gutteridge et al. for two other chalcones. Both the *E*- and *Z*-isomer of C16 showed a comparable short half-life in rat whole blood. The extent to which this isomerization occurs *in vivo* is unknown, yet it does not cause tolerability issues (*see below*). ## Single-dose PO and IP MTD study The MTD of C16 in mice, using 10% Solutol HS-15 / 90% PEG 600 as a high-dose vehicle, was determined for both PO and IP administration. PO administration (oral gavage) was evaluated in male and female ICR mice (*N* = 2 per group) using a starting dose of 10 mg/kg. The compound was tolerated at 10–300 mg/kg though mild adverse effects such as vocalization and increase in sensitivity to touch were observed at 100 and 300 mg/kg (*see also* Section D). All animals survived during the 72-h observation period and no gross lesions were discovered after necropsy. Body weight gain was not affected between cohorts of the same sex (Gamess-Howell). C16 was also administered intraperitoneally (IP) to a group of 5 male ICR mice. The vehicle 10% Solutol HS-15 / 90% PEG 600 at 5 mL/kg induced adverse effects such as decreased exploratory behavior and decreased muscle tone within 30 min after administration ( and Section D). Additional effects such as decreased sensitivity to touch, decreased exploratory behavior, decreased spontaneous activity, decreased muscle tone, deep respiration, decreased palpebral size and hunch back were elicited by C16 at the three doses levels (10, 30 and 50 mg/kg), with more severe effects observed at higher doses. The effects were reversible, and all animals survived without significant changes in food consumption across groups (Kruskal-Wallis) and with moderate weight decrease during the 24-hour experimental period. Only between the 10 and 50 mg/kg cohorts, a significant difference in average weight decrease was noted (*p* = 0.037, Kruskal-Wallis). In addition, no significant changes were observed during gross necropsy. ## Repeat-dose toxicity study Finally, an assessment was conducted of the suitability of the medicated gel as a vehicle for oral delivery, and of the toxicity of C16 on continuous dosing via this route of administration. Animals were given the equivalent of *ter in die* (TID, three times per day) dosing of 0, 100 and 300 mg/kg of C16 in Medigel Sucralose (1.64% DMSO) over a 7-day period. The 300 mg/kg value was chosen as it was the highest evaluated dose in the single-dose MTD study. As a control, a fourth cohort received regular drinking water during the same period. Overall, the Medigel vehicle and compound were well-tolerated (raw data is available in Section E). All animals survived during the 7-day observation period. As seen in the single-dose MTD study, mild behavioral changes were observed in the C16-treated cohorts. From day 2 on, animals exhibited hyperactivity and increased anxiety on opening of the cages. This behavior was absent in the 0 mg/kg and the water groups. The effects were mild, ranging from slightly noticeable in the 100 mg/kg cohort to noticeable in the 300 mg/kg cohort. The magnitude of these effects did not increase in time. No significant difference in average weight between the four cohorts was noted during the study (Tukey HSD or Kruskal-Wallis). A slight but non-significant decrease in weight was observed during the acclimatization period for all three Medigel cohorts with respect to the water cohort, indicating slight neophobia towards the non-medicated gel. Animals recovered fast from day 0–1 on, despite the start of the treatment (no neophobia towards C16 or DMSO). Average food intake per animal was around 6 g/day with a slight upward trend over the test period in all cohorts. Throughout the study, no significant differences in average food consumption were observed between study groups (Tukey HSD). Decrease of gel per day was between 8 and 16 g/animal, and did not differ significantly between the gel cohorts (Tukey HSD). An identical, slowly increasing trend in consumption was observed. The actual gel consumption lies a fraction lower than the reported amounts because of loss of gel due to sanitation of the cups pre-weighing. We estimate that the actual consumption will be close to that of the water cohort, which is commensurate with the manufacturer’s findings. No gross lesions were discovered on detailed necropsy. Liver and spleen weights were identical across the Medigel-treated cohorts, while a slight but significant decrease in kidney weight was noted for the 300 mg/kg cohort with respect to the 100 mg/kg and to the 0 mg/kg cohort (, ANOVA with post-hoc Tukey HSD, *p* = 0.001). No comparison was made with the water cohort, since these animals were fasted overnight prior to necropsy. Histology of the kidney and spleen revealed no abnormalities. A slight hepatocyte toxicity (centrilobular) was noted in some animals of the Medigel-treated groups, but not in the water cohort. Clinical chemistry and hematology parameters were determined in all cohorts at the end of the study. All measurements fell within or were close to normal ranges (*see* Table Z). A (marginally) statistically significant difference between cohorts was only noted for the glucose measurements of the treatment groups versus the blank. This result was expected, since only the blank animals were fasted overnight prior to analysis. On the last evening (day 6) and last morning (day 7) of the study, blood samples were taken from the Medigel-treated cohorts to determine the systemic availability of C16 via this route of administration. At both time points, whole blood C16 concentrations were above its projected EC₅₀. No difference was found in C16 levels between the 100 mg/kg and 300 mg/kg cohort, in both the evening (ANOVA) and morning (Mann-Whitney U test) samples. As expected from their dietary pattern, average C16 levels were higher in the morning than in the evening. Nevertheless, this difference was not statistically significant for the 100 mg/kg cohort (*p* = 0.200, Mann-Whitney U test) and only borderline significant for the 300 mg/kg group (*p* = 0.057, Mann-Whitney U test). A final aspect of this 7-day study was a comparison between attainable plasma levels of C16 on administration *via* gel and oral gavage. On the final morning of the study, animals in the water cohort received a single PO administration of 300 mg/kg of C16 in 10% Solutol HS-15 / 90% PEG 600 via oral gavage. Blood samples were collected after 30 min, 1 h and 3 h. Maximum attainable concentrations via this route of administration were not higher than via the medicated gel. Extrapolation of the time-concentration curve obtained for oral gavage reveals the need for a dosing frequency of at least three times per day in order to obtain sustained levels above the projected EC₅₀ of the compound. # Discussion 4-Fluoro-3’,4’,5’-trimethoxychalcone or C16 is a small molecule with nanomolar anti-invasive activity, a defined structure-activity relationship and a satisfactory ADME and toxicity profile *in vitro*. In order to examine its potential as a pharmacological tool, prior exploration of its *in vivo* pharmacokinetic profile, toxicity and tolerability was required. The obtained data is also relevant to several other members of the chalcone chemical class, which have been attributed biological activities (mostly at elevated concentrations), but for which associated DMPK profiling is lacking. ## Vehicle development DMSO is an excellent solvent for C16 but its applicability *in vivo* is limited. Other formulations were therefore evaluated. For low single dose studies, DMSO/Cremophor/water was identified as a suitable vehicle, while for high single-dose evaluation, Solutol HS-15 / PEG 600 proved appropriate. C16 is stable when dissolved in Solutol HS-15 / PEG 600, and the vehicle is well tolerated in the mouse when given *via* oral gavage. It however generated moderate adverse effects on IP administration, which prohibits its IP use in chronic dosing studies. The suboptimal DMPK properties of C16 (*see* below) would necessitate at least three oral gavage administrations per day to maintain the desired plasma levels in chronic experiments. We therefore turned to self-medication, as this reduces animal handling and stress, and does not disrupt the diurnal rhythm of the animals. It also reduces the burden on the investigator significantly. Administration of C16 via regular drinking water was not possible, as the compound crashes out of solution and moreover is unstable in an aqueous environment upon prolonged dissolution. At the start of this study, a limited number of reports on the use of medicated gels as a route of administration were available. Medicated gels have advantages over mixing in the feed, as new batches can be prepared on a flexible basis and without the need for special equipment. Daily gel intake is comparable to that of regular drinking water, the gel can thus serve as a complete replacement of drinking water over prolonged periods of time. A major aim of this study was the development of a C16-doped gel as a vehicle for chronic dosing. A suitable formulation was developed in the form of a stable suspension of C16 in Medigel Sucralose. The compound proved stable in this preparation over 144 h, and was shown to be uniformly spread. Earlier reports had validated the tolerability of Medigel Sucralose in rodents. Validation of C16-doped Medigel Sucralose for use in mice was conducted in a repeat-dose toxicity study (described below). ## Pharmacokinetics The IV, IP and PO pharmacokinetic profile of C16 was determined in rat plasma and mouse whole blood. Regardless of the vehicle (DMSO or DMSO/Cremophor/water), sampling (plasma or whole blood) and species (rat or mouse), the compound shows low (PO) to moderate (IP) bioavailability. In both cases, slow dissolution is a limiting factor, resulting in a similar t_max for IP and PO dosing, prolonged exposure and rather flat (rat) or multiple peak (mouse) concentration- time profiles on PO dosing. The use of cassette dosing may have accentuated this phenomenon, though the main cause can be found in the physicochemical profile of C16. It should be noted that the significance of the average t_max for PO dosing in rats is limited, as a flat average concentration-time profile was obtained. Despite its good permeability, the low A→B recovery of C16 in the Caco-2 assay suggests cellular metabolism, which also limits oral bioavailability. While the high extent of protein binding may also necessitate elevated oral dosing, the high recovery levels suggest reversible binding. Regardless of the sampling mode (plasma or whole blood), both rat and mouse half-life on IV dosing are short. This is in line with data published for other chalcones. Nevertheless, an important difference in initially attained IV concentrations was noted, which can be attributed to species and sampling factors. Clearance in both species is much higher than hepatic blood flow (rat: 3.3 L/h/kg, mouse: 5.4 L/h/kg), which points towards extra-hepatic elimination. C16 shows high apparent blood partitioning and a large difference between plasma and whole blood half-life. Rather than irreversible binding to erythrocytes, the compound was found to be metabolized rapidly with glutathione conjugation as the major mechanism. Erythrocytes are a major site of biotransformation and contain cytosolic glutathione S-transferases. The conjugates are then actively transported outward across the red-cell membrane. Metabolism in red blood cells thus seems the major route of extra-hepatic elimination for C16, explaining its much shorter half-life in blood versus plasma. This route of metabolism was not observed for other rapidly metabolized chalcones such as cardamonin. As a further consequence, the higher reported blood-to-plasma ratio is not accurate because it was not based on measurements of both blood and plasma fractions. ## Tolerability, toxicity and chronic systemic availability The maximum tolerated single dose of C16 in the mouse on administration via oral gavage is \>300 mg/kg. The compound was well tolerated, and induced only mild behavioral effects at 100 and 300 mg/kg (vocalization and increased sensitivity to touch). The pre-defined humane end points (*see* *section*) were not reached. No other treatment-induced phenomena were noted on observation or necropsy. The animals proved much more sensitive to intraperitoneal delivery of a single dose of the same vehicle and compound, though humane end points were again not reached. These observations can be correlated to a low tolerability of the vehicle itself, and the earlier observed higher C_max for C16 on IP delivery with respect to oral administration. Moderate to severe behavioral, neurologic and autonomic effects were noted in a dose-response pattern. These results indicate that the vehicle and C16 up to 50 mg/kg may be tolerated for single-dose IP administration in the mouse, but not for chronic experiments. Despite its low oral bioavailability, the high oral tolerability of C16 offers opportunities to obtain sufficiently high systemic levels for efficacy testing. This hypothesis was tested in a repeat-dose experiment. The 7-day continuous dosing experiment validated Medigel sucralose as a vehicle for self-medication of C16 in a chronic setting. Substitution of drinking water for blank gels caused a slight but non-significant decrease in weight during the acclimatization period due to neophobia. During the further course of the experiment, however, treatment with solvent-doped gels (DMSO) did not generate significant differences in weight, water and food intake, and hematological and biochemical parameters with respect to the water cohort. No gross lesions or differences in organ weights were noted on necropsy. A slight hepatocyte toxicity, however, was noted for some gel-treated animals, which was absent in the water cohort. This effect may be due to the relatively high DMSO content of the gel. DMSO hepatotoxicity has been reported in chronic trials in laboratory animals. In our setting the effect was not major, nor is it reflected in the clinical chemistry panel. The DMSO dose was also well below the LD₅₀ in the mouse (21–28 g/kg, PO). Overall, Medigel Sucralose thus proved a suitable vehicle for chronic studies. Oral C16 dosing via medicated gels over a 7-day period caused sustained systemic levels above its projected EC₅₀ value (6.32 ng/mL). This is a significant result, as a major reason for poor correlation between *in vitro* and *in vivo* efficacy of chalcones has been their poor pharmacokinetics and bioavailability, yielding plasma levels below the EC₅₀ of the compounds. Gel intake was close to the anticipated amount for all cohorts, resulting in an actual daily oral dose close to the intended. C16 showed nonlinear uptake as systemic levels did not differ significantly between the 100 mg/kg and 300 mg/kg cohort. Given the physicochemical profile of C16, this is likely due to limits in dissolution. Lower dosing regimens may thus still generate sufficient systemic levels, but were not evaluated in this work. Importantly, as expected from the low-dose PK studies, the high-dose concentration-time profile obtained after a single administration of 300 mg/kg C16 via oral gavage confirms the need for TID dosing in order to reach sustained levels above the EC₅₀. This underpins the value of the developed medicated gel formulation in the present research and for others working on related molecules. The confirmed seven-day exposure to the molecule at levels above its EC₅₀ did not engender treatment-associated differences between the 0, 100 and 300 mg/kg cohorts, except for a slightly lower kidney weight in the 300 mg/kg cohort. Nephrotoxicity was however not detected in any of the cohorts during histological examination. Hematological parameters were all satisfactory. Some individual measurements fall slightly outside of reported reference ranges, but these ranges itself vary strongly upon the cited source and underlying animal-related variables. The only valid comparison of therapy response thus is to the age-, strain-, and sex-matched controls of the water cohort. No differences were found, except for a lower average glucose reading in the water cohort due to fasting prior to sampling. Oral delivery using our medicated gel formulation thus is a suitable tool for evaluating the efficacy of C16 in a range of invasion and metastasis models. Depending on the models and more specifically on the interrogated phenomena (local invasion, regional or distal metastasis), compatible dosing schedules may include prophylactic treatment over a 7-day period before inoculation (e.g. in a peritoneal metastasis model), intermittent or low-dose chronic treatment in long-term experiments (months) or high-dose treatment in short-term, more aggressive models (weeks). From the data gathered in this report, adequate exposure and good tolerability of C16 can be expected in these settings. # Conclusions C16, like other members of the chalcone family, shows suboptimal pharmacokinetics in the rat and the mouse. Its limited solubility and cellular metabolism causes poor intestinal uptake and limited oral bioavailability. The compound is rapidly metabolized, with a clearance rate exceeding hepatic blood flow. Glutathione conjugation in erythrocytes was identified as a major route of extra-hepatic elimination, explaining the much shorter half-life of C16 in blood versus plasma. An oral dosing routine using medicated gels was developed that solved these issues. A schedule providing amounts equivalent to oral gavage at 100 mg/kg three times per day yielded sustained whole blood levels above the EC₅₀ in a 7-day chronic study. The compound showed good tolerability during the latter experiment. Evaluation of C16 as a pharmacological tool in animal models of metastatic disease, using the medicated gel formulation, is currently ongoing. Besides, we are preparing analogs with a superior DMPK profile in search of improved clinical relevance. # Supporting information The authors thank prof. Marc E. Bracke for helpful discussions regarding the repeat-dose toxicity experiment. ACN acetonitrile ADME absorption, distribution, metabolism, excretion AUC area under the time-concentration curve AUMC area under the moment curve C16 4-fluoro-3’,4’,5’-trimethoxychalcone CL clearance DMA N,N-dimethylacetamide DMPK drug metabolism and pharmacokinetics DMSO dimethyl sulfoxide EC₅₀ half maximal effective concentration EDTA ethylenediaminetetraacetic acid HPLC-MS/MS high-performance liquid chromatography-mass spectrometry/mass spectrometry IP intraperitoneal IS internal standard IV intravenous JVC jugular vein cannula MRM multiple reaction monitoring MRT mean residence time MTD maximum tolerated dose NMR nuclear magnetic resonance PEG polyethylene glycol PG propylene glycol PK pharmacokinetics PO *per os* SAR structure-activity relationship TID *ter in die* V_ss steady-state volume of distribution [^1]: The authors have declared that no competing interests exist.

# Introduction Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia in clinical practice. AF is a well-known risk factor for thromboembolic events, silent cerebral infarcts, strokes, congestive heart failure and mortality in the general population and in high stroke risk patients, and it has become a major public health problem worldwide. Although the underlying mechanisms of AF are not yet fully understood, autonomic dysfunction, unbalanced inflammation/oxidative stress and renin-angiotensin system activation have all been shown to be related to AF. However, the pulmonary vein (PV) is confirmed to be the most important and critical trigger for AF, and PV isolation (PVI) using catheter ablation is the cornerstone therapy for symptomatic AF refractory to antiarrhythmic drugs. The 2016 European Society of Cardiology (ESC) AF guidelines also recommend catheter ablation as a more effective therapy than antiarrhythmic drugs (AADs) for restoring and maintaining sinus rhythm (SR) in patients with symptomatic paroxysmal AF (PAF) and persistent AF (PerAF). Radiofrequency (RF) ablation is a well-established treatment for AF that achieves PVI by creating consecutive, transmural ‘point-to-point’ lesions with heat energy. Recently, cryoballoon (CB) ablation, which achieves PVI by a single-shot deployment of a CB with frozen energy, has become a substitute for RF ablation, as CB ablation has the advantage of being an easier and faster ablation procedure than RF ablation. Several previous studies have shown that the efficacy and safety of CB ablation therapy are comparable to those of RF ablation in patients with PAF. Compared with first-generation equipment, second- generation CB (2G-CB) ablation devices have preponderant cooling capacity and seem to reduce the procedure duration. However, 2G-CB ablation to achieve PVI has some shortcomings, including an inability to perform ablation of roof linear (RL) lesions, complex fractionated atrial electrograms (CFAEs) and non-PV triggers. On the other hand, early studies of PVI involving only patients with PerAF revealed suboptimal success rates. Therefore, the ‘PVI-plus’ ablation strategy that combines 2G-CB ablation to achieve PVI and RF ablation and addresses additional cardiac substrate modification and extra-PV lesions during the same surgery might be a better strategy for treating PerAF and long-standing PerAF. However, the effectiveness and safety of the ‘PVI-plus’ ablation strategy have not been sufficiently discussed. To this end, we executed a pooled analysis and meta-analysis of data from existing studies and trials investigating the efficacy and safety of ‘PVI-plus’ ablation vs. ‘PVI-only’ ablation in patients with PerAF. # Methods Our systematic literature search was performed according to the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) guidelines and conducted using a predetermined protocol by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement. ## Literature search strategy The search strategy was conducted in the PubMed, EMBASE and Cochrane library databases until September 1, 2018. The search terms were as follows: (“Catheter Ablation” OR “Cryosurgery” OR “Second-Generation Cryoballoon”) AND (“Atrial Fibrillation” OR “Persistent Atrial Fibrillation”). No restrictions were applied on regions or languages. We also manually searched the reference lists of all publications and review articles to identify other relevant studies. ## Study selection Two investigators (JS and LXS) independently scanned all the titles and abstracts to identify studies that met the inclusion criteria and extracted data from these studies. Discrepancies between reviewers were resolved by discussion with a third reviewer (MJS). The inclusion criteria were as follows: (1) involved patients with drug- refractory symptomatic PerAF who underwent 2G-CB; (2) involved patients who were treated with a ‘PVI-only’ strategy or a ‘PVI-plus’ strategy using 2G-CB (during surgery, PVI and PVI-plus had to be carried out simultaneously); (3) reported 1-year clinical success rates; (4) involved follow-up (FU) periods longer than 12 months; (5) published as a full-text article; and (6) provided data regarding efficacy and safety. Studies were excluded for the following reasons: (1) conference abstracts, case reports, case series, editorials, or review articles. (2) They did not report clinical success rates. (3) The maximum FU period was shorter than 12 months. (4) They were animal or in vitro studies. ## Data extraction and quality assessment Two primary investigators (JS and LXS) independently evaluated and extracted data from each study. Data on the first author, publication year, study population, age, sex ratio, CHA2DS2-VASc score, underlying disease, medication usage and ablation strategy were collected using a predesigned electronic form. All studies were reviewed twice, and disagreements were discussed and resolved by consensus in a meeting with a third investigator (MJS). We defined the primary outcome criterion as no episode of AF or any atrial arrhythmia lasting longer than 30 s without administration of AADs after a single ablation procedure using 2G-CB with an FU period of at least 12 months (considering an initial blanking period (BP) of 3 months). Group analysis was performed based on the ablation strategy: ‘PVI-only’ strategy versus ‘PVI-plus’ strategy, the latter of which involved PVI plus another substrate modification. The secondary outcomes were complication rates, including the phrenic nerve (PN) palsy (PNP) rate, the rates of cardiac and vascular assess site complications (including hematomas, pseudoaneurysms, and fistulas), and the rates of other complications (pericardial effusions and/ or tamponade, PV stenosis or atrioesophageal fistula). Whenever the data of interest were not available in the literature, the investigators tried to contact the authors by email to obtain the data. We described incomplete data as “not reported” in our manuscript. We used the Newcastle-Ottawa Scale (NOS) to further evaluate the quality of the observational studies, and a NOS score ≥ 7 was considered good quality. ## Statistical analysis Statistical analysis was performed using the Cochrane RevMan version 5.3 software (The Cochrane Collaboration, UK). The results are reported as weighted mean differences and relative risks (RRs) with 95% confidence intervals (CIs) for continuous and dichotomous outcomes, respectively. Heterogeneity was assessed using the *I*^*2* test, Cochran’s Q statistic and the *I*^*2* index. *I*^*2* values of 25%, 25–50%, or 50% indicated low, moderate, or high heterogeneity, respectively. Funnel plot analysis was used to evaluate potential publication bias. In all analyses, a *P* value \< 0.05 was considered statistically significant. # Results ## Study characteristics Finally, as illustrated in, four studies involving a total of 879 patients who underwent ablation with 2G-CB for PerAF were enrolled in this study. The safety and efficacy of a combination of extra-PV lesions (linear ablations and/or atrial substrate modifications) or PVI alone with 2G-CB for PerAF ablation were identified. The characteristics of the included studies are described in. All of the included patients completed at least one year of FU. Five studies were analyzed. Unfortunately, we did not find randomized controlled trials meeting the inclusion criteria. The quality and bias of the included studies are shown in. Four full-text studies had an NOS score of 8, and the other study had an NOS score of 7. The patients were predominantly male (69.5%); 70.1% had hypertension, 15.6% had diabetes, and 13.35% had structural heart disease. In all the studies, 87.6% of patients were treated with oral anticoagulants for PerAF. The mean CHA2DS2-VASc score was 2.1. The mid-term FU (≥ 12 months, considering an initial BP of 3 months) occurred at 27 months. Regarding monitoring during FU, 2 studies implemented 24-h Holter recordings during FU, while two studies recorded data from 7-day Holter or external cardiac event recorders; the other studies recorded data from 2-week mobile cardiac telemetry monitoring performed at 6 weeks, 3\~6 months, 13 months, and 18 months. ## Publication bias Regarding the proportion of patients who were free from AF recurrence during FU, all the included studies had publication bias that was approximately symmetrical on visual inspection of the funnel plot **.** ## Ablation strategy In 3 studies, a bonus freeze protocol was implemented with an additional freezing cycle applied to each PV after successful PVI. Subgroup analysis to compare freezing protocols was not performed because the majority of patients were treated with a bonus freeze protocol. All studies (including 543 patients) performed PVI plus other linear ablations or substrate modifications. There was no significant methodological heterogeneity in terms of patient characteristics and ablation strategies (*I*^*2* = 0%, *P* = 0.82). The result of the study by Aryana et al. demonstrated the ablation of non-PV triggers in 12.8% of all patients treated with 2G-CB without further specifications. The mean procedure time was 131.9 ± 7.2 min in the ‘PVI-plus’ group and 124.2 ± 13.2 min in the ‘PVI-only’ group. No interaction was found between the ablation approach and clinical success rate (*P* = 0.47). However, the results of the study by Su et al. confirmed that application of 2G-CB could achieve large-area atrial substrate modification mostly with left atrial (LA) roofline and Coumadin-ridge ablation (atrial tissue between the LA appendage and left superior PV). Akkaya et al. showed that among patients with PerAF and LA enlargement, PVI with 2G-CB in addition to the creation of RLs may be a viable alternative to point-by-point RF ablation, as indicated by a lower arrhythmia recurrence rate with this strategy than with the ‘PVI-only’ strategy. Aryana et al. demonstrated that ablation of PVI in conjunction with isolation of the posterior LA wall is associated with improved clinical outcomes in certain patients with AF. ## Primary outcome ### Recurrence Data from 5 studies were included. At a mid-term FU of 27 months, the overall success rate of 2G-CB in patients with PerAF was 66.1% ****. In the ‘PVI-plus’ group, the success rate of 2G-CB in patients with PerAF was 73.8%. In the ‘PVI- only’ group, the success rate of 2G-CB in patients with PerAF was 53.6% \[RR: 0.52; 95% CI 0.42\~0.63, *P*\<0.00001\]. The heterogeneity among studies was not significant (*I*^*2*  =  0%, *P* = 0.82). The rate of AAD use at the FU was 48.96%. The rate of AAD use at the FU was reported in 4 studies. However, because of the lack of correlated studies, we could not analyze subgroups in detail. Meanwhile, the heterogeneity among studies was not significant. Categories including PerAF (short-term PerAF, long-standing PerAF or unspecified PerAF), FU strategy (24-h Holter monitoring vs. 7-day Holter monitoring or event recording), and AF population (PerAF alone, mixed AF, such as PAF and PerAF) could not be included in further subgroup analyses. ### One-year recurrence At the one-year FU, compared with that in the ‘PVI-plus’ group, the one-year success rate associated with 2G-CB in the ‘PVI-only’ group was 55.1% (vs. 75.1% in the ‘PVI-plus’ group) \[RR 0.49; 95% CI 0.40\~0.61, *P*\<0.00001\] ****. The heterogeneity among studies was not significant *(I*^*2** * =  0%, *P* = 0.61). ## Secondary outcomes ### Overall complications The overall complication rate was 5.2% ****. Phrenic nerve palsy (PNP)/PN injury were the most frequent complications. Compared with that in the ‘PVI-plus’ group, the complication rate associated with 2G-CB in the ‘PVI-only’ group was 5.4% (*vs*. 5.2% in the ‘PVI-plus’ group) \[RR 0.98; 95% CI 0.57\~1.67, *P* = 0.93\]. ### Phrenic nerve palsy PNP was found in all studies. Transient PNP was defined as PNP that resolved by discharge. Persistent PNP was defined as PNP that persisted beyond discharge. The overall rate of PNP was 2.8% ****. Compared with that in the ‘PVI-plus’ group, the rate of PNP associated with 2G-CB in the ‘PVI-only’ group was 1.8% (*vs*. 3.5% in the ‘PVI-plus’ group) \[RR 1.84; 95% CI 0.83\~4.10, *P* = 0.14\]. Two studies reported persistent PNP that recovered during FU (the maximal time to PN recovery was not reported). ### Vascular access site complications The overall rate of vascular access site complications was 1.6% ****. Compared with that in the ‘PVI-plus’ group, the PNP complication rate associated with 2G-CB in the ‘PVI-only’ group was 2.1% (*vs*. 1.3% in the ‘PVI-plus’ group) \[RR 0.61; 95% CI 0.22\~1.65, *P* = 0.33\]. In one study, 5 patients had a groin complication with a hematoma that required more than 2 weeks for full recovery, and 3 patients had a pseudoaneurysm that required a thrombin injection for patient recovery. Another study reported that 3 patients had minor groin hematomas. ### Other complications No death, myocardial infarction, PV stenosis or atrioesophageal fistula related to the procedure was reported. One patient had a perforation that resulted in cardiac tamponade requiring pericardiocentesis and immediate surgical treatment. One patient had a stroke after ablation and was treated conservatively. # Discussion ## Principal findings In this detailed meta-analysis and systematic review of trials, we evaluated different approaches to 2G-CB in patients with PerAF. The main findings are as follows: (i) the ‘PVI-plus’ strategy had comparable clinical efficacy and safety as the ‘PVI-only’ strategy; (ii) we detected a reduction in the proportion of recurrent AF when either RL ablation, ablation of CFAEs or substrate modification was executed in addition to ablation with PVI; and (iii) compared with ‘PVI-plus’ involving 2G-CB, ‘PVI-only’ involving 2G-CB had similar rates of severe complications. ## Efficacy We used different approaches for 2G-CB. In the ‘PVI-plus’ group, the success rate of 2G-CB in patients with PerAF was superior to that in the ‘only PVI’ group. Additional ablation can isolate other targets and reduce recurrence. Previous studies have shown that the expression of circulating miRNAs changes after ablation of AF, and these miRNAs participate in atrial electrical remodeling and fibrosis, which might be one of the reasons why additional matrix ablation reduces the recurrence of PerAF. However, such ablation may be arrhythmogenic. The STAR AF II trial reported no added advantage from a more extensive ablation than PVI with RF for PerAF. This study showed a success rate of 68.9% at a mean FU of 16.7 months. Compared with the results of recent studies involving cohorts treated with RF energy, 2G-CB may be the optimal treatment strategy. In our study, the mid-term FU was 27 months, and there was a greater than 70% success rate at the mid-term FU. However, in other studies evaluating PVI alone for PerAF, such as a previous meta-analysis, a success rate of only 51.9% was observed. The improved success rates noted in our meta-analysis are likely multifactorial and related to differences in the patient population and the use of wide RL ablation, which incorporates some substrate and technological changes. ## Ablation strategy Patients with PerAF were recognized to have more advanced atrial remodeling than those with paroxysmal AF. Until now, the main strategy for ablation of PerAF was atrial substrate modification (in addition to PVI) to achieve acceptable success rates. Importantly, similar to the results of our study, some meta-analyses have demonstrated improved ablation outcomes when substrate ablation (CFAE or linear ablation) was performed in addition to PVI in PerAF patients. Linear ablations (the most common type of ablation) were successful in this study when cryoenergy freezes lasting 180 s were applied along the LA roof by caudocranial ascending activation and a conduction delay next to the ablation line of \> 120 s when pacing at the right atrial upper septum during SR. In another study involving AF without conversion to SR after PVI, additional LA rooflines were created with point-by-point lesions using contact force (CF) catheters. The acute endpoint was the elimination of local bipolar electrograms, with 20- to 40-s lesions or with a force-time integral \> 400 s. CF ablation was performed for documented atrial flutter prior to or during the procedure. This study reported that when AF ablation required more than CB PVI, an ablation catheter and technique were used on additional lesions. Catheter maneuvers necessary to achieve electrical blockage, substrate modifications and clinical outcomes were recorded. The 11 methods of CB catheter substrate modification that are described in this study included all the extra-PV lesion sets applied for 120 s\~180 s at each location. In the other study, LA linear ablations and substrate modifications, including cavotricuspid isthmus ablation and substrate ablation of CFAEs and other non-PV triggers, were performed. RF ablation was performed using an externally irrigated, non-force-sensing catheter with a 3.5-mm tip. Power was limited to 40 W over the anterior walls and 30 W over the posterior walls. At the end of the study, using a high-density mapping catheter and the available data, between one and two 120\~180 s cryoapplications were delivered to each PV with guidance by time to PVI. Once PVI was achieved, the cryoballoon was used to perform PWI in patients undergoing this treatment. PVI+PWI can be achieved safely and effectively using the cryoballoon. However, the latter required adjunct RF ablation for completion of PWI in approximately one-third of the patients. Therefore, we evaluated PVI, which is the cornerstone treatment for ablation of AF, and atrial substrate modification. This combination had a higher success rate than other methods. We found that ‘PVI-plus’ with 2G-CB for the treatment of PerAF was an effective ablation strategy that potentially reduced the procedure and fluoroscopy times. The improved effect with 2G-CB is likely due to structural improvements in the device, which optimized the refrigerant flow and distribution, providing a larger, more uniform freezing zone at the entire distal hemisphere of the balloon irrespective of balloon orientation and enabling a shorter application time. ## Safety In terms of overall outcomes, no deaths, myocardial infarctions or clinical cerebral emboli were reported in a total of 879 patients. The overall complication rate was 5.2% and mainly included PNPs (2.8%) and vascular access site complications (1.6%). Previous studies have reported similarly high rates of PNP (up to 5.4%) and vascular complication. Casado-Arroyo et al. reported that 2G-CB was more likely to cause PNP than other methods, possibly due to the larger cooling surface area, ablation area being more adjacent to the PN, and deeper damage foci. Moreover, the percentage of complications excluding PNP seemed to be lower with 2G-CB than with RF ablation. On the other hand, the incidence of pericardial effusion and/or tamponade was very low with 2G-CB and was consistently lower than the previously reported incidence with RF. ## Heterogeneity analysis In this study, significant heterogeneity was observed in the incidence of total complications and the rate of PNP. Due to the lack of correlated studies, we could not analyze detailed subgroups. On the other hand, the enrolled populations in the included studies may have been affected by the selection bias associated with a single-center experience and the preferences of different centers that refer patients for AF ablation because different centers used different protocols and tools, which ultimately resulted in substantial heterogeneity. In addition, there was no significant heterogeneity in recurrence and one-year recurrence in the overall population, and funnel plot analysis did not provide evidence of significant publication bias. Therefore, it was believed that the included studies had sufficient similarities. In conclusion, the outcome of the meta-analysis was reliable. ## Previous meta-analysis A previous and similar meta-analysis summarized data on the safety and mid-term efficacy of PVI using 2G-CB in patients with PerAF. A total of 11 studies were analyzed. After FU, 68.9% of patients were free from recurrence. Complications occurred in 5.5% of patients. Compared with the abovementioned meta-analysis, our meta-analysis used different inclusion criteria and required ‘PVI-only’ and ‘PVI-plus’ to be performed simultaneously. However, studies that included patients treated with other substrate modifications were permitted as long as PVI was performed using 2G-CB at least once during surgery. Similarly, we used the inclusion criterion that PVI was performed using 2G-CB. Thus, only 4 studies were included in this review. ## Limitations The limitations of this article include the following: 1. We did not find randomized trials that were eligible for this analysis, and the total sample size was not sufficient. 2. Heterogeneity among studies was significant in total complications and the rate of PNP, and we could not perform a more detailed subgroup analysis. In addition, the inclusion of only published studies may have resulted in publication bias toward more favorable ablation outcomes from more experienced centers with variable FU and assessment of arrhythmia recurrence. For example, in 2/4 studies, postablation monitoring was performed with only 24-hour monitors, and previous studies have shown that continuous monitoring and telemonitoring to detect clinical and subclinical AF events are more effective than 24-hour monitoring. Moreover, the use of AADs varied between studies, and postablation monitoring was performed with only 24-hour monitors. <!-- --> 1. Two studies included in this meta-analysis may have involved the same study population. We tried to contact the authors via email to evaluate the data. Unfortunately, we did not receive any response from those authors. 2. The studies we included were not direct studies of PVI-only *vs*. PVI- plus adjunctive ablation strategies for PerAF; rather, we extracted the data that we were interested in. In addition, some of the data needed to be calculated based on the results of the studies. # Conclusions In conclusion, this meta-analysis suggests that ‘PVI-plus’ involving 2G-CB for the treatment of patients with PerAF seemed to have a superior success rate and similar rates of severe complications compared with ‘PVI-only’ involving 2G-CB. However, we recommend that large, prospective, randomized, controlled studies should be performed in the future to validate our results. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Despite advances in vaccine development, vaccine distribution infrastructure remains inadequate in many parts of the world, and it is estimated that up to 40–50% of all vaccine doses are wasted in certain countries. Crucial to current vaccine transport is the idea of the cold-chain–a series of refrigerated enclosures with tight temperature control that allows for stable transport of vaccine from manufacturer to patient. Strict control of temperature is important for whole pathogen vaccines, as these are particularly prone to stability losses. In addition, pathogens with lipid membranes, such as bacteria and certain viruses including influenza, are especially prone to osmotic stress, and changing salt concentrations due to temperature-driven solvent evaporation can lead to pathogen shrinkage and destruction. The development of vaccines that can maintain viability outside of the cold chain would decrease vaccine wastage and increase immunization rates in regions of the world with underdeveloped vaccine distribution infrastructure. Although recombinant, subunit protein vaccines have been proposed as a more stable alternative to whole pathogen vaccines, issues with low immunogenicity and appropriate adjuvant choice have slowed their development as a viable option. Protein nanoparticles, made entirely of crosslinked protein antigens, are a means of delivering antigen and adjuvant in the same delivery vehicle, and are thus an excellent candidate for testing cold-chain-independent vaccine stability\[–\]. Monomeric influenza hemagglutinin is a 63 kDa protein responsible for mediating viral entry into host cells. We have shown previously that protein nanoparticles made from trimerized, H7 hemagglutinin (3HA) were able to protectively immunize mice against a 10xLD₅₀ H7 influenza challenge. Given the immunogenicity of these nanoparticles, as well as the *in vitro* hemagglutination assay that can provide a simple measurement of protein conformation in nanoparticles, we have examined the viability of cold- chain-independent storage of 3HA nanoparticles. We sought to assess whether storing nanoparticles at room temperature (\~25°C) or 37°C for several months resulted in a loss of hemagglutinating capability or immunogenic potential. We found that nanoparticles stored at room temperature retained both hemagglutinating activity and immunogenicity, while nanoparticles stored at 37°C retained hemagglutinating activity for 2 weeks. # Materials and methods ## 2.1 Nanoparticle synthesis and characterization Trimerized H7 hemagglutinin (3HA) protein was produced and purified from Sf9 cell culture, and nanoparticles were synthesized and characterized exactly as described previously. Briefly, 400 μL ethanol was added to 100 μL of a 1.6 mg/mL 3HA solution at a rate of 1 mL/min under constant stirring at 600 rpm. The particles were collected by centrifugation, and resuspended in sterile phosphate-buffered saline (PBS) with sonication. 800 μg soluble 3HA protein was added at a final concentration of 1.6 mg/mL to 480 μg desolvated 3HA nanoparticles and an amine crosslinking reaction was performed using 3 mM 3,3´-Dithiobis\[sulfosuccinimidylpropionate (DTSSP, Thermo Scientific, Waltham, MA) for 12 hours while stirring to coat the nanoparticles. Coated nanoparticles were collected by centrifugation, and protein concentration was measured by a bicinchoninic acid (BCA) assay according to the manufacturer’s instructions (Thermo Scientific) to estimate the total protein content in nanoparticles. Dynamic light scattering (DLS) was performed in PBS with a Malvern Zetasizer Nano ZS (Malvern Instruments, Westborough, MA) to assess nanoparticle size distributions. Hemagglutinating capability of 3HA nanoparticles was tested as previously described. Briefly, 5 μg of 3HA protein or 3HA protein nanoparticles in 100 μL of PBS was serially diluted by half across 11 wells of a 96-well plate. 50 μL of a 0.5% turkey RBC suspension in saline was then added to each well, and incubated at room temperature for 1 hour. The hemagglutination titer was read as the last well in the serial dilution that did not form a red button of settled RBCs. ## 2.2 Extended storage Extended storage of 3HA nanoparticles was performed in PBS at room temperature (25°C) or 37°C. Nanoparticles and soluble protein were diluted to 200 μg/mL, the concentration for vaccination, in 100 μL aliquots in 2 mL centrifuge tubes. The tubes were sealed with parafilm and incubated for up to 1 month at 37°C, and up to 112 days at room temperature. At each time point, one 100 μL aliquot was used to measure hemagglutination activity of the nanoparticles, while two 300 μL aliquots were stored at room temperature for *in vivo* immunizations. ## 2.3 Immunization and sample collection All animal work was done in accordance with the IACUC guidelines of Georgia State University, which specifically approved this study under IACUC approval number A16024. Female, 6 week old Balb/c mice (Charles River, Wilmington, MA) were intra-muscularly immunized with 10 μg aged 3HA nanoparticles, 20 μg aged 3HA nanoparticles, 10 μg freshly prepared 3HA nanoparticles, 10 μg soluble 3HA protein, or PBS as a control. Five mice per group were immunized twice i.m. at a 3-week interval. To compare antibody responses, sera were collected 2 weeks after each immunization by submandibular venipuncture. No anesthesia or analgesia was used in this study. All animals were euthanized by CO₂ asphyxiation according to the American Veterinary Medical Association Guidelines for the Euthanasia of Animals. ## 2.4 Serum IgG titer Serum IgG titer was assessed as previously described. Briefly, ELISA plates were coated with 1 μg/mL 3HA protein in PBS and incubated overnight at room temperature. Plates were blocked with 1% bovine serum albumin (BSA) in PBS for 1 hour. Mouse serum samples were initially diluted 1:100 in 1% BSA, and serial half-dilutions were made in 1% BSA across the 96-well ELISA plate. Serum sample dilutions were incubated for 1 hour, and 1 μg/mL HRP-conjugated goat-anti-mouse IgG (Life Technologies, Grand Island, NY) in 1% BSA was used as a detection antibody. Chromogenic quantification was performed by the oxidation of tetramethylbenzidine by hydrogen peroxide (R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions. Two times the absorbance of naïve group’s serum samples was considered the cutoff for measuring the endpoint titer. ## 2.5 Hemagglutination inhibition assessment Serum hemagglutination inhibition (HAI) activity was assessed according to a protocol adapted from the World Health Organization. 10 μL mouse serum was incubated with 30 μL of receptor-destroying enzyme (RDE) (Denka Seiken Co, Tokyo, Japan) at 37°C overnight, followed by 56°C incubation for 30 minutes to inactivate non-specific agglutinating proteins. Inactivated serum was diluted with 60 μL room temperature PBS, and centrifuged at 500xg for 8 minutes to collect treated serum samples. Eight, 25 μL serial half-dilutions of 1:10 treated sera were mixed in a round-bottom 96-well plate with 25 μL of 2 μg/mL 3HA protein and incubated at room temperature for 30 minutes. This concentration of 3HA corresponded to 8 Hemagglutination Units (HAU) in 50 μL of PBS, as prescribed by the WHO protocol. To this mixture, 50 μL of a 0.5% turkey RBC suspension was added, and the wells were incubated for 1 hour to develop. HAI titer was read as the inverse dilution of the last well able to prevent hemagglutination, or in which a red button of settled RBCs was formed. ## 2.6 Statistics IgG titers were assessed by the Kruskal-Wallis one-way analysis of variance (ANOVA) for non-parametric samples. Hemagglutination inhibition titers were analysed by comparing the number of wells neutralized to the standard HAI titer of 40 (2 wells) established by the US Food and Drug Administration (FDA) as protective, using a one-sample t test. P values less than 0.05 were considered significant. # Results H7 hemagglutinin nanoparticles were synthesized and characterized as described previously. The resulting nanoparticle size distribution was similar to that of previously synthesized batches of nanoparticles. A pilot hemagglutination study of three replicates showed identical hemagglutinating titers, and one replicate per timepoint was used for further hemagglutination assays to conserve materials. Nanoparticles stored in PBS at room temperature did not lose agglutinating activity or appreciably change in size after 112 days, while nanoparticles stored at 37°C retained full hemagglutinating activity for 2 weeks, and lost activity at 1 month. Soluble 3HA in PBS showed similar stability to nanoparticles when stored at room temperature; no hemagglutinating activity losses were observed after 56 days. To compensate for any potential loss in hemagglutinating activity, we immunized mice with a single dose (10 μg) or double dose (20 μg) of nanoparticles stored for 112 days in PBS at room temperature. As controls, mice were immunized with 10 μg freshly-prepared nanoparticles or 10 μg soluble 3HA protein. Serum samples were collected 2 weeks after priming and boosting immunizations and were assessed for anti-3HA IgG titers. No significant differences in IgG titer were observed between the nanoparticles stored at room temperature and freshly made nanoparticles. The double dose of aged nanoparticles induced significantly higher titers compared to soluble protein after the priming immunization, but that significance disappeared after the boost immunization. Hemagglutination inhibition (HAI) titer is a measurement of the ability of serum antibodies to block hemagglutinin binding to sialic acid residues. This measurement is directly correlated to the ability of an animal’s antibodies to block influenza viruses from infecting host cells, and is a means of indirectly predicting neutralizing antibody production. The World Health Organization has defined a serum HAI titer of 40 or above to provide good protection against influenza infection. Groups immunized with 10 μg stored and fresh nanoparticles had significantly higher HAI titers than 40, while groups immunized with soluble protein or PBS did not have significantly higher titers than 40. # Discussion Nanoparticles were stored at room temperature for 112 days in PBS with no loss in hemagglutinating activity. According to a report by the United Nations Children’s Fund (UNICEF) and the WHO, approximately one month is needed for cold-chain transport of vaccines from a manufacturer in the developed world to a developing region. Storage of protein nanoparticles for almost 3 times this length without additional excipients is especially promising, given that viral and subunit vaccine storage typically requires the addition of polyethylene glycol or sucrose to achieve weeks-long cold chain-independent storage. Surprisingly, soluble 3HA was also able to retain hemagglutinating activity for 56 days. This could be due to the inherent stability of this particular protein, and of recombinant subunit antigens generally. Nanoparticles stored at 37°C in PBS were able to retain hemagglutinating activity for 2 weeks, but started to lose activity at 1 month. Although we did not immunize mice with nanoparticles stored at 37°C, we hypothesize that a similarly high antibody titer would be achieved based on the correlation between *in vitro* vaccine hemagglutinating activity and *in vivo* immunization efficacy. Two weeks is a sufficient amount of time to reach rural health clinics from a distribution center, but may not be enough time to store the vaccine there for a prolonged period of time. Future studies should investigate the mechanism of elevated-temperature-related activity loss and whether stabilizing additives could mitigate any potential losses in immunogenicity. Hemagglutination inhibition data suggested that our room temperature-stored nanoparticles could be just as effective at inducing neutralizing antibodies as freshly made nanoparticles. The FDA benchmark HAI titer of 40 allows us to compare our vaccine to an accepted standard of immunity, and our data suggests that our nanoparticles stored at room temperature can confer protective immunity. Given the ability of freshly-made 3HA nanoparticles to protect mice against a 10xLD₅₀ challenge of H7N9 influenza, we relied on antibody measurements in this work. However, future challenge studies should examine whether nanoparticles stored at warm temperatures can also confer protective immunity. The ability of 3HA nanoparticles to induce better HAI titers than soluble 3HA concurs with current nanoparticle vaccine literature theorizing that multivalent epitope presentation is a mechanism of recombinant antigen nanovaccine adjuvancy. Our previous results demonstrating enhanced dendritic cell TNF-α and IL-1β responses to OVA-coated OVA protein nanoparticles also support this hypothesis. In conclusion, we have shown that 3HA nanoparticles can be stored for up to 112 days at room temperature with no loss in *in vitro* hemagglutinating activity or *in vivo* antibody and hemagglutination-inhibiting responses. The fact that 3HA nanoparticles can remain immunogenic outside of cold-chain storage is a promising sign that nanoparticle vaccines made from recombinant protein antigens can survive both transport to and storage in the developing world. Future work in assessing and engineering protein nanoparticle vaccine stability should aim to (1) corroborate these results with other types of antigen nanoparticles, (2) examine the mechanism of elevated temperature-related antigen nanoparticle instability, and (3) extend viable storage at elevated temperatures beyond 2 weeks. # Supporting information The authors would like to thank Dr. Jong Rok Kim for the synthesis and purification of the 3HA protein used for this study. [^1]: The authors have declared that no competing interests exist.

# Introduction The placenta is the key organ for successful pregnancy and fetal growth. It performs key transport, metabolic, and secretory functions to support fetal development. Term placental villi are covered by the multinucleated syncytiotrophoblast that shares a basement membrane with a subjacent, discontinuous layer of cytotrophoblast. The syncytiotrophoblast is in direct contact with maternal blood and regulates maternal-fetal exchange. To allow the efficient supply with oxygen and other key molecules to the placenta and hence the fetus, the invasive cell population of the extravillous trophoblast (EVT) invades the uterine decidua and myometrium (interstitial trophoblast). A subgroup of the interstitial trophoblast migrates towards the uterine spiral arteries, reaches the walls of such arteries (intramural trophoblast) and can be found in the lumen of such vessels (endovascular trophoblast), transforming them into large conduit vessels of low resistance. This physiological transformation is characterized by a gradual loss of the normal musculo-elastic structure of the arterial wall and replacement by amorphous fibrinoid material in which trophoblast cells are embedded. The interstitial route of trophoblast invasion through the placental bed has been well described by Kaufmann et al.). These physiological changes are thought to be required for a successful pregnancy. At the same time, proliferation, migration, and invasion of extravillous trophoblast are regulated by a large number of locally derived molecules including members of the VEGF and the angiopoietin families to maintain a healthy uteroplacental homeostasis. Extravillous trophoblast dysfunction has been implicated in IUGR, one of the leading syndromes causing preterm delivery and perinatal morbidity. This dysfunction is characterized by reduced numbers of both, interstitial and endovascular trophoblast. Severe early onset IUGR is often associated with preeclampsia, a leading cause of maternal death worldwide. In preeclampsia, hypertension is associated with widespread maternal endothelial dysfunction, leading to significant maternal morbidity. Oxidative stress of the placenta is considered to be a key intermediate step in the pathogenesis of preeclampsia, but the cause for this stress remains unknown. In about 80% of all preeclampsia cases the extravillous trophoblast is not affected, while the other 20% of preeclampsia cases also suffer from IUGR with trophoblast malinvasion. In any case, placental hypoxia does not occur in cases with trophoblast malfunction such as IUGR and preeclampsia. Hence, the cause for the presence of oxidative stress with increased lipid peroxidation products, in preeclampsia remains unknown. It is also still unknown whether in cases with IUGR and preeclampsia, the malfunctional extravillous trophoblast also suffers from oxidative stress. A sensitive marker of oxidative damage and lipid peroxidation is 4-hydroxy-2-nonenal (4-HNE), a highly toxic aldehyde product of lipid peroxidation which can be evaluated by immunohistochemical staining. Significantly higher levels of 4-HNE have been consistently reported in pathological placentas associated with oxidative stress. A battery of genes encoding antioxidant enzymes is orchestrated upon exposure to reactive oxygen species (ROS). This coordinated response is regulated via the antioxidant response element (ARE) contained within the regulatory regions of so-called “safeguard” genes such as glutathione peroxidase, and heme oxygenase-1 (HO-1). Activation of the nuclear factor erythroid 2-related factor-2 (Nrf2) as a consequence of oxidative stress initiates and enhances transcription of these safeguard genes, thus protecting cells against oxidative stress as well as a wide range of other toxins. Mann et al. were the first to discuss a link between Nrf2, vascular homeostasis, and preeclampsia. Our laboratory provided the first experimental data that Nrf2 is active exclusively within villous cytotrophoblast of the preeclamptic placenta, strongly suggesting that these cells suffer from oxidative stress. Loset et al. reported that the Nrf2-mediated oxidative stress response was overrepresented in the decidua of patients with preeclampsia, without indicating the presence or absence of IUGR. Furthermore, in the human placenta VEGF, PlGF and their two receptors are differentially expressed throughout gestation. VEGF expression in the placenta and placental bed declines as pregnancy advances, while PlGF and Flt-1 increase towards term. Moreover, a recent study has shown that VEGF prevents oxidative damage via activation of the Nrf2 pathway in the choriocarcinoma cell line BeWo. This suggests that that decreased VEGF bioavailability during preeclampsia could result in reduced basal defense against oxidative stress. As an in vitro interplay was established between Nrf2 and VEGF, we hypothesized that severe early onset IUGR and preeclampsia could be associated with alterations in Nrf2 expression in the placental bed, particularly in the extravillous trophoblast, since it is well known that VEGF and its receptor VEGFR-1 (Flt-1) are expressed in these cells. To test this hypothesis we used immunohistochemistry to examine the expression of Nrf2, VEGF, and the oxidative stress marker 4-HNE in third trimester placental bed tissues in cases of severe early onset IUGR/PE and control pregnancies. # Materials and Methods ## Tissues All samples were collected from caesarean hysterectomy subjects received for pathological examination at the Medical Institute, Ashgabat, Turkmenia (by M.K.) apart from 2 control subjects collected in Aachen, Germany. These materials were used in a former study by Kadyrov et al.. In all instances, permission was granted for the histological studies, regarding the samples that were collected in Turkmenia, approval was obtained from the Ethics Committee of the mother and child medical centre in Ashgabat; protocol Nr. 047/1991 and Nr. 101/1992. For the other samples our protocol was approved by the Ethics Committee of the Medical Faculty of the University of Technology, Aachen, Germany (EK 512). Written informed consent was obtained from each patient enrolled in this study. Uterine tissues of control subjects were derived from caesarean hysterectomies performed in five healthy normotensive women delivering term infants (38–40 weeks gestation) for reasons unrelated to placental development (fibroids, cervical carcinoma, or uterine atony after caesarean section). Uterine tissues of pathological subjects were collected from six women with severe early onset IUGR and preeclampsia (29–34 weeks gestation). The selected criterion for complicated pregnancies was severe preeclampsia with IUGR. Severe Preeclampsia was defined following the criteria of the American College of Obstetricians and Gynecologists, new onset hypertension (systolic blood pressure ≥160 mmHg or diastolic blood pressure ≥110 mmHg at least twice, measured six hours apart) and proteinuria (5 g or higher per 24-hour period) after 20 weeks of gestation. IUGR was defined as birth weight below the 10th centile of customized birth weight for gestational age. The clinical characteristics of patients enrolled in this study are summarized in. ## Immunohistochemistry The placental bed was identified macroscopically and later verified immunohistochemically by identification of extravillous trophoblast expressing cytokeratin 7. At least five samples measuring about 2×2×2 cm per uterus, covering the full thickness of the uterine wall of the placental bed, were dissected, and fixed in 4% neutrally buffered formalin. The tissues were oriented during embedding in such a way that the uterine layers were perpendicular to the plane of section. To prevent bias each tissue block was rotated randomly around a virtual axis from the endometrium to the perimetrium and embedded vertically in paraffin at 56°C. All the uterine layers can be seen in A) revealing the distribution of extravillous trophoblast (positively stained) that invaded perpendicularly from the endometrial-myometrial border into the myometrium. Routine immunohistochemical procedures were performed on serial sections of 5 µm thickness with antibodies against cytokeratin 7 (trophoblast marker; clone OV_TL 12/30, dilution 1∶200, DAKO, Denmark), VEGF (sc- 7269, 1∶30, Santa Cruz, USA), Nrf2 (ab31163, 1∶50, Abcam, UK), and 4-HNE (ab46545, 1∶200, Abcam, UK). Binding of species-specific biotinylated secondary antibodies was visualised with AEC substrate chromogen (AEC) (Invitrogen, Germany). Sections were counterstained with hematoxylin. ## Evaluation of Immunohistochemical Staining Each immunostained section was analyzed semi-quantitatively using a modification of the “quick score” method described by Detre et al.. In brief: An intensity score was made on the basis of the average intensity of staining: 0 =  negative, 1 =  weak, 2 =  intermediate and 3 =  strong, then the percentage of positive cells (endovascular and interstitial, mononuclear and multinuclear extravillous trophoblast cells) for each staining was rated as: 1 = 0–25%, 2≤25–50%, 3≤50–75%, and 4≤75–100%. The whole of the section was assessed. Two independent pathologists examined the immunohistochemical slides while blinded to the clinical history of the patients. The intensity score and the proportion score were then multiplied and scores summed to give a range of the possible score of 0 to 15. For example, negative staining in 25% of the extravillous trophoblast (0×1 = 0), weak staining in 50% (1×2 = 2) and strong staining in 25%(3×1 = 3) would give a total score of 0+2+3 = 5. ## Statistical Analysis Statistical analyses were performed using Student’s unpaired t test for dual comparisons. Mean differences were considered to be significant when p\<0.05. All statistical graphs and analyses were created with GraphPad Prism 5.0 (GraphPad Software, La Jolla, CA). # Results Several different cell types in the placental beds and uterine wall biopsies were positive for Nrf2, 4-HNE and VEGF, including extravillous trophoblast B, 4 B, 5 A), decidual stromal cells A arrows), myometrial cells, leukocytes in decidua and myometrium B arrows), and vascular endothelial cells (A, B arrows). Since the aim of this study was to evaluate the differential immunostaining of the aforementioned proteins in extravillous trophoblast, other cells were not analyzed in detail. Semi quantitative analysis was confined to endovascular/intramural and interstitial extravillous trophoblast. In control pregnancies, extravillous trophoblast expressed Nrf2 to a similar extend as has been described previously for villous trophoblast by Wruck et al.. However, immunostaining for Nrf2 was stronger in endovascular and interstitial trophoblast in pregnancies complicated by early onset IUGR/PE B). Immunostaining scores showed that immunostaining of cytoplasmic Nrf2 in extravillous trophoblast was significantly increased in the IUGR/PE group (p\<0.0001) when compared with the control group E). In control pregnancies, interstitial and endovascular trophoblast, which stained positive for cytokeratin 7 C), were virtually negative for 4-HNE A). By contrast, these cytokeratin 7-positive cells D) showed stronger 4-HNE immunoreactivity in IUGR/PE cases B) in comparison with the same cell type of the control group. Immunostaining scores showed that immunostaining of 4-HNE in extravillous trophoblast was significantly increased in the IUGR/PE group (p\<0.005) when compared with the control group E). VEGF was immunolocalized in both populations of the extravillous trophoblast, interstitial and endovascular trophoblast of the placental bed of the control group A). VEGF immunostaining was almost absent on extravillous trophoblast from IUGR/PE cases B). Semi quantitative analysis of VEGF immunostaining revealed a statistically significant difference between the control and IUGR/PE groups (p\<0.05) E). # Discussion The feto-placental circulation in severe early onset IUGR/PE cases is characterized by abnormal umbilical blood flow velocity waveforms, thought to be indicative of increased placental resistance. In such cases, the uteroplacental blood is altered as well showing a ten-fold increased blood flow velocity from the spiral arteries into the intervillous space. The distorted blood flow through spiral arteries that have not been transformed adequately may also induce changes in oxygen delivery to the surrounding decidual tissues. This in turn may result in a marked overproduction of ROS (reactive oxygen species), generated mainly in mitochondria, and will cause oxidative stress. Increased production of ROS will trigger a cascade of events to enhance the cellular defense against oxidative stress, mainly by the Nrf2/ARE system. This is the first study to examine the expression of Nrf2 in the placental bed of IUGR/PE samples. As the transformation of spiral arteries is mediated by invasive extravillous trophoblast, we focused on the expression of Nrf2 in these cells. Although our data were limited because of the shorter duration of gestation in the early onset IUGR/PE group, it is very difficult to obtain a normal control group of the same gestational age. At the same time, a comparison between specimens from early onset IUGR/PE and normal term controls does not necessarily limit the significance of the data because transformation of the myometrial spiral arteries largely occurs during the transition between the first and second trimester. Consequently, uterine artery Doppler waveforms can identify a high proportion of women who develop early onset IUGR/PE already at 12 weeks , while they show low resistance patterns in normal pregnancies at 22–24 weeks. Therefore, trophoblast populations and spiral artery modifications are very unlikely to change further between 30 and 40 weeks of gestation. Endovascular and interstitial extravillous trophoblast populations show decreased densities in early onset IUGR/PE as already described by Kadyrov et al.. In these pathological samples the extravillous trophoblast revealed a significant increase in staining intensity for Nrf2 in the cytoplasm of these cells (B, E). This data suggests that the extravillous trophoblast in IUGR/PE suffers from (oxidative) stress leading to increased Nrf2 expression. Activation of Nrf2 has been correlated with transfer of this protein into the nucleus which does not become obvious in our study. It seems as if Nrf2 is not activated in these cells similar to what has been described for Nrf2 in preeclamptic villous trophoblast These authors described lower transfer of Nrf2 into the nucleus, indicative for a lower activity of Nrf2. Hence, although the extravillous trophoblast in IUGR/PE upregulates Nrf2 expression it seems as if activation of Nrf2 fails under these conditions and thus a defense system to combat oxidative stress in the extravillous trophoblast is not effective. Vascular endothelial growth factor (VEGF)-A is expressed by extravillous trophoblast and binds to the fms-like (Flt1) and kinase-insert domain-containing tyrosine kinase receptors, which are expressed on extravillous and villous trophoblast. In cases with early onset IUGR/PE, which are associated with a failure in spiral artery invasion, there is down-regulation of VEGF in the extravillous trophoblast and placental up-regulation of soluble Flt1. In an earlier study we found that VEGF induces Nrf2 activation leading to prevention of oxidative stress. Therefore, we tested whether increased Nrf2 expression is associated with higher immunostaining for VEGF in these cells. In contrast to Nrf2, VEGF expression was reduced in extravillous trophoblast in the placental bed of IUGR/PE pregnancies B) confirming already published data. This data suggests that Nrf2 activation is not a consequence of increased VEGF expression, but may be a secondary adaptive response to ROS signaling. Since Nrf2 upregulates the expression of antioxidative and detoxifying enzymes, we further tested the level of oxidation-mediated changes in lipids (4-HNE) in extravillous trophoblast in term controls and IUGR/PE by immunohistochemistry (A, B). There was a marked increase in 4-HNE immunostaining in extravillous trophoblast of IUGE/PE compared with the control group B). The picture that develops from our data is as follows: In IUGR/PE extravillous trophoblast experience oxidative stress (increased 4-HNE) and try to counteract by increased expression of Nrf2. However, since these cells fail to upreguate VEGF at the same time, activation of Nrf2 does not occur. In addition, several studies have shown that IUGR/PE is associated with reduced levels of antioxidant enzymes, which lead to further trophoblast damage. This in turn may result in increased apoptosis of extravillous trophoblast and decreased densities of such cells in the placental bed. Taken together, we suggest that the extravillous trophoblast at late stages of IUGR/PE pregnancies demonstrate an impairment of the Nrf2 signalling pathway, in spite of the increased cytoplasmic Nrf2 expression, related to the cellular oxidative damage occurring at earlier stages of the syndrome. In conclusion, it can be hypothesized that decreased VEGF bioavailability during early stages of preeclampsia results in insufficient Nrf2 activation, reduced basal defense against oxidative stress and a higher vulnerability of trophoblast to oxidative cell damage. This does not seem to be true for the villous trophoblast only but may be extended to the extravillous trophoblast in cases which combine IUGR and preeclampsia. The resulting damage causes increased apoptosis and will further speed the vicious circle of shallow invasion in such cases. Consequently, one would expect that these disturbances will limit trophoblast invasion into the walls of spiral arteries of women destined to develop early onset IUGR/PE. Specific attempts to strengthen the fetal endogenous defence system against oxidative stress during early gestation could prove to be a possible treatment option and may in turn reduce the risk of the combination of IUGR and preeclampsia and associated perinatal complications. The authors thank Dr. rer. nat Lars Ove Brandenburg and Dr. rer. nat Mersedeh Tohidnezhad for performing the IRS scoring in this work. We would like to thank Michaela Nicolau for her excellent technical assistance. We thank Wolfgang Graulich for the production of the illustration figure (1). We also gratefully thank the Department of Histology, Faculty of Medicine at Damascus University because of the scholarship of Nisreen Kweider. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: NK BH MK. Performed the experiments: NK RB MK. Analyzed the data: NK BH CJW WR TP MK. Contributed reagents/materials/analysis tools: NK BH CJW RB WR TP MK. Wrote the paper: NK BH MK.

# Introduction Microarray technology allows the capture of diverse aspects of genetic, environmental, oncogenic and other factors as reflected in global mRNA expression and opens the possibility of personalizing treatment of disease. Multiple studies have taken a “top-down” approach to profiling gene expression in human cancers, and this has led to the identification of tumor subtypes unrecognized previously as well as gene signatures predicting various clinical phenotypes. Alternatively, other studies have taken a “bottom-up” approach to determine the change of gene expression caused by specific manipulations of cultured cells *in vitro*. In these studies gene expression serves as a common phenotype to recognize similar features in human cancers *in vivo* and to provide a direct linkage between the known biological perturbation and the clinical contexts. Though many such studies have shown promise in using *in vitro* cell manipulations to understand *in vivo* biology, this approach cannot fully reflect the enormous phenotypic variation seen in human cancers. From such studies, one can derive *signatures*. These we define to be lists of genes that are differentially expressed along with their associated levels of differential expression (which we call weights). However, there is nearly always a poor match between these signatures and expression patterns of the same genes *in vivo*. Therefore, a conceptual framework is needed to further dissect, enhance and extend the *in vivo* utility of the *in vitro* derived signature. Here, we present a technique for achieving this purpose. We propose deriving multiple factors, based on human cancer gene expression studies, from an experimentally defined signature. These derived factors will retain their relationship to the original signature but represent distinct biological processes. Importantly, we show that different derived factors can be combined to provide much better predictive values for the clinical outcomes. Different factors also reflect different biological processes and are linked to various aspects of molecular and clinical features of human cancers. There are a number of possible approaches to this problem. One popular approach has been to compare the identities of the differentially expressed probes to databases of pre-defined pathways. Descriptions of such approaches can be found in. While these approaches are appealing for their interpretability, they rely on the appropriately pre-defined pathways rather than the structure of the data under study. Alternatively, one may simply define the signature activity level for a sample as the weighted average of *in vivo* expression levels (where the genes over which to compute the weights and the weights themselves are drawn from the original signature). Although some studies have shown the power of this concept, it is clear that one can not hope to capture the heterogeneity of *in vivo* biology from the one-dimensional controlled biological response the *in vitro* signature reflects. The inherent heterogeneity of environment and cell type in tissue samples means that the genes in a signature may potentially involve many additional activities not evident *in vitro*. Further, experiments on cloned cell lines of a single cell type grown under tightly controlled conditions for a fixed (and relatively short) length of time may contrast starkly with clinical samples extracted from living organisms containing multiple cell types that have been in a dynamic environment for months or years. There is no clearly “correct” method for taking what is learned by microarray experiment in culture and applying it to assess pathway activity in tissue samples. Some genes may be poorer representatives of pathway activity *in vivo* because they are more likely to be involved in other pathways, because they react to environmental conditions that are not present *in vitro*, or for a myriad of other reasons. It is, therefore, important to provide a statistical and conceptual framework which can allow us to use the *in vivo* expression data to further dissect, refine and enhance the *in vitro*-derived gene signatures. *Signature Factor Profiling Analysis* (SFPA), based on sparse statistical factor models, is a framework for mapping *in vitro* signatures to a collection of *in vivo* factors. While this sounds similar to hierarchical clustering (which has become the default method for this type of problem), there are important distinctions. First, while hierarchical clustering can be used to break a set of samples into groups, within which expression patterns are similar in some way, it does not quantify that similarity. Second, hierarchical clustering requires that each observation (gene) be a member of just one cluster. This precludes assigning clusters to biological pathways, because many combinations of pathway activity are possible. Lastly, because the factors are generated within a statistical model, it is possible to identify the levels of activity in each of the factors on a newly measured sample without redoing the statistical analysis. While there are techniques other than hierarchical clustering which address some of these issues, for example soft-clustering and k-means clustering, our algorithm addresses them all within a single coherent statistical framework. SFPA provides: - Robust statistical modeling of both experimental gene expression and tissue sample expression. - Identification and correction of assay artifacts, which are known to be a significant issue associated with the use of microarray technologies. - A mapping from a single signature, generated *in vitro*, to a collection of factors that retain the pertinent characteristics of the signature while better reflecting heterogeneity *in vivo* associated with the biological perturbation the signature represents. - A model for imputing the values of factors in new collections of tissue samples even though these samples may originate from different groups and at different times. We explore this analysis approach in translating a collection of gene signatures reflecting cellular response to five known tumor microenvironmental factors, discovered *in vitro*, with particular emphasis on the signature associated with response to lactic acidosis. We demonstrate that multiple factors arising in a breast cancer context remain representative of the individual microenvironmental pathway responses from which they are derived. Furthermore, these factors differentiate key biological phenotypes in breast cancer, are able to improve clinical predictions across multiple cancer data sets, and retain their predictive ability even when applied to samples taken at vastly different times or at different study centers. # Results ## Context, Data and Analysis Strategy We begin with five signatures defined by the transcriptional responses of cultured human mammary breast epithelial cells to five microenvironmental perturbations: hypoxia, lactic acidosis, hypoxia plus lactic acidosis, lactosis, and acidosis. Each of these is seen in human cancers and carries prognostic information with respect to clinical outcomes. The signatures represent changes in expression of genes between a set of control observations and cells grown in the presence of lactic acidosis (25 mM lactic acid, pH 6.7), hypoxia (2% O2), lactic acid plus hypoxia, lactosis (25 mM sodium lactate, neutral pH), and acidosis (pH 6.7 without lactate). Expression assays used Affymetrix U133+ 2.0 microarrays and signatures reflecting each of the microenvironmental factors have been described. As shown in, hypoxia, lactic acidosis and acidosis have strong prognostic significance in several studies of breast cancers. Our aim here is to explore the various components of the original gene signatures to evaluate the opportunity for further enhancing their prognostic values and dissecting them into distinct biological pathway-relevant factors with clinical relevance. We use Bayesian Factor Regression Modeling (BFRM) to define and estimate factors based on a given signature. This begins with a small collection of genes that are highly responsive to the original intervention (highly differentially expressed between control and experimental groups in cell culture) and then iteratively refines the gene set, based on co-expression in an in vivo data set, in the context of a statistical factor analysis. First, common patterns of expression (factors) are discovered within the subset of genes currently under consideration. Next, the association between these factors and the full set of genes on the array allows us to identify additional genes to be included in a revision of the factor analysis. The rationale for this is that, while evaluating factors underlying the initial selected signature genes allows us to elucidate *in vivo* variability that is not present *in vitro*, adding genes from outside the original signature can improve the characterization of these factors while providing linkages to other relevant pathways. Running SFPA on each of the five signatures independently, we obtain 11 hypoxia factors, 10 lactic acidosis factors, 20 hypoxia plus lactic acidosis factors, 17 lactosis factors and 9 acidosis factors. SFPA stops discovering factors once most of the variability in the original gene set has been explained. ## Signature-Factor Relationships We will focus, for now, on the ten lactic acidosis factors. Examining the genes in each of the factors shows that all factors have representatives from the original signature in addition to genes added during the process of fitting the factor model. It is important to be sure that in the discovery of these ten factors, we have not lost our original signature. We check this by regressing the 10 sets of derived factor scores on the lactic acidosis signature scores. (Calculation of a signature score is described in the section.) Witin a single multivariate regression model, we find that 7 of the 10 are significant at the.01 level, and that when we eliminate the remaining three factors from the multivariate regression, those seven remain significant. Thus, at least seven of the factors show a significant association to the original signature. shows the fitted values from the regression of the lactic acidosis signature score on the lactic acidosis factors from the analysis of the 251 tumor sample data set from. The for this regression is high (.74), but it is possible these ten factors might be able to explain many different signatures. In order to show that this is not a spurious association, we test the hypothesis that this level is independent of which genes are assigned which weights. We re-sampled the weights 10,000 times, each time regressing the signature score vector computed from these weights on the 10 lactic acidosis factors and computing an value. Of the 10,000 values of so computed under the null hypothesis, the maximum was.48 ensuring that the p-value ≪10⁻⁴. If we approximate the distribution of values by a beta distribution (calculated by method of moments) we get a very close fit and estimate the p-value to be ≈10⁻¹³. Because only the list of highly differentially expressed genes from the lactic acidosis signature, and not the weights, are used in the factor discovery, and because the weights are critical for the computation of the lactic acidosis signature scores, the ability to recover signature scores from factors is strong evidence of the relationship between the two. The three factors derived from the lactic acidosis signature that were not important in the prediction of signature scores may still represent activity relevant to the presence of lactic acid, but they are not strongly predictive of the original signature. They may also simply represent the activity of biological pathways that involve very large sets of genes, and are thus discovered from many different possible starting points. Nonetheless, they represent significant structure in expression of the expanded signature gene set in tumor data, and none of these factors would be detectable from studying the signature alone as a phenotype. Factors can reflect distinct aspects of biological activity. shows which of the 67 factors (all factors discovered from each of the five starting signatures) have high correlation with the 10 lactic acidosis factors from the Miller breast data analysis. Notice that no two of the lactic acidosis factors are highly correlated, thus these factors seem to describe distinct processes. Some of the 10 factors, such as lactic acidosis factor 8 for example, are highly correlated with multiple other factors, indicating that these factors have been identified from multiple initial signatures. Most, however, show low levels of pairwise correlation. Among the 67 factors, 40 principal components are required to account for 95% of the observed variability (supplementary) implying that a relatively high biological “dimension” underlies the 67 factors – they reflect a diverse set of biological activities, and presumably pathways altered in the cellular responses to lactic acidosis within human breast tumors. shows the connections between genes and the 10 lactic acidosis factors in the analysis. The genes include the initial selected signature genes and those added through the iterative enrichment analysis. The SFPA-derived factors retain a high percentage of genes that have been shown to exhibit a change in expression when cells are exposed to the presence of lactic acid *in vitro*, showing in another way that these factors still maintain their connection with the original signature. The cross-talk between factors, in terms of genes defining more than one factor, is also evident. ## Factors Predict Molecular Features SFPA-derived factors can represent distinct aspects of biological processes associated with clinical phenotypes. To evaluate this, we explored subset regression models to predict a number of clinical phenotypes in the Miller data set - the phenotypes including ER and PgR status, p53 status and survival times. The molecular status indicators were modelled with binary probit regressions on the factors, and survival with standard Weibull survival models. We utilized the Shotgun Stochastic Search (SSS) method, to identify small subsets of the factors showing predictive value with respect to each of these phenotypes. SSS is a variable selection model which allows the use of model averaging (based on posterior likelihood) for prediction. Model averaging has been shown to perform better than algorithms which use the single best model for prediction (such as AIC or BIC) because it gives a truer estimation of uncertainty. This analysis was performed on the data set from, and then the resulting fitted/trained regression models were used to predict phenotypes in each of five separate and biologically diverse breast cancer data sets. All data sets are available from the Gene Expression Omnibus (GEO). ### Factors predict ER status The analysis indicates that highly scoring regression models for the prediction of ER status utilize one of the factors – Acidosis 1, Hypoxia 4, Lactic Acidosis 2, or Lactosis 5. From, one can see that the correlation between any two of these factors is high, so we will refer to them collectively as the ER factors. demonstrates the ability of this factor to predict ER status on the training set and 3b shows prediction on a distinct and completely unrelated test set. To examine the gene ontology (GO) composition of the list of genes involved in the ER factors, we applied the GATHER analysis and find that GO terms associated with cell cycle, proliferation and and mitosis are greatly enriched in these factors, corroborating well-known connection between cell progression and ER. It is also expected that the presence of lactic acid or hypoxia acts to shut down the cell cycle and the ER factor appears to directly link the two processes. ### Factors predict PgR status Estrogen and progesterone are known to be antagonists, so it is expected that ER factors can predict progesterone receptor status. Using SSS we find that the highly scoring regression models for PgR status involve the ER factor in addition to Lactic Acidosis factor 10 – we label this the PgR specific factor. show the fitted and predictive ability of these two factors used in a binary regression model fit to progesterone receptor status. There is no significant correlation in tumor expression between the PgR and ER factors. Gene ontology for the genes in the PgR specific factor bear out some of the known links between progesterone and RNA metabolism in breast cancer. ### Factors predict p53 status The third binary phenotype, wild type versus mutant p53 gene, is present in only the data set from. SFPA was re-run on a randomly selected 50% of these data and used to predict the other 50%. Highly scored models for p53 involve the ER factor, the PgR specific factor, and one of either Hypoxia 1 or Lactic Acidosis 3. The correlation between these latter two factors is 99%, so we label them collectively as the p53 specific factor. Gene ontology for this factor is identical to that for the ER factor with the exceptions that “cell proliferation” and “DNA replication initiation” are replaced by “nuclear division” and “M phase”. For all gene ontologies listed in the top eight for these two factors, the Bayes factors are ≥10. Because of the high degree of similarity in the gene ontology, it is tempting to try to equate these two factors. shows a scatterplot of the activity of the tumors in the data from on each of the two factors. The p53 factor is significantly bimodal, and the mild correlation one can see is due entirely to this bimodality, as tumor samples with high ER factor activity are more likely to be in the second mode of the p53 factor. We theorize that this bimodality is associated with a particular subtype of the p53 mutation. However, there is no evidence of multimodality in the ER factor, and the p53 specific factor predicts ER status poorly. Because of these differences, and because cell replication is a complex process, it is likely that these two factors are related to distinct features of cell development. We stress that, if we restrain ourselves to considering the original *in vitro* lactic acidosis signature, we have no ability to fit or predict any of these biological phenotypes. Additionally, these factors were generated entirely without regard to the ER status, PgR status, or p53 status of the samples. This is in contrast to a more typical design in which signatures associated with phenotypes are defined strictly based on genes with expression profiles that match those phenotypes (for example). This type of design is plagued with difficulties that arise from the large number of genes, out of the tens of thousands on an array, with expression patterns that match any arbitrary phenotype. With SFPA, we search for genes that are expressed together without regard to phenotype, and we are therefore much less likely to be plagued by false discovery (as proven by our out of sample predictive accuracy). ## Factors Predict Clinical Phenotypes SFPA offers a technique for interrogating a single independent tumor sample against any number of biologically determined signatures, and then consequent linking of factors to phenotypes may include clinically relevant outcomes such as patient survival outcomes and drug response. ### Factors improve prediction of breast cancer survival Subsets of the 67 factors were evaluated in Weibull survival regression models using the SSS method to identify and score models predicting survival. Each model in a resulting set of highly scoring models produces fitted survival curves and also may be used to predict survival for new samples. Bayesian analysis mandates averaging predictions from such a set of models, and this was done to result in. This shows fits of survival curves for the training data set, together with out of sample predictions in four of the other data sets for which information regarding survival exists. Recall that these are data sets from quite distinct and diverse studies, so we are assessing a model fitted to one data set on four quite challenging out of sample validation data sets. Though not described further here, the BFRM statistical model analysis used by the SFPA also addresses issues of gene-sample-study specific effects within the analysis and is able to correct enough of the idiosyncracies and bias inherent in microarray assays to retain predictive accuracy. The results demonstrate that the factorprofiles of these *in vitro* environmental signatures can improve survival prediction significantly across several test data sets. Similar results are obtained for the prediction of metastasis-free survival. ### Factors predict Tamoxifen response Four of the breast cancer data sets have clinical annotation pertaining to treatment with Tamoxifen. Though the 67 factors are in no way specifically targeted at Tamoxifen, we do know they are associated with relevant biological pathways. From our 67 factors, we found that Lactic Acidosis 1 is predictive of Tamoxifen resistance. It differentiates metastasis-free survival in patients who received the drug and shows no predictive ability in patients who did not (the analysis underlying this followed the same approach as for survival discussed above). Because all of the patients who received Tamoxifen were ER positive, drug resistance associated with this factor must be independent of the antagonistic action of the drug on estrogen receptors. Since none of these data sets were used in the training of the factor model, the ability of these factors to distinguish resistance to Tamoxifen is remarkable and demonstrates that they are robust to the collection biases often seen in microarray experiments. We again used GATHER to study the ontology of the genes included in this factor. This connects with the known association of Tamoxifen with phosphate transport, as well as cell adhesion. In particular, Cowell et al. report that p130Cas/BCAR1 is a cell adhesion molecule that promotes resistance to Tamoxifen via a particular phosphorylation pathway. In addition to these connections to the secondary effects of Tamoxifen is the well-known connection between survival of patients on Tamoxifen and toxicity associated with blood coagulation. Further study of the genes in this factor may lead to insight into the mechanism behind Tamoxifen resistance in ER positive breast cancer. ### Discovery of organ-specific factors from lactic acidosis signatures While the same biological processes may contribute to tumor phenotypes in different cancers, the process by which this happens may be entirely different given the particular cellular context, tissue-specific gene expression and epigenetic influences. Since SFPA can utilize *in vivo* cancer gene expression to dissect the *in vitro*-generated gene signature, it offers the possibility of identifying tissue and organ-specific factors associated with the same gene signatures. This application has the potential to distinguish sub-pathways that are conserved across many tissue types from those that are organ-specific. To illustrate this point, we utilize the lung cancer data set published in and the ovarian cancer data set from. We obtained the lung cancer data from GEO and the ovarian cancer data from the Duke Integrative Cancer Biology Program (ICBP) web site (<http://data.cgt.duke.edu/platinum.php>). We performed SFPA analysis of the same lactic acidosis signature as a starting point for factor discovery from the lung and ovarian cancer data sets. In the case of the lung cancer, the analysis discovered 20 factors associated with lactic acidosis. When we compared the expression levels of the 10 lactic acidosis factors in the breast cancer data with the 20 lactic acidosis factors discovered in the lung cancer data, we found that several factors are highly conserved, including the tamoxifen factor, the p53 specific factor, as well as factors 7 and 8. In contrast, the ER and PgR factors are only found in breast cancers. If we look specifically at standardized raw expression levels for the genes in the ER factor in the breast data as compared to that for the lung data we see that there is consistent variation in the breast data which is not present in the lung data. In contrast, the standardized raw expression for the conserved tamoxifen factor shows a coherent expression pattern in both breast and the lung cancers. Additionally, within this data set, our newly discovered factors also possess significant prognostic value, being able to distinguish between the two types of cancers as well as distinguish between high and low risk patients. Similar observations are also present in ovarian cancer since the model averaged survival using the 8 lactic acidosis factors discovered in the ovarian cancer dataset can clearly differentiate high versus low acuity patients. Additionally, we see the same patterns of loss of the ER factor and conservation of the tamoxifen factor. Finally, we find that the exact same three factors, the p53 specific factor, factor 7, and factor 8, have analogous factors in the ovarian cancer data with greater than 90% correlation (as computed on the 251 breast cancer samples). # Discussion It is increasingly common for investigators to use gene expression signatures directly as phenotypes to link various biological processes and perturbations to disease phenotypes and chemical agents. Although these signatures derived *in vitro* offer a way to understand the *in vivo* biology, there is still considerable limitation due to the differences between these two settings. Here, we have exemplified a statistical approach to further improve the *in vitro* gene signatures based on the gene expression in *in vivo* cancer samples. Elaborating the factor profile underlying the original signatures can, as we have seen, improve the *in vivo* relevance by more fully describing the diversity of *in vivo* expression patterns. This may enhance prognostic value and provide mechanistic insights into how biological processes affect clinical phenotypes. As an example, we have found direct links between factors generated by the use of SFPA on the lactic acidosis signature. Such links are intriguing, and open questions regarding causation as well as questions about the biological associations of the remaining factors. However, regardless of links to known biology, this strategy and analysis seem to provide an advance in our ability to obtain consistent results across many different data sets collected at different times by different groups. This is a significant advance, as data collection inconsistencies are one of the main roadblocks to the use of microarrays in a clinical setting. There are several possible explanations for the enhancement of the prognostic values achieved with SFPA. It is possible that certain genes or pathway components in the original gene signatures are simply noise or artifact due to their *in vitro* origins. These genes may offer no or even negative prognostic values for *in vivo* biology. By using SFPA to separate different components, it is possible to enhance the prognostic value by selecting only the relevant components or genes for predictive purposes. By so doing, it is also possible to examine the genes comprising those factors with strong links to clinical phenotypes which will lead to biological insights into this association. Another opportunity this analysis raises is the ability to uncover the pathways which would be “hidden” in the *in vitro* signature. In our current study, factor one was not immediately recognizable as a clinically relevant list of genes, but the ability of this factor to predict patient resistance to Tamoxifen points to an important connection which would not have been possible to discern otherwise. This observation will lead to efforts in investigating the biological roles of this factor and how it is related to Tamoxifen treatment and cellular response to lactic acidosis. For example, it is well known that tumor hypoxia negatively impacts clinical outcomes, but the actual mechanism by which this occurs is complex and may include radiation resistance, increased tumor invasion, migration, increased survival and decreased drug sensitivities. Although these hypoxia-induced effects occur in cancer patients, many of these events cannot be replicated or modeled in any particular *in vitro* setting. It is possible to uncover these processes via of the use of SFPA for the cancer gene expression. In a similar fashion, it is unclear how lactic acidosis responses are linked to good prognosis, and SFPA will allow us to explore *in vivo* gene expression to dissect this response and develop testable biological hypotheses. Equally importantly, the mechanisms by which hypoxia and lactic acidosis link to different clinical outcomes may vary among different cancer types, and the use of SFPA can specifically pinpoint the relevant biological processes to target or intervene to modulate clinical courses of cancer patients. Tremendous resources continue to be expended on the discovery of biomarkers for drug susceptibility. The ability to predict susceptibility to a given drug has the potential to significantly increase efficacy while decreasing morbidity and mortality in the relevant patient population. Additionally, it opens the possibility of facilitating the process of bringing new drugs to market. We have demonstrated the efficacy of SFPA for translating signatures discovered *in vitro* into factors which are clearly related to specific biological processes and which can be used to assess important clinical outcomes. The factors may be applied to just one observation (an important consideration for use in a clinical setting), and remain consistent across many different data sets. We do view this as a useful step ahead in thinking about how gene expression genomics will advance us towards the goals of personalized medicine. # Methods A total of five signatures were derived from two different experiments on Human Mammary Epithelial Cells (HMEC). The details of the collection of gene expression data from these cell lines are in. Signatures from these experiments were derived using the Bayesian Factor Regression Modeling (BFRM) software detailed in, and that has been used in multiple previous analyses of similar data. The workspaces used for BFRM are available in the supplementary, and the software is publicly available. In designed experiments such as, BFRM provides a sparse ANOVA framework for studying changes associated with environmental stresses. It includes functionality for correcting systematic laboratory bias which come about due to differences in conditions under which the microarray data is collected. These systematic differences are reflected in the doping control genes, which are used to construct the correction factors. ## Sparse Regression for Experiments with Known Variation BFRM is a Bayesian modeling framework. As such, we assume that all of the parameters of our model are random variables. In order to learn more about the values of these parameters, we specify prior distributions, which are subsequently updated based on the data. The result of fitting the model to data in this way is a joint posterior distribution for all of the model parameters. In our case, the parameters of interest are the coefficients of the regression. The general model implemented in BFRM is as follows. Let be a matrix of expression values where (row, column) is the expression of gene from sample where. Denote the design matrix (describing known sources of variability) by having elements on sample and design or regression variable. The model may be written as a separate linear regression for each probe on the array: Or alternatively in matrix notationwhere is a matrix of regression coefficients having elements and is a diagonal covariance matrix with non-zero diagonal elements. We allow the regression coefficients to vary across both genes and design vectors,but assign them a sparsity prior. We define to be the posterior mean for, to be the posterior probabilities on non-zero values of the, and to be the posterior mean of. All of these parameters are computed automatically by BFRM (along with many others). We have used a prior distribution for the coefficients of the regression that has a point mass at zero. This reflects our belief that, for any particular intervention, there will be relatively few genes (of the over ten thousand that are measured in a microarray experiment) that are affected. For the case outlined in this paper, we argue that growing mammary epithelial cells in the presence of mild lactic acidosis has led to changes in the expression of some of the genes on the array, but that most remain unchanged. Thus our posterior distribution for each will consist of a probability that the parameter is non- zero (corresponding to a probability that the gene is differentially expressed in the lactic acidosis experimental group versus the control group), along with a distribution on the magnitude of that possible change. The prior on is assumed to be a diffuse inverse gamma distribution (which is a standard conjugate prior), and the prior on is also given a point mass mixture prior, reflecting the belief that we must maintain significant mass around the extremes (zero and one) even after updating with all of the probes on the chip (50,000+). The precise values of all hyper-parameters are available in the parameter files in the supplementary section. We define a signature to be a list of genes and associated weights. Using the posterior parameters from above we define the weight of gene for experimental group (design variable) to be. Calculation of the level of activity of a known signature within a tumor sample requires that we initially subtract mean expression levels and laboratory biases. These are computed with BFRM exactly as above with the exception that the design matrix contains only the intercept vector and correction factors (no design vectors). If is a *p*-dimensional vector of corrected expression values associated with tumor sample *m*, then the signature score of signature in sample *m* is the weighted average. ## Analysis of Breast Cancer Data Sets We use six cancer data sets with Affymetrix U133+ expression samples available on the Gene Expression Omnibus (GEO) web site. Details of the collection and measuring are contained in,. For all but Wang, Affymetrix.CEL files were available, and we computed RMA normalized values in these cases. For the data set from, we used MAS5 normalized data which was obtained from the authors. Statistical factor analysis using BFRM estimates latent factors that represent common, underlying aspects of covariation of subsets of genes, typically representing expression gene-by-gene in terms of contributions from possibly several factors. The iterative analysis to expand on an initial set of signature genes that we used here then revises the gene list by adding in genes apparently associated with estimated factors, and then refitting the model. Full details of this algorithm are available in. To choose a collection of seed genes associated with experimental group *j*, we identify all genes, *g*, such that. From this list of genes, we take the 25% with the highest absolute change in expression level between the control group and the corresponding experimental group (as measured by the posterior). Following this procedure, we obtain between 20 and 200 “seed” genes for factor analysis. Given a signature, we must choose a collection of tissue samples on which to train the factor model. Because of its relatively large size, the availability of CEL files, and the wealth of clinical and phenotypic information, we chose the data set from for the identification of factors. We have five sets of seed genes, obtained as described above, from experiments on HMEC's. For each of these five sets of genes, we independently use BFRM to obtain the factors that are represented. We limited the number of genes to recruit into factors to a total of 500. To fit our binary regression and survival models, statistical analysis used Shotgun Stochastic Search (SSS) routines from. Initialization files used by SSS for these searches are included in the supplementary,. All Kaplan-Meier curves showing differential survival are drawn by separating samples at the median of the score that is relevant for that figure. ## Statistical Factor Models for Tumor Expression Data Factor models are structured as in. If represents the column vector of gene expression measures on *p* genes as assayed in a single tumor, is regressed linearly on a combination of an overall intercept term and assay correction factor, plus a set of latent (i.e., to be estimated) factors. If is the column vector of known regressors (the intercept and assay correction terms) on tumor *i*, the model is of the formwhere is a column vector of unknown latent factor values on tumor *i*, and *A*,*B* are coefficient matrices. In the BFRM context, both *A* and *B* are large “tall and skinny” matrices with many more rows (genes) than columns (the number of regressors and factors), and are described by the same sparsity probability models introduced above for the elements of (inducing many zeros). Implicit in this formulation is the assumption that there is a set of vectors, equivalent to design vectors, which describe some part of the variation observed in the matrix of expression values. This leads to the grouping of probes in a way that is comparable to clustering, whereby we assign genes corresponding to non-zero values in any particular column of to the same group. This is exactly parallel to sparsity in the coefficients associated with the design vectors in that we are assuming that most genes are not differentially expressed with any single latent factor. Calculation of the activity of a set of factors, on each tumor *i*, and estimation of the factor loadings, is then a problem of statistical estimation of the overall model. Details of these calculations are available in. The issue of projecting factors to a new sample, is then one of prediction that is immediately available from the BFRM analysis framework. For completeness, we present the formula here: Where are approximations of the factor scores for a new observation, with gene expression values and design variables. # Supporting Information [^1]: Conceived and designed the experiments: JTC. Performed the experiments: JLYC JTC. Analyzed the data: JEL CC MW. Wrote the paper: JEL MW. [^2]: The authors have declared that no competing interests exist.

# Introduction Juvenile-onset systemic lupus erythematosus (JSLE) is a heterogeneous systemic autoimmune disease characterized by onset before 18 years of age and multi-organ involvement. The occurrence of JSLE was estimated at approximately 20% of all systemic lupus erythematosus (SLE) cases. The disease course in JSLE is relatively more severe than that in adult-onset SLE due to increased rates of renal, neuropsychological, and hematological manifestations. Lupus nephritis (LN) and central nervous system involvement are predictors of poor prognosis in JSLE. Differences in clinical phenotypes related to underlying immune mechanisms were reported to involve innate and adaptive immune responses. T and B cell abnormalities contribute to the process of immune dysregulation and loss of self-tolerance in this disease. Imbalance of lymphocyte subsets has been demonstrated in SLE patients and showed some correlations with clinical manifestations and disease activity. Reduced CD4⁺ T cell frequency and elevated CD8⁺ T cell frequency were observed in JSLE patients with high disease activity. An elevated B cell proportion was related to increased incidence of arthritis, in line with a previous study indicating that musculoskeletal involvement improved after receipt of B cell-targeted therapy. Small subsets of peripheral blood lymphocytes, such as gamma delta T cells (γδ T cells) and natural killer (NK) cells, are notably involved in the pathogenesis of many autoimmune diseases. γδ T cells play an essential role in the pathogenesis of SLE by acting as antigen-presenting cells, producing pro- inflammatory cytokines, having immunoregulatory functions together with regulatory T cells (Tregs), and enhancing autoantibody production by B cells. Previous studies demonstrated that SLE patients had a lower frequency of γδ T cells in their peripheral blood compared with healthy controls (HC), which may result from infiltration of γδ T cells into target tissues such as the skin and kidneys or from the disease itself. Tregs have a critical function in maintaining self-tolerance through suppression of autoreactive lymphocytes. Previous studies reported contradictory results regarding the use of different markers and gating strategies to identify Tregs. The frequency of CD4⁺CD25⁺FoxP3⁺ cells was lower in SLE patients compared with HC and was associated with the SLE disease activity index (SLEDAI). Moreover, the reduction in CD4⁺CD25⁺FoxP3⁺ cells was linked to kidney damage in active SLE patients. However, other studies reported unchanged CD4⁺CD25^+/hi cell frequency with or without FoxP3⁺ cells or even increased frequency of these cells in SLE patients compared with HC. NK cells are innate immune system cells responsible for cytotoxic functions and act as regulatory cells in the context of inflammation. Natural killer T (NKT) cells have shared characteristics of NK cells and T cells (CD3⁺). Previous reports described decreased frequencies of NK cells and NKT cells in SLE patients related to high disease activity and renal activity index. Previous studies on SLE have mostly focused on the cellular components in adult patients, with findings that the lymphocyte subsets varied according to age, ethnicity, and environmental factors. Therefore, the findings have limited applicability in JSLE. In addition, the complexity of the pathogenesis for JSLE suggests the need for research targeting young patients rather than adult patients. In the present study, we aimed to investigate the lymphocyte subsets (CD4⁺ T cells, CD8⁺ T cells, γδ T cells, Tregs, CD19⁺ B cells, NK cells, and NKT cells) in JSLE patients and determine their associations with clinical phenotypes and disease activity, as well as long-term outcomes. These findings may fill the gap in knowledge on the biological process of the disease and identify subgroups of patients with different phenotypes and prognoses, leading to better personalized treatment strategies in JSLE. # Methods ## Study design and patients This was a prospective cohort study. A total of 60 patients aged \<18 years who were diagnosed with JSLE according to the Systemic Lupus International Collaborating Clinics classification criteria at the Pediatric Rheumatology and Pediatric Nephrology clinics of our hospital between May 2015 and December 2018 were included in the study. Blood samples were obtained from all 60 JSLE patients during active disease with or without medications, and follow-up samples were obtained in 34 of the 60 JSLE patients during inactive disease. For a longitudinal follow-up study, 49 of the 60 JSLE patients were followed up at 0, 3, and 6 months. The HC group comprised 42 age-matched children from our previous study. Baseline characteristics and clinical information, including concurrent medications, were routinely collected during the follow-up visits. Written informed consent was obtained from the legal guardians of the study participants before enrollment. The study was approved by the Ethics Committee of Ramathibodi Hospital (ID 055806) and conducted in accordance with the Declaration of Helsinki. ## Disease activity measurement We used the SLE disease activity index 2000 (SLEDAI-2K) to measure disease activity. Inactive disease was defined as clinical SLEDAI-2K (cSLEDAI-2K), excluding anti-double-stranded DNA (anti-dsDNA) and complement, equal to zero regardless of medication. Clinical remission was defined as no clinical activity (cSLEDAI-2K = 0) for 1 year. Three levels of remission were defined as follows: (i) remission on therapy, patients with clinical remission, physician global assessment of disease activity \<0.5 on a 0–3 visual analog scale, daily prednisolone dose ≤5 mg/day or ≤0.2 mg/kg/day for body weight \<25 kg, with or without antimalarials and immunosuppressants; (ii) remission off therapy, patients with clinical remission without any medications or with only maintenance antimalarials; and (iii) complete remission, patients with clinical remission and serological remission (normalization of anti-dsDNA and complement levels), without any medications or with only maintenance antimalarials. Laboratory parameters including complement component (C)₃, C₄, erythrocyte sedimentation rate (ESR), anti-dsDNA, complete blood count, direct Coombs test (DCT), urinalysis, and urine protein-to-creatinine ratio were collected. ## Immunophenotyping Peripheral blood mononuclear cells were stained with fluorochrome-conjugated antibodies (eBioscience, San Diego, CA, USA). Briefly, 10–20 μL of EDTA-treated peripheral blood was incubated with fluorochrome-conjugated antibodies against cell-surface antigens for 15 minutes at room temperature in the dark. The antibodies used in the staining process were fluorescein isothiocyanate (FITC)-conjugated anti-human CD3, allophycocyanin (APC)-conjugated anti-human CD4, APC-eFluor 780-conjugated anti-human CD8, phycoerythrin (PE)-conjugated anti-human γδ TCR, APC-conjugated anti-human CD19, PE-conjugated anti-human CD56, and PE-Cy7-conjugated anti-human CD45. Red blood cells were lysed at room temperature using Lysing Buffer (BD Biosciences, San Jose, CA, USA) for 10 minutes before measurement in a flow cytometer (BD FACSVerse; BD Biosciences). For Treg staining, 10–20 μL of EDTA-treated blood was stained with PE- Cy7-conjugated anti-human CD25 and APC-conjugated anti-human CD4 (1:100 dilution) for 15 minutes at room temperature, and then incubated with Fixation/Permeabilization Solution (eBioscience) for 15 minutes in accordance with the manufacturer’s protocol. Finally, the stained cells were incubated with Permeabilization Buffer (eBioscience) containing 1:50 dilution of FITC- conjugated anti-human FoxP3 at room temperature for 45 minutes and analyzed by flow cytometry. The gating strategies for the lymphocyte subsets were as follows: CD4⁺ T cells (CD3⁺CD4⁺); CD8⁺ T cells (CD3⁺CD8⁺); γδ T cells (CD3⁺CD4⁻CD8⁻γδ TCR⁺); B cells (CD3⁻CD19⁺); NK cells (CD56⁺); NKT cells (CD3⁺CD56⁺); and Tregs (CD4⁺CD25⁺FoxP3⁺). The percentages of the T cell subsets among the total lymphocytes were analyzed using FlowJo v.10 software (FlowJo LLC, Ashland, OR, USA). ## Statistical analysis We used IBM SPSS statistics 25 software (IBM Corp., Armonk, N.Y., USA) and GraphPad Prism 8.3 software (GraphPad Software Inc., La Jolla, California, USA) for the data analyses. Demographic, clinical, and laboratory parameters were presented as median, interquartile range (IQR), and percentage as appropriate. Categorical data were compared using the chi-square test or Fisher exact test. For continuous data, the Mann–Whitney U test (unpaired data) or Wilcoxon matched-pairs signed-rank test (paired data) was used for comparisons between two groups. For comparisons between three or more groups, the Kruskal–Wallis test (unpaired data) or Friedman test (paired data) was used with a post-hoc Dunn test. The probability of remission on therapy was analyzed by the Kaplan–Meier method. Predictive factors were evaluated by Cox proportional hazards regression analysis and presented with hazard ratio and confidence interval (CI). Statistical significance was accepted for a two-sided p-value of \<0.05. # Results ## Patient characteristics The demographic, clinical, and laboratory parameters of the JSLE patients are shown in. The JSLE patients in the study cohort had a median age of approximately 12 years, which was comparable to that in HC. Of the 60 JSLE patients, 34 provided blood samples for longitudinal follow-up during both active disease (active group) and inactive disease (inactive group). The majority of the total JSLE patients and longitudinal follow-up JSLE patients were female. The median SLEDAI-2K at enrollment in all JSLE patients was 12 (IQR 6.00–18.00). For the longitudinal follow-up JSLE patients, the median SLEDAI-2K was 10 (IQR 6.00–16.25) in the active group and 0 (IQR 0.00–2.00) in the inactive group. The median C₃ and C₄ levels were low in all JSLE patients and the active group. The median ESR was quite similar in all JSLE patients and in the active group, but was nearly normal to normal in the inactive group. Low median values of white blood cells, absolute lymphocyte count (ALC), platelet count, and hematocrit were found in all JSLE patients and the active group. The percentages of patients with anti-dsDNA positivity and DCT positivity were 73.33% and 40% in all JSLE patients, 64.71% and 38.24% in the active group, and 8.82% and 11.76% in the inactive group, respectively. Skin involvement, mucosal ulcer, and fever were the most common clinical manifestations. Approximately 35% of JSLE patients were included before starting treatment. Prednisolone was administered with the highest frequency among the medications. ## Differences among lymphocyte subsets between JSLE patients and HC We first assessed the percentages of lymphocyte subsets in JSLE patients compared with age-matched HC. The percentages of CD4⁺ T cells, γδ T cells, and NK cells were significantly lower in JSLE patients compared with HC. In contrast, the percentages of CD8⁺ T cells, NKT cells, and CD19⁺ B cells were significantly higher in JSLE patients compared with HC. The percentages of Tregs and Tregs within CD4⁺ cells did not differ significantly between JSLE patients and HC. The data for the percentages of lymphocyte subpopulations are provided in. ## Differences in lymphocyte subsets among non-LN JSLE patients, LN-JSLE patients, and HC JSLE patients with LN have different disease severity and outcomes compared with non-LN JSLE patients. Therefore, the frequencies of lymphocyte subsets in these two groups of JSLE patients were evaluated and compared with HC. Disease activity measured by SLEDAI-2K differed significantly between non-LN JSLE patients and LN JSLE patients (median 10 \[IQR 6–16\] vs. median 21 \[IQR 11–26\], p = 0.0011). The frequency of Tregs in LN JSLE patients was significantly lower than that in non-LN JSLE patients and HC. The frequencies of CD4⁺ T cells and CD8⁺ T cells differed significantly between non-LN JSLE patients and HC but did not differ significantly between LN JSLE patients and non-LN JSLE patients. The frequencies of γδ T cells, NK cells, and CD19⁺ B cells differed significantly between HC and LN JSLE patients and between HC and non-LN JSLE patients. However, there were no significant differences in these cell percentages between LN JSLE patients and non-LN JSLE patients. ## Associations between lymphocyte subsets and clinical manifestations in JSLE We further evaluated whether the differences in lymphocyte subsets were associated with other clinical manifestations. We stratified the patients into high and low groups by the median frequency of each subset and compared the clinical manifestations between the sets of two groups. The γδ T cells high group and NK cells high group were significantly related to higher frequency of mucosal ulcer, and the CD4⁺ T cells high group was significantly associated with higher rate of arthritis. The NKT cells high group was substantially linked with higher number of patients with autoimmune hemolytic anemia (AIHA). The CD8⁺ T cells low group was mainly related to higher frequency of vasculitis, and the Tregs low group was significantly associated with higher rate of LN. ## Lymphocyte subset distribution during longitudinal follow-up of JSLE patients Thirty-four of the 60 JSLE patients had follow-up samples until inactive disease. At enrollment, 15 patients with active disease were treatment-naïve and 19 were treated with glucocorticoid therapy. After treatment, even in the inactive disease state, the percentages of CD4⁺ T cells, CD8⁺ T cells, γδ T cells, NK cells, Tregs within CD4⁺ cells, and CD19⁺ B cells did not reach similar percentages to HC, even though the median SLEDAI-2K in the inactive group was 0 (IQR 0–2). The percentages of Tregs within CD4⁺ cells and CD8⁺ T cells were substantially increased in the inactive group compared with the active group. Meanwhile, the percentage of CD19⁺ B cells was significantly lower in the inactive group compared with the active group. In contrast, the percentages of Tregs, CD4⁺ T cells, γδ T cells, NK cells, and NKT cells did not differ significantly between the active and inactive groups. ## Tregs in JSLE patients during the 6-month follow-up period For 49 of the 60 JSLE patients, follow-up blood samples for the Tregs study were collected at 0, 3, and 6 months. The percentage of Tregs in all JSLE patients was significantly increased at 6 months of follow-up compared with 0 months (median 1.25 \[IQR 0.53–2.70\] vs. median 0.85 \[IQR 0.31–1.82\], p = 0.0259) (Fig). In the subgroup analysis, LN JSLE patients had a lower frequency of Tregs than non-LN JSLE patients at 6 months of follow-up (median 0.99 \[IQR 0.39–1.53\] vs. median 1.53 \[IQR 0.65–2.84\]) (Fig). The percentage of Tregs between treatment-naïve patients and treated patients at enrollment is shown in . There was a significant increase of Treg percentage at 6 months in treatment- naïve patients with LN (Table). During the 6-month follow-up, SLEDAI-2K decreased over time in all JSLE patients, particularly in non-LN JSLE patients. ## Long-term outcomes in JSLE patients At the end of the study, JSLE patients had a remission on therapy rate of 70%, remission off therapy rate of 21.67%, and complete remission rate of 18.33%. As shown in, the median time to achieve remission on therapy was longer in the Tregs low group compared with the Tregs high group (38.80 \[CI 20.06–57.54\] months vs. 15.27 \[CI 14.51–16.03\] months, p = 0.008). The number of patients in remission on therapy at the end of the study was 18 (60%) in the Tregs low group and 24 (80%) in the Tregs high group. Moreover, we also performed the Kaplan-Meier analysis only on the treatment-naïve patients. This result corresponded with the finding from all JSLE patients and showed the difference between Tregs low and high groups more distinct. In the treatment-naïve patients, the median time to achieve remission on therapy was longer in the Tregs low group (58.40 \[CI 20.51–96.29\] months) compared with the Tregs high group (14.70 \[CI14.16–15.24\] months) with more statistical significance (p\<0.001) as shown in. In addition, 50% of treatment-naïve patients in the Tregs low group and 100% of Tregs high group achieved remission on therapy. Furthermore, multivariate analysis was performed by selecting four significant covariates in univariate analysis, and it showed that predictors of remission on therapy were high Tregs, high ALC (\>1.5×10⁶/L), positive DCT, and non-LN JSLE at enrollment. Other lymphocyte subsets, clinical manifestations, disease activity, anti-dsDNA positivity, medications, age, and sex at enrollment were not related to remission on therapy. # Discussion The present study showed that dynamic changes in the percentages of lymphocyte subsets occurred during the disease course of JSLE. The frequency of Tregs in LN JSLE patients was substantially lower than that in non-LN JSLE patients and HC, even though no significant difference was observed between all JSLE patients and HC. The present findings further demonstrated relationships between lymphocyte subsets and clinical phenotypes. For the longitudinal follow-up study at 0, 3, and 6 months, the percentage of Tregs in all JSLE patients was significantly increased at 6 months of follow-up. In the subgroup analysis, the LN JSLE group had a lower frequency of Tregs compared with the non-LN JSLE group at 6 months of follow-up. Predictors of remission on therapy were high Tregs, high ALC, positive DCT, and non-LN JSLE at enrollment. Consistent with previous studies, the percentages of CD4⁺ T cells, γδ T cells, and NK cells were low in JSLE patients. However, there were some discrepancies between the percentages of CD8⁺ T cells, Tregs, NKT cells, and CD19⁺ B cells in JSLE patients and HC in both the present study and previous studies. This variation may be explained by the wide spectrum of disease phenotypes and disease mechanisms that vary by age, ethnicity, and environmental factors. An imbalance of lymphocyte subsets is strongly involved in the immune response process in SLE patients. CD4⁺ T cell functions and cytokine production are impaired in SLE patients. Moreover, CD8⁺ T cells and NK cells have impaired cytotoxic functions. NK cells contribute to the inflammatory response through type I and type II interferon (IFN) via interactions with plasmacytoid dendritic cells. Furthermore, activated NKT cells produce various cytokines and chemokines that regulate T cells, B cells, NK cells, and dendritic cells. B cells are crucial for the production of autoantibodies, resulting in immune complex activation and organ inflammation. Not all SLE patients respond well to targeted B cell therapy, implying that the immune cell dysfunction in SLE is not specific for B cells alone. In a previous study, γδ T cells exhibited a regulatory function by inhibiting activated CD4⁺ T cells and dendritic cells. Notably, γδ T cells were related to disease activity and disease progression in SLE patients. From initiation of treatment to the achievement of inactive disease, the percentages of CD4⁺ T cells, CD8⁺ T cells, γδ T cells, NK cells, Tregs within CD4⁺ cells, and CD19⁺ B cells in JSLE patients did not recover to the normal levels. Persistently abnormal frequencies of CD8⁺ and CD4⁺ T cells may be associated with intrinsic defects irrespective of medications received. An environment containing high levels of circulating immune complexes could stimulate NK cell apoptosis, leading to a decreased number of NK cells. Regarding γδ T cells, their levels in patients with adult-onset SLE who responded to treatment gradually returned to the normal level within 12 weeks. It will be interesting to determine whether the persistently abnormal percentages of lymphocyte subsets during inactive disease reflect subclinical ongoing systemic inflammation and are related to a flare of the disease in JSLE. Tregs have roles in immune regulation and maintenance of self-tolerance. Defects in these cells can result in autoimmune diseases. Previous studies showed that the percentage of Tregs was significantly decreased in the active SLE patients compared with the inactive SLE patients and HC. In contrast, the present study found that the percentage of Tregs did not differ significantly among active JSLE patients, inactive JSLE patients, and HC. Moreover, we found a significantly lower frequency of Tregs in LN JSLE patients compared with non-LN JSLE patients and HC. These findings indicate that the role of Tregs may be more pronounced in LN JSLE patients. In the longitudinal follow-up study, the percentage of Tregs increased substantially over 6 months, corresponding to the decrease in disease activity. An imbalance between Tregs and excessive T cell and B cell activation has been demonstrated in SLE disease development. The possible mechanism behind this finding is that Tregs are suppressed by activated immune cells and various inflammatory cytokines in an inflammatory environment. This hypothesis could explain the increased Tregs after treatment due to reduced inflammation. Moreover, immune complexes, the main pathogenesis in LN JSLE patients, induce a high IFN-α response that leads to Treg suppression, and this could be another mechanism for the altered numbers of Tregs in these patients. In a previous study, SLE patients under immunosuppressive medications had significantly increased Tregs levels compared with untreated patients. We demonstrated percentage of Tregs between treatment-naïve and treated JSLE with LN during the 6-month follow-up period and showed a clearer alteration pattern of Treg percentages in treatment-naïve JSLE patients. LN patients with naïve treatment also had a lower percentage of Tregs than treated patients at baseline. In addition, they showed significantly increased Treg percentage at 6 months, implying that lower Treg percentage at baseline derived from the active disease more than immunosuppressive medications. However, further study with a larger sample size regarding the treatment affecting Tregs should be performed. The contradictory results of Tregs in various studies are from multiple factors. First, since there are several markers of Tregs with multiple phenotypic features, the variable Treg markers in each study cause the difference of Tregs results. The earlier studies found that the percentage of Tregs was significantly lower in active SLE compared to controls when CD25⁺ or CD25^high cells were used as Treg markers. In contrast, other studies that used FoxP3⁺ or CD127^low staining showed a comparable percentage of Tregs between active SLE and controls. A previous study suggested that CD25 alone should not be classified as Tregs because many of these cells were FoxP3 negative, and other activated T cells can express CD25. Regarding extracellular staining CD4⁺CD25⁺CD127^- cells, a previous study reported that CD127 is also downregulated during early activation of effector T cells, and around one-third of CD127^low cells did not express FoxP3. Therefore, low expression of CD127 might not be a good marker that represents the Treg population. Unlike an important regulator in Treg development, a transcription factor FoxP3 is a more specific marker and remains the best marker of Tregs up to this point. Second, each study had a different definition of active SLE disease, and the studies with higher cut-off SLEDAI scores tended to have a lower percentage of Tregs. Our study results also supported this finding that the percentage of Tregs had increased while the disease activity had decreased. Third, since SLE is a heterogeneous disease, it is difficult for all studies to have the same patients’ baseline characteristics, especially the frequency of lupus nephritis, which might have an influence on the percentage of Tregs the most. The present results demonstrated associations between clinical phenotypes and alterations to lymphocyte subsets. We found that γδ T cells and NK cells were linked with mucosal ulcer, CD8⁺ T cells were associated with vasculitis, CD4⁺ T cells were associated with arthritis, NKT cells were related to AIHA, and Tregs were associated with LN. In previous studies, SLE patients with predominant skin damage were related to γδ T cells, while SLE patients with LN were associated with γδ T cells, NK cells, CD8⁺ T cells, and Tregs and the pathology of affected tissues such as the skin and kidneys confirmed local damage arising from cellular immunity. Abnormal activation of B cells and other lymphocyte subsets contributes to the pathogenesis of SLE and is related to certain features. Therefore, targeted therapies restricted to these cells may show promise for JSLE patients. The discrepancy of the results in the above studies could be explained by heterogeneity in the cellular defects and differences in the disease phenotypes among the JSLE patients. In the present study, the most frequent clinical feature of JSLE patients at enrollment was skin disease, followed by mucosal ulcer, fever, AIHA, and LN, which may affect the altered lymphocyte subsets. Regarding the relationships between lymphocyte subsets and long-term outcomes, we found that assignment to the Tregs high group at enrollment could predict clinical remission on therapy, and this subgroup of patients required a shorter time to achieve clinical remission on therapy. These results were supported by the finding that the Tregs low group was associated with LN, and the fact that LN is a known factor related to morbidity and mortality. Tregs tend to exhibit less proliferation in LN JSLE patients compared with non-LN JSLE patients due to the extensive inflammatory environment in LN, leading to more suppression of their function and proliferation. There is evidence from lupus-prone mice that adoptive transfer of Tregs effectively ameliorated glomerulonephritis and increased survival. The present findings strengthen the role of Tregs in the pathogenesis of JSLE and suggest that Treg-based therapies may have a benefit for the treatment of JSLE patients, particularly JSLE patients with LN. The limitations of the study were the small number of enrolled patients, including both treatment-naïve and treated patients, and the single-center design. Despite the relatively small number of patients, particularly those under longitudinal follow-up, we observed significant changes in the lymphocyte subsets in individual patients and revealed more aspects of the cellular involvement in JSLE. # Conclusions We demonstrated that changes in percentages of lymphocyte subsets were strongly associated with clinical phenotypes and the disease course in JSLE. Due to the heterogeneity of JSLE, patients with different clinical phenotypes had different dynamic changes in lymphocyte subsets. Therefore, stratification of JSLE patients based on lymphocyte subsets could help toward the establishment of personalized treatment strategies, leading to better outcomes. Furthermore, low levels of Tregs could help physicians identify a subgroup of JSLE patients with severe clinical manifestations and worse long-term outcomes who require more intense regimens. In particular, this subgroup of patients may be candidates for Treg-based therapies. # Supporting information We would like to thank the patients who participated in the study. We also thank Chompunuch Klinmalai, MSc, Division of Infectious Disease, Department of Pediatrics, Faculty of Medicine Ramathibodi Hospital, Mahidol University, Bangkok, Thailand, for performing the flow cytometry. Finally, we thank Alison Sherwin, Ph.D., from Edanz (<https://www.edanz.com/ac>) for editing a draft of the manuscript. 10.1371/journal.pone.0263536.r001 Decision Letter 0 Kuwana Masataka Academic Editor 2022 Masataka Kuwana This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Dec 2021 PONE-D-21-35318Associations of lymphocyte subpopulations with clinical phenotypes and long-term outcomes in juvenile-onset systemic lupus erythematosusPLOS ONE Dear Dr. Vilaiyuk, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Our reviewers found some interests in this manuscript, but one of them pointed out a number of comments that require improvement. I ask the authors to fully respond to all comments made by the reviewer in the revised manuscript. Please submit your revised manuscript by Jan 20 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Masataka Kuwana, MD, PhD Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  [" ext-link-type="uri" xlink:type="simple"\>https://journals.plos.org/plosone/s/ file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf"](https:// journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors _affiliations.pdf) 2\. Thank you for stating the following in the Funding Section of your manuscript:  SV received funding support from the Faculty of Medicine Ramathibodi Hospital, Mahidol University, Bangkok, Thailand (RF_59002). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form.  Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows:  SV received funding support from the Faculty of Medicine Ramathibodi Hospital, Mahidol University, Bangkok, Thailand (RF_59002). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. (URL funder website <https://www.rama.mahidol.ac.th/en>) Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 3\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Lerkvaleekul B et al. evaluated the association of lymphocytes subsets with clinical manifestations in patients with JSLE. The manuscript sounds interesting, but I have some concerns which authors need to answer. 1\. Previous studies have demonstrated quite contradictory results regarding with frequency of Treg cells. Some reported a reduced frequency of circulating Foxp3+ Treg cells, but others found an increased or comparable frequency of circulating Foxp3+ Treg cells. I suggest authors to discuss this issue in discussion section. 2\. Since the subsets of peripheral lymphocytes were dramatically changed by the background treatment, including glucocorticoid dose, and immunosuppressive agents. I suggest author to add background treatment in Table 1 and 2. 3\. Authors compared the clinical manifestations between high and low frequency of each subset in Figure 3. I wonder how did authors select the manifestations, only mucosal ulcer, arthritis, AIHA, vasculitis and LN? Please show all manifestations listed in BILAG and compare them between high and low frequency of each lymphocyte subset. 4\. Before showing Table 4, additional Table needs to be created comparing clinical characteristics depending on achievement of remission on therapy and select covariates for multivariate analysis using items having a P-value 0.05 in univariate analysis. There was no data supporting the items were properly selected as covariates for multivariate analysis in Table 4. 5\. In Figure 4 and Figure 5 showed the serial change of each subset and Treg proportion and Figure 6 showed the probability of clinical remission depending on the frequency of Treg cells. These results were influenced by background treatment, and I suggest authors to show these results were independent of background treatment. Reviewer \#2: There are no reports of lymphocyte subsets in Asian children with SLE, and this is an important finding that correlates with the phenotype. Of particular importance is the significantly lower proportion of regulatory T cells (Tregs) found only in patients with lupus nephritis (LN), a major cause of refractory disease. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0263536.r002 Author response to Decision Letter 0 19 Jan 2022 Response to reviewers 19 January 2022 Dear Editor, Thank you for allowing us to revise this manuscript. We have responded and revised this manuscript as reviewers' suggestions. We highlighted the revised sentences in yellow color throughout the entire manuscript. We also revised the format of the manuscript and added minimal data set as the journal's recommendation (2 supplement figures and 4 supplement tables). We removed the funding information from the manuscript and did not amend the previous funding statement given to the journal. With Regards, Soamarat Vilaiyuk Comments from reviewers Reviewer \#1: Lerkvaleekul B et al. evaluated the association of lymphocytes subsets with clinical manifestations in patients with JSLE. The manuscript sounds interesting, but I have some concerns which authors need to answer. 1\. Previous studies have demonstrated quite contradictory results regarding with frequency of Treg cells. Some reported a reduced frequency of circulating Foxp3+ Treg cells, but others found an increased or comparable frequency of circulating Foxp3+ Treg cells. I suggest authors to discuss this issue in discussion section. Response: We would like to thank the reviewer for your valuable comments on this manuscript. We have discussed this issue and added it in the discussion section as follows. The contradictory results of Tregs in various studies are from multiple factors. First, since there are several markers of Tregs with multiple phenotypic features, the variable Treg markers in each study cause the difference of Tregs results. The earlier studies found that the percentage of Tregs was significantly lower in active SLE compared to controls when CD25+ or CD25high cells were used as Treg markers. In contrast, other studies that used FoxP3+ or CD127low staining showed a comparable percentage of Tregs between active SLE and controls \[20,22\]. A previous study suggested that CD25 alone should not be classified as Tregs because many of these cells were FoxP3 negative, and other activated T cells can express CD25 \[20\]. Regarding extracellular staining CD4+CD25+CD127- cells, a previous study reported that CD127 is also downregulated during early activation of effector T cells, and around one-third of CD127low cells did not express FoxP3 \[24\]. Therefore, low expression of CD127 might not be a good marker that represents the Treg population. Unlike an important regulator in Treg development, a transcription factor FoxP3 is a more specific marker and remains the best marker of Tregs up to this point \[20,22\]. Second, each study had a different definition of active SLE disease, and the studies with higher cut-off SLEDAI scores tended to have a lower percentage of Tregs \[22\]. Our study results also supported this finding that the percentage of Tregs had increased while the disease activity had decreased. Third, since SLE is a heterogeneous disease, it is difficult for all studies to have the same patients' baseline characteristics, especially the frequency of lupus nephritis, which might have an influence on the percentage of Tregs the most. 2\. Since the subsets of peripheral lymphocytes were dramatically changed by the background treatment, including glucocorticoid dose, and immunosuppressive agents. I suggest author to add background treatment in Table 1 and 2. Response: We have added more details about medications in Table 1 and Table 2 as your suggestion. (Highlight in yellow color) 3\. Authors compared the clinical manifestations between high and low frequency of each subset in Figure 3. I wonder how did authors select the manifestations, only mucosal ulcer, arthritis, AIHA, vasculitis and LN? Please show all manifestations listed in BILAG and compare them between high and low frequency of each lymphocyte subset. Response: We analyzed all clinical manifestations but showed only the clinical manifestations that were significantly different as stacked bar chart. More details about lymphocyte subsets in each clinical manifestation are in supplement table (S1 Table). We also highlighted the yellow color in the clinical presentations and cells that were statistically different between the high and low frequency of each subset. 4\. Before showing Table 4, additional Table needs to be created comparing clinical characteristics depending on achievement of remission on therapy and select covariates for multivariate analysis using items having a P-value 0.05 in univariate analysis. There was no data supporting the items were properly selected as covariates for multivariate analysis in Table 4. Response: We have added the supplement table (S4 Table) that showed the comparison of clinical characteristics between patients who had achievement and non-achievement of clinical remission on therapy. We found that absolute lymphocyte count and lupus nephritis were significantly different between both groups (yellow highlight). The presence of the direct Coombs test also tended to have a significant difference. Therefore, we selected these three factors with our interesting factor, high Tregs, and performed univariate analysis. All of them showed significance in univariate analysis. Furthermore, we performed the multivariate analysis using these four factors as covariates, which showed significance. We also added other factors that readers might be interested in and showed in the univariate analysis. However, those factors were not significant. We also added more details as your suggestion in the result section of the manuscript, as shown below. Furthermore, multivariate analysis was performed by selecting four significant covariates in univariate analysis, and it showed that predictors of remission on therapy were high Tregs, high ALC (1.5×106/L), positive DCT, and non-LN JSLE at enrollment (Table 4). 5\. In Figure 4 and Figure 5 showed the serial change of each subset and Treg proportion and Figure 6 showed the probability of clinical remission depending on the frequency of Treg cells. These results were influenced by background treatment, and I suggest authors to show these results were independent of background treatment. Response: We have changed the presentation of Figure 4 and Figure 5 per your suggestion. Instead of presenting all JSLE patients, we classified patients into treatment-naïve and treated patients at enrollment. Overall results corresponded with the initial one (all JSLE patients), and it showed a clearer pattern, especially a significant increase of Treg percentage at 6 months in treatment- naïve patients with lupus nephritis. We also added more details in the result and discussion sections below (yellow highlight). In the result section. The percentage of Tregs between treatment-naïve patients and treated patients at enrollment is shown in Fig 5A. There was a significant increase of Treg percentage at 6 months in treatment-naïve patients with LN (Table in S3 Table). During the 6-month follow-up, SLEDAI-2K decreased over time in all JSLE patients, particularly in non-LN JSLE patients (Fig 5B). In the discussion section. In a previous study, SLE patients under immunosuppressive medications had significantly increased Tregs levels compared with untreated patients \[55\]. We demonstrated percentage of Tregs between treatment-naïve and treated JSLE with LN during the 6-month follow-up period and showed a clearer alteration pattern of Treg percentages in treatment-naïve JSLE patients. LN patients with naïve treatment also had a lower percentage of Tregs than treated patients at baseline. In addition, they showed significantly increased Treg percentage at 6 months, implying that lower Treg percentage at baseline derived from the active disease more than immunosuppressive medications. However, further study with a larger sample size regarding the treatment affecting Tregs should be performed. Regarding Fig 6, we have added another graph of Kaplan Meier analysis, which was performed only in treatment-naïve patients (Fig 6B). The graph of treatment- naïve patients demonstrated the same pattern with all JSLE patients (Fig 6A). However, the graph in treatment-naïve patients showed a more significant difference. We have added more details in the result section as below. Moreover, we also performed the Kaplan-Meier analysis only on the treatment- naïve patients (Fig 6B). This result corresponded with the finding from all JSLE patients (Fig 6A) and showed the difference between Tregs low and high groups more distinct. In the treatment-naïve patients, the median time to achieve remission on therapy was longer in the Tregs low group (58.40 \[CI 20.51-96.29\] months) compared with the Tregs high group (14.70 \[CI14.16-15.24\] months) with more statistical significance (p0.001) as shown in Fig 6B. In addition, 50% of treatment-naïve patients in the Tregs low group and 100% of Tregs high group achieved remission on therapy. Reviewer \#2: There are no reports of lymphocyte subsets in Asian children with SLE, and this is an important finding that correlates with the phenotype. Of particular importance is the significantly lower proportion of regulatory T cells (Tregs) found only in patients with lupus nephritis (LN), a major cause of refractory disease. Response We would like to thank the reviewer for your valuable comments on this manuscript. 10.1371/journal.pone.0263536.r003 Decision Letter 1 Kuwana Masataka Academic Editor 2022 Masataka Kuwana This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 21 Jan 2022 Associations of lymphocyte subpopulations with clinical phenotypes and long-term outcomes in juvenile-onset systemic lupus erythematosus PONE-D-21-35318R1 Dear Dr. Vilaiyuk, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Masataka Kuwana, MD, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No 10.1371/journal.pone.0263536.r004 Acceptance letter Kuwana Masataka Academic Editor 2022 Masataka Kuwana This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 28 Jan 2022 PONE-D-21-35318R1 Associations of lymphocyte subpopulations with clinical phenotypes and long-term outcomes in juvenile-onset systemic lupus erythematosus Dear Dr. Vilaiyuk: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Prof. Masataka Kuwana Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction It is well documented that the prevalence of adult hypertension is different by race in the US with Non-Hispanic Blacks having a higher prevalence compared to Non-Hispanic Whites. Considering the tracking of blood pressure (BP) levels from childhood to adulthood, it is critical to examine racial disparities among children and adolescents and to better understand the associated risk factors in order to narrow the racial gaps among adults. Relatively few studies have investigated racial differences in pediatric hypertension (PHT), with mixed findings. Most of the earlierstudies characterized racial differences in hypertension among children and adolescents as non-significant. However, Rosner and colleagues reported a significant racial difference between Hispanic and White boys, which persisted after controlling for body mass index (BMI). Body size and sex appeared to modify the association between race and PHT. Significant differences were also observed in subsequent studies by Lo et al. and Rosner et al. for Black and by Shay et al. for both Black and Hispanic youth compared to their White counterparts. In addition, low birth weight (LBW) has been linked with high risk of adult hypertension, although the evidence was less consistent among children and adolescents The role of LBW in racial disparities in PHT is unclear. The objectives of this study were to identify racial differences in the prevalence of PHT in the US and to explore whether such differences can be explained by body size and birth weight using the most updated nationally representative data from the National Health and Nutrition Examination Survey (NHANES) 1999–2010. # Methods ## Study Population The study population included children and adolescents aged 8–17 years who participated in the NHANES 1999–2010. Details of the NHANES design and procedures are available on the Center for Disease Control and Prevention (CDC) website. In brief, the 1999–2010 NHANES included 6 cross-sectional surveys of the non-institutionalized US population. These surveys applied a stratified multistage probability sampling design. Low-income populations, adolescents, seniors, Non-Hispanic Blacks, and Hispanics were over-sampled. Each survey consisted of a household interview and a physical examination in a fully equipped mobile examination center. The procedures and protocols of NHANES 1999–2010 were approved by the institutional review board of the CDC/ National Center for Health Statistics (NCHS). Informed consent was obtained from study participants or their guardians where appropriate. This study was a secondary analysis using the de-identified data released from the CDC, therefore, no additional consent procedures were carried out. For the current study, individuals who self-identified as “Other Hispanics” (6.37%) or “Other Race” (6.73%) were excluded due to small sample size. All the data are available at the CDC NHANES website and are open to public access **(**<http://wwwn.cdc.gov/nchs/nhanes/search/nhanes11_12.aspx>**).** ## Measurement of BP and Definition of PHT For children and adolescents aged 8 years or above, BP was measured in a seated position following 5 minutes of quiet rest during the physical examination. For each individual, up to three completed BP measurements were taken at one visit by a physician using a mercury sphygmomanometer with an appropriate size cuff. The measuring procedure followed the American Heart Association standard protocol. Quality control of BP measurements was reinforced in the NHANES. The mean of the three measurements were used to determine PHT status. PHT was defined as a systolic blood pressure (SBP) or diastolic blood pressure (DBP) ≥ 95th percentile of age-, sex-, and height-specific reference level for the US population, according to the Fourth Task Force Report. A height z-score relative to the 2000 (most recent) CDC growth charts was calculated for each child according to the CDC Box-Cox transformation estimation procedure. The age-, sex-, and height-specific BP percentile was then calculated using the methods provided by the Fourth Task Force Report, applying the following steps: 1) compute the expected BP value for a given age, sex, and height z score; 2) convert the participant’s observed BP into a Z score; and 3) convert the BP Z score into a percentile. Individuals who were taking antihypertensive medications were also considered as having PHT. Antihypertensive medication use was determined by utilizing information from the Prescription Medications questionnaire of NHANES, which was verified by survey interviewers who asked to see the medication containers when possible. It was defined as reported use of at least one of the following classes of medications: angiotensin converting enzyme inhibitor, beta-adrenergic blocking agents, calcium channel blocking agents, diuretics, angiotensin II inhibitors, other antihypertensive agents (other antiadrenergic agents, vasodilators, renin inhibitors, agents for hypertensive emergencies, agents for pulmonary hypertension, and antihypertensive combinations). Antihypertensive medications were classified using the Multum Lexicon Drug Database **(**<http://www.multum.com/Lexicon.htm>**)**. ## Measurements of Race, Age, Body Size, and Birth Weight Race was self-reported during the household interview. It was derived by combining responses to questions on race and Hispanic origin and classified into the following categories: Non-Hispanic Whites, Non-Hispanic Blacks, Mexican Americans, Other Hispanics, and Other Race. Age was grouped into two categories: children aged 8–12 years and adolescents aged 13–17 years. Anthropometric measurements were conducted by a trained examiner following the standard protocol in the mobile examination center during the physical examination. Body weight was measured to the nearest 0.05 kg using the Toledo digital scale while wearing underwear, a disposable gown, and foam slippers. Standing height was measured to the nearest 0.1 cm by using a fixed stadiometer with a vertical backboard and a moveable headboard. BMI was calculated as weight (kilogram) divided by height (squared meters) and classified into three categories: normal weight (\<85.0^th percentile of sex-and-age- specific BMI), overweight (85.0^th -94.9^th percentile of sex-and-age-specific BMI) and obesity (≥95.0^th percentile of sex-and- age-specific BMI). Birth weight (in pounds and ounces) was self-reported and collected during the interview from participants aged 15 years or younger and their guardians. Previous studies have shown that maternal- and self-reports of birth weight and LBW status have high validity. The specificity of self-reported birth weight below 3,000 grams was 93% among middle-aged and elderly women participating in the Danish Nurse Cohort Study. The specificity is likely to be higher among children and adolescents given a shorter recall period. Birth weight at or below 5.5 pounds or approximately 2,500 grams was defined as LBW. ## Statistical Analyses Results of descriptive analysis were expressed as mean or percent, with the corresponding standard error (SE) presented. Odds ratio (OR), 95% confidence interval (CI) and *P* values based on multivariable logistic regressions were conducted to evaluate the independent association between race and PHT after controlling for BMI, LBW and their interactions. This was realized through a stepwise modeling approach, in which race, BMI categories, LBW, and the interaction between BMI categories and birth weight categories were added to the regressions sequentially. Age categories were also adjusted for in all multivariable analyses where needed. The analyses were conducted separately for boys and girls following previous practices. For each sex, interactions between race and age and race and BMI categories were also examined. Results were reported for the following age-and- sex-specific subgroups: (a) boys aged 8–12 years, (b) boys aged 13–17 years, (c) girls aged 8–12 years, and (d) girls aged 13–17 years based on the evidence of potential marginally significant interactions (P = 0.08) between race and age among boys. Significant interactions between race and BMI categories were also observed in boys aged 13–17 years and stratified analyses by BMI categories were further performed among this subgroups. Interactions between race and LBW were not explored considering the small sample size of participants born at a LBW with PHT. Analyses involving birth weight were only performed among children aged 8–15 years, the ages for which birth weight was available in NHANES. All analyses took account of the complex survey design including stratification, clustering, and weighting to consider the oversampling of subgroups, unit non- response and non-coverage in the NHANES. This was realized by applying the set of survey commands (with prefix ‘svy-’) in STATA 12 with the sampling errors estimated by the Taylor linearization method. Statistical significance was set at the level of P ≤ 0.05 (2 tailed). # Results ## Background Characteristics Of 9,250 children and adolescents aged 8–17 years who were included in this analysis, 31.24% were Non-Hispanic Whites, 32.18% were Non-Hispanic Blacks, and 36.57% were Mexican Americans. On average, Black and Mexican American participants had significantly higher mean BMI and BMI z scores compared to their White peers (mean BMI: 22.58 and 22.48, vs. 21.28, respectively, both P\<0.001; BMI z score: 0.71 and 0.75, vs. 0.45, respectively, both P\<0.001). Black and Mexican American youth also had significantly higher prevalence of obesity than Whites (22.88% and 24.68%, vs. 15.84%, respectively, both P\<0.001). Although Blacks and Mexican Americans on average were born at significantly lower birth weight than Whites, only Blacks had a significantly higher prevalence of LBW (14.56% vs. 5.89%, P\<0.001). Overall, Blacks had a higher SBP compared to Whites (107.80 vs. 106.32, P\<0.001), and Mexican Americans had a lower DBP than Whites (57.90 vs. 59.25, p = 0.001). Antihypertensive medication use was low in general. ## Prevalence of PHT among children and adolescents by race and other factors In 1999–2010, the prevalence (SE) of hypertension in U.S. children and adolescents aged 8 to 17 years was 6.07% (0.39) among both sexes, 5.58% (0.49) in girls, and 6.56% (0.58) in boys. No significant difference was observed between boys and girls (P = 0.20). Mexican Americans had the lowest PHT prevalence at 5.36% (0.57), followed by 5.60% (0.55) in Non-Hispanic Whites and 7.10% (0.59) in Blacks. Compared to non-Hispanic White youth, the unadjusted prevalence of PTH was significantly higher in non-Hispanic Blacks (P = 0.04), but the difference in Mexican Americans did not approach significance (P = 0.77). When stratified by sex alone, no racial differences were observed among boys or girls. When further stratified by age and sex, among boys aged 13–17 years Blacks had significantly higher PTH than Whites 9.59% (1.02) vs. 6.59% (1.04) (P = 0.04). ## Racial Differences in PHT Prevalence and Their Association with Body Size and Birth Weight It was further assessed whether observed racial differences in PTH can be explained by body size and birth weight using multivariate logistic regressions. After controlling for age and sex, Black youth had a 29% increased odds of hypertension (model 1, Odds ratio (OR) = 1.29; 95% CI: 1.01–1.65; P = 0.044) compared to White youth. The Black-White difference in hypertension rate showed little change (model 2, OR = 1.25; P = 0.047) after additional adjustment for birth weight, but was attenuated (model 3, OR = 1.16; P = 0.19) to the degree that it lost statistical significance with further adjustment for current BMI. When stratified by sex and age group combined, no significant racial differences were identified in age adjusted models and after controlling for BMI categories, LBW status and their interaction among boys aged 8–12 years. However, among boys aged 13–17 years Blacks were at significantly higher odds of PHT compared to Whites in the age adjusted model (OR: 1.51; 95% CI: 1.03–3.43; P = 0.041) and with the addition of control for LBW (OR: 2.00; 95% CI: 1.09–3.71; P = 0.027), and then for BMI categories (OR: 1.50; 95% CI: 1.01–2.14; P = 0.048). There was no significant racial difference among girls for both 8–12 and 13–17 age groups. No significant differences in PTH were observed between White and Mexican youth in any models including those for each sex-age group. Further analysis stratified by the BMI categories among boys aged 13–17 years shows that significant Black-White differences were only found in boys with normal body size (OR = 2.16; 95% CI: 1.22–3.80; P = 0.008), but not among those who were overweight or obese. No significant differences in PTH were observed between Whites and Mexican across all BMI categories. When the overweight and obese categories were combined to potentially increase statistical power, the results remain the same (OR: 1.03; 95% CI: 0.65–1.66; P = 0.91 in Black and OR: 0.71; 95% CI: 0.44–1.13; P = 0.15 in Mexicans). # Discussion The study investigated racial differences in the prevalence of PTH among children and adolescents aged 8–17 years and examined their association with body size and birth weight using the nationally representative sample of NHANES 1999–2010. The prevalence of hypertension in U.S. children and adolescents aged 8 to 17 years was 6.1%. We found that Black youths had a 29% increased odds of hypertension compared to their White peers. When stratified by sex and age groups, the Black-White difference was only significant in boys aged 13–17. Compared to Whites, Black youths also had an elevated prevalence of obesity and lower birth weight. However, the racial difference in PTH among boys aged 13–17 was not explained by their obesity and low birth weight status. No significant differences in PTH were observed between White and Mexican s youth taken as a whole or by sex-age group. It is well documented that in the US Black adults have higher prevalence of hypertension than White adults. However, it is unclear whether such racial disparity also exists in US children and adolescents. It is also unknown at which age the Black-White differences in BP emerge. While most previous studies found no significant racial differences in hypertension among US youths, Rosner and colleagues reported a significant racial difference between Hispanic and White boys, but not between Black and White boys. For girls, both Blacks and Hispanics were found to have higher prevalence as compared to their White peers. In contrast, we found a Black-White difference in PTH among US youths in the current study. When stratified by age and sex, such a racial difference was only significant among boys aged 13–17, which persisted after controlling for birth weight and BMI. Although Mexican American youth had a higher rate of LBW and obesity compared to Whites, they had little difference in PTH after adjusting for age, sex, birth weight and obesity. The differences in BP measurement and hypertension definition might partially explain the differences in findings between the current and Rosner’s study. The timing of when the study data were collected may also influence the two studies' results. The Rosner study dataset encompassed all children in the Pediatric Task Force Database which included surveys and studies conducted between 1973 and 2000. Our study, however, applied the most recent national representative sample collected between 1999 and 2010. Our results are consistent with 3 recent studies in US populations and provided additional data stratified by age and gender. Nevertheless, more studies are warranted to shed further light on the racial differences in pediatric hypertension prevalence in the US. Another important finding of this study is the Black-White disparity was observed among adolescent boys with normal weight, but not among overweight and obese boys. Similar findings that Blacks at normal weight were more likely to have PHT than Whites have been reported previously. However, the racial difference was significant among female adolescents in Rosner et al., and male children and adolescents in Rosner et al.. It is unclear why among Black adolescents only those at normal weight, but not those overweight or obese were at increased risk of PTH than their White peers. Although we cannot rule out the possibility that the lack of significant results in overweight and obese groups is due to smaller sample size, results were similar when we combined the overweight and obese youths together. The current study has several limitations. First, the 3 BP measurements were all obtained in one visit during physical examinations in the NHANES. Ideally, hypertension should be defined according to 3 repeated BP measures obtained on 3 separate occasions. Secondly, the reliability of our findings in analyses stratified by age-sex group could be compromised by small sample size, although each age-sex group consisted of at least 600 individuals. Thirdly, NHANES is a cross-sectional survey by design, which is not the best design to examine the age onset of racial disparity in PTH. Lastly, data on LBW were self-reported and only available among participants aged 8–15 years. Misclassification in self- reported birth weight, although likely small, could introduce bias into the association between LBW and PHT. Further, when controlling for LBW the study sample was limited to those aged 8–15 years. A sensitivity analysis demonstrated that restricting all analyses to that age range yield qualitative unchanged results. Future studies are warranted to control for a better measured LBW status. Despite these limitations, this study contributes to the literature by examining the most recent nationally representative samples of the US population including extensive relevant data with PTH defined using three measurements of BP. Other notable advantages of this study include the relatively large sample size available in each race group, objective measure of weight and height, and the combination of stratified and multivariate analyses guided by both previous studies and empirical data (test for interactions). # Conclusions In current analysis of the most recent nationally representative samples of US children and adolescents aged 8–17 years, we found Non-Hispanic Black children and adolescents to have a significantly higher prevalence of hypertension than their non-Hispanic White counterparts. This racial difference in pediatric hypertension was only shown in boys aged 13–17 when stratified by age and sex. The racial disparity of PTH among boys aged 13–17 could not be explained by low birth weight and current obesity of the Black youth. No significant differences in PTH were observed between White and Mexican Americans in all models and in each age-sex group. Our findings are not entirely consistent with previous findings. Future research is needed to further investigate racial differences in hypertension in children and adolescents. We thank the NHANES participants and staff for their contributions to the study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LC LL NS. Performed the experiments: LC LL. Analyzed the data: LC LL. Contributed reagents/materials/analysis tools: LC LL. Wrote the paper: LC LL NS.

# Introduction Myotonic dystrophy type 1 (MD₁), or Steinert's disease, is the most common form of dystrophy in adolescents and adults, and the second most common neuromuscular disease. The classic form of MD₁ is characterized by weakness and atrophy in skeletal muscles, myotonia, cardiac conduction abnormalities, cognitive impairment and myocardial changes. Additionally, the involvement of respiratory muscles is common with reduction of respiratory muscles strength, especially expiratory muscles, resulting in death, specifically due to respiratory insufficiency or pneumonia,. The respiratory impairment in MD₁, is rather complex because progressive muscle weakness is associated to different degrees of central respiratory control abnormalities causing alveolar hypoventilation. In fact, CO₂ insensitivity occur in MD₁ independently of lung function impairment and respiratory muscle weakness. In addition the participation of myotonia in DM1 respiratory restriction is not fully clarified. Myotonia is characterized by a delayed relaxation of skeletal muscles after an intense contraction and is one of the most important clinical features of MD₁. The presence of myotonia determines the appearance of symptoms such as muscle stiffness and cramps. Although few studies have been published, there are indications of its presence in the respiratory muscles. In 1954, Benaim & Worster-Drougth showed, through electromyography, the presence of myotonia in the intercostal muscles, but not in diaphragm. In 1959, Kilburn et al. observed in two cases a delayed return of the diaphragm to its resting position. Smorto et al. found high amounts of high frequency electrical activity, considered myotonia, in MD₁ patients. In another case study, Estenne et al. were able to detect abnormal muscle relaxation and registry of myotonia of the diaphragm and abdominal muscle by performing electromyographic and nasal inspiratory pressure test (SNIP) studies in a patient with congenital myotonia (Thompsen disease). However, due to the presence of abnormalities in relaxation of the diaphragm, which was confirmed by variations in abdominal pressure, the presence of myotonia was attributed to the abdominal muscles and not to the diaphragm. In contradiction, Fitting et al. using similar methodology, found no changes in the abdominal muscles, but in the diaphragm. Finally, Rimmer et al. in a study with 11 MD patients, found only two patients with electrical activity consistent to the presence of myotonia, as measured by surface electromyography (sEMG) of the scalene, parasternal and external abdominal oblique muscles. Therefore, the presence of myotonia in the inspiratory muscles of patients with MD is doubtful, since most of the studies involve few patients with MD and, in some cases, inadequate techniques. For some years, the maximum relaxation rate (MRR) of the inspiratory muscles has been used as an indirect marker of fatigue and overload of the inspiratory muscles. A decreasing MRR of skeletal muscle means muscle overload and precedes the failure of power generation. Considering the muscle relaxation as an active process that consumes energy that can be affected by different conditions, including fatigue, muscle weakness and atrophy of muscle fibers, a decreasing MRR during the sniff test could be useful for evaluating the presence of inspiratory muscles overload, signs of myotonia, and fatigue. Therefore, the objectives of this study were to evaluate the MRR and electrical activity of the inspiratory muscles in patients with MD₁. We hypothesized that markers of myotonia could be found in this population by analyzing different parameters describing the strength, the relaxation and the electrical activity of respiratory muscles. More specifically, we have considered maximal inspiratory pressure (PImax), maximal expiratory pressure (PEmax) and SNIP test as indexes of strength and MRR and sEMG as indexes of muscle relaxation and contraction, respectively. # Methods ## Subjects and study design This is an observational, quasi-experimental study, with control group. Subjects were individuals diagnosed with MD₁ recruited from a clinical follow- up with a neurologist at a University Hospital from January 2015 to June 2016. All patients and healthy subjects were assessed in a single day after receiving initial information and recommendations on the preparation for the assessments. Inclusion criteria for the MD₁ group were: patients aged between 18 and 65 years, properly diagnosed and accompanied by a neurologist. Those who were smokers (or ex-smokers) or presented obstructive respiratory disease, musculoskeletal comorbidities or impaired intellectual ability (established by medical diagnosis) that would prevent the assessments, who failed to perform tests for other reasons or quit from the study were excluded. The control group included self-reported healthy subjects matched for gender and age. Individuals who were smokers (or ex-smokers) or presented cardiac, respiratory or musculoskeletal comorbidities, deviated septum, rhinitis or nasal obstruction, as well as those who failed to perform the tests or gave up from participating of study were excluded. Additionally, healthy subjects should not use any type of drug that could interfere with the test results, such as antihypertensive or bronchodilator drugs. All subjects agreed to participate in the study by signing an informed consent form, which was approved by the Ethics Committee of the Federal University of Rio Grande do Norte/Brazil (protocol n^o: 752.322) according to the Declaration of Helsinki. ## Clinical phenotype and muscle impairment MD₁ classification The MD₁ patients were classified in relation to degree of muscle impairment and clinical phenotypes by the neurologist according to the Muscle Impairment Rating Scale (MIRS), and the MD₁-specific form according to age and onset and clinical symptoms. The MIRS is structure in five degrees of impairment in relation to muscle involvement, progressing from distal to proximal: grade 1, no muscular impairment; grade 2, minimal signs (myotonia and digit flexors and craneal muscle wasting and weakness); grade 3, distal weakness (no proximal weakness except isolated elbow extensor weakness); grade 4, mild to moderate proximal weakness; grade 5, severe proximal weakness. The phenotype classification includes four categories: 1- congenital, 2- juvenile, 3- classic adult onset and 4- mild late-onset. ## Lung function Spirometry was performed through a KoKo DigiDoser^® spirometer (nSpire Health, Inc. Longmont, USA) and considered complete when three acceptable curves were produced (with variation equal to or lower than 5% of the highest value). The technical procedure, the acceptance and reproducibility criteria as well as the standardization of equipment followed the recommendations of the ATS/ERS. The reference values used in this study had been previously published for the Brazilian population. ## Respiratory muscle strength Respiratory muscle strength was assessed from measurements of PImax, PEmax and SNIP by a digital manometer (NEPEB-LabCare/UFMG, Belo Horizonte, Brazil), following the procedures previously published by the Brazilian Society of Pneumology. PImax was measured from residual volume and PEmax from total lung capacity, using a nose clip. The highest value obtained in a maximum of five tests (with variation equal to or lower than 10% of the highest value) was used for each assessment. SNIP was obtained from functional residual capacity, with the subject having one nasal cavity occluded by a plug coupled to a catheter connected to a digital manometer, while maintaining the other nostril open. Previously published reference values were used. In order to identify respiratory muscle weakness, cutoff values were adopted and calculated as the mean of the normal population, published previously, minus 1.96 multiplied by the standard deviation. The reference values of Neder et al. were used for PImax and PEmax; and the reference values of Araujo et al. were used for SNIP. The cutoff values established for men and women, respectively, were: 59.1 cmH₂O and 70.4 cmH₂O for PImax; 58.2 cmH₂O and 53.9 cmH₂O for SNIP; 101.1 cmH₂O and 69.2 cmH₂O for PEmax. ## Assessment of maximum relaxation rate of inspiratory muscles MRR was derived from SNIP test and calculated using LabChart Reader 8.0 software (ADInstruments, New Zealand). The values were obtained as the first derivative of pressure-time curve (dP/dT) over the first half of the relaxation curve by calculating the tangent at the steepest part of the pressure curve. To exclude the effect of pressure oscillation amplitude on MRR, dP/dT was normalized and expressed as percentage of the pressure drop in every 10 milliseconds (dP/dt) / Psniff\*100 (%10ms). The following criteria were established for choosing the best SNIP value for calculation of MRR: (1) sniff maneuver performed from FRC; (2) peak pressure maintained for less than 50 milliseconds; (3) total duration of the maneuver less than 500 milliseconds; and (4) shape of SNIP curve showing soft peaked curves with upward and downward movement,. ## Surface electromyography Capture and processing of myoelectric signals were performed using TeleMyo DTS Desk Receiver^® (Noraxon, USA, Inc., Scottsdale, USA) electromyograph and 4 wireless Clinical DTS (Noraxon, USA, Inc., USA) sensors with 16-bit resolution and common-mode rejection ratio \> 100 dB. The sampling frequency of the captured signals was 1500 Hz, with 500 Hz low pass filter and signals pre- amplified 1000 times. The software used to capture and store the myoelectric signals during PImax, PEmax and SNIP tests was MR 3.2 (Noraxon, USA, Inc., USA). Ag/AgCl bipolar surface electrodes were placed along the direction of the muscle fibers, all in the right side of the body to avoid contamination by the cardiac electrical signals. The skin of the volunteers was properly prepared in order to reduce impedance, favoring the capture of a signal with greater amplitude, with less interference and noise. To capture electromyographic signals, all recommended procedures were strictly followed. Electrodes were placed over the parasternal muscle on the second intercostal space (2^ndIS), 3 cm from the sternum; the sternocleidomastoid muscle (SCM), on the muscle belly, 5 cm from the mastoid process; the rectus abdominis (RA), 4 cm from the umbilical scar; and the scalene muscle (SCA), on the muscle belly, 5 cm from the sternoclavicular joint and 2 cm above that mark. Electromyographic signals were analyzed using the RMS (root mean square) and normalized from the average of three maximum voluntary isometric contractions for SCM, SCA, RA; and three SNIP and PImax tests for 2^ndIS. In order to quantify possible delayed muscle relaxation, the duration of electrical activity of each muscle studied was measured and normalized by the duration of the different maneuvers (SNIP, PImax and PEmax). The duration of each respiratory muscle contraction (i.e., the difference between onset and offset time during the maneuver) was automatically calculated by analyzing the time course of pressure curves using the MR3.2 software (MyoMuscle Module, Noraxon, USA). ## Sample size and statistical analysis Sample size was calculated through a pilot study with five MD₁ subjects. The standard deviation of the MRR variable was used for a hypothetical t test. Twelve subjects in each group were required to achieve 80% power with P\<0.05. Data normality were checked using Shapiro-Wilk test. Parametric data were compared using the Unpaired t-test, while non-parametric data were compared using the Mann-Whitney test. Significant levels were set at a 2-sided P\<0.05. Receiver operating characteristic curves (ROC) were calculated using the MRR, SNIP%, PImax%, PEmax% between MD₁ patients and healthy subjects. ROC curve was calculated as described previously by Hanley and McNeil. Prism^® version 5 (GraphPad Software Inc, USA) was used for data analysis and G\*Power version 3.1.9.2 (Franz Faul—Universität Kiel, Germany) for sample size, effect-size and power calculation. # Results ## Sample size Initially, 74 individuals were invited to participate in the study: 44 patients with MD₁ and 30 healthy subjects. After exclusion criteria a total of 18 MD₁ subjects and 11 healthy were included in the study. Considering the mean of each group and the difference between standard deviation of MRR variable, a Cohen’s d = 3.89 (considered large) was found. The power of the study (1-ß), considered the effect size found, was 1.0. ## Symptoms, severity and rate of progression of the disease Demographic features of the patients are reported in. The mean age of the patients at time of evaluation was 42.3 ± 11.3 and the mean time elapsed from diagnosis were 6.83 ± 5.98 years. In relation to phenotype, 88.8% (n = 16) were classified as a Classic form and 11.2% (n = 2) as a Mild type. Regarding the classification on MIRS we found that 11,11% (n = 2) were classified as MIRS 1, 55.5% (n = 10) classified as MIRS 2, 22.22% (n = 4) classified as MIRS 3 and 11,11% (n = 2) classified as MIRS 4. ## Pulmonary function and respiratory muscle strength The spirometric values for the MD₁ group showed, on average, a moderate restrictive pattern. Regarding inspiratory muscle strength in the control group, 25% of women were below the cutoff point, while in the MD₁ group was 40%. Regarding male gender, 14.3% of control group and 54.5% of MD₁ group were below the cutoff point. In the SNIP test, all subjects were above the cutoff point in the control group, while in the MD₁ group 63.6% of men and 80% of women were below that point. In the PEmax assessment, where the biggest difference was observed between the two groups, all healthy subjects met or were above the cutoff point for muscle weakness. In the MD₁ group, 90.9% of men and 80% of women did not reach that value. For all maneuvers previously mentioned, the differences between the two groups were statistically significant: in absolute values SNIP (P = 0.0014), PImax (P = 0.0006), PEmax (P = 0.0002) and in percentage of predict values SNIP (P = 0.0015), PImax (P = 0.0016) and PEmax (P\<0.0001). Spirometric and respiratory muscle strength data are shown in and. ## Electromyography of the respiratory muscles In the MD₁ group, 2 subjects were excluded due to the poor quality of the signal captured by the sEMG. Obtained signals from the surface electromyography showed increased muscle activity in MD₁ subjects in most of the maneuvers carried out, with significant differences found at rest for the SCM (P = 0.002) and SCA (P = 0.004). It was also found significant difference during the PEmax test for the 2^ndIS (P = 0.003), and during SNIP for the SCM (P = 0.02) and SCA (P = 0.03). Electrical activity time of respiratory muscles in patients with DM₁ was significantly increased compared to controls. ## Maximum relaxation rate of inspiratory muscles To assess MRR, four DM₁ subjects were excluded from the analysis because their SNIP test did not meet the criteria for proper data acquisition. The MRR was significant lower in MD₁ group compared to control group (P = 0.001; 5.7±1.2 vs. 7.9±1.7, respectively). ## ROC analysis The results, considering the Area Under the Curve, showed, for all variables, a good ability to discriminate between those individuals with the disease and those without it. The thresholds of sensitivity/specificity, confidential interval of area and p values are shown in. The area under the curve (AUC) above 0.80 associated with a P\<0.05, for all the variables (maneuvers), lead to the conclusion that these variables show a good ability to discriminate between those individuals with the disease and those without it. # Discussion The study results showed that the MRR is reduced in MD₁ subjects compared to healthy. The MRR of inspiratory muscles was considered sensitive to identify the presence of reduced MRR in MD₁ patients and specific to discard it in healthy individuals. According to the results of the ROC curve analysis, which determined a cutoff point of 5.65; the electrical activity of the accessory muscles of respiration (SCM, SCA and RA) was increased in MD₁ compared to controls (P = 0.004; P = 0.009; P = 0.045, respectively). In addition, lung function was also altered in MD₁ group as well as reduced respiratory muscle strength showing considered as muscle weakness. The MRR of inspiratory muscles has been studied for about 25 years. This is an index of fatigue and overload of inspiratory muscles, developed from sniff maneuvers obtained by measuring esophageal or transdiaphragmatic pressure, which are invasive methods. Subsequent studies have observed a high relationship between the MRR obtained through sniff maneuvers associated with esophageal pressure and with nasal pressure, providing a non-invasive measurement of that variable. Despite the technical characteristics of MRR as a measure capable of identifying the delay in inspiratory muscle relaxation in MD₁ and, therefore, suitable for assessing the presence of myotonia, we have not found previous studies of MRR obtained through nasal inspiratory pressure in MD₁. Fitting & Leuenberger were the first authors to study MRR in a single case in 1989. These authors evaluated the effects of procainamide, an anti-myotonic drug, in the diaphragm through transdiaphragmatic pressure. Due to the type of study and the technique used, comparisons are inappropriate. Another study published by Jammes et al., aimed to analyze the electrical activity of the diaphragm and intercostal muscle during loaded inspiratory and expiratory breathing. The authors suggest that the absence of relaxation in diaphragm muscle during expedition may be related to the persistence of repetitive discharge due to a failure o neural inhibitory circuits or to myotonia. More recently, Garcia-Rio et al were the first to assess noninvasively the MRR of inspiratory muscles using SNIP curve in 20 patients with different neuromuscular diseases (5 amyotrophic lateral sclerosis, 3 duchenne muscular dystrophy, 2 myotonic dystrophy, 7 myasthenia gravis, 2 congenital myopathy and brachial plexus palsy) and 10 healthy subjects. These authors used a similar methodology to calculate MRR and found a significantly lower MRR in the neuromuscular group compared to healthy (7.21 ± 0.68 vs. 9.31 ± 1.22). Regarding our results, we also found a significantly lower MRR in MD₁ patients however; the study of Garcia-Rio et al. used a heterogeneous sample with several different neuromuscular diseases. Neuromuscular disorders, despite having some common physiopathological characteristics, have varying degrees of severity and progression, which makes unsuitable grouping these diseases for assessment of MRR and its subsequent comparison to a control group. It is important to report that the MD₁ is characterized by the presence of myotonia in the skeletal muscles and possibly, in the inspiratory muscles, which could potentially affect the relaxation of these muscles and reduce the MRR. Regarding the ROC curve, the high sensitivity and specificity found for MRR quantifies with high statistical value the overall capacity of the MRR to discriminate the pattern of inspiratory muscle relaxation in MD₁ compared with healthy subjects. Regarding pulmonary function, our study confirmed previous results as it was found the presence of moderate restrictive disorder in MD₁. In relation to respiratory muscle strength, Gillam et al presented, in a series of 10 patients, values considered below the cutoff point for muscle weakness of the expiratory muscles in 50% of subjects, with values less than 80 cmH₂O for men and 60 cmH₂O for women. Recently, similar results have been reported in the literature, in which approximately 50% of patients presented expiratory muscle strength with values below the cutoff point for expiratory muscle weakness. These same authors also showed PEmax, PImax and SNIP percentage of predicted values of about 64%, 70% and 80%, respectively. Additionally, in a study analyzing respiratory muscle strength involving different neuromuscular diseases, patients with MD₁ have shown a reduction in respiratory muscle strength and PEmax/PImax ratio below normal. In the present study, changes in respiratory muscles strength were reinforced by the changes observed in the sEMG. A major limitation of our study was the small sample size due to the limited time and financial support. However, we may consider the high power and effect size found and the low prevalence of the disease. Additionally, the results found in our study add new perspectives regarding the MRR assessment (through the SNIP test) of subjects with MD₁. We must interpret the results of non-invasive MRR of inspiratory muscles in MD₁ with caution, considering that non-invasive MRR of inspiratory muscles in MD₁ should be validated with golden standard methods as an invasive measurements of MRR during assessment of SNIP. # Conclusion The study results may provide information that can contribute to a better understanding of myotonic dystrophy type 1 and its impact on the respiratory system (either in its functional or muscular aspect). These results may also contribute to the development of new evaluation methods that may be routinely used in clinical practice. Moreover, we have brought new approaches to the surface electromyography, which is still underused to assess respiratory muscles and may be a complementary method for neuromuscular diseases researches. The study was performed at the PneumoCardioVascular Lab/HUOL, Empresa Brasileira de Serviços Hospitalares (EBSERH), Universidade Federal do Rio Grande do Norte (UFRN), Natal, Rio Grande do Norte, Brasil and PneumoCardioVascular Lab/HUOL, Empresa Brasileira de Serviços Hospitalares (EBSERH), Universidade Federal do Rio Grande do Norte (UFRN), Natal, Rio Grande do Norte, Brasil. The study received financial support of Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES), PROCAD 2013 88881.068409/2014-01, Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) processo: 400316/2012-9. Guilherme Augusto de Freitas Fregonezi is fellow from CNPq number 307353/2015-0 and Vanessa Resqueti is fellow from CNPq number 310091/2015-2. The authors have declared that no competing interests exist. 2^ndIS Parasternal muscle on the second intercostal space MD₁ Myotonic dystrophy type 1 MRR Maximum relaxation rate PEmax Maximum expiratory pressure PImax Maximum inspiratory pressure RA Rectus abdominis SCA Scalene SCM Sternocleidomastoid sEMG Surface electromyography SNIP Sniff nasal inspiratory pressure [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** FALD AA GAdFF. **Data curation:** MdAE METDJ. **Formal analysis:** MdAE AS LPG VR GAdFF. **Funding acquisition:** VR GAdFF. **Investigation:** MdAE METDJ GAdFF. **Methodology:** MdAE FALD VR AA GAdFF. **Project administration:** VR GAdFF. **Resources:** GAdFF METDJ FALD. **Supervision:** VR GAdFF. **Visualization:** MdAE FALD AS LPG VR AA GAdFF GCdN. **Writing – original draft:** MdAE FALD AS LPG VR AA GAdFF. **Writing – review & editing:** AS LPG VR AA GAdFF GCdN.

# Introduction As you read these words, you may be experiencing the presence of a familiar speechlike companion. This internal speech production may accompany daily activities such as reading (see, but see), writing (), memorising (), future planning, problem solving or musing (for reviews see). Several studies using experience sampling or questionnaires have shown that by deliberately paying attention to this internal speech, one can examine its phenomenological properties such as identity (whose voice is it?) or other high-level characteristics (e.g., is it gendered?). Moreover, it is often possible to examine lower-level features like the tone of the inner speech, its pitch or its tempo. This set of basic observations leads to some important insights about the nature of inner speech. The simple fact that we can make sensory judgements about our inner speech tautologically reveals that inner speech is accompanied by sensory percepts (e.g., speech sounds, kinaesthetic feelings). Some of these introspective accounts have been examined, tested and complemented using empirical methods from cognitive neuroscience. As summarised in, behavioural and neuroimaging data reveal that some variants of inner speech are associated with auditory and/or somatosensory sensations that are reflected by auditory and/or somatosensory cortex activations. Visual representations may also be at play, typically for inner language in the deaf population. Inner verbalising therefore involves the reception of imaginary multisensory signals. This leads to other fundamental questions: where do these percepts come from? Why do they sound and feel like the ones we experience when we actually (overtly) speak? Two main classes of explanatory theories have been offered to answer these questions. A first class of theories, that derives from Vygotky’s views on language and thought, and that we describe as the *abstraction view*, suggest that inner speech is profoundly internalised, abbreviated and condensed in form. Vygotsky suggested that inner speech evolved from so-called egocentric speech (i.e., self-addressed overt speech or private speech), via a gradual process of internalisation during childhood. According to him, the properties of speech are transformed during this internalisation, and inner speech cannot merely be described as a weakened form of overt speech (as claimed for instance by). This has led some scholars to conceive of inner speech as predominantly pertaining to semantics, excluding any phonological, phonetic, articulatory, or even auditory properties. The property of abbreviation and condensation is supported by several psycholinguistic experiments on the qualitative and quantitative differences between overt and covert speech, as concerns rate and error biases (e.g., but see). Such condensation implies that the auditory qualities mentioned above would only rarely be observed during introspection and would merely be the result of learned associations between abstract linguistic representations and auditory percepts. A second class of theories is described under the umbrella term of *motor simulation view*. These theories suggest that inner speech can be conceived as a kind of action on its own, produced in the same way as overt speech is, except that the last stage of articulatory execution is only simulated. Most theories under this view share the postulate that the speech motor system is involved (to some extent) during inner speech production and that the auditory and somatosensory consequences of the simulated articulatory movements constitute the inner speech percepts referred to in subjective studies. As explained in the ConDialInt model, these two views can be reconciled if various degrees of unfolding of inner speech are considered. Fully condensed forms of inner speech only involve semantics, and are deprived of the acoustic, phonological and syntactic qualities of overt speech. Expanded forms of inner speech, on the other hand, presumably engage prosodic and morpho-syntactic formulation as well as phonological specification, articulatory simulation and the perception of an inner voice. Between the fully condensed abstract forms and the expanded articulation-ready form, it can be assumed that various semi- condensed forms may exist, with morphosyntactic properties and perhaps even phonological features, depending on the stage at which the speech production process is truncated. Such a view was also taken by who has suggested that inner speech varies with cognitive demands and emotional conditions on a continuum between extremely condensed and expanded forms (see also). Therefore, the two views of inner speech (abstraction vs. simulation) can be construed as descriptions of two opposite poles on the condensation dimension. On the most expanded side of the continuum, inner speech entails full phonetic specification and articulatory simulation. It might therefore be expected that speech motor activity could be detectable. If the motor simulation view is correct, then motor activity could be recorded during expanded forms of inner speech. If, on the other hand, the abstraction view applies to all forms of inner speech, then no motor activity should be present, even in phonologically-expanded forms. Previous research has demonstrated that it is possible to record muscle-specific electromyographic correlates of inner speech (e.g.,). However, these studies mostly focused on small samples of participants and sometimes used invasive intramuscular electromyography. In contrast, more recent research studies using surface electromyography lead to mixed results. Building upon previous work, we describe an experimental set-up using surface electromyography with the aim of testing the involvement of specific speech muscle groups during the covert production of phonologically expanded speech forms. ## Inner speech as motor imagery of speech Speech production is a complex motor action, involving the fine-grained coordination of more than 100 muscles in the upper part of the body. In adult humans, its covert counterpart (referred to as *inner speech* or *verbal imagery*) has developed to support a myriad of different functions. In the same way as visual imagery permits to mentally examine visual scenes, *verbal imagery* can be used as an internal tool, allowing –amongst other things– to rehearse or to prepare past or future conversations. Because speech production results from sequences of motor commands which are assembled to reach a given goal, it belongs to the broader category of motor actions. Therefore, a parallel can be drawn between verbal imagery and other forms of motor imagery (e.g., imagined walking or imagined writing). Accordingly, studies on the nature of inner speech might benefit from insights gained from the study of motor imagery and the field of motor cognition. Motor imagery can be defined as the mental process by which one rehearses a given action, without engaging in the physical movements involved in this particular action. One of the most influential theoretical accounts of this phenomenon is the *motor simulation theory*. In this framework, the concept of simulation refers to the “offline rehearsal of neural networks involved in specific operations such as perceiving or acting”. The MST shares some similarities with the theories of embodied and grounded cognition in that both account for motor imagery by appealing to a simulation mechanism. However, the concept of simulation in grounded theories is assumed to operate in order to acquire specific conceptual knowledge, which is not the concern of the MST. In other words, we should make a distinction between *embodiment of content*, which concerns the semantic content of language, and *embodiment of form*, which concerns “the vehicle of thought”, that is, proper verbal production. A second class of explanatory models of motor imagery are concerned with the phenomenon of *emulation* and with *internal models*. Internal model theories share the postulate that action control uses internal models, that is, systems that simulate the behaviour of the motor apparatus. The function of internal models is to estimate and anticipate the outcome of a motor command. Among the internal model theories, motor control models based on robotic principles assume two kinds of internal models (that are supposed to be coupled and regulated): a forward model (or simulator) that predicts the sensory consequences of motor commands from efference copies of the issued motor commands, and an inverse model (or controller) that calculates the feedforward motor commands from the desired sensory states. Emulation theories borrow from both simulation theories and internal model theories and provide operational details of the simulation mechanism. In the emulation model proposed by, the *emulator* is a device that implements the same input-output function as the body (i.e., the musculoskeletal system and relevant sensory systems). When the emulator receives a copy of the control signal (which is also sent to the body), it produces an output signal (the emulator feedback), identical or similar to the feedback signal produced by the body, yielding mock sensory percepts (e.g., visual, auditory, kinaesthetic) during motor imagery. By building upon models of speech motor control, a recent model describes wilful (voluntary) expanded inner speech as “multimodal acts with multisensory percepts stemming from coarse multisensory goals”. In other words, in this model the auditory and kinaesthetic sensations perceived during inner speech are assumed to be the predicted sensory consequences of simulated speech motor acts, emulated by internal forward models that use the efference copies of motor commands issued from an inverse model. In this framework, the peripheral muscular activity recorded during inner speech production is assumed to be the result of *partially* inhibited motor commands. It should be noted that both simulation, emulation, and motor control frameworks can be grouped under the *motor simulation view* and altogether predict that the motor system should be involved to some extent during motor imagery, and by extension, during inner speech production. We now turn to a discussion of findings related to peripheral muscular activity during motor imagery and inner speech. ## Electromyographic correlates of covert actions Across both simulationist and emulationist frameworks, motor imagery has consistently been defined as the mental rehearsal of a motor action without any overt movement. One consequence of this claim is that, in order to prevent execution, the neural commands for muscular contractions should be blocked at some level of the motor system by active inhibitory mechanisms. Despite these inhibitory mechanisms, there is abundant evidence for peripheral muscular activation during motor imagery. As suggested by, the incomplete inhibition of the motor commands would provide a valid explanation to account for the peripheral muscular activity observed during motor imagery. This idea has been corroborated by studies of changes in the excitability of the motor pathways during motor imagery tasks. For instance, measured spinal reflexes while participants were instructed to either press a pedal with the foot or to simulate the same action mentally. They observed that both H-reflexes and T-reflexes increased during motor imagery, and that these increases correlated with the force of the simulated pressure. Moreover, the pattern of results observed during motor imagery was similar (albeit weaker in amplitude) to that observed during execution, supporting the *motor simulation view* of motor imagery. Using transcranial magnetic stimulation, several investigators observed muscle-specific increases of motor evoked potentials during various motor imagery tasks, whereas no such increase could be observed in antagonist muscles. When considered as a form of motor imagery, inner speech production is also expected to be accompanied with peripheral muscular activity in the speech muscles. This idea is supported by many studies showing peripheral muscular activation during inner speech production \[, –\], during auditory verbal hallucinations in patients with schizophrenia, or during induced mental rumination. Some authors also recently demonstrated that it is possible to discriminate inner speech content based on surface electromyography (EMG) signals with a median 92% accuracy. However, other teams failed to obtain such results. Many of these EMG studies concluded on the involvement of the speech motor system based on a difference in EMG amplitude by contrasting a period of inner speech production to a period of rest. However, as highlighted by, it is usually not enough to show an increase of speech muscle activity during inner speech to conclude that this activation is related to inner speech production. Indeed, three sorts of inference can be made based on the studies of electromyographic correlates of inner speech production, depending on the stringency of the control procedure. The stronger sort of inference is permitted by highlighting a discriminative pattern during covert speech production, as for instance when demonstrating a dissociation between different speech muscles during the production of speech sounds of different phonemic class (e.g, contrasting labial versus non-labial words). According to, other (weaker) types of control procedures include i) comparing the EMG activity during covert speech production to a baseline period (without contrasting phonemic classes in covert speech utterances), or ii) comparing the activity of speech-related and non-speech related (e.g., forearm) muscle activity. Ideally, these controls can be combined by recording and contrasting speech and non-speech related muscles in different conditions (e.g., rest, covert speech, overt speech) of pronunciation of different speech sounds classes (e.g., labial versus non-labial). Previous research studies carried out using the preferred procedure recommended by suggest a discriminative patterns of electromyographic correlates according to the phonemic class of the words being covertly uttered, which would corroborate the *motor simulation view* of inner speech. However, these studies used limited sample sizes (often less than ten participants) and worked mostly with children. These factors limit the generalisability of the above findings because i) low-powered experiments provide biased estimates of effects, ii) following the natural internalisation process, inner speech muscular correlates are expected to weaken with age and iii) a higher sensitivity could be attained by using modern sensors and signal processing methods. The present study intends to bring new information to the debate between the *motor simulation view* and the *abstraction view* of inner speech, by focusing on an expanded form of inner speech: wilful nonword covert production. This work can be seen as a replication and extension of previous works carried out by McGuigan and collaborators. We aimed to demonstrate similar dissociations by using surface electromyography recorded over the lip (*orbicularis oris inferior*, OOI) and the *zygomaticus major* (ZYG) muscles. More precisely, given that rounded phonemes (such as /u/) are articulated with orbicular labial contraction, whereas spread phonemes (such as /i/) are produced with zygomaticus contraction, if the *motor simulation view* is correct, we should observe a higher average EMG amplitude recorded over the OOI during both the overt and inner production of rounded nonwords in comparison to spread nonwords. Conversely, we would expect a lower average EMG amplitude recorded over the ZYG during both the inner and overt production of rounded nonwords in comparison to spread nonwords. In addition, we would not expect to observe content-specific differences in EMG amplitude concerning the non speech-related muscles (i.e., forehead and forearm muscles). # Methods In the *Methods* and *Data analysis* sections, we report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study. A pre-registered version of our protocol can be found at: <https://osf.io/czer4/>. ## Participants As previous studies of the electromyographic correlates of inner speech were mostly carried out with samples of children or young adults, used different kinds of EMG measures (surface EMG or needle EMG), and different kinds of signal processing methods, it was impractical to determine the effect size of interest for the current study. Therefore, we used sequential testing as our sampling procedure, based on the method described in and. We fixed a statistical threshold to *BF*₁₀ = 10 and *BF*₁₀ = 1/10 (i.e., *BF*₀₁ = 10), testing the difference between the inner production of labial items versus the inner production of non-labial items on the standardised EMG amplitude of the lower lip (*orbicularis oris inferior*). In order to prevent potential experimenter and demand biases during sequential testing, the experimenter was blind to BFs computed on previous participants. All statistical analyses have been automatised and a single instruction was returned to the experimenter (i.e., “keep recruiting participants” or “stop the recruitment”). We fixed the maximum sample size to 100 participants. As a result of the above sampling procedure, a total of 25 French-speaking female undergraduate students in Psychology from the Univ. Grenoble Alpes (mean age = 19.57, SD = 1.1). took part in this experiment, in exchange for course credits. It should be noted that this procedure did not work optimally because we later spotted an error in the EMG signal processing workflow. Thus, the sequential testing stopped earlier than it should have. These participants were recruited via mailing list, online student groups, and posters. Each participant provided a written consent and the present study was approved by the local ethics committee (Grenoble CERNI agreement \#2016-05-31-9). ## Material ### EMG recordings EMG activity was recorded using TrignoTM Mini sensors (Delsys Inc.) with a sampling rate of 1926 samples/s, a band pass of 20 Hz (12 dB/oct) to 450 Hz (24 dB/oct) and were amplified by a TrignoTM 16-channel wireless EMG system (Delsys Inc.). These sensors consist of two 5 mm long, 1 mm wide parallel bars, spaced by 10 mm, which were attached to the skin using double-sided adhesive interfaces. The skin was cleaned by scrubbing it with 70% isopropyl alcohol. EMG signals were synchronised using the PowerLab 16/35 (ADInstrument, PL3516). Raw data from the EMG sensors were then resampled at a rate of 1 kHz and stored in digital format using Labchart 8 software (ADInstrument, MLU60/8). EMG sensors were positioned over five muscles: the *corrugator supercilii* (COR), the *frontalis* (FRO), the *zygomaticus major* (ZYG), the *orbicularis oris inferior* (OOI), and the *flexor carpi radialis* (FCR). Given that the activity of the *orbicularis oris inferior* and *orbicularis oris superior* muscles has previously been observed to be strongly correlated and that the activity of the OOI was more strongly affected by the experimental manipulation, we decided to record only the activity of the OOI in this study. The two speech- related muscles (OOI and ZYG) were chosen to show speech-specific EMG correlates, whereas the two non-speech related facial muscles (COR and FRO) were chosen to control for overall facial muscular activity. We also recorded the activity of the FCR of the non-dominant forearm to control for overall (body) muscular activity. As reviewed in, the dominant side of the face displays larger movements than the left side during speech production, whereas the non-dominant side is more emotionally expressive. Therefore, we recorded the activity of control and emotion-linked muscles (i.e., COR and FRO) that were positioned on the non- dominant side of the face (i.e., the left side for right-handed participants), while sensors recording the activity of the speech muscles (i.e., ZYG and OOI) were positioned on the dominant side of the face. The experiment was video-monitored using a Sony HDR-CX240E video camera to track any visible facial movements. A microphone was placed 20–30 cm away from the participant’s lips to record any faint vocal production during the inner speech and listening conditions. Stimuli were displayed using the OpenSesame software on a 19-inch colour monitor. ### Linguistic material We selected ten rounded and ten spread bi-syllabic nonwords (cf.). Each class of nonwords was specifically designed to either induce a greater activation of the lip muscle (rounded items) or a greater activation of the zygomaticus muscle (spread items). These stimuli were selected based on phonetic theoretical constraints, with the aim of maximising the differences between the two classes of non-words in their involvement of either the OOI or the ZYG muscle. More precisely, rounded items consisted in the repetition of a syllable containing a bilabial consonant followed by a rounded vowel, whereas spread items consisted in the repetition of a syllable containing a lingual consonant followed by a spread vowel. ## Procedure Participants were seated in front of a computer screen while audio stimuli (when applicable) were presented through speakers on both sides of the screen. A video camera was positioned on one side of the screen to monitor facial movements. A microphone was positioned at approximately 10 cm of the participant to record possible speech sounds. After positioning of the EMG sensors, each participant underwent a relaxation session aiming to minimise pre-existing inter-individual variability on facial muscle contraction (approximate duration was 330 s). This relaxation session was recorded by a trained professional sophrology therapist. Baseline EMG measurements were performed during the last minute of this relaxation session, resulting in 60 s of EMG signal at baseline. By using this relaxation period as a baseline, we made sure that participants were all in a comparable relaxed state. In addition, several previous EMG studies have argued for the use of a relaxation period as a baseline, since mere resting periods may include some inner speech production (e.g., for a review). Subsequently, participants went through a training session, during which they could get familiar with the main task. They trained with 8 stimuli in total (4 rounded nonwords and 4 spread nonwords, cf.). Each training stimulus appeared in three conditions (for all participants): overt speech, inner speech and listening. Nonwords to be produced (covertly or overtly) were visually presented on the screen. Then, a central fixation dot appeared on the screen, indicating to the participant that s•he should utter the nonword (either overtly or covertly). This aimed to ensure that participants were actually producing a nonword, not just simply visually scanning it. In the overt speech condition, participants were asked to produce the nonword “just after the word disappeared from the screen”, with “the most neutral intonation possible”. In the inner speech condition, participants were asked to “innerly produce the nonword” (cf. the for precise instructions in French), with “the most neutral intonation possible” and while remaining as still as possible. In the listening condition, the order of these two screens was reversed. A fixation dot was first presented (for 1 second), followed by a blank screen (for 1 second). The audio stimulus was presented when the blank screen appeared, while participants were asked to remain as still as possible. After the training, participants moved to the experimental part, that included a novel list of 20 nonwords (cf.). Each nonword was presented 6 times in each condition for each participant. The EMG activity was recorded during the entire experiment. The periods of interest consisted in one-second portions, after each stimulus presentation and during either production or listening. This resulted in 60 observations (60 periods of 1 second) for both classes of nonword in each test condition. The total duration of the experiment ranged between 30 min and 40 min. ## EMG signal processing EMG signal pre-processing was carried out using Matlab r2014a (Version 8.3.0.532, [www.mathworks.fr](http://www.mathworks.fr)). We first applied a 50Hz frequency comb filter to eliminate power noise. Then, we applied a 20 Hz—450 Hz bandpass filter to the EMG signals, in order to focus on the 20–450 Hz frequency band, following current recommendations for facial EMG studies. Although participants were explicitly asked to remain still during inner speech production or listening, small facial movements (such as swallowing movements) sometimes occurred. Such periods were excluded from the final sample of EMG signals. To remove these signals, we first divided the portions of signals of interest into periods of 1 second. The baseline condition was therefore composed of 60 trials of 1 second. The periods of interest in all the speech conditions consisted of the 1-second interval during which the participants either produced speech or listened to speech. It is possible that the nonword took less than 1 second to be produced, but since there was no way to track when production started and ended in the inner speech condition, the entire 1-second period was kept. Therefore, the overt speech condition was composed of 6 repetitions of each nonword, that is 6x20 trials of 1 second. The “inner speech” and “listening” conditions were similarly composed of 6x20 trials of 1 second. Then, we visually inspected the EMG signals recorded during each trial and listened to the audio signal simultaneously recorded. In all conditions, any time a non- speech noise (such as coughing or yawning) was audible in a trial, the trial was discarded (i.e., we did not include this trial in the final analysis, for any of the recorded muscles). In the listening and overt speech conditions, if a burst of EMG activity was present after the relevant audio speech signal, then the trial was discarded. In the overt speech condition, if the participant started too early or too late and only part of the nonword was recorded in the audio signal, then the corresponding trial was discarded. In the inner speech and listening condition, if a large EMG burst of activity was present, potentially associated with irrelevant non-speech activity, we excluded the trial. The fact that the artefact rejection procedures slightly differ in the various conditions is not an issue since we do not directly compare between conditions. Instead, we compare the EMG correlates of the two classes of nonwords within each condition. This inspection was realised independently by two judges. Subsequently, we only kept the trials that were not rejected by any of these two judges (i.e., we removed a trial as soon as it was rejected by at least one judge). The agreement rate between the two judges was of 87.82% (with a moderate Cohen’s *κ* of approximately 0.48). The overall procedure led to an average (averaged over participants) rejection rate of 22.96% (SD = 6.49) in the baseline condition and 18.49% (SD = 6.48) in the other conditions. After pre-processing and artefact rejection, we computed the by-trial average amplitude of the centered and rectified EMG signal. This provided a score for each muscle of interest (OOI, ZYG, FRO, COR, FCR) in each condition (Baseline, Overt Speech, Inner Speech, Listening) and for each participant. Absolute EMG values are not meaningful as muscle activation is never null, even in resting conditions, due in part to physiological noise. In addition, there are inter- individual variations in the amount of EMG activity in the baseline. To normalise and standardise for baseline activity across participants, we thus expressed the EMG amplitude as a z-score from baseline activity (i.e., we subtracted the mean amplitude of the centred and rectified baseline signal and divided the result by the standard deviation of the centred and rectified baseline signal), thereafter referred to as *δ*. ## Data analysis Statistical analyses were conducted using `R` version 3.5.3, and are reported with the `papaja` and `knitr` packages. To assess the effects of the condition and the class of nonwords on the standardised EMG amplitude, we analysed these data using *Condition* (3 modalities: speech, inner speech, and listening) and *Class* of nonwords (2 modalities, rounded and spread, contrast-coded) as within-subject categorical predictors, and the standardised EMG amplitude as a dependent variable in a multivariate (i.e., with multiple outcomes) Bayesian multilevel linear model (BMLM). An introduction to Bayesian statistics is outside the scope of this paper. However, the interested reader is referred to for an introduction to Bayesian multilevel modelling using the `brms` package. In order to take into account the dependencies between repeated observations by participant, we also included in this model a varying intercept by participant. Contrary to what we pre-registered, we used a multivariate model (instead of separate models by muscle). This allowed us to estimate the correlation between each pair of muscles. Models were fitted with the `brms` package and using weakly informative priors (see the for code details). Two Markov Chain Monte- Carlo (MCMC) were run for each model to approximate the posterior distribution, including each 5.000 iterations and a warmup of 2.000 iterations. Posterior convergence was assessed examining trace plots as well as the Gelman-Rubin statistic $\hat{R}$. Constant effect estimates were summarised via their posterior mean and 95% credible interval (CrI), where a credible interval can be considered as the Bayesian analogue of a classical confidence interval. When applicable, we also report Bayes factors (BFs), computed using the Savage-Dickey method, which consists in taking the ratio of the posterior density at the point of interest divided by the prior density at that point. These BFs can be interpreted as an updating factor, from prior knowledge (what we knew before seeing the data) to posterior knowledge (what we know after seeing the data). # Results The *Results* section is divided into two parts. First, we present results from confirmatory (preregistered) analyses, aiming to test whether it is possible to dissociate the activity of the OOI and the ZYG during inner speech production, according to the content of inner speech (here, the class of nonword). More precisely, we expected an increased EMG activity of the OOI during the inner production of rounded nonwords in comparison to spread nonwords. Conversely, we expected elevated EMG activity of the ZYG during the inner production of spread nonwords in comparison to rounded nonwords. Second, we present results from exploratory (non-preregistered) analyses. To foreshadow the results, we did not observe such a clear dissociation between the EMG activity of the OOI and the ZYG muscles, neither in the inner speech condition nor in the overt speech condition. Contrary to theoretical expectations based on phonetics and speech production theory, the activity of both muscles was of higher amplitude for the pronunciation of rounded nonwords (as compared to spread nonwords) during overt speech production. Additionally, the EMG amplitude on both muscles of interest was similar during the inner production (or listening) of the two classes of nonwords. However, in the exploratory analyses section, we report results from supervised machine learning algorithms (classification using random forests), showing a reasonable accuracy to classify EMG signals according to the class of nonwords during overt speech production. This strategy was however unsuccessful for the inner speech and the listening conditions. Before moving to the statistical results, we represent the distribution of the whole dataset, by class, by condition and by muscle for the two main muscles of interest (OOI and ZYG) in. More precisely, the first row of this figure represents the distribution of the standardised EMG scores in the inner speech condition, the second row depicts the distribution of these scores in the listening condition, whereas the third row depicts the distribution of the standardised EMG scores in the overt speech condition. The first column depicts the distribution of the standardised EMG scores recorded over the OOI muscle whereas the second one represents the distribution of the standardised EMG scores recorded over the ZYG muscle. Each individual data point is represented as a vertical bar along the x-axis of each panel whereas the vertical coloured line represents the class-specific median. Additionally, a vertical dashed line is plotted at zero, which represents the baseline level. Thus, a positive value on the x-axis represents EMG standardised scores that are higher than baseline. In, we report the mean standardised EMG amplitude of all recorded muscles in each condition. Given the skewness of the distribution of these scores, the mean and the standard deviation (SD) are not the best indicators of the central tendency and dispersion of these distributions. Therefore, we also report the median, the median absolute deviation (MAD), and the inter-quartile range (IQR). We also created a `shiny` application allowing for further visual exploration of the data by muscle, by condition, and by participant, in the 3D space formed by three (to be chosen) muscles. This application is available online (at <https://barelysignificant.shinyapps.io/3d_plotly/>) and the associated code is available in the OSF repository (<https://osf.io/czer4>). ## Confirmatory (preregistered) analyses ### Bayesian multivariate multilevel Gaussian model We then compared the standardised EMG amplitude *δ* for each muscle in each condition (Overt Speech, Inner Speech, Listening) by fitting a multivariate multilevel Gaussian model (as detailed previously in the Methods section). We predicted a higher increase of OOI activity during the inner production of rounded items in comparison to spread items and conversely, a higher increase of ZYG activity during the inner production of rounded items in comparison to spread items. These predictions should also apply to the overt speech condition (and to the listening condition). We should not observe any by-class differences of FRO and COR activity in any condition. The results of the Bayesian Gaussian multivariate model are summarised in. This table reports the estimated average EMG amplitude in each condition and the corresponding BF. As they are not the main focus of interest here and for the sake of clarity, descriptive results for the other two facial muscles and for the forearm muscle are reported in the. This analysis revealed that the EMG amplitude of the OOI was higher than baseline (the standardised score was above zero) in every condition whereas it was only the case in the overt speech condition for the ZYG. Moreover, in all conditions, the EMG amplitude of the ZYG was lower than that of the OOI. Crucially, we did not observe the hypothesised difference according to the class of nonwords on the OOI during inner speech production (*β* = 0.071, 95% CrI \[-0.204, 0.342\], BF₀₁ = 64.447) nor on the ZYG (*β* = 0.005, 95% CrI \[-0.031, 0.041\], BF₀₁ = 532.811). depicts these results by representing the distribution of the raw data (coloured dots) along with the predictions from this model. The black dots and vertical intervals represent the predicted mean and associated 95% credible interval for each class of non-word, each condition and for the OOI and the ZYG. Coherently with, this figure shows that the fitted model predicts no noticeable differences between the two classes of non-words in any condition for the OOI muscle. However, it predicts a higher average EMG amplitude associated with the rounded item as compared to the spread items in the overt speech condition for the ZYG muscle. Before proceeding further with the interpretation of the results, it is essential to check the quality of this first model. A useful diagnostic of the model’s predictive abilities is known as *posterior predictive checking* (PPC) and consists in comparing observed data to data simulated from the posterior distribution. The idea behind PPC is that a good model should be able to generate data that resemble the observed data. In this vein, represents the distribution of the whole dataset (across all participants and conditions) by muscle (the dark blue line) along with the distribution of hypothetical datasets generated from the posterior distribution of the model (the light blue lines). As can be seen from this figure, the distributions of the data generated from the model differ considerably from the distribution of the observed data. Therefore, in the next section, we turn to a more appropriate model for these data. ### Bayesian multivariate multilevel distributional Skew-Normal model reveals an important failure of the first model, as it fails to generate data that look like the data we have collected. More precisely, the collected data look right-skewed, as it usually happens with physiological measurements. To improve on the Gaussian model, we then assumed a Skew-normal distribution for the response variable (the standardised EMG amplitude *δ*). The Skew-normal distribution is a generalisation of the Gaussian distribution with three parameters *ξ* (xi), *ω* (omega), and *α* (alpha) for location, scale, and shape (skewness), respectively (note that the Gaussian distribution can be considered a special case of the Skew-normal distribution when *α* = 1). In addition, we also improved the first model by turning it into a *distributional model*, that is, a model in which we can specify predictor terms for all parameters of the assumed response distribution. More precisely, we used this approach to predict both the location, the scale, and the skewness of the Skew-Normal distribution (whereas the first model only allowed predicting the mean of a Gaussian distribution). As can been seen in, this second model seems better than the first one at generating data that fit the observed data. The estimates of this second model are summarised in and. According to this model, the EMG amplitude of the OOI was higher than baseline (the estimated standardised score was above zero) in every condition whereas, for the ZYG, it was only the case in the overt speech condition. We did not observe the hypothesised difference according to the class of nonwords during inner speech production, neither on the OOI (*β* = 0.025, 95% CrI \[-0.012, 0.064\], BF₀₁ = 64.447) nor on the ZYG (*β* = 0.004, 95% CrI \[-0.007, 0.014\], BF₀₁ = 532.811). Predictions from this model are visually represented in. This figure differs from (showing the predictions of the Gaussian model) in that the second model (the Skew-normal model) predicts shifts in location for both the OOI and the ZYG muscles according to the class of non-word in overt speech prediction. In contrast, the first model (the Gaussian model) predicted a by-class difference only for the ZYG muscle. ## Exploratory (non-preregistered) analyses In the previous section, we tried to predict the average EMG amplitude by condition on each single muscle. Although this approach was appropriate to tackle our initial research question (i.e., can we distinguish muscle-specific EMG correlates of inner speech production?), it is not optimal to answer more general questions such as “can we predict the content of inner speech based on the available EMG data?”. In, we depict the distribution of the by-word averaged EMG scores in the 2D space formed by the OOI and the ZYG muscles. This figure reveals that although different nonwords produced in overt speech seem difficult to discriminate on the basis of a single muscle (cf.), it seems easier to discriminate them in the space formed by two muscles (here OOI and ZYG). More precisely, the two classes of nonwords seem to form two separate clusters in the overt speech condition, but these clusters do not seem discriminable in the inner speech or in the listening condition. In other words, it is easier to discriminate these signals in the multidimensional space of all speech muscles, rather than by considering each muscle independently. Thus, we used a supervised machine learning algorithm aiming to classify speech signals according to the class of nonwords. Broadly, the machine learning approach seeks to find a relationship between an input *X* (e.g., EMG recordings over the four facial muscles) and an output *Y* (e.g., the class of nonwords). Once trained, it allows predicting a value of the output based on some input values, whose prediction can be evaluated against new observations. We used a random forest algorithm, as implemented in the `caret` package. Random forests (RFs) represent an ensemble of many decision trees (a forest), which allow predictions to be made based on a series of decision rules (e.g., is the score on predictor *x*₁ higher or lower than *z*? If yes, then …, if not, then …). The specificity of RFs is to combine a large number of trees (usually above 100 trees), and to base the final conclusion on the average of these trees, thus preventing overfitting. We used three separate RFs to classify EMG signals in each condition (Overt Speech, Inner Speech, and Listening). To evaluate the performance of this approach, we report the raw accuracy (along with its resampling-based 95% confidence interval), or the proportion of data points in the test dataset for which the RF algorithm predicted the correct class of nonwords. First, we randomly split the entire dataset into a training (80%) and a test set (20%). The training set was used for the learning whereas the test set was used to evaluate the predictions of the algorithm. To prevent overfitting, we used repeated 10-fold cross-validation during the learning phase. ### Predicting the class of nonwords during overt speech production We first tried to predict the class of nonwords produced in overt speech, based on the activity of the four facial muscles (OOI, ZYG, COR, FRO). Each predictor was centred to its mean and standardised before the analysis. This analysis revealed an overall classification accuracy of 0.847, 95% CI \[0.814, 0.876\] (cf. confusion matrix). Examining the relative importance of each feature (i.e., each muscle) for prediction revealed that the muscles containing most information to discriminate the two classes of nonwords were the ZYG and the OOI, whereas, as predicted, forehead muscles did not seem to strongly contribute to predictive accuracy in the overt speech condition. ### Predicting the class of nonwords during inner speech production and listening We then applied the same strategy (the same algorithm) to the signals recorded in the inner speech and listening conditions. The results of these analyses are reported in Tables and. This analysis revealed an overall classification accuracy of 0.472, 95% CI \[0.426, 0.52\] in the inner speech condition, which indicates that the RF algorithm did not allow discriminating the two classes of nonwords better than random guessing. As the classification accuracy in the inner speech and listening conditions was not better than chance, we do not report the relative importance of the predictors. Indeed, it would be difficult to interpret the importance of predictors for a classification task at which they do not perform better than chance. This analysis similarly revealed an overall classification accuracy of 0.46, 95% CI \[0.413, 0.507\] in the listening condition. # Discussion In the present study we aimed to replicate and extend previous findings showing that facial electromyography can be used to discriminate expanded inner speech content. As these studies used small samples of children, our study aimed to examine whether such results can be reproduced using surface electromyography and modern signal processing methods in an adult sample. To this end, it was crucial to first show that the EMG correlates of our two classes of nonwords were discriminable during overt speech production. Surprisingly, the data we collected during overt speech production do not corroborate the hypothesis according to which the average EMG amplitude of the OOI should be higher during the production of “rounded” nonwords as compared to “spread” nonwords. For both orofacial speech muscles (OOI and ZYG), the average EMG amplitude was higher for rounded nonwords than for spread nonwords during overt speech production. Moreover, whereas the average EMG amplitude recorded over speech muscles was higher than baseline in both the inner speech and listening conditions, we did not find differences of activation according to the content of the material (the class of nonword). An automatic classification approach, using the four facial muscles (OOI, ZYG, COR, FRO), revealed that although it was possible to discriminate EMG signals related to the two classes of nonwords with a reasonable accuracy during overt speech production, this approach failed in discriminating these two classes during inner speech production or during listening. We also observed a higher EMG amplitude recorded over the facial (both orofacial and non-orofacial) muscles during inner speech production and during the listening of speech production than during rest. However, as pinpointed by, this observation is not sufficient to conclude that these activations were actually related to inner speech production, because i) both orofacial speech-related muscles and forehead non-speech related muscles showed similar EMG amplitude changes from baseline and ii) we did not observe different changes in EMG amplitude depending on the content of inner speech (i.e., depending on the class of nonword to be uttered). Before discussing the theoretical implications of these results, two main issues are worth discussing. First, how can we explain that rounded nonwords were associated with higher EMG amplitude during overt speech on both OOI and ZYG muscles? Second, how can we explain the indiscriminability of inner speech content, which seems to contradict classic as well as recent findings in the field? We turn to each of these questions in the following. To answer the first question, we began by comparing our results to results obtained by another group. The authors of this study recorded surface EMG activity from five participants while they were producing seven facial expressions and five isolated vowel sounds (/a/, /e/, /i/, /o/, /u/), repeated five times each. They recorded EMG activity over eight facial muscles (the zygomaticus major (ZYG), the risorius (RIS), the orbicularis oris superior (OOS) and inferior (OOI), the mentalis (MEN), the depressor anguli oris (DAO), the levator labii superioris (LLS) muscles, and the digastric muscle (DIG)). We divided these vowels in two classes to fit our own classes of nonwords. More precisely, we have created the following two classes: a *rounded* class, composed of the vowels /o/ and /u/, and a *spread* class, composed of the vowels /e/ and /i/ (note that we did not include the vowel /a/ because it theoretically does not fit in one of these two categories). We present the average EMG amplitude recorded over the OOI and the ZYG according to the vowel class in. We notice that have indeed observed the dissociation we initially predicted, that is, that the EMG amplitude recorded over the OOI was higher during the pronunciation of rounded vowels than during pronunciation of spread vowels, whereas the reverse pattern was observed concerning the ZYG. Paired-samples Wilcoxon signed rank tests revealed a shift in location (pseudomedian) between rounded and spread items for the OOI (*β* = 24.12, 95% CI \[15.19, 40.77\], V = 1184, p \<.001) with rounded items being associated with a higher location than spread items. This analysis also revealed a shift in the inverse direction concerning the ZYG (*β* = -1.51, 95% CI \[-2.94, -0.48\], V = 275, p \<.001). However, one crucial difference between design and ours is the complexity of the linguistic material. Whereas used single phonemes, we chose to use bisyllabic nonwords to increase the ecological validity of the paradigm. Although these nonwords were specifically created to theoretically increase the engagement of either the OOI or the ZYG (cf. the “Linguistic material” section), it is reasonable to expect differences in the average EMG patterns between isolated phonemes and nonwords. More precisely, we expect the *average* EMG amplitude associated with the production of a given phoneme (e.g., /y/) to be impacted by the production of the consonant (e.g., /b/) it is paired with, due to coarticulation. More generally, we could hypothesise that the difference between the *average* EMG amplitude recorded during the production of the phoneme /i/ and during the production of the phoneme /y/ could be reduced when these phonemes are coarticulated in CV or CVCV sequences like /byby/ or /didi/ (as in our study). In other words, we might expect an interaction effect between the structure of the to-be produced speech sequence (either a single vowel or a CV/CVCV sequence) and the class of the vowel. This is coherent with previous findings showing that the muscular activity associated with vowel production is strongly influenced by the surrounding consonants in CVC sequences. Thus, further investigations should focus on how the average EMG amplitude is impacted by coarticulation during the production of CVCV sequences. With regards to inner speech, our results do not support theoretical predictions of the *motor simulation view*, according to which it should be possible to discriminate classes of nonwords produced in inner speech based on EMG signals. Whereas this outcome is consistent with some recent results, it also stands in sharp contrast with classical results in the field as well as more recent developments. For instance, developed a wearable device composed of seven surface EMG sensors that can attain a 92% median classification accuracy in discriminating internally vocalised digits. There are a few crucial differences between’s work and ours that stand as good candidates to explain the discrepancies between our results. First, the strategy adopted to position the sensors was radically different. Following guidelines from the field of psychophysiology, our strategy was to position sensors precisely over the facial muscles of interest, aligned with the direction of the muscle fibers and in theoretically optimal positions to record activity of this muscle while reducing cross-talk. However, precisely because of pervasive cross-talk in facial surface EMG recordings, this strategy, whereas maximising the probability of recording activity from a given single muscle, was also (as a result) reducing the probability of recording activity from potentially speech-relevant neighbour muscles. Therefore, this strategy might work sub-optimally when the goal of the experiment is to extract the maximum amount of (relevant) EMG information to discriminate inner speech content. However, this problem might be mitigated by using more sensors and a more lenient sensor-positioning approach. Whereas we recorded the EMG amplitude over only two lower facial muscles (OOI and ZIG), analysed EMG data from seven different sensors, whose position and number was defined iteratively in order to maximise the classification accuracy. In other words, the parameters of the experiment were iteratively optimised to maximise a certain outcome (classification accuracy). This strategy is radically different from the classical approach in experimental and cognitive psychology where experimental conditions are defined to test theoretically derived hypotheses. Whereas the first approach is arguably more efficient at solving a particular problem at hand, the second approach might be more efficient in tackling theoretical questions. For instance, a recent study reported a greater EMG amplitude of laryngeal and lip muscles during auditory verbal tasks (covert singing) than during visual imagery tasks. By coupling EMG recording with demographic and psychological measures, they were able to show that these correlates were related to the level of accuracy in singing, thus shedding light upon the nature and functions of peripheral muscular activity during covert singing. However, adding more sensors (e.g., on the risorius), or better optimising sensor placement, could improve the sensitivity of the present approach. Putting aside considerations related to methodological aspects of the present study, these results do not corroborate the *motor simulation view* of inner speech production. Instead, it seems to support the *abstraction view*, which postulates that inner speech results from the activation of abstract linguistic representations and does not engage the articulatory apparatus. However, individual differences in discriminability highlight that the abstractness of inner speech might be flexible, as suggested by. Indeed, although for most participants it was not possible to decode the phonetic content of inner speech, rounded and spread nonwords were in fact distinguishable based on OOI and ZYG information only (by visual inspection of the 2D plot), for two of them (S_15 and S_17, cf.). This suggests either that the extent to which inner speech production recruits the speech motor system might vary between individuals or that it might vary within individual depending on the properties of the ongoing task (these two suggestions are not mutually exclusive). For instance, we know from early research on the EMG correlates of inner speech that the average amplitude of these correlates tend to be higher when the task is more difficult. As such, the extent to which inner speech production recruits the speech motor system could be moderated by manipulating the difficulty of the ongoing task. In addition, the electromyographic activity recorded during motor imagery could be modulated by the perspective taken in motor imagery. A distinction is made between first-person perspective or *internal imagery* (i.e., imagining an action as we would execute it) and third-person perspective or *external imagery* (i.e., imagining an action as an observer of this action), that may involve different neural processes. It has been shown that a first-person perspective may result in greater EMG activity than motor imagery in a third- person perspective. Therefore, we hypothesise that the involvement of the speech motor system during inner speech production may be modulated by the specific instructions given to the participants. For instance, by instructing participants to focus on *inner speaking* (imagining speaking), instead of *inner hearing* (imagining hearing), and by asking them to focus on the kinaesthetic feelings related to speech acts (rather than on auditory percepts), we could expect to find a higher average EMG amplitude recorded over the speech muscles. In addition, by specifically asking the participants to mentally articulate the nonwords, as if they were dictating them to someone, rather than just read and visually scan them, we may expect stronger articulatory involvement. Of course, the current study and the above discussion should be interpreted with a few words of caution in mind. For each class of nonwords, we collected around 6 x 10 = 60 observations by condition and by participant. For 25 participants and two classes of nonwords, this results in 25 (participants) x 120 (individual trials) x 3 (conditions) = 9000 observations. However, after rejecting trials with movement artefacts, we had 7285 observations in total. Although the number of observations reported in the present study is reasonable, the sensitivity of the experiment could be improved by increasing the number of observations and/or by reducing two important sources of variation. More precisely, one could reduce the variance related to the item (the specific nonword being uttered) by selecting nonwords that are more similar to each other in the way they are uttered, by selecting less items or simpler items. Similarly, particular attention should be devoted to reducing inter-participant variability, which could be done by using more guided and specific instructions, as well as a longer training phase to familiarise the participant with the task. In summary, we have demonstrated that whereas surface electromyography may lead to reasonable accuracy in discriminating classes of nonwords during overt speech production (using signals recorded over only two speech-related muscles), it did not permit to discriminate these two classes during inner speech production across all participants (only for two participants). These results, in comparison with results obtained by other teams, highlight that depending on the aim of the research, different strategies might be more or less successfully pursued. More precisely, if the goal is to attain high classification accuracy (problem-solving approach), then the parameters of the experiment (e.g., number of repetitions, number of sensors, position of the sensors, parameters of the signal processing workflow) should be optimised based on the desired outcome (i.e., classification accuracy). However, the classical laboratory strategy used in experimental and cognitive psychology, aiming to compare specific conditions (or muscles) to each other in a controlled environment, is deemed to be more appropriate when the aim of the research is to sharpen our understanding of the psychological phenomenon under study. # Supporting information 10.1371/journal.pone.0233282.r001 Decision Letter 0 Sulpizio Simone Academic Editor 2020 Simone Sulpizio This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Jan 2020 PONE-D-19-27197 Can we decode phonetic features in inner speech using surface electromyography? PLOS ONE Dear Mr. Nalborczyk, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. I have received two reviews from experts in the field and I have also read the paper. Both reviewers and I think the paper is interesting and deserves consideration. The reviewers give helpful suggestions that I would ask you to implement with care. I only add one indication: I agree with reviewer 2 that Vygotskian theory should be cited and discussed in the paper, but I would suggest to include it in the Introduction without re-writing it. We would appreciate receiving your revised manuscript by Mar 07 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Simone Sulpizio Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> 2\. We note you have included a table to which you do not refer in the text of your manuscript. Please ensure that you refer to Table 5 in your text; if accepted, production will need this reference to link the reader to the Table. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Nalborczyk and colleagues present a study in which they tested the ‘motor simulation’ model of inner speech by measuring EMG to overt speech, inner speech and listening in the OOI and ZYG muscles (and control muscles) in healthy adult volunteers. They did not observe the expected difference in EMG activity between the OOI and ZYG muscles to overt speech, which somewhat complicated the comparison with the inner speech condition. They did find that it was possible to discriminate between the phonetic content of inner speech in some individuals, but this did not maintain at the level of the group. Overall, I think this this paper represents a nice piece of scholarship. The literature review was comprehensive, well written and argued, and the hypotheses were clear and well justified. The statistics and analyses were sophisticated and thoroughly and carefully presented. I particularly liked how the (key) analyses were pre-registered – IMO this is a good habit and one which should be encouraged in the field. The interpretability of the results was limited somewhat by the study’s failure to replicate the key finding of differential EMG activity in the OOI and ZYG muscles during the (overt) production of rounded vs. spread nonwords. But that’s the way science goes, and it would be good to have these data available in the literature, and I believe that the publication of null results should be encouraged in order to provide a more accurate reflection of the true state of the field. I have a few minor comments that the authors could consider, primarily regarding the experimental task itself: \- Probably my main point is that I do not understand is the ‘artefact removal procedure’ described on p.8. The authors state: “To remove these signals, we first divided the EMG signals into periods of 1 second” – were these epochs centred around something (e.g., fixation dot onset), or was the start/end location arbitrary? How did the authors distinguish between ‘unwanted’ EMG activity (caused by swallowing, etc.) from ‘of-interest’ EMG activity caused by movement of the articulators? It is possible that the artefact removal procedure removed some of the signal of interest? Could this not be problematic, particularly if the authors only applied the artefact removal procedure to the listening and inner speech conditions? (as seems to be implied on p.8, but I could be wrong here – please clarify). \- P.7. To clarify: am I right in understanding that the EMG activity in the 1 second following the presentation of the fixation dot was used as the dependent variable? If so, were participants instructed to produce the inner/overt speech as soon as possible after the fixation dot appeared? What was the delay (on average) between the dot and speech production in the overt speech condition? Do the authors assume that this delay was consistent between the inner and overt speech conditions? If so, can they justify this assumption? \- It appears as though the ‘baseline’ period (i.e., against which the standardized scores were calculated) was calculated during the relaxation task that occurred prior to the talk itself, is that correct? I don’t understand the benefit of doing this as opposed to, say, calculating the baseline from the 1 second prior to the presentation of the fixation dot – can the authors clarify? \- Figure 1: I think it would be appropriate to mention in the caption that the scales between the figure panels differ markedly – these differences are understandable (i.e., EMG activity in the overt speech condition will obviously be higher than the inner speech / listening conditions), but still appropriate to mention. Reviewer \#2: The present article addresses an important but overlooked question about the external behavioural traces of inner speech. Its reliance on preregistered methods is really impressive, particularly in light of the complexity of possible analyses and the risk of data ‘fishing’. We have several main concerns. One is the theoretical framework in which the study has been conceived, which over-emphasises a problematic distinction between simulation and abstraction views, and pays insufficient attention to the key work in this area. Abstract. Second sentence should be revised; the authors do not need to describe the steps in detail. The last sentence should also be removed as this information should be submitted separately from the abstract (and also should be included in the Data Availability section at the end of the manuscript). 2\. A strong claim is made about the automatic elicitation of inner speech in reading. The picture is not quite so clear-cut; Russell Hurlburt for example has recently published a study showing very little inner speech during reading. 9\. The discussion of the featural properties of inner speech is a little thin and could be enhanced. 13\. The term ‘inner voice’ is problematic – does it mean ‘inner speech’? In our view they have different connotations (see Fernyhough’s recent book on inner speech, The Voices Within, for a discussion). 18 – 28. This section is incomplete and does not include the essential Vygotskian theory that inner speech is internalised external speech. This is a serious omission. See Alderson-Day and Fernyhough (2015) for a full description of this theory. This is important not just because it is the most developed theory of inner speech, but because it is also highly relevant to the question of whether any behavioural traces of inner speech will be observed (see the discussion in Jones & Fernyhough, 2007, cited here as ref 4, and also extensive discussion in Fernyhough’s book). The fact that Vygotsky’s name is not mentioned in this manuscript is surprising. The introduction will need to be rewritten to represent these important views, which are significant for all of the main hypotheses in this study and for the interpretation of the findings. Furthermore, since Vygotsky’s is a developmental theory, it is very relevant to the question of different methods and findings for EMG studies of inner speech in children and adults (line 547). Giving some attention to Vygotsky’s theory would also make a lot of sense of the authors’ surprising findings on EMG activity during inner speech. See the discussions cited above about whether, on an internalisation view, you would expect any motoric trace of inner speech, if the latter is fully internalised and transformed outer speech – which also brings in the important issue of semantic and syntactic condensation. 188\. The authors describe the sampling procedure in detail, but the Participants section would benefit from including more demographic information on the sample. 214\. Studies looking at EMG correlates of lip muscle activity often investigate the orbicularis oris superior muscle alongside the orbicularis oris inferior. The authors should discuss in more detail their justification for studying the OOI, but not OOS. 243\. ‘during rumination’ – we think this is a mistake, and an alternative term should be used, e.g., covert speech condition. 245\. Please give more detail here on how the stimuli were selected. Were the stimuli piloted before being selected? 253\. More detail is needed on the instructions given in the three conditions. Hurlburt et al. (PLOS, 2016) have recently shown a difference in brain activations between elicited and spontaneous inner speech, questioning whether inner speech elicited by the method apparently used here can be taken as a reliable proxy for genuine, spontaneous inner speech. This needs some acknowledgement and discussion, especially as it addresses many of the issues raised in the Jones & Fernyhough (2007) paper. More explanation is needed on the instruction given to participants during training, e.g. how the visual cues were explained to indicate the start of the task. 349\. Figures 6 & 7 could be moved to Supplementary Information. 583-628. The paragraph on stimuli selection should be revised. The detailed information should be moved to the Method section, where it would have been useful to have more information on how the stimuli were selected. More general reflection on this process should be included in the Discussion instead. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0233282.r002 Author response to Decision Letter 0 10 Mar 2020 Response to editorial revision requests We are very grateful to the editor and reviewers for their suggestions and comments which substantially contributed to the improvement of our paper. We provide a point-by-point response in the following and have included appropriate changes in the revised version of our manuscript. Reviewer \#1 Nalborczyk and colleagues present a study in which they tested the ‘motor simulation’ model of inner speech by measuring EMG to overt speech, inner speech and listening in the OOI and ZYG muscles (and control muscles) in healthy adult volunteers. They did not observe the expected difference in EMG activity between the OOI and ZYG muscles to overt speech, which somewhat complicated the comparison with the inner speech condition. They did find that it was possible to discriminate between the phonetic content of inner speech in some individuals, but this did not maintain at the level of the group. Overall, I think this this paper represents a nice piece of scholarship. The literature review was comprehensive, well written and argued, and the hypotheses were clear and well justified. The statistics and analyses were sophisticated and thoroughly and carefully presented. I particularly liked how the (key) analyses were pre-registered – IMO this is a good habit and one which should be encouraged in the field. The interpretability of the results was limited somewhat by the study’s failure to replicate the key finding of differential EMG activity in the OOI and ZYG muscles during the (overt) production of rounded vs. spread nonwords. But that’s the way science goes, and it would be good to have these data available in the literature, and I believe that the publication of null results should be encouraged in order to provide a more accurate reflection of the true state of the field. I have a few minor comments that the authors could consider, primarily regarding the experimental task itself: \- Probably my main point is that I do not understand is the ‘artefact removal procedure’ described on p.8. The authors state: “To remove these signals, we first divided the EMG signals into periods of 1 second” – were these epochs centred around something (e.g., fixation dot onset), or was the start/end location arbitrary? How did the authors distinguish between ‘unwanted’ EMG activity (caused by swallowing, etc.) from ‘of- interest’ EMG activity caused by movement of the articulators? It is possible that the artefact removal procedure removed some of the signal of interest? Could this not be problematic, particularly if the authors only applied the artefact removal procedure to the listening and inner speech conditions? (as seems to be implied on p.8, but I could be wrong here – please clarify). We thank the reviewer for their comment. These elements are indeed crucial in interpreting our data. The 1-second epochs were not chosen arbitrarily but correspond to the periods of interest, that is, the periods during which participants were either 1) doing nothing (that is the baseline period, composed of 60 periods of 1-second), 2) producing overt speech (the “speech” condition, composed of 6 repetitions of each nonword, that is overall 6\*20 periods of 1 second per participant), 3) producing inner speech (the “inner speech” condition, composed of 6 repetitions of each nonword, that is overall 6\*20 periods of 1 second per participant), or 4) listening to overt speech (the “listening ” condition, composed of 6 repetitions of each nonword, that is overall 6\*20 periods of 1 second per participant). Of course, the artefact rejection procedure could not be applied similarly in the overt speech condition, since it is impossible to visually distinguish speech-related activation from non- speech related activities such as swallowing. But, in addition to the artefact rejection procedure applied to the inner speech and listening conditions, further checks were carried out. In all conditions, since we had recorded the audio signal, any time a non-speech noise (such as coughing or yawning) was present, the EMG signal for that trial was discarded. In the listening and overt speech conditions, if a burst of EMG activity was present after the relevant audio speech signal, then the trial was discarded. In the overt speech condition, if the participant started too early or too late and only part of the nonword was recorded in the audio signal, then the corresponding trial was discarded. Moreover, the fact that the artefact rejection procedure differ in the various conditions is not an issue, since we do not directly compare between conditions. Instead, we compare the EMG correlates of the two classes of nonwords within each condition. However, we agree it is a possibility that this procedure might remove some of the signal of interest in the inner speech and listening conditions. The reason we still applied it is because the kind of activation we can record in the inner speech condition may be orders of magnitude lower than parasite activations such as activation related to breathing, swallowing, facial expressions, or coughing (amongst others). Therefore, although these “parasite activities” are expected to be distributed uniformly across the two classes of non-words, they add a significant amount of noise to the signal and could diminish the ability to detect inner-speech-related EMG correlates. To clarify this, we have made the following change in the manuscript (cf. page 9): Although participants were explicitly asked to remain still during inner speech production or listening, small facial movements (such as swallowing movements) sometimes occurred. Such periods were excluded from the final sample of EMG signals. To remove these signals, we first divided the portions of signals of interest into periods of 1 second. The baseline condition was therefore composed of 60 trials of 1-second. The periods of interest in all the speech conditions consisted of the 1 second interval during which the participants either produced speech or listened to speech. It is possible that the nonword took less than 1 second to be produced, but since there was no way to track when production started and ended in the inner speech condition, the entire 1-second period was kept. Therefore, the overt speech condition was composed of 6 repetitions of each nonword, that is 6\*20 trials of 1 second. The “inner speech” and « listening » conditions were similarly composed of 6\*20 trials of 1 second. Then, we visually inspected the EMG signals recorded during each trial and listened to the audio signal simultaneously recorded. In all conditions, any time a non-speech noise (such as coughing or yawning) was audible in a trial, the trial was discarded (i.e., we did not include this trial in the final analysis, for any of the recorded muscles). In the listening and overt speech conditions, if a burst of EMG activity was present after the relevant audio speech signal, then the trial was discarded. In the overt speech condition, if the participant started too early or too late and only part of the nonword was recorded in the audio signal, then the corresponding trial was discarded. In the inner speech and listening condition, if a large EMG burst of activity was present, potentially associated with irrelevant non-speech activity, we excluded the trial. The fact that the artefact rejection procedures slightly differ in the various conditions is not an issue, since we do not directly compare between conditions. Instead, we compare the EMG correlates of the two classes of nonwords within each condition. \- P.7. To clarify: am I right in understanding that the EMG activity in the 1 second following the presentation of the fixation dot was used as the dependent variable? If so, were participants instructed to produce the inner/overt speech as soon as possible after the fixation dot appeared? What was the delay (on average) between the dot and speech production in the overt speech condition? Do the authors assume that this delay was consistent between the inner and overt speech conditions? If so, can they justify this assumption? In both the “speech” and the “inner speech” conditions, the nonword was first visually presented on the screen (for 1 second) and then a fixation dot appeared (for 1 sec), indicating to the participants that they had to produce the nonword that they had just seen (i.e., participants had to produce the nonword in the 1 second time window during which the fixation dot was on the screen). In the “listening” condition however, the order of the fixation dot and the blank screen was reversed. First, the fixation dot appeared for 1 second (so that participants get ready for the task, which consists, in this condition, in listening to an overtly uttered nonword), followed by a fixation dot (for 1 second), during the presentation of which the nonword was delivered through the speakers. Regarding the instructions that were given to the participants, participants were asked to produce the nonword “when the fixation dot appears” in both the “speech” and “inner speech” conditions. Due to inattention or anticipation effects, we might expect that some participants may indeed sometimes “miss” the 1-second time window and produce speech either too late or too soon, respectively. We did not analyse the average “onset time” (i.e., the delay between the apparition of the fixation dot and the beginning of the movement) because these are not available in the inner speech condition (as the EMG signal does not show discernible bursts of activations). Although we generally assume that the dynamics of overt and covert speech may be roughly similar, it is indeed possible that production in the inner speech condition may start earlier (for instance because there is less articulatory constraints on the production of inner speech, that is, participants do not have to move their articulators to produce sounds). However, given that the purpose of this study is to highlight feature-specific EMG correlates (i.e., it aims at discriminating two classes of non-words), and because we do not have any reason to think that the “dynamics” of (overt and covert) speech production may be affected by this variable (i.e., the non-word class) in any way that could affect our main measure (i.e., the average EMG amplitude), we do not think this could be an issue in the interpretation of our data. We have therefore clarified this section in our revised manuscript and we have corrected a mistake in this paragraph (cf. lines 361-366). Moreover, we have added this sentence in the description of the periods of interest: The periods of interest in all the speech conditions consisted of the 1 second interval during which the participants either produced speech or listened to speech. It is possible that the nonword took less than 1 second to be produced, but since there was no way to track when production started and ended in the inner speech condition, the entire 1-second period was kept. \- It appears as though the ‘baseline’ period (i.e., against which the standardized scores were calculated) was calculated during the relaxation task that occurred prior to the talk itself, is that correct? I don’t understand the benefit of doing this as opposed to, say, calculating the baseline from the 1 second prior to the presentation of the fixation dot – can the authors clarify? We indeed standardised the EMG amplitude in each experimental condition by subtracting the baseline value to it. The main reason we chose to standardise EMG amplitude by baseline activity (rather than pre-stimulus activity) is because in both the “speech” and “inner speech” conditions, the second preceding the overt/inner production of the nonwords was the second during which the nonwords was displayed on the screen, and during which the participants were reading the nonword. Because (some) previous studies have shown an increase in the EMG activity of the lips during reading (e.g., Faaborg-Andersen et al., 1958; Sokolov, 1972), standardising the EMG amplitude during inner production by the activity during reading may obfuscate the activity that could be observed during the task. In other words, subtracting reading (during which inner speech may be produced) to inner speech may result in null activity. Moreover, several previous works have argued for the use of a relaxation period as a baseline, since mere resting periods may include some inner speech production (see e.g., Jacobson, 1931; Vanderwolf, 1998, for a review). To make this more explicit, we have modified the text as follows: Baseline EMG measurements were performed during the last minute of this relaxation session, resulting in 60s of EMG signal at baseline. By using this relaxation period as a baseline, we made sure that participants were all in a comparable relaxed state. In addition, several previous EMG studies have argued for the use of a relaxation period as a baseline, since mere resting periods may include some inner speech production (e.g., Jacobson, 1931; Vanderwolf, 1998, for a review). Faaborg-Andersen, Edfeldt, K. Å. W. & Nykøbing, F. (1958). Electromyography of Intrinsic and Extrinsic Laryngeal Muscles During Silent Speech: Correlation with Reading Activity: Preliminary Report, Acta Oto-Laryngologica, 49(1), 478-482. Jacobson, E. (1931). Electrical measurements of neuromuscular states during mental activities. VII. Imagination, recollection, and abstract thinking involving the speech musculature. American Journal of Physiology, 97, 200-209. Sokolov, A. (1972). Inner speech and thought. New York: Springer-Verlag. Vanderwolf, C.H. (1998). Brain, behavior, and mind: what do we know and what can we know? Neuroscience and Biobehavioral Review, 22,125-142. \- Figure 1: I think it would be appropriate to mention in the caption that the scales between the figure panels differ markedly – these differences are understandable (i.e., EMG activity in the overt speech condition will obviously be higher than the inner speech / listening conditions), but still appropriate to mention. We agree with the reviewer’s comment and we have added a word of caution about this issue in the caption of Figure 1. Reviewer \#2 The present article addresses an important but overlooked question about the external behavioural traces of inner speech. Its reliance on preregistered methods is really impressive, particularly in light of the complexity of possible analyses and the risk of data ‘fishing’. We have several main concerns. One is the theoretical framework in which the study has been conceived, which over-emphasises a problematic distinction between simulation and abstraction views, and pays insufficient attention to the key work in this area. Abstract. Second sentence should be revised; the authors do not need to describe the steps in detail. The last sentence should also be removed as this information should be submitted separately from the abstract (and also should be included in the Data Availability section at the end of the manuscript). We thank the reviewer for their suggestion. We have removed the second and last sentences of the abstract. 2\. A strong claim is made about the automatic elicitation of inner speech in reading. The picture is not quite so clear-cut; Russell Hurlburt for example has recently published a study showing very little inner speech during reading. We thank the reviewer for this important comment. We only used “silent reading” as a way to introduce the notion of inner speech in a practical way, as several studies reveal that reading may be accompanied by inner speech (e.g., Yao et al., 2011). However, we fully agree that not everyone produces inner speech during reading (and typically not expert readers, who may well be the typical readers of a scientific paper). We have therefore added a nuance to this introduction, and we inserted the reference suggested by the reviewer (Brouwers et al., 2018), as well as the discussion by Hurlburt (2018). We have also expanded our presentation of the occurrences of inner speech. The manuscript now reads: As you read these words, you may be experiencing the presence of a familiar speechlike companion. This internal speech production may accompany daily activities such as reading (see \[1–4\], but see \[5,6\]), writing (\[7\]), 4 memorising (\[8,9\]), future planning \[8\], problem solving \[9,10\] or musing (for reviews see \[11–14\]). 9\. The discussion of the featural properties of inner speech is a little thin and could be enhanced. We have provided a more thorough description of the featural properties of inner speech and have added references to the literature on this subject. The text is now modified to: Several studies using experience sampling or questionnaires (e.g.,\[14,15\]) have shown that by deliberately paying attention to this internal speech, one can examine its phenomenological properties such as identity (whose voice is it?) or other high-level characteristics (e.g., is it gendered?). Moreover, it is often possible to examine lower-level features like the tone of the inner speech, its pitch or its tempo. This set of basic observations leads to some important insights about the nature of inner speech. The simple fact that we can make sensory judgments about our inner speech tautologically reveals that inner speech is accompanied by sensory percepts (e.g., speech sounds, kinaesthetic feelings). Some of these introspective accounts have been examined, tested and complemented using empirical methods from cognitive neuroscience. As summarised in \[17\], behavioural and neuroimaging data reveal that some variants of inner speech are associated with auditory and/or somatosensory sensations that are reflected by auditory and/or somatosensory cortex activations. Visual representations may also be at play, typically for inner language in the deaf population. Inner verbalising therefore may involve the reception of imaginary multisensory signals. This leads to other fundamental questions: where do these percepts come from? Why do they sound and feel like the ones we experience when we actually (overtly) speak? 13\. The term ‘inner voice’ is problematic – does it mean ‘inner speech’? In our view they have different connotations (see Fernyhough’s recent book on inner speech, The Voices Within, for a discussion). We fully agree that “inner voice” and “inner speech” should not be equated. When we used the term of “inner voice”, we meant the audible voice signal that (may or may not) accompany inner verbal production. When we use the terms of “inner speech”, we mean the activity of talking silently to oneself (where “silently” is to be understood as from the perspective of an external observer, not from the perspective of the inner speaker, for which inner speech may (or may not) be accompanied by sounds). To avoid confusion, we have modified the text and only used the term voice to refer to acoustic voice quality. 18 – 28. This section is incomplete and does not include the essential Vygotskian theory that inner speech is internalised external speech. This is a serious omission. See Alderson-Day and Fernyhough (2015) for a full description of this theory. This is important not just because it is the most developed theory of inner speech, but because it is also highly relevant to the question of whether any behavioural traces of inner speech will be observed (see the discussion in Jones & Fernyhough, 2007, cited here as ref 4, and also extensive discussion in Fernyhough’s book). The fact that Vygotsky’s name is not mentioned in this manuscript is surprising. The introduction will need to be rewritten to represent these important views, which are significant for all of the main hypotheses in this study and for the interpretation of the findings. Furthermore, since Vygotsky’s is a developmental theory, it is very relevant to the question of different methods and findings for EMG studies of inner speech in children and adults (line 547). Giving some attention to Vygotsky’s theory would also make a lot of sense of the authors’ surprising findings on EMG activity during inner speech. See the discussions cited above about whether, on an internalisation view, you would expect any motoric trace of inner speech, if the latter is fully internalised and transformed outer speech – which also brings in the important issue of semantic and syntactic condensation. We thank the reviewer for this thoughtful comment. We are well aware of the Vygotskian developmental theory of inner speech and the various extensions that have been proposed in the recent years (see for instance a more exhaustive historical overview in the introduction of Nalborczyk, 2019). However, we do not think the Vygotskian theory of inner speech is directly relevant here, as it does not (contrary to the opposition between the motor simulation view and the abstraction view) lead to testable predictions regarding the central question of our experiment, that is, whether the two classes of non- words will be associated with distinct orofacial EMG correlates. In other words, it does not clearly predict whether inner speech will be associated with phonetically-specific “observable” (using EMG) traces in our situation. We agree, however, that the Vygotskian notion of condensation should be mentioned, as well as the view of Charles Fernyhough and his colleagues that inner speech can vary between two extremes, depending on cognitive demands and emotional conditions. We are aware that condensation is, as claimed in the ConDialInt model by Grandchamp et al. (2019), one of the important dimensions of inner speech. We have therefore substantially modified our text, into the following paragraph in the introduction: Two main classes of explanatory theories have been offered to answer these questions. A first class of theories, that derives from Vygotky’s views on language and thought, and that we describe as the abstraction view \[18\] suggest that inner speech is profoundly internalised, abbreviated and condensed in form. Vygotsky suggested that inner speech evolved from so- called egocentric speech (i.e., self-addressed overt speech or private speech), via a gradual process of internalisation during childhood \[19\]. According to him, the properties of speech are transformed during this internalisation, and inner speech cannot merely be described as a weakened form of overt speech (as claimed for instance by \[20\]). This has led some scholars to conceive of inner speech as predominantly pertaining to semantics, excluding any phonological, phonetic, articulatory or even auditory properties (e.g., \[19–22\]). The property of abbreviation and condensation is supported by several psycholinguistic experiments on the qualitative and quantitative differences between overt and covert speech, as concerns rate and error biases (e.g., \[20,21,23,24\], but see \[25\]). Such condensation implies that the auditory qualities mentioned above would only rarely be observed during introspection and would merely be the result of learned associations between abstract linguistic representations and auditory percepts. A second class of theories is described under the umbrella term of motor simulation view. These theories suggest that inner speech can be conceived as a kind of action on its own \[26,27\], produced in the same way as overt speech is, except that the last stage of articulatory execution is only simulated. Most theories under this view share the postulate that the speech motor system is involved (to some extent) during inner speech production and that the auditory and somatosensory consequences of the simulated articulatory movements constitute the inner speech percepts referred to in subjective studies. As explained in the ConDialInt model \[26\], these two views can be reconciled if various degrees of unfolding of inner speech are considered. Fully condensed forms of inner speech only involve semantics, and are deprived of the acoustic, phonological and syntactic qualities of overt speech. Expanded forms inner speech, on the other hand, presumably engage prosodic and morpho-syntactic formulation as well as phonological specification, articulatory simulation and the perception of an inner voice. Between the fully condensed abstract forms and the expanded articulation-ready form, it can be assumed that various semi- condensed forms may exist, with morphosyntactic properties and perhaps even phonological features, depending on the stage at which the speech production process is truncated. Such a view was also taken by \[29\] who has suggested that inner speech varies with cognitive demands and emotional conditions on a continuum between extremely condensed and expanded forms (see also \[11,28\]). Therefore, the two views of inner speech (abstraction vs. simulation) can be construed as descriptions of two opposite poles on the condensation dimension. On the most expanded side of the continuum, inner speech entails full phonetic specification and articulatory simulation. It might therefore be expected that speech motor activity could be detectable. If the motor simulation view is correct, then motor activity could be recorded during expanded forms of inner speech. If, on the other hand, the abstraction view applies to all forms of inner speech, then no motor activity should be present, even in phonologically- expanded forms. Nalborczyk, L. (2019). Understanding rumination as a form of inner speech. PhD thesis. Univ. Grenoble Alpes & Ghent University. Retrieved from <https://thesiscommons.org/p6dct/> 188\. The authors describe the sampling procedure in detail, but the Participants section would benefit from including more demographic information on the sample. We have added information about the sex and age of our participants (cf. lines 254 -256). 214\. Studies looking at EMG correlates of lip muscle activity often investigate the orbicularis oris superior muscle alongside the orbicularis oris inferior. The authors should discuss in more detail their justification for studying the OOI, but not OOS. We have included more details about this choice in the Methods section (cf. footnote n°1). Namely, we chose to record only the OOI in this study because we previously observed (e.g., in Nalborczyk et al., 2017; Rapin et al., 2013) that the OOS and OOI were strongly correlated but that the OOI activity was more affected by the experimental manipulation. Given that the information contained in these two muscles was therefore slightly redundant and given material limitations (i.e., we have a limited amount of sensors), we therefore opted for recording additional control muscles (i.e., COR and FRO) instead of the OOS. Nalborczyk, L., Perrone-Bertolotti, M., Baeyens, C., Grandchamp, R., Polosan, M., Spinelli, E.... Lœvenbruck, H. (2017). Orofacial electromyographic correlates of induced verbal rumination. Biological Psychology, 127, 53–63. <https://doi.org/10.1016/j.biopsycho.2017.04.013> Rapin, L., Dohen, M., Polosan, M., & Perrier, P. (2013). An EMG Study of the Lip Muscles During Covert Auditory Verbal Hallucinations in Schizophrenia. JSLHR, 56, 1882–1893. https:// [doi.org/10.1044/1092-](http://doi.org/10.1044/1092-) 4388(2013/12- 0210)and 243\. ‘during rumination’ – we think this is a mistake, and an alternative term should be used, e.g., covert speech condition. We thank the reviewer for spotting this mistake, we corrected it. 245\. Please give more detail here on how the stimuli were selected. Were the stimuli piloted before being selected? Stimuli were mostly selected based on theoretical constraints, with the aim of maximising the differences between the two classes of non-words (+ informal piloting on master students). We clarified this in the manuscript (in the “Linguistic material” section). Please note that these theoretical constraints translate well in practice as the two classes of non-words appear to form two clearly distinct clusters in the OOI-ZYG space in the overt speech condition (cf. leftmost panel of Figure 6 in the updated manuscript). 253\. More detail is needed on the instructions given in the three conditions. Hurlburt et al. (PLOS, 2016) have recently shown a difference in brain activations between elicited and spontaneous inner speech, questioning whether inner speech elicited by the method apparently used here can be taken as a reliable proxy for genuine, spontaneous inner speech. This needs some acknowledgement and discussion, especially as it addresses many of the issues raised in the Jones & Fernyhough (2007) paper. More explanation is needed on the instruction given to participants during training, e.g. how the visual cues were explained to indicate the start of the task. We have clarified the instructions given to the participants in this section (cf. pages 8-9). As made clearer in the introduction (see above), this study only pertains to elicited, wilful inner speech. 349\. Figures 6 & 7 could be moved to Supplementary Information. We agree that Figure 7 does not contain information that could not be understood in text format and we therefore removed it. However, we do see value in having Figure 6 in the main manuscript, as it allows the reader to immediately see how the EMG amplitude can be clustered in the 2D space formed by the OOI and the ZYG muscles. 583-628. The paragraph on stimuli selection should be revised. The detailed information should be moved to the Method section, where it would have been useful to have more information on how the stimuli were selected. More general reflection on this process should be included in the Discussion instead. Details concerning the selection of stimuli (of our study) are already provided in the “Linguistic materials” section of our manuscript. It is not clear from the reviewer’s comment what supplementary information should be added to this section. Please note that the details provided in the paragraph commented on by the reviewer (i.e., lines 583-628 of the previous submission) concern additional analyses performed on the data from Eskes et al. (2017). As this analysis is not central to the purpose of the current study and rather subserves a general reflection (cf. discussion in the subsequent paragraph), we feel this belongs to the discussion section. 10.1371/journal.pone.0233282.r003 Decision Letter 1 Sulpizio Simone Academic Editor 2020 Simone Sulpizio This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 4 May 2020 Can we decode phonetic features in inner speech using surface electromyography? PONE-D-19-27197R1 Dear Dr. Nalborczyk, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Simone Sulpizio Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: (No Response) Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: (No Response) Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: (No Response) Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: (No Response) Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) Reviewer \#2: The authors have thoroughly addressed the concerns raised in the original version of the manuscript. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No 10.1371/journal.pone.0233282.r004 Acceptance letter Sulpizio Simone Academic Editor 2020 Simone Sulpizio This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 May 2020 PONE-D-19-27197R1 Can we decode phonetic features in inner speech using surface electromyography? Dear Dr. Nalborczyk: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Simone Sulpizio Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Formation of blood vessels in the mammalian eye involves extensive tissue reorganization including regression of embryonic vascular structures. The developing murine eye is initially supplied with oxygen and nutrients by the hyaloid vasculature (HV), which is later replaced by the retinal vasculature. The HV is formed in the primitive vitreous body between embryonic days (E) 10.5 and E13.5. Concomitant with the postnatal (P) formation and maturation of the intraretinal vasculature, the HV degenerates via apoptosis, beginning on P4 and culminating on P7–8. On P10 most of the HV vessels have regressed and although complete regression of the hyaloid takes a few weeks, the vitreous body is completely avascular by P16. Vascularization of the retina is preceded by colonization with Pax2-positive astrocyte precursors that form a network, which becomes covered by endothelial cells. As they differentiate, these precursor cells begin to express GFAP as well, and change their morphology. Failure of the HV to regress, results in a congenital condition known as Persistent Fetal Vasculature Syndrome (PFVS), or persistent hyaloid vasculature (PHV). The consequences can be severe intraocular hemorrhage, cataract and retinal detachment due to forces exerted on the neural retina by contractile cells associated with the abnormal vessels in the vitreous. Although transgenic mouse models have shed some light on possible pathways the precise molecular and cellular mechanisms underlying the failure of the HV to regress are not yet fully understood. Disturbance of hyaloid vessel regression was reported in mice deficient in both Wnt7b-dependent and Wnt7b-independent Fzd4 signaling and formation of the deeper plexus is also disrupted in these mutant mice. Wnt7b is believed to be produced by the macrophages that play key roles in the regression of capillaries of the HV as indicated by the finding that in heterozygous BMP4 +/−, which lack macrophages in the vitreous, the HV persists. Moreover, Arf knockout mice and certain p53-null strains, both of which proteins are tumor suppressors, also display persistent HV, as do Ang-2 knockout mice. Platelet-derived growth factor (PDGF) is essential for proper development of the retina and has been associated with proliferative retinopathies. The PDGF family consists of four ligands, designated A, B, C and D that function as homodimers or in the case of AB, also as a heterodimer. PDGF-AA, -AB, -BB and -CC activate the PDGF receptor-α (PDGFRα) while PDGF-BB and –DD bind to PDGFRβ. In the normal eye, PDGF-A is expressed by both neurons and astrocytes and, together with PDGFRα, regulates the recruitment of astrocyte precursors to the retina and their subsequent development at this location. In this manner, interactions between PDGF-A and PDGFRα determine the number and distribution of astrocytes in the retina. Maintenance of the retinal vasculature depends on signaling by PDGF-B via the PDGFRβ. Pericytes express PDGFRβ, and their attachment to vessels is dependent on PDGF released from endothelial cells. Transgenic over-expression of PDGF-A in retinal ganglion cells (RGCs) results in a dose-dependent increase in the proliferation of GFAP-immunoreactive (+) cells in the retina, as well as inhibiting the migration and spread of these cells across the retina, thus producing a thick carpet of GFAP+ cells close to the exit of the optic nerve. Furthermore, over-expression of PDGF-B under control of the rhodopsin promoter also enhances astrocyte proliferation in the retina. In this case, folding of the retina occurs, a phenomenon also observed in MBP- PDGF-B transgenic mice that, in addition, exhibit disorganization of capillaries in the retina. HV cells express PDGFRβ but not PDGFα. In the case of Arf-deficient mice with a persistent HV, it has been proposed that inadequate repression of the PDGFRβ promoter in mural cells stimulates their proliferation at the expense of differentiation. This proposal implies that interactions between PDGF and its receptors are important not only for the maintenance, but also for regression of the HV. To characterize the effects of over-expression of PDGF in neural progenitor cells on the developing retina, we developed transgenic mice that express PDGF-B under the control of a nestin enhancer element, that is active in progenitor cells during development, but not in adult neural stem or progenitor cells. These nes/tk-pdgfb-lacZ mice are viable and fertile but develop enlarged lateral ventricles and mild behavioral abnormalities. In the present investigation we report that the eyes of these mice display severe defects in retinal development. # Materials and Methods ## Animals and genotyping Our experimental model, nes/tk-PdgfB-lacZ mice over-express a 730-bp PDGF-B cDNA of human origin under control of the second intron enhancer of the human nestin gene (nes) joined to a basic thymidine kinase promoter (tk). An internal site for ribosomal entry (IRES) and a LacZ reporter gene were cloned downstream of the PGDF-B gene. The nes element enhances expression in nestin-expressing cells during embryonic development, but not in the adult cells that express this protein. The generation and characterization of the nes/tk-PdgfB-lacZ strain have been described elsewhere and all animals used here belongs to the \#310 line described in this earlier report. A back-cross to C57Bl/6 mice for 5 generations was performed prior to the onset of the experiment. All animal protocols were pre-approved by the local ethics committee for laboratory animals. Embryonic day 0.5 (E0.5) was defined on the basis of the presence of a copulatory plug in the morning and postnatal day 0 (P0) was defined as the morning of the day of birth. Animals were sacrificed by CO₂ gas, cervical dislocation (pregnant dams and animals \>15 days of age) or by decapitation (embryos and pups \<14 days old). The transgenic mice were identified by PCR analysis as described elsewhere. In brief, genomic DNA extracted from tail biopsies was subjected to PCR (95°C for 10 min, 40 cycles×\[94°C for 30 sec, 55°C for 30 sec, 72°C for 1 min\], 72°C for 7 min) utilizing: 5′-TGCTGCTACCTGCGTCTG as the forward primer and 5′-TTCTTCCGCACAATCTCG as the reverse primer. ## Tissue preparation and immunohistochemical analysis For our purposes, we chose to routinely analyze the eyes and retinas of mice at the following stages of development in utero and post-natal life: E13.5, E15.5, E17.5, P1, P5, P10, P15, P20, P30, P60 and P90. On occasion, other time-points were examined as well. To obtain cryosections, eyeballs were dissected out, fixed in 4% PFA (4°C for 15–20 minutes), washed with phosphate-buffered saline (PBS) (10 minutes at 4°C), cryoprotected in 30% sucrose (at 4°C for 3–4 hours), embedded in OCT (Sakura, Alphen aan den Rijn, The Netherlands) and frozen. 10 µm sections were cut, collected onto SuperFrost Plus objective glasses (Menzel-Gläser, Braunschwieg, Germany) and subsequently stored at −20°C or −80°C until being analyzed. Specimens were either stained with hematoxylin and eosin (H&E) or processed for immunostaining. For immunohistochemical staining, these sections were first re- hydrated in PBS for 15 minutes and then blocked and permeabilized for 30 minutes in blocking solution (PBS, 1% FCS and 0.1% Triton-X supplemented with 0.02% thimerosal for purposes of preservation). Thereafter, the sections were incubated with primary antibodies in blocking solution overnight at 4°C and subsequently with secondary antibodies in the same solution for 2–4 hours at room temperature. Following each such incubation with antibodies, the samples were rinsed 3 times with PBS for 5 minutes each. Finally, the sections were then mounted and coverslipped using VectaShield HardSet (with or without DAPI, H-1400/H-1500, Immunkemi, Järfälla, Sweden). In the case of retinal flatmounts, eyeballs were fixed in 4% PFA (at 4°C for 3 hours) prior to dissecting out the retinas, which were then washed for 10 minutes in PBS and subsequently stored in blocking solution at 4°C until being analyzed. For staining, the retinas were incubated successively with primary and secondary antibodies overnight at 4°C, with washing in PBS (4–6 times, 1 hr each time after each such incubation). After the final wash, perpendicular cuts were made at the peripheral margins of the retina to facilitate flattening. Finally, the retinas were mounted onto SuperFrost Plus objective glasses and coverslipped as described above. ## Antibodies and image acquisition A list of the primary antibodies employed is presented in. The secondary antibodies (diluted 1∶200) were obtained from Vector Laboratories, Jackson Immunoresearch Laboratories and Molecular Probes. Samples were examined under a Zeiss Axioplan2 microscope and images acquired using the Axiovision software. These images have been formatted, resized, combined, enhanced and arranged for publication in Axiovision or Adobe Photoshop. ## Cell counting in P10 retinas For quantification of the proportions of different cell populations in the INL and GCL, P10 retinas were co-stained for Pax6 and Isl1 and the resulting fractions of Isl1+ only, Pax6+ only and Isl1+/Pax6 double positive cells were manually counted in images taken using the 20× or 10× objective. Four images from two individuals were counted and the fractions of cells single or double positive for the markers were expressed as percentages of the total number of stained cells per retina image. Student's t-test was performed to test for significant differences between wild-type and transgenic mice, where p\<0.05 was considered significant. ## Staining for X-Gal In order to monitor the localization and temporal pattern of the activity of the transgene via the LacZ reporter, cryosections from E13.5- P90 animals (prepared as described above) were stained for X-gal. After re-hydration in PBS for 10 minutes and post-fixation in PBS containing 1% formaldehyde, 0.2% glutaraldehyde, 0.02% NP40, 2 mM MgCl₂, and 5 mM EGTA, pH 8.0, for 15 minutes at room temperature, these sections were washed 3×5 minutes in X-Gal wash buffer (PBS containing 2 mM MgCl₂, 0.01% Na-deoxycholate and 0.02% NP40). Visualization of X-galactosidase activity was carried out by incubation at 37°C for 4 hours in X-Gal wash buffer supplemented with 1 mg X-Gal/ml, 5 mM K₃Fe(CN)₆ and 5 mM K₄Fe(CN)₆. The sections were washed 2×5 minutes in X-Gal wash buffer, then washed once in PBS and finally mounted in Entellan (VWR, Gothenburg, Sweden). ## Determination of proliferation utilizing EdU To examine proliferation during the earliest phases of transgene activation, pregnant dams carrying E18.5 pups were injected intraperitoneally with 200 µg of the nucleotide analogue 5-ethynyl-2′-deoxyuridine (EdU) and the pups analyzed 2 days later, at P1. The eyes were fixed, frozen and sectioned as described above and the Click iT EdU imaging kit (C10083/C10337, Invitrogen, Paisley, UK) employed in accordance with the manufacturer's protocol to visualize cells that had incorporated EdU during S-phase. ## Measurement of intraocular pressure The intraocular pressure (IOP) in 3–4 months-old mice was measured using the TonoLab rodent tonometer (Icare Finland Oy, Espoo, Finland) in accordance with the manufacturer's instructions, following sedation by i.p. injection of Rompun (10 mg/kg, Bayer Animal Health, Copenhagen, Denmark) and Ketalar (60 mg/kg, Pfizer, Sollentuna, Sweden). For each mouse the IOP in both eyes (with the exception three transgenic mice in which one eye was too small to allow reliable measurements) was determined consecutively 6 times and the means for the transgenic (9 animals, 15 eyes measured) and wild-type mice (11 animals, 22 eyes measured) analyzed for statistically significant differences utilizing the Mann- Whitney test. ## Administration of STI571 STI571 was administered (100 µg/g body weight, diluted in dH₂O) by gavage once daily during three different time intervals: the first group received this inhibitor between E17.5 and P0 or P1 (samples taken for analysis at P1 or P2, respectively), the second group from P0–P4 (samples being taken for analysis on P5), and the third group between P7 and P14 (samples taken on P15). The body weights and general health of the mice was monitored daily. The eyes were fixed, frozen and sectioned as described above. ## Statistical analyses The data were analyzed in Excel and graphs created in GraphPad Prism (v3.02, GraphPad Software Inc., La Jolla, USA). # Results ## Morphological alterations in eyes of mice over-expressing PDGF-B In the transgenic mice employed, PDGF-B is over-expressed in nestin-expressing cells of the CNS during development. The macroscopic appearance of the nes/tk- PdgfB-lacZ mice and their wild type littermates was generally similar, except that the transgenic mice displayed enlarged lateral ventricles, smaller eyes and had abnormal behavior as described previously. At the time of eye opening, the transgenic mice displayed uni- or bilateral reduction in the size of their eyes and/or irises and we frequently observed blood inside the eyeball in transgenic mice. Retinal folding with frequent attachment to the lens was observed in all the transgenic animals after birth. With increasing age of the transgenic mice, the retinas degenerated progressively and after 2–3 months they were markedly thinner than those of the wild-type animals (–D). Around one year of age, the retinas had degenerated completely and consisted of only a thin sheet of cells. Histological examination revealed a severe distortion of the transgenic retinas, with disorganization of retinal lamination discernable already in the earliest days of post-natal life. One prominent feature of these retinas was rosette-like structures in the outer nuclear layer (ONL) containing the nuclei of photoreceptors (arrow), which appeared between P5 and P10 The other nuclear layers of transgenic retina; the inner nuclear layer (INL) and the ganglion cell layer (GCL) were also distorted, with local variations in thickness and folding. In general, the nuclear and plexiform layers could be distinguished from one another although with a varying degree of distortion. Among the general laminar disarray, local regions in some retinas appeared normal. The progressing deterioration of the neural retina was visualized by H&E staining. ## The temporal pattern of transgene expression in the developing retina We have previously shown that X-gal staining in this mouse strain faithfully depicts transgenic PDGF-B expression. In order to monitor changes in the expression of the transgene with time, the LacZ reporter gene and staining of X-gal in sectioned tissue were therefore employed. In contrast to the brain and spinal cord, where X-gal staining is most intense between E10.5 and E14.5 retinal expression occurred later during development. In this tissue, staining first appeared as a faint signal distributed throughout the neuroblastic layer (NBL) and in some cells within the putative ganglion cell layer (GCL) on E17.5. On P1 the pattern of staining was similar, but much darker and between P5 and P15, the robust staining persisted, albeit now restricted to a band of cells located in the middle of the INL (note the localization of X-Gal staining in the ‘near-normal’ region shown). By P20, the intensity of the staining, although still detectable, had attenuated significantly (arrowheads) and after P30 no X-Gal staining could be seen ( and not shown). We conclude that transgene expression in the retina commenced on E17.5 and attained its maximal level between P1 and P15, before declining on P20. ## Neural development and cell death in the retinas of mice over-expressing PDGF-B Since the onset of expression of the transgene in the nes/tk-PdgfB-lacZ retina is concurrent with retinal neurogenesis, lamination and developmental cell death we examined whether over-expression of PDGF-B affected these processes. First, we analyzed the generation of the different types of retinal neurons and their laminar positions by staining with antibodies for the five major markers: Pax6 (expressed by horizontal cells, HCs, amacrine cells, ACs, and retinal ganglion cells, RGCs), Isl1 (bipolar cells, BPs, RGCs), calbindin (HCs and certain ACs), neurofilament 165 kD (neurites in the plexiform layers) and the photoreceptor (PR) markers recoverin and rhodopsin. Co-labeling with Pax6 and Isl1 revealed that over all, the different cell types were present in the retina, and despite the apparent laminar distortion at the correct positions relative to one another. To determine if there were detectable differences in cell number, we next analyzed markers of proliferation and apoptosis. We first used an antibody directed against cleaved caspase 3, and found that in P5 eyes the number of labeled cells in the transgenic retinas was increased by 10% compared to corresponding normal tissue. We also analyzed the effects of over-expression of PDGF-B on the proliferation by using the proliferation marker Ki67 (not shown) and incorporation of the nucleotide analogue EdU. Pregnant mice were injected with EdU on E18.5 and analysis was performed two days later. No differences between transgenic and wild-type mice were discernable at the ages tested. Finally, by counting cells expressing Isl1 and/or Pax6 in P10 retinas we estimated that here was a slight but significant decrease in the Isl1-/Pax6+ cell population. ## Altered photoreceptor development and maturation by over-expression of PDGF-B Analysis of rhodopsin and recoverin immunoreactivity at P5 revealed that transgenic retinas stained differently from the wild-type. Rhodopsin staining appeared displaced relative to its normal position. In the adult, the staining of the inner and outer segments of the photoreceptors in nes/tk-PdgfB-lacZ mice was irregular and frequently restricted to isolated patches on the outer side of the outer nuclear layer (ONL) or located within the lumen of the rosettes Horizontal cells (HC) make synapses with photoreceptors within the outer plexiform layer (OPL) of the retina. Staining for the HC markers NF165 kD and calbindin revealed that the appearance of cells and neurites in transgenic retinas were normal up until P5 (not shown), but exhibited abnormalities thereafter. On P10, a time when many synapses are formed in the mouse OPL, the neurites of transgenic HC developed ectopic extensions outside of the OPL (not shown). These extensions reached into and across the ONL and on P30 they surrounded photoreceptor rosettes. These ectopic HC neurites persisted, showing little or no sign of regression even on P60. ## Glial activation in the retinas of mice over-expressing PDGF-B Müller glia and astrocytes are the major two types of glia cells in the murine retina. Glial fibrillary acidic protein (GFAP) is widely used as a marker for the latter, whereas nestin is commonly employed to identify activated Müller glia cells. Ischemia or other retinal damage triggers the expression of both nestin and GFAP in Müller cells. Since antibodies towards Pax2 stain the precursor cells to retinal astrocytes, a combination of antibodies directed against nestin, GFAP and Pax2 were used to monitor glial development in the transgenic retina. Although the expression of nestin by Müller glia cells in transgenic retinas appeared relatively normal during early postnatal development (not shown), on P5 GFAP-positive cells were more abundant in transgenic retinas than in wild-type tissues. In the transgenic retinas Pax2+ and GFAP+ cells did not spread out evenly on the nerve fiber layer between E17 and P10 as wild type astrocytes do. Instead, on P1 and P5, astrocyte precursors staining positively for Pax2 remained at the vitreal side, and did not populate the retina. Although the transgenic retina contained scarce GFAP-expressing cells with normal morphology (insert), these cells were not evenly distributed in contrast to the wild type retina. Signs of glial activation in transgenic retinas became more manifest with time: elevated expression of GFAP in the inner parts of the retina and of nestin on P30 indicates the presence of either ischemia and/or mechanical forces exerting traction on the retina. At the age of P60, the glial activation was exacerbated. ## Over-expression of PDGF resulted in failure to vascularize the retina During postnatal development in the mouse, retinal vascularization takes place concomitantly with regression of the hyaloid vasculature, which initially supplies the eye with oxygen. Upon staining cross-sections and flatmounts with antibodies against CD31 (an endothelial cell marker) and NG2 (a pericyte marker) on P5 virtually no CD31+ or NG2+ cells were detected in the retinas of transgenic mice, indicating complete failure of retinal vascularization and angiogenesis. In their wild-type littermates newly formed blood vessels had covered more than 70% of the retina by this time (arrowhead point at the border of CD31+ cell migration). In contrast, remnants of the embryonic vasculature was more abundant in the transgenic mice. In addition, GFAP staining showed that the transgenic astrocyte precursors did not spread out to populate the retina in a normal fashion. Consequently, there was a delay in population of the retina by vascular cells, and the network of vessels in the inner retina of transgenic mice that finally formed was abnormal. ## Abnormal development of the retinal capillary network At P20, the transgenic retina was vascularized, although this took place in an irregular fashion. At P60, a notable feature of the transgenic vascular network was the absence of large trunk vessels (compare), a finding confirmed by the low level or absence of α-SMA immunoreactivity, characteristic of smooth muscle cells on arteries. The NG2 and CD31 staining patterns revealed that at P60, vessels resembling capillaries of varying diameters and thickness grew irregularly into the retina, populating the different retinal layers in an apparently random manner. Staining of retinal whole mounts for CD31 and NG2 at P9 to visualize interactions between endothelial cells and mural cells show that there was some pericyte coverage of transgenic vessels albeit irregular. Some NG2+ cells were attached to CD31+ cells while others were found separated from the endothelium. This demonstrated that over-expression of PDGF did not completely abolish the interactions between endothelial and mural cells. ## The transgenic mice exhibited reduced intraocular pressure We observed that blood and other fluid often (∼80% of the mice examined) oozed out of nes/tk-PdgfB-lacZ eyes following puncture for retinal dissections, indicating that the retinal blood vessels were leaky and perhaps that the circulation of ocular fluids in these mice is abnormal. The apparent lack of arteries also suggested that the ability to stabilize circulatory pressure in the eyes should be reduced. When the intraocular pressure in 3–4 month-old transgenic and wild-type mice was determined using a rodent tonometer, the transgenic animals displayed significantly lower pressure (9.55 versus 12.4 mmHg for the wild-type, p = 0.0006, Mann Whitney test). ## Partial restoration of astrocyte precursor migration by the tyrosine kinase inhibitor STI571 In an attempt to pinpoint the period during which over-expression of PDGF affected the formation of various cells in the retina and possibly even reverse the defects observed, the mice were administered the tyrosine kinase inhibitor STI571 (Imatinib, Glivec or Gleevec) systemically by oral gavage once daily during three different intervals: from E17.5 until P0 or P1 with samples taken for analysis on P1 or P2, respectively, from P0–P4 (samples taken for analysis on P5), or between P7 and 14 (samples taken on P15). STI571 blocks signaling via PDGF receptors, and in addition, inhibits other tyrosine kinases, such as c-Kit and Bcr-Abl. We found that E17.5 was the earliest time point at which pregnant females could be treated with STI571 since earlier treatment resulted in the premature birth of non-viable pups. STI571 treatment during the late embryonic period partly restored colonization of the transgenic retina by astrocyte precursors. In untreated transgenic P2 mice the Pax-2 positive and CD31+ cells exhibited the characteristic compact and irregular cell mass that associates with the HYV (see also). However, in STI571 treated pups Pax2-positive astrocyte precursor cells had not been trapped by the hyaloid to the same extent, and were more spread apart (compare 9A with 9B and 9C). Concomitantly, CD31+ cells also had an organization that was less dense than in non-treated mice (compare 9A′ with 9B′ and 9C′). Suppression of PDGF signaling in wild-type by STI571 during the early postnatal period (P0–P4) inhibited colonization of the retina by vessels (–B). However, no difference between treated and un-treated transgenic mice with respect to CD31 staining was detected (–D). P7–P14 was the latest time window during which phenotype rescue was attempted. The hyaloid in un-treated nes/tk-PdgfB-lacZ mice was very often attached to both the lens and retina, making it difficult to dissect out these structures damaging the retina. Moreover, prominent folding of the retina often made flatmounting challenging. In contrast, on P15, the lens of transgenic mice treated with STI571 was easier to dissect out, and indeed, actually fell out not being stuck in the hyaloid, indicating partial reversal of these effects. In the case of both transgenic and wild-type mice, treatment with this inhibitor resulted in less coverage of deep CD31+ vessels by NG2 positive cells. # Discussion We report here that in mice expressing elevated levels of PDGF-B in nestin positive cells of the retina during development, astrocytes fail to populate the retina and vascular progenitors do not colonize this tissue to produce a network of vessels. Interestingly, with the exception of their eyes, nes/tk-PdgfB-lacZ mice do not exhibit any profound alteration in phenotype. The transgenic eyes were smaller than normal with an iris of a reduced diameter and with frequent occlusions. The transgenic retinas had folds and adhered to the back of the lens, forming a retrolental mass that made dissection difficult. This morphology was accompanied by frequent bleeding and by the age of one year the transgenic retina had deteriorated completely. Many of these defects in the eye are likely to be associated with abnormal traction forces within the structures that eventually also leads to retinal detachment. Patients with persistent fetal vasculature syndrome (PFVS) display similar problems with a large individual variability that was also seen among the nes/tk-PdgfB-lacZ mice. Many of the transgenic mice developed uni-lateral microphtalmia (around 40–50% pups per litter) and/or a cataract. These abnormalities in retinal structure of the transgenic mice are similar to PFVS, which results from failure to degrade the hyaloid vasculature. The crucial role of PDGF signaling in connection with the normal formation of retinal blood vessels is well-established. The functions of PDGFRα-signaling include regulation of the migration and proliferation of astrocyte precursor cells, while signaling by PDGFR-β recruits mural cells, which allows for their proper attachment to the endothelial cells of the blood vessels. Seo and coworkers have proposed that gain-of-function mutations in PDGF-B exert more profound effects on retinal development than mutations in PDGF-A, due to the ability of the former to activate both α and β receptors, present on different types of cells. This suggestion was confirmed in studies involving overexpression of different isoforms of PDGF in photoreceptors and further supported by the findings of Vinores et al. In our previous study involving transgenic expression of PDGF-B under control of the myelin basic protein promoter a disorganized neural retina with an under-developed capillary network was observed. The transgenic mice employed here display more severe alterations in eye phenotype, probably because the transgene is expressed more widely, both spatially and during the period of development examined. By binding to both the α and β receptors the B isoform of PDGF could affect the hyaloid as well as astrocytes and pericytes, cells that form and provide support for the vasculature during a critical period of retinal development. The defects in lamination discernible shortly after birth are consistent with the onset of transgene expression on E17.5 with peak expression on P1. Since no major effects on the representation of retinal cells, on their relative spatial orientation or on their proliferation were seen, despite the overall disorderly structure, we conclude that the transgene expression did not affect the general formation, differentiation or migration of retinal progenitors. The elevation of the number of caspase positive cells observed on P5 occurred at a time when this tissue normally undergoes developmental cell death. Cell death is part of the normal development of a functional nervous system and half of all postmitotic retinal ganglion cells die during the 2 first post-natal weeks in mice. This period of cell death is associated with trophic interactions within the retina and with the central targets for the retinal ganglion cells, and the extent of death is determined in part by what connections the cells can establish. The fact that we see more apoptosis in the transgenic mice during this period suggests that the increased cell death is associated to the structural changes in the retina and is not a direct effect of the transgene. Moreover, the deterioration of the retina, as seen after one year, is probably not related to this increased apoptosis but rather related to the malfunctioned blood supply to the eye. The inner and outer segments of the photoreceptor layer in the transgenic animals did not develop normally. The rhodopsin localized in patches or inside the rosette formations and the horizontal cells, which normally form synapses with photoreceptors, developed ectopic neurites that extended outside the outer plexiform layer and into rosettes. This pattern persisted in the adult retina for several months indicating that these cells may have formed synapses with the PR, despite their anomalous location. The up-regulation of nestin and GFAP in Müller glia cells indicated that the transgenic retina experienced a gliotic response that is likely to be triggered by ischemia due to the missing retinal vasculature. Mechanical traction may be an alternative explanation since Müller glia is known to exert tractional forces when stimulated by PDGF in vitro. It seems likely that traction also occurred in the transgenic retinas. During normal retinal vascularization, astrocyte precursor cells migrate from the optic nerve to populate the inner portion of the retina, where they form a scaffold on which vascular progenitor cells can migrate and form vessels. In the nes/tk-PdgfB-lacZ mice most of the Pax2/GFAP positive cells remained associated with the vitreal side of the retina, and consequently, no astrocyte network on which vessels could expand was laid down. The lack of proper formation of new vessels was also apparent from the staining of CD31 and NG2 in both flat-mounts and cross-sections of the nes/tk-PdgfB-lacZ retina. This staining demonstrated defective vascularization, with few capillary-like structures, in a seemingly random fashion and devoid of large trunk vessels. Our results suggest that over-expression of PDGF-B delayed regression of the hyaloid, and prevented APCs from migrating and spreading across the retina, which in turn prevented endothelial and mural cells from populating the retina to form a normal vascular network. Endothelial cells recruit pericytes by secreting PDGF-B and it is well established that pericytes require PDGF-B to remain attached to the vessel wall. The opposite situation, i.e. an excess of PDGF-B is apparently not as detrimental, since many transgenic pericytes remained attached to the endothelial cells. Therefore, the abnormal vessel properties cannot mainly be related to altered coverage by mural cells. Intraocular pressure (IOP) in mice varies between strains, underscoring the importance of using non-transgenic littermates as controls when congenic strains are not available. The mice examined here had been back-crossed with C57Bl/6 mice for 5 generations, but the remaining contribution of the CBA genome is unknown. Nonetheless, the wild-type control mice, which were always siblings, exhibited an IOP close to the reference value for C57Bl/6. The attenuated IOP in our transgenic mice is consistent with our hypothesis that the perfusion of the eye is severely affected in the transgenic animals. The lack of proper retinal vessels prevented these mice from controlling the ocular circulation. Furthermore, traction on the ciliary body can lead to acute and chronic hypotony, followed by retinal detachment, as has been reported in patients with PFVS. The effect of STI571 on the retinal phenotype of the transgenic mice depended on the time when this inhibitor was administrated. Administration prior to E17.5 caused abortion, demonstrating that PDGF signaling, and/or other tyrosine kinases sensitive to STI571 are critical for embryo survival. From E17.5 to the time of birth, exposure to STI571 did not affect the birth of live offspring. Since this time period coincided with the onset and early phase of transgene expression in the eye, the earliest effects of transgenic PDGF could be monitored. The timing of inhibition was of importance for APC maturation and spread. APCs were less compacted upon treatment with STI751, which apparently allowed CD31-positive cells to follow, since these cells were also less tightly packed in the presence of the inhibitor. However, the retinal vascularization was not greatly improved despite the better APC colonization. The observation that vessel distribution was not influenced at all by administration of STI571 from birth to P4 suggests that at this point it is too late to reverse the abnormal development of the vascular network. In contrast, if STI571 was administered to the mice during the second week of postnatal life, the hyaloid regressed in part. Thus, even though STI571 blocks all PDGF signaling, not only that originating from the transgene, we could discern partial rescue of the phenotype with this inhibitor. The present investigation reveals that timely regression of the hyaloid vasculature and development of the adult retinal vascularization are prevented by over-expression of PDGF-B in nestin expressing cells during development. Failure of astrocyte precursors to form a scaffold precluded retinal vessel formation and gave rise to a defective retina that deteriorated with time. We also showed these mice to be useful for testing pharmacological intervention by the ability of a small-molecular inhibitor to partially restore retinal vascularization. # Supporting Information We thank the personnel at the Rudbeck Animal Facility for excellent animal care, and Tobias Bergström for help with genotyping. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: PDE FH KFN. Performed the experiments: PDE MN. Analyzed the data: PDE FH MVS KFN. Contributed reagents/materials/analysis tools: KFN FH MVS. Wrote the paper: PDE FH KFN.

# Introduction Proximal femoral fractures are one of the most common fractures treated surgically. The common denominator in surgical technique of fixation of these fractures, both intracapsular and extracapsular, is the placement of a large diameter cannulated screw or blade in the cancellous bone of the femoral head. Their correct placement is important as misplaced implants tend to destabilize, especially in intertrochanteric fractures, less so the in treatment of femoral neck fractures with cannulated screws. The placement of the implant into the femoral head is ensured by the insertion of a guide wire under fluoroscopic guidance. Misplaced wires are corrected until satisfactory position is achieved, and the final implant is inserted. Every insertion of the guide wire creates a canal in the bone, often in the direct proximity to the final screw, what raises concerns about weakening of the bone. It has been well recognized that screw holes in cortical bone are sites of increased fracture incidence, what has been proved biomechanically for predominantly cortical bone, but not for cancellous bone. This study was designed to compare the relative compressive strength of paired cancellous bone specimens retrieved from the femoral head with and without guide wire hole. A comparison of purely cancellous bone specimens has not been investigated before in similar intact and drilled samples. The aim was to recreate the compression of the cancellous bone that occurs in the femoral head between the lag screw and the dense subchondral bone. The holes simulate additional drill holes performed during guide wire positioning or small bone cysts around the lag screw. Moreover, the comparison of compressive strength was made between human and artificial cancellous bone, as synthetic bone models are commonly used to evaluate implant stability. Our hypothesis was that guide wire drill holes decrease the strength of cancellous bone samples. Moreover, the artificial bone samples should have similar force at failure, stiffness as well as the decrease of those parameters after drilling as the human samples. # Materials and methods ## Materials Femoral heads were retrieved from 38 consecutive patients (30 female, 8 male) who sustained femoral neck fractures and underwent a standard bipolar hip arthroplasty in the course of their treatment. They were operated on within 72 hours from injury. Mean age was 80 (62–96) years old (SD = 10.4). Femoral neck fractures sustained in males above 65 years of age and in females above 50 years of age meet the diagnostic criteria of osteoporosis, according to national guidelines. The femoral head retrieval did not affect decision making in their treatment and did not change the course of the surgical procedure. All patients gave their consent to participate in the study. Femoral heads were fresh frozen right after surgery. Patients with suspected metastatic neoplasms, arthrosis, rheumatoid arthritis or any other condition that could affect bone quality were excluded from the study. This study was approved by the Independent Bioethics Committee at the Medical University of Gdansk, Poland (issued 21.05.2018, NKBBN/228/2018). For synthetic cancellous bone samples, Synthetic osteoporotic left femora (LD2350.01, cortical low density/soft cancellous bone, Synbone AG, Neugutstrasse 4, 7208 Malans, Switzerland) were used. ## Sample preparation The femoral heads were cut into cuboids using custom made templates with a surgical oscillating saw. The cuboid was then cut in half along the sagittal plane using templates, leaving two matching samples, an anterior and a posterior, from the same femoral head. The size of the cuboid was described as follows: size 1 is the size of the base of the cuboid resulting from the cut in the sagittal plane, size 2 is the length of the base in the coronal plane, and size 2 is the longest side of the cuboid that was directed supero-inferiorly with regard to the long axis of the patients’ body (the sample height) to ensure that the force applied to the sample would have the same vector as hip contact forces for a standing patient. One of the matching samples was chosen at random from each pair. It was drilled with an original 3.2 mm partially threaded guide wire for cancellous screws used in femoral neck osteosynthesis (Ansis III Threaded Guide Wire, Stryker GmbH). show the relative size, shape and direction of the cut bone samples and the direction of the drill hole was directed as it would have been during actual surgery. The synthetic bone samples were prepared in a similar manner from the foam that forms the cancellous bone in femoral head. After sample preparation a set of 33 pairs of samples of human and 9 pairs of synthetic bone were available for further investigation in this study. The size of the samples as well as their weight were then recorded and the apparent density was calculated from the volume and weight of the sample. The mean percentage loss of cross-sectional area due to drilling was estimated as follows: Mean loss = (size 1 x 3.2 mm)/(size 2 x size 3) x 100% The samples were compared statistically (Wilcoxon signed-rank test for comparison of the samples) to ensure that the evaluated physical parameters were similar in both sample groups. Before testing, human bones were bathed in 37°C saline for 10 minutes after cutting to achieve body temperature. The samples were not cleared of the bone marrow to make sure that the bone resembles as closely as possible its intracorporeal state. ## Mechanical testing The uniaxial compression tests of bone blocks were carried out at room temperature using the Zwick-Roell Z020 testing machine. The grips separation and actual force were recorded during the test. In the preloading protocol, 3 loading-unloading cycles of the strain range 0–0.1% were applied. The compression loading with the strain rate of 0.5 1/s was continued until the failure of the specimen. A small toe effect being a consequence of adapting of the cuboid samples between testing machine grips and adaptation of the testing machine at the beginning of the experiment, was ignored in further analysis. The results were not recalculated to get their representation in terms of stress and strain relations, as the half of samples were drilled and therefore area of their cross section was not constant along the whole specimen length. The whole bone blocks of exactly same sizes in paired samples, and of slightly differing sizes between pairs, were compared. Consequently, the only reliable measures taken for further analysis were the force at failure and stiffness of the sample. The highest load level obtained before fracture of the specimen was defined as the force at failure. Stiffness was calculated as the slope of the linear part of the force-displacement curve with skipping the very initial part of the curve, where the toe effect occurred. The percentage difference between the mean values *X* of the obtained characteristics for the drilled and not-drilled samples was calculated as follows: *Difference = (X*_{*NOT DRILLED*}*−X*_*DRILLED**) / X*_{*NOT DRILLED*} × 100% Statistical analysis was performed using Statistica PL v. 13.3 software. The Wilcoxon Signed-Rank Test was used for statistical comparison. Results were considered statistically significant with *p* \< 0.05. # Results The comparison of physical parameters of bone samples are shown in Tables and. There were no significant differences regarding size of the sample, weight, volume and density. The estimated mean loss of sample cross section area due to drilling was 25% (median 24%, SD = 5.3) for human bone and 24% (median 24%, SD = 0.73) for synthetic bone. The strength characteristics of the drilled and not drilled human bone samples are presented in. The results for synthetic bone are presented in. The differences between the force at failure were statistically significant, but the comparison of sample stiffness were not. # Discussion The successful treatment of proximal femoral fractures depends, among other things, on correct positioning of the implant. In proximal femoral fractures, optimal placement of the guide wire for the lag screw is mandatory to achieve good results. Often, multiple trials are needed to introduce it correctly. It seems safe to assume that multiple drill holes around the final screw or blade of an implant may weaken the bone and lead to the more common screw cutout than when the bone around the implant is intact. This study has shown human cancellous bone weakening after drilling of a single hole in a bone specimen. Drilling eliminates a portion of the cancellous bone from the transmission of the force applied to the bone and therefore weakens it. The mean and median force at failure are similar to the loss of cross section area of the sample due to drilling. This suggests that the loss of strength of cancellous bone due to drilling may be roughly proportional to the loss of cross section area. The major drawback of this study however is that the effect of the drill hole would be dependent on the size of the bone specimen, since the diameter of the guide wire is constant. It is safe to presume that the smaller the specimen, the greater the impact of the 3.2 mm hole in the bone. In literature, the effect of weakening of the cancellous bone was not studied before in the past the way it is done in this study. In studies of whole bone specimens (cortico-cancellous bone) the decrease in the force to failure in drilled specimens is between approximately 20% to 40%. The fractures occur through the drill holes, but the tests were performed as a three-point bending test or a torsional strength test, unlike the current study. Comparisons between cortical and cancellous bone fracture models after drilling should be cautious. Cortical bone is on average 20–30% stronger than cancellous bone, as well as there are differences in elastic modulus and yield strain between those two types of bone. Moreover, there are obvious anatomical differences between whole bone models and a pure cancellous bone sample. The effect of weakening of the cancellous bone in previous studies is not as obvious as shown for cortical bone. The surprisingly high “safe” number of drill holes (between 14 and 40) found in a recent final element analysis study suggests that a femoral head cancellous bone can withstand multiple drilling attempts without compromising its strength. After anterior cruciate ligament reconstruction acute fractures usually, but not always occur through the drilled canals. In an investigation of composite femur bones, all specimens failed through the femoral canals of the double-bundle ACL reconstruction technique, but in the single bundle technique the fracture patterns were similar to the intact groups and only half occurred through the drill canal. In a similar study of tibial fractures, it was found that load to failure was similar in all groups (intact, single bundle and double bundle reconstructions). In a biomechanical study of the core decompression of an equine navicular bone, which is predominantely cancellous, it was found that the presence of unicortical bone canals significantly decrease the strength of bone by about 20%, what is similar to the current study. However, the size and number of drill holes did not affect the strength of the drilled specimens. The studies mentioned above suggest that the weakening of the bone due to drilling may not be directly proportional to the size and number of drill holes, and therefore to the volume of the bone stock lost because of drilling. The fact that the loss of strength of cancellous bone that is roughly proportional to the loss of cross-section area of the sample is a rather unexpected finding in this study. In the current study, no fracture occurred through the drill hole as was expected considering other biomechanical studies. This indicates that failure occurred rather by microfractures of the bone trabeculae of the specimens. It must be noted, however, that in the current study a purely compressive force was applied, and not a torsional or bending force as in others. A compressive force would rather crush the specimen, while a torsional or bending force would fracture the specimen through the point of least resistance, that is, through the hole in the bone. There are no biomechanical studies that evaluate the human cancellous bone strength after drilling without any cortical component. The implants used in the proximal femoral fractures are fixed in a cancellous bone, and the screw migration occurs within the femoral head, without a typical fracture of the femoral head. Therefore for the purpose of this study, there was a need to examine the cancellous bone without the surrounding cortex. This study attempts to resemble true in vivo conditions as closely as possible, but generally the tests used to determine the effect of a hole in a bone use a simplified protocol. In true clinical setting, the femoral head is compressed cyclically in changing force vectors because of femoral head rotation in the acetabulum. The site of compression in a fixed intertrochanteric fracture is between the articular surface and the femoral head implant, which is commonly round and threaded, or has several blades. This simplification is another major limitation of the study. In our study, we used a compression test, as was used in testing femoral head cancellous bone samples, as the best possible approximation of the forces acting on the cancellous bone in this region. Care was taken to use paired specimens of the same shape, size and origin, and the drilling was assigned to one of the paired specimens randomly. The bone in the anterior and posterior part of the femoral head have the same mechanical properties. The number of paired specimens is similar to other studies. The direction of force in this study is different than the principal compressive region or the main anatomical trabecular direction to ensure similar force direction as occurs in vivo between the hip contact forces of a standing patient and the implant screw. This study has shown statistically significant cancellous bone weakening after drilling of a single hole in a cancellous bone specimen. The effect of weakening cancellous bone by a guide wire hole cannot be considered yet as consistent as the effect of screw holes in the cortical bone, but should be taken into account in proximal femoral fracture surgery. Further studies, however, are needed to weigh the value of ideally placed implant against weakening of the bone with guide wire holes created by multiple attempts to insert it perfectly. This study suggests that drill holes in the femoral head weaken the bone stock, what may compromise the stability of the fixation. Moreover, since it has been found in other studies that the optimal position of the screws in intracapsular femoral neck fractures has no significant influence on the outcome, the authors of the current study no longer attempt multiple times to place the screws in a perfect configuration fearing that multiple drill holes around final implants may weaken the fixation strength. The synthetic bone model, that was similar in size and shape, shows comparable similarities between loss of force at failure and loss of cross section area as human bone. However, force at failure of human bone is almost twice that of the synthetic bone, and stiffness is approximately 4 times greater in human bone. This shows that biomechanical experiments performed on a synthetic bone cannot be directly translated to human bone, when it comes to absolute measures of the force to failure or stiffness, but relative loss of strength in both types of bones that sustained some kind of damage (fracture, cyst, or implant placement) may be comparable between human and synthetic bone. The experiments performed on synthetic bone probably give an accurate view of the mechanical properties of the fractures or implants compared in biomechanical studies, that is an implant that is inferior in synthetic bone studies most likely will also be inferior in the human bone. However, the magnitudes of forces recorded in synthetic bones may not be accurate in human bones or in patients. Other studies show some conflicting results regarding mechanical parameters of human v/s synthetic cancellous bone. The screw pullout force, which is probably the only mechanical parameter that can be attributed solely to cancellous bone, was found to be similar to natural bone in one study, nearly 4 times greater with standard synthetic bone, or exhibited only 40% of pullout strength of natural bone when an osteoporotic type of artificial bone was used. A study that compared stability of identical fracture fixations between human and standard synthetic bone showed that the artificial samples were significantly more stable. Those differences between types of bones and studies can be attributed to the type of material used for comparison. The human bones may have a different degree of osteoporosis and also different types of artificial bones are used (standard v/s osteoporotic type, as well as different generations from different manufacturers). In this study, we used a new type of osteoporotic bone, and compared it to femoral heads from elderly patients who sustained femoral neck fracture, what meets the criteria for osteoporosis. This study, as well as previous studies suggest that a perfect replacement for human bone in mechanical studies has not been found, but the relative differences found in those studies between samples are most likely accurate. In scientific research, this study supports a good scientific practice to prepare a control group sample (for example, a perfectly reduced fracture fixed with a standard implant, comparable with other studies), and compare differences in force parameters of the other tested groups relative to the standard control group. The absolute values of force to fracture (in N) should be considered with great caution when comparing to other studies, because mechanical properties may vary between artificial bones of different types or between bones from different manufacturers. The findings of this study support the hypothesis that drill holes decrease the strength of cancellous bone, moreover, the decrease is roughly proportional to the loss of cross section area of the sample caused by the drill hole. The comparison between human and artificial bone shows that forces measured in biomechanical studies on artificial bone are different than in human samples and cannot be directly attributed to humans, but the relative differences in mechanical properties of fractured and fixed synthetic samples may be accurate and resemble that of human bones. # Supporting information The authors wish to thank prof. Andrzej Ceynowa for correcting English language. [^1]: The authors have declared that no competing interests exist.

# Introduction Glutathione *S*-transferases (GSTs, EC 2.5.1.18) comprise a large superfamily of enzymes whose soluble members primarily function as detoxification enzymes, facilitating the conjugation of a diverse array of hydrophobic electrophilic xenobiotics by the nucleophilic attack of glutathione. GSTs have long been known as important components of cellular defense mechanisms in mammalian systems and recent studies are revealing their significance in mediating allelochemical tolerance in invertebrate-host interactions (reviewed). The diversity of GST isoforms and their capacity to detoxify allelochemicals in consumers has been correlated with diet breadth. For example, in a survey of GST isoforms from five lepidopteran species, specialist herbivores expressed only one major GST isoform, while generalists expressed multiple forms. Furthermore, GST isoforms purified from polyphagous herbivores that regularly consume isothiocyanate-containing cruciferous plants were able to metabolize a broader range of isothiocyanate allelochemicals in comparison to GSTs from specialists that did not consume crucifers and lacked the ability to conjugate isothiocyanates. These findings imply that the evolution of generalist GST forms favors promiscuous catalytic activity presumably needed to cope with the breadth of dietary toxins encountered; this mirrors results for other consumer counter- defense proteins (i.e., cytochrome P450s) whereby more catalytically flexible detoxification enzymes may promote a greater degree of polyphagy. Moreover, while polyphagous insects have little to no ability to increase their GST expression upon allelochemical exposure, they often possess a higher constitutive level of GST activity in comparison to oligophagous and monophagous species. The difference in constitutive and inducible GST expression in consumers may ultimately be a reflection of the non-specific role of GSTs as antioxidant enzymes. GSTs are known to be under regulatory control by antioxidant response elements found in their promoters. Due to the diverse range of dietary pro-oxidants encountered in polyphagous species, sustained transcriptional activation of GST enzymes by dietary compounds may result in the near maximal expression of GST enzymes. Consequently, possessing constitutively expressed GST enzymes that are catalytically versatile may confer a selective advantage to those consumers that regularly encounter unpredictable host chemistry. Similar to their terrestrial counterparts, marine consumers that regularly feed on allelochemically-rich prey may have evolved a parallel suite of biochemical resistance mechanisms (reviewed in). The induction or high constitutive activity of GSTs seen in several marine molluscs after allelochemical exposure has been suggested as a protective mechanism against dietary intoxication. High cytosolic GST activity was observed from the digestive gland of the generalist marine gastropod *Cyphoma gibbosum*, which feeds solely on a diet of chemically- defended gorgonians, and whose GST activity levels rival those of terrestrial invertebrates that feed solely on allelochemically-rich prey. This gastropod predator utilizes three families of gorgonian corals as hosts and in doing so encounters a range of lipophilic allelochemicals that includes diterpenes, sesquiterpenes, acetogens, highly-functionalized steroids and eicosanoids. One gorgonian diet, *Plexaura homomalla*, has been suggested to be a favorite of *C. gibbosum*, even though this gorgonian is known to contain impressive quantities of the cyclopentenone prostaglandin, PGA₂, which is known to serve as a feeding deterrent against generalist reef predators. Perhaps related to their anti-predatory properties in marine systems, the cytotoxic nature of cyclopentenone prostaglandins occurs, in part, due to the reactive α,β-unsaturated carbonyl group in the cyclopentenone ring, which can undergo nucleophilic addition with electrophilic moieties, resulting in protein and DNA adduct formation. Cyclopentenone prostaglandins of the A and J series have been shown to be inducers of GST enzymatic activity as well as mRNA expression in mammalian, and invertebrate cell lines. Furthermore, certain vertebrate alpha-, mu-, and pi-class GSTs were found to enhance PGA₂ conjugation with glutathione, suggesting that the overexpression of GST forms could modulate the cytotoxic effects of cyclopentenone prostaglandins. Mammalian GSTs also have the ability to non-catalytically bind lipophilic, amphipathic ligands, including PGJ₂, via noncovalent interactions, which effectively sequester these ligands in the cytosol away from their nuclear targets (i.e., peroxisomal proliferator-activated receptor, PPAR). Given that *C. gibbosum* neither avoids *P. homomalla* nor adjusts its feeding rates to reduce toxin exposure, this snail likely possesses effective detoxification mechanisms, possibly GST- mediated, to contend with the high concentrations of dietary prostaglandins. In a previous study, proteomic analysis of affinity-purified cytosolic GST fractions from *C. gibbosum* revealed that two major GST mu-class isoforms were responsible for the high GST activity observed in the digestive gland. Here, in a controlled laboratory feeding study, we allowed snails to feed ad libitum for four days on one of seven gorgonian species or a diet devoid of gorgonians, and monitored GST activity levels and isoform expression in response to the different suites of gorgonian allelochemicals. To investigate whether gorgonian extracts contained possible substrates for *Cyphoma* GSTs, we used a bioassay- guided fractionation approach, screening gorgonian crude organic extracts and subsequent fractions of different polarities for their ability to inhibit the 1-chloro-2,4-dinitrobenzene (CDNB)-conjugating activity of GST. Selected HPLC fractions found to inhibit GST activity were further characterized by ¹NMR and LC-MS spectral analyses to identify possible bioactive compounds. In addition, a series of commercially-available prostaglandins representing a range of eicosanoids previously described from *P. homomalla* were examined for their ability to inhibit *Cyphoma* GST activity. # Results ## Gorgonian Dietary Influence on GST Activity and Subunit Expression GST specific activity levels measured from digestive gland cytosolic preparations ranged from 1930 to 2957 nmol min⁻¹mg protein⁻¹. GST activity levels were within the range reported by Vrolijk and Targett, but did not differ significantly between snail diets. HPLC separation of affinity-purified GSTs identified fourteen unique peaks. HPLC peak 1 was previously identified as a theta-class GST, while HPLC peaks 2 thru 14 were identified as mu-class GST subunits. HPLC peaks 4 and 8 represented the majority of expressed GST subunits at 25% and 68%, respectively. The relative proportion of each GST subunit, represented by separate HPLC peaks and calculated based on HPLC peak area, did not differ significantly as a function of gorgonian diet when expressed either as percent of all subunits present or when normalized to the amount of affinity-purified GST sample injected on to the HPLC column. These results indicate that while GST activity is constitutively expressed at high levels in *Cyphoma* digestive gland, both GST activity and subunit abundance are unaffected by gorgonian diet. ## Inhibition of GST Activity by Gorgonian Extracts Crude organic extracts from *B. asbestinum*, *E. mammosa*, *G. ventalina*, *P. acerosa*, *P. americana*, *P. blanquillensis*, *P. elisabethae*, and *P. homomalla* tested at 5% natural volumetric concentration (NC) inhibited ≥70% of the GST activity in *Cyphoma* digestive gland cytosol compared to solvent controls. Chloroform-soluble fractions from all gorgonian species examined consistently showed ≥80% inhibition of GST activity compared to controls. Aqueous fractions generally displayed minor inhibitory effects, with the exception of fractions from *P. acerosa* and *P. homomalla*, which inhibited GST activity by 85% and 99%, respectively. Hexane-soluble fractions exhibited intermediate and wide-ranging inhibitory effects depending on gorgonian species. To further investigate the source of the putative gorgonian GST substrates/inhibitors, we used HPLC to fractionate the chloroform-soluble extracts of each gorgonian and tested their ability to inhibit the activity of affinity-purified *Cyphoma* GSTs. For all eight gorgonian species, HPLC fraction 1 (compounds eluting from 3 to 6 mins) exhibited the greatest inhibitory activity, causing \>80% GST inhibition at 10% NC compared to paired solvent controls. Inspection of HPLC chromatograms for all gorgonian species indicated that HPLC fraction 1 consisted of a mixture of compounds. For four of these gorgonian species (*B. asbestinum*, *E. mammosa*, *P. acerosa*, and *P. homomalla*) affinity-purified GST activity was completely inhibited by HPLC fraction 1 at 10% NC. Diluting HPLC fraction 1 from *B. asbestinum*, *E. mammosa*, *P. acerosa*, and *P. homomalla* to 0.05% NC decreased the inhibitory effect of the compound(s); however, in all cases diluted fractions still retained the ability to inhibit \>65% of affinity-purified GST activity compared to solvent controls. ## Identification of Prostaglandins in *P. homomalla* Extracts Because *P. homomalla* is a favored diet of *Cyphoma*, and possesses extracts determined to significantly inhibit the CDNB-conjugating activity of *Cyphoma* GSTs, we focused our subsequent efforts on elucidating the compound(s) responsible for this inhibition. LC-MS and ¹H-NMR spectral analysis of HPLC fraction 1 from the chloroform-soluble extract revealed the presence of PGA₂. The ¹H-NMR spectrum showed a 3H triplet at 0.85 ppm, consistent with the presence of a terminal methyl group in prostaglandins. The mass spectrum of HPLC fraction 1 displayed a parent ion at *m/z* 333 with fragment peaks at *m/z* 315, 271, 233, and 189, characteristic of PGA₂. For HPLC fraction 2 from *P. homomalla*, the LC-MS signal at ∼2.5 min showed an *m/z* 371.2, which corresponds to \[M+Na\]⁺ of 5Z and 5*E*-prostaglandin B₂ methyl ester. The ¹H NMR spectrum of *P. homomalla* HPLC fraction 2 showed a dominant prostaglandin-like compound that matched the literature values (ca. \<0.25 ppm difference between literature and experimental values) for 5*Z* and 5*E*-prostaglandin B₂ methyl ester; however, this compound showed a doublet of doublets at 7.6 ppm which was substantially further downfield than expected. The dominant compound in HPLC fraction 2 also displayed a singlet at 2.0 ppm, suggestive of acylation. A second peak at ∼16.6 min in the LC-MS signal for HPLC fraction 2 showed an *m*/*z* 413.2, which corresponds to \[M+Na\]⁺ of either 5*Z* or 5*E*-acetyl-prostaglandin B₂ methyl ester. ¹H NMR spectral data for the dominant compound were in agreement with those from 5*Z* and 5*E*-acetyl-prostaglandin B₂ methyl ester, with the major exception of the signal at 7.6 ppm, which is further downfield than expected for these known compounds. The *m/z* of 413.2 also corresponds to 15-epi-prostaglandin A₂ diester, whose ¹H NMR spectrum matched very closely to the dominant compound in HPLC fraction 2, including the signal at 7.6 ppm. However, further comparison was made difficult due to an absence of a complete set of NMR spectral data in the literature for 15-epi-prostaglandin A₂ diester. Overall, ¹H NMR and LC-MS spectral data support the presence of a dominant prostaglandin-derivative in *P. homomalla* HPLC fraction 2; however, because this fraction is a mixture of compounds, the exact identity cannot be established. Quantification of PGA₂ from *P. homomalla* HPLC fraction 1 by LC-MS revealed an approximate whole tissue concentration of 1.6 mM (or 530 µg PGA₂/mL of wet gorgonian tissue). HPLC fraction 2 (eluting at 6–9 mins) from *P. homomalla* showed a selected ion recording (*m/z* 333.3) at the expected retention time; however, peak intensities were below the limit of quantification and subsequent NMR spectral analysis indicated that PGA₂ was not present. ## Inhibition of GST Activity by Pure Prostaglandins When commercially available prostaglandins representing a diversity of classes found in *P. homomalla* were screened, those compounds containing a cyclopentenone ring (e.g., PGA₂) caused the greatest inhibition of GST activity, whereas the methyl ester forms of PGE₂ and PGF_2α displayed little to no inhibitory activity in comparison to solvent controls. The potencies of the four most inhibitory prostaglandins (15(S)-PGA₂, 15(R)-15-methyl PGA₂, 15(S)-PGE₂, 15(S)-PGF_2α) were further evaluated at a range of concentrations (0.2–2000 µM). All prostaglandins displayed concentration-dependent inhibition of enzyme activities, with IC₅₀ values ranging from 75.4 µM for 15(S)-PGA₂ to 334.6 µM for 15(S)-PGF_2α. Those prostaglandin series with the greatest inhibitory potencies (e.g., PGA₂) are known to be in the highest abundance in gorgonian tissues. The K_i values for 15(S)-PGA₂, 15(R)-15-methyl PGA₂, 15(S)-PGE₂, and 15(S)-PGF_2α calculated using the Cheng-Prusoff equation, ranged from 21.7 to 96.4 µM. # Discussion The exploitation of allelochemically-defended gorgonian corals by the co-evolved predator, *Cyphoma gibbosum*, is likely to be facilitated by this predator's ability to biotransform and/or sequester dietary allelochemicals using detoxification enzymes, such as soluble glutathione *S*-transferases. GSTs are integral components of the cellular xenobiotic defense system and have been documented to mediate allelochemical tolerance in terrestrial consumers. Our results suggest that they may also be important for marine predators that consume chemically defended prey; *C. gibbosum*'s high, constitutive expression of GSTs may protect this consumer from the abundance of deterrent lipophilic compounds found in its gorgonian diet. In a controlled feeding assay we determined that digestive gland tissues from *C. gibbosum* constitutively express high levels of GST activity regardless of the gorgonian diet. This finding differs from that of a previous study that noted differences in GST activity from field-collected *C. gibbosum* from different gorgonian hosts. The apparent differences could reflect differences in experimental design between the two studies. Vrolijk & Targett (1992) noted differences in GST activity among field-collected individuals for which no data were available on the residence time of snails on their respective hosts. In contrast, snails in the present study were subject to controlled, four-day feeding assays. It is possible that GST enzymes could show significant levels of induction if snails were allowed to feed on gorgonian diets longer than four days. However, studies of GST induction in other invertebrates, suggest that four days is sufficient for induction to occur; thus, it seems unlikely that our design, which included those gorgonian species examined by Vrolijk & Targett (1992), would have missed significant induction of GSTs. Snails could conceivably extend their exposure to the same suite of allelochemicals beyond the average 3.3 day residence time migrating to another colony of the same species. This scenario would be favored if *B. asbestinum* and *G. ventalina*, the two gorgonian diets eliciting increased GST activity in Vrolijk & Targett (1992), were in higher abundance on reefs, because prey selection by *C. gibbosum* is in proportion to gorgonian species abundance. Alternatively, geographical or within colony differences in allelochemical content could account for the differences between the two studies. Although GST activity did not vary by gorgonian diet, cytosolic digestive gland GSTs were further purified by affinity chromatography to investigate if GST subunit composition was influenced by allelochemical exposure. *Cyphoma* GST subunits were separated by HPLC, resulting in the identification of two major mu-class GST subunits, which accounted for 93% of the total GST subunit abundance. Quantification of GST subunit composition indicated that the relative abundance of GST subunits did not differ among snails feeding on different gorgonian diets. Interestingly, GST activity was maintained at a high level and subunit composition did not vary in snails fed control diets devoid of allelochemicals as compared to snails fed gorgonian diets. The presence of high GST activity in control-fed snails could indicate that some lipophilic gorgonian compounds and/or their metabolites may persist in snail tissues even after feeding has ceased, causing the expression of GSTs to be maintained. Alternatively, the constitutive expression of *Cyphoma* GSTs could be regulated by an allelochemical-independent mechanism. In either case, having a constant supply of ‘all-purpose’ GST enzymes may prove advantageous for predators that consistently feed on prey containing allelochemical GST substrates. The majority of GST substrates are hydrophobic compounds that react with the thiol moiety of glutathione. In our bio-assay guided screening approach, we used the ability of extracts/compounds to inhibit GST activity as an indirect measure of their potential to act as GST substrates. The results of GST inhibition assays indicated that the chloroform-soluble fractions from gorgonian extracts contained the bulk of inhibitory compounds. However, in addition to containing potential GST substrates, gorgonian extracts may also contain electrophilic compounds that could act as potent GST inhibitors, binding to free cysteine residues on the protein resulting in enzyme inactivation. The presence of high affinity GST inhibitors in gorgonian tissues may represent specific counter- adaptations of prey to thwart consumer GST-mediated metabolism of co-occurring allelochemicals. Although our initial screening approach of gorgonian extracts was not able to distinguish between GST substrates and inhibitors, this result did substantiate the hypothesis that all gorgonian species contained significant quantities of compounds capable of interacting with *Cyphoma* GSTs, which could account for the high constitutive activity of digestive gland GSTs identified here. The gorgonian *Plexaura homomalla* is a favored diet of *C. gibbosum*, , despite having high tissue concentrations of deterrent cyclopentenone prostaglandins. Electrophilic eicosanoids, like the cyclopentenone prostaglandin PGA₂, have been recognized as high affinity substrates/inhibitors for vertebrate GSTs,. Furthermore, NMR and LC-MS analysis showed that PGA₂ was present in the *P. homomalla* HPLC fractions demonstrating the greatest inhibitory potential. Therefore, the potential importance of *P. homomalla*'s allelochemicals in the co-evolution of *C. gibbosum* detoxification enzymes, coupled with the interesting ecological and biological activities of eicosanoids, prompted us to determine if prostaglandins could serve as substrates for *C. gibbosum* GSTs. *P. homomalla* tissues predominately contain the fully esterified form of PGA₂ (∼2% dry weight of the gorgonian), which is related to a larger group of eicosanoids that includes the coral-derived halogenated marine clavulones, and puniglandins, all of which display cytotoxic activities thought to be related to the presence of a reactive α,β-unsaturated ketone. While the exact mechanism of toxicity is unknown, the prostaglandins are transported into the nucleus – where the electrophilic α,β-unsaturated carbonyl is free to bind with nucleophilic sulfhydryl residues on target proteins, unless rapidly conjugated by cytosolic GSH and transported out of the cell by glutathione- conjugate transporters. In this study, the α,β-unsaturated carbonyl-containing prostaglandins (15(R)-15-methyl PGA₂ and 15(S)-PGA₂) were the most potent inhibitors of CDNB-conjugating activity of *Cyphoma* GSTs in both the initial screening of eight prostaglandin compounds and upon comparison of IC₅₀ values, establishing the order of potency of prostaglandins to be 15(S)-PGA₂\>15(R)-15-methyl PGA₂≫15(S)-PGE₂≈15(S)-PGF_2α. The K_i values for cyclopentenone-containing prostaglandin A series were also 2.3- to 4-fold lower (greater affinities) for *Cyphoma* GSTs in comparison to those of either PGE₂ or PGF_2α. If we assume PGA₂ is a substrate for *Cyphoma* GST(s), possibly binding with high affinity in the active site (H-site) once occupied by CDNB, it is reasonable to compare K_i values (i.e., apparent K_m for PGA₂) obtained here to K_m values for PGA₂ cited in other studies. The apparent K_m (∼21.7 µM) for 15(S)-PGA₂ described here is in line with values identified for the conversion of PGA₂ to its glutathione conjugate by human mu-class GST M1a-a (26 µM), rat alpha-class GST A4-4 (12 µM) and human mu-class GST M2-2 (7.6 µM). The rank order of GST affinity for prostaglandins (15(S)-PGA₂\>15(R)-15-methyl PGA₂\>15(S)-PGE₂\>15(S)-PGF_2α), is also positively correlated with the relative abundance of each prostaglandin series in *P. homomalla* tissues. This finding may suggest that *Cyphoma* GSTs have evolved to efficiently catalyze the conjugation of prostaglandins found in the greatest abundance in its diet (PGA₂), yet still retain a broad enough substrate specificity to metabolize additional prostaglandin classes (PGE₂, PGF_2α). The apparent K_m (Ki) values reported here indicate that dietary prostaglandins could be high affinity substrates for *Cyphoma* GSTs *in vivo*. However, the physiological relevance of this value would depend on the concentration of prostaglandins occurring in digestive gland tissues of *Cyphoma*. To obtain an estimate of these concentrations, we first calculated the volume of *P. homomalla* tissue consumed per snail per day based on feeding scar measurements. *Cyphoma* feeding scars on *P. homomalla* colonies averaged 12 cm in length, did not exceed 1 cm in width, and penetrated to the gorgonian skeleton 66% of the time, a depth of 0.4 cm (K. Whalen, pers. observation). Therefore, conservative estimates of scar volume averaged 1.44 cm³ per snail or 1.44 mL of gorgonian tissue. This tissue volume was divided by the mean residence time of snails feeding on *P. homomalla* (2.9 d, n = 50 snails) to yield an estimate for the volume of *P. homomalla* tissue consumed by each snail per day (0.66 mL/snail/day). Pawlik & Fenical determined that 1 mL of wet *P. homomalla* tissue (excluding the gorgonian axial skeleton) was equivalent to 0.86 g of dry gorgonian tissue. If 2% of the dry weight of the gorgonian is prostaglandins, then 0.66 mL of gorgonian tissue would contain 0.011 g of prostaglandins. Assuming the majority of prostaglandins are in the PGA₂ form (FW = 348.5 g) and are completely retained within the digestive gland upon ingestion (ave. dig. gland weight = 0.25 g, K. Whalen pers. obs., with a density comparable to human liver ∼1 g/mL), then the upper limit of PGA₂ concentrations in digestive gland would be 0.13 M. The single day grazing rates of *P. homomalla* colonies by *C. gibbosum* of 0.17 mL/snail/day reported in were used to obtain the lower bound of PGA₂ tissue concentration (0.03 M). Ciereszko and Schneider reported the fecal pellets of *C. gibbosum* contain no appreciable amounts of recognizable prostaglandins, suggesting that the majority of prostaglandins are being metabolized or sequestered within the snail. If we conservatively assume that only 1% of the ingested prostaglandins are retained within the digestive gland during feeding (e.g., 1% concentration of PGA₂∼0.3–1.3 mM PGA₂), the *in vivo* concentration of prostaglandins in this tissue would still be 7- to 59-fold higher than the apparent K_m (21.7–39.4 µM) obtained for PGA₂ and its methylated derivative. Even at the lower bound of *in vivo* PGA₂ concentration (∼0.3 mM), *Cyphoma* GSTs would be operating at near physiological capacity (\>93%) according to the fractional velocity (v/V_max) estimates. Glutathione *S*-transferases are most well known for their ability to conjugate electrophilic toxicants; however, their capacity to bind and sequester non- substrate ligands may also be an important protective mechanism. Certain human GST isoforms have been shown to exert their protective effects through this ligandin-like behavior by binding with high affinity to inhibitory prostaglandins (e.g., PGJ₂), effectively sequestering them in the cytosol away from target nuclear proteins and preventing effects on gene regulation. A comprehensive screening of allelochemicals from host plants of the fall armyworm *Spodoptera frugiperda* found that many act as non-competitive inhibitors of GST activity. Overexpression of GSTs in *S. frugiperda* may serve as a detoxification strategy by facilitating the sequestration of non-substrate ligands and thereby preventing their interference with essential cellular functions. The same strategy might be used by the marine chiton *Cryptochiton stelleri*, which consumes a red algal diet rich in the feeding deterrent lanosol, a noncompetitive inhibitor of this chiton's GST activity. Similarly, high constitutive GST activity was observed in *Cyphoma* independent of allelochemical diet and all of the gorgonian extracts examined contained potent inhibitory compounds. While the type of inhibition was not quantified for gorgonian extracts and all compounds, it is likely that gorgonian diets contain both substrates (e.g., PGA₂) and non-substrate ligands. Therefore, the constitutive expression of GSTs may be indicative of a more *general* biochemical resistance strategy that is capable of responding to a diversity of compounds in the diet of a generalist consumer. The results of this study provide the first comprehensive evaluation of the influence of dietary allelochemicals on the expression and function of glutathione transferases in a generalist marine consumer. Together with our companion studies on the gorgonian diet-mediated expression of cytochrome P450 expression in *Cyphoma*, the present results add substantial knowledge regarding the role of detoxification enzymes in determining macroevolutionary patterns of diet preference among consumers. Controlled feeding assays showed that *Cyphoma* digestive gland GST composition and activity did not differ with gorgonian diet. This result in combination with evidence from *in vitro* inhibition studies with *Cyphoma* GSTs by gorgonian extracts, suggests that the high constitutive expression of GST enzymes in *Cyphoma* digestive gland may be necessitated by the presence of numerous potent inhibitors/substrates in their gorgonian diets. Furthermore, all three prostaglandin classes (A, E, F) found in the gorgonian *P. homomalla* were able to inhibit *Cyphoma* CDNB-conjugating GST activity, with relative potencies positively correlated with their abundance in gorgonian tissues. Together, these findings suggest that *C. gibbosum* detoxification enzymes may have evolved to enable the conjugation and sequestration of a broad range of lipophilic allelochemicals resulting from this predator's close association with chemically diverse gorgonian diets. Given the importance of allelochemicals in shaping patterns of predation and herbivory in marine systems, these findings suggest that co-evolved consumers have the capacity to detoxify allelochemicals in their prey, providing these consumers with a competitive advantage in ecosystems where allelochemically-rich prey species abound. # Methods ## Materials CDNB, DTT, potassium phosphate, potassium chloride, EDTA, protease inhibitor cocktail (4-(2-aminoethyl)benzenesulfonyl fluoride, aprotinin, bestatin hydrochloride, E-64, leupeptin, and pepstatin A), GSH-agarose (G4510) was purchased from Sigma (St. Louis, MO). Bradford reagents were purchased from Bio- Rad (Hercules, CA). PD-10 size exclusion columns were purchased from GE Healthcare (Piscataway, NJ). Amicon Ultra-4 centrifugational filters were purchased from Millipore (Billerica, MA). NanoOrange protein quantitation kit was purchased from Molecular Probes (Eugene, OR). Prostaglandins (15(S)-PGA₂, 15(R)-15-methyl PGA₂, 15(S)-PGE₂, 15(R)-PGE₂, 15(S)-PGE₂-methyl ester, 15(S)-PGF_2α, 15(R)-PGF_2α, 15(S)-PGF_2α-methyl ester) were purchased from Caymen Chemical (Ann Arbor, MI). All solvents used for extract and chemical analysis were purchased from Fisher Scientific (Pittsburgh, PA). ## Animal Collection and Feeding Assay Design A total of 39 adult *Cyphoma gibbosum* (ca. 2–3 cm length) were collected from four shallow reefs (\<20m) (Big Point – 23°47.383′N, 76°8.113′W; Rainbow Gardens – 23°47.792′N, 76°8.787′W; Shark Rock – 23°45.075′N, 76°7.475′W; Sugar Blue Holes – 23°41.910′N, 76°0.23′W) surrounding the Perry Institute of Marine Science (PIMS), Lee Stocking Island, Exuma Cays, Bahamas in January 2006. Snails were immediately transported to wet laboratory facilities provided by PIMS where a series of feeding assays were conducted with six gorgonian species (*Briareum asbestinum*, *Eunicea mammosa*, *Gorgonia ventalina*, *Pseudopterogorgia acerosa*, *Pseudopterogorgia americana*, *Plexaura homomalla*) observed to serve as hosts for *C. gibbosum* in the field. Individual snails were housed separately in 3-L polycarbonate tanks which were placed in a 12′×20″ raceway supplied with filtered, continuous-flow seawater at a flow rate of approximately 1L min⁻¹. This design allowed for a common water source to feed each tank but prevented mixing between tanks. Snails collected from the same reefs were housed separately in the same raceways and randomly assigned to one of seven groups – one of six gorgonian diets or a control diet – at the start of the feeding assays. Snails were allowed to feed ad. libum on either a control diet devoid of gorgonian compounds (i.e., alginic acid and freeze-dried squid) (n = 12) or one of six gorgonian diets (n = 27) for four days. A minimum of ten colonies for each gorgonian species were collected from shallow reefs (\<20m) surrounding PIMS and housed in a separate raceway prior to introduction into the tanks containing *C. gibbosum*. The maximum amount of time between gorgonian field collection and introduction into the feeding assay was 12 hours. Gorgonian colonies were cut into 2–3 inch pieces and allowed to recover for four hours before addition to *C. gibbosum* tanks. The control diet consisted of alginic acid and freeze-dried squid powder prepared as described in and mirrored the average nutritional quality of gorgonian tissue. The squid- alginate paste was pressed into sixteen 3-mm deep wells drilled into a 3″×1″ piece of Formica® resembling a domino. The domino was then placed into a 0.25 M calcium chloride solution allowing the squid-alginate paste to harden. Both control and gorgonian diets were replaced every 24 hours for four days and feeding activity was monitored by the presence of feeding scars on their gorgonian prey and empty wells on control dominos. Following the completion of the feeding assay digestive glands were immediately dissected, weighed, frozen in liquid nitrogen and maintained at −80°C until further processing. ## GST Purification and HPLC Subunit Analysis Cytosolic and affinity-purified GSTs were isolated from *Cyphoma* digestive gland samples as described in. Briefly, cytosolic GSTs were isolated by homogenizing digestive glands (n = 39) separately in buffer (0.1 M potassium phosphate, 1 mM EDTA, 1 mM DTT, 1.15% potassium chloride, protease inhibitor cocktail (1×); pH 7.5), and differentially centrifuging the homogenates to obtain the cytosolic fraction containing the soluble GST pool. Individual cytosolic GST fractions were then applied to both a PD-10 size exclusion and GSH-agarose affinity column in series to obtain the affinity-purified fraction of GSTs. GST fractions were buffer exchanged to low salt concentration and concentrated with Amicon Ultra-4 centrifugational filters (5K NMWL membrane) and protein concentrations of Amicon concentrates were determined using the NanoOrange protein quantitation kit. A 30 µL aliquot of the affinity-purified GST concentrate from each of the 39 digestive gland samples was injected onto a reverse phase Vydac protein/peptide column (model \#218 TP 52; C18 µm 250 mm×2.1 mm) and separated using a Waters 600 MultiSolvent Delivery System, with a flow rate of 0.5 mL/min. Peaks were detected using a Waters 2487 Dual Wavelength Absorbance Detector (λ = 214 nm). Mobile phase A consisted of 38% acetonitrile, 62% water and 0.1% trifluoroacetic acid (TFA). Mobile phase B consisted of 80% acetonitrile, 20% water and 0.1% TFA. The initial mobile phase consisted of 100% A. GST subunits were separated using a linear gradient from 0 to 40% B in 22 min, and 40 to 100% B in 37 min. The column was re-equilibrated with 100% A from 37–50 mins prior to the next injection. Integration of HPLC peak area was achieved using the Empower 2 Chromatography Data Software package (Waters, Milford, MA) and converted to GST subunit percent composition for each digestive gland sample. ## GST Activity Assay Enzyme activity was measured using CDNB as a substrate by the method of optimized for *C. gibbosum*, in a microplate format. The reaction mixture (in a final volume of 200 µL) contained 0.1 M potassium phosphate buffer, 1.0 mM EDTA, pH 7.5, 1 mM CDNB, 1 mM reduced GSH and 2 µg of cytosolic protein or 3.3–6.4 ng of affinity-purified GST sample. CDNB was solubilized in ethanol and constituted 1% of the final reaction mixture volume. The reaction incubated at 25°C was initiated by the addition of CDNB and performed in triplicate. The conjugation of CDNB with GSH was measured as the increase in absorbance at 340 nm (Δε₃₄₀ 0.00503 µM⁻¹ cm⁻¹) using a tunable microplate reader (Versamax, Molecular Devices, Sunnyvale, CA). Activity was calculated using protein concentrations determined via the Bradford assay with BSA as a standard. ## Extraction and Isolation of Gorgonian Compounds A minimum of ten colonies for each gorgonian species were collected from shallow reefs (\<20 m) surrounding Lee Stocking Island, Exuma Cays, Bahamas. A portion of the gorgonian colonies, prior to their introduction into the feeding assay, was immediately removed after field collection and immersed in seawater to determine volumetric displacement, frozen at −80°C, and lyophilized for subsequent chemical extraction. *Pseudopterogoria blanquillensis* was also collected for chemical analysis; however, this gorgonian species did not participate in the feeding assays. A 50 mL volumetric equivalent of pooled tissue for each of the eight gorgonian species was extracted twice at room temperature in 250 mL reagent grade acetone overnight with agitation. Resulting crude organic extracts were vacuum-filtered through celite, dried by rotary evaporation, and recombined into a 20 mL scintillation vial using a minimum volume of solvent. The crude organic extracts were then completely dried using a vacuum concentrator. Gorgonian crude organic extracts were assayed at 5% natural concentration by volume (i.e., the extract from 0.05 mL of gorgonian was diluted into 1 mL of assay buffer) for the ability to inhibit *Cyphoma* cytosolic GST activity as described below. Those crude organic extracts that were able to inhibit ≥80% of GST activity were subjected to further fractionation using a bioassay-guided fractionation approach. Gorgonian crude organic extracts were separated by partition between hexane and methanol-water (9∶1) followed by partition of the methanol-water fraction (adjusted to 6∶4) against chloroform. The chloroform-soluble, hexane-soluble, and aqueous (i.e., methanol/water-soluble) fractions were reduced *in vacuo* and assayed for their ability to inhibit cytosolic GST activity as described below. Based on patterns of GST enzyme inhibition, chloroform fractions from all eight gorgonian species were separated further using a reverse-phase semi-prep Zorbax SB-C18 column (5 µm, 9.4 mm×2.5 cm) attached to a Waters Breeze HPLC system (515 pump) with a Waters 2487 UV detector at 215 and 254 nm. Compounds were eluted over 33 mins at a flow rate of 3 mL/min with methanol/water (9∶1) with linear ramping to 100% methanol. HPLC fractions were collected at three minute intervals over 33 min, yielding ten fractions per gorgonian species. Each chloroform fraction was assayed for GST inhibition at 10% natural volumetric concentration. Chloroform fractions yielding 100% inhibition were further assayed at 0.5% natural volumetric concentration. ## Inhibition Assays GST activity measurements were performed as described above. Gorgonian crude organic extracts and partitions were dissolved in the appropriate solvent (e.g., acetone, *n*-propanol, or methanol), HPLC fractions were dissolved in methanol, and prostaglandins were dissolved in DMSO. Solvent concentrations did not exceed 5% of the experimental volume and had no effect on GST activity when compared to non-solvent controls (data not shown). Immediately prior to the start of the assay, inhibitor solutions were added to the buffer/GSH mixture and homogenized to ensure equal distribution of inhibitor in all microplate wells. The data were corrected for the non-enzymatic reaction rates and the effect of the inhibitors on catalytic activity was measured by comparing the initial rate of reaction in the presence and absence of the inhibitor. Eight commercially available prostaglandins representing a diversity of forms present in gorgonian tissue (15(S)-PGA₂, 15(R)-15-methyl PGA₂, 15(S)-PGE₂, 15(R)-PGE₂, 15(S)-PGE₂-methyl ester, 15(S)-PGF_2α, 15(R)-PGF_2α, 15(S)-PGF_2α-methyl ester), including both enantiomers (R and S) forms when possible, were screened at 600 µM for their ability to inhibit crude cytosolic GST activity. From this initial screening, only those prostaglandin compounds that demonstrated an ability to reduce GST activity by 50% or greater were further evaluated at a range of concentrations (0.2–2000 µM) in order to estimate the concentration producing 50% inhibition of enzyme activity (IC₅₀). Prostaglandin IC₅₀ values were calculated and 95% confidence intervals were estimated using Prism 5.0 software (GraphPad) by fitting the log transformation of the response variable by nonlinear regression to the variable slope equation (1) and constraining the bottom to zero but allowing the Hill Slope to vary. The variable slope equation is:where Top is the maximum percent GST activity remaining, Bottom was constrained to zero, IC₅₀ is the concentration of inhibitor that produces inhibition half-way between the Top and Bottom, and \[I\] is the logarithmic concentration of the inhibitor. ## Determining the Dissociation Constant for Inhibitor Binding (K_i) Because the IC₅₀ depends on the substrate concentration used in the experiment, this value is only useful for comparing inhibitors within experiments and not between laboratories unless identical assay conditions were used. However, calculated K_i (the dissociation constant of the enzyme-inhibitor complex) values can be used to directly compare inhibitor affinity for the enzyme between studies. K_i estimates were calculated using the IC₅₀ values obtained for 15(S)-PGA₂, 15(R)-15-methyl PGA₂, 15(S)-PGE₂, 15(S)-PGF_2α with the Cheng-Prusoff equation (Eq. 2), where K_m is the Michaelis- Menten constant for CDNB (see below), \[S\] is the substrate concentration (1 mM CDNB), K_i is the equilibrium dissociation constant for the inhibitor, and IC₅₀ is as defined above.To obtain an estimate of K_m for CDNB, initial-rate measurements using CDNB as the concentration-variable substrate were performed. GST activity was measured at five concentrations of CDNB ranging from 0.5 to 3 mM in the absence of inhibitors with 6 ng of affinity-purified GST protein at 25°C in 0.1 M potassium phosphate buffer, 1.0 mM EDTA, pH 7.5, containing 1 mM GSH and 4% (v/v) DMSO. An affinity-purified GST preparation from a single digestive gland was used as the protein source with a specific activity (mean±SE) of 561±25 µmol min⁻¹ mg protein⁻¹. The reaction was initiated by the addition of 2 µL of CDNB and performed in duplicate. The data from two independent experiments were corrected for the non-enzymatic reaction rates and globally fitted to the Michaelis-Menten equation to yield an estimate of K_m = 0.41±0.14 mM CDNB (mean±SD). ## Chemical Analysis of HPLC Fractions Proton NMR spectra for *P. homomalla* HPLC fractions 1 and 2 were recorded in deuterated DMSO (Cambridge Isotope Laboratories, Andover, MA) with a Bruker DRX-500 instrument using a 5 mm inverse detection probe, and referenced to residual DMSO (δ 2.49 ppm). Spectra collected for chromatographic fractions were compared with the ¹H NMR spectrum obtained for authentic 15(S)-PGA₂ (Cayman Chemical, Ann Arbor, MI). LC-MS analyses of *P. homomalla* HPLC fractions were completed using a Waters 2695 HPLC with a Waters 2996 photodiode array UV detector and Micromass ZQ 2000 mass spectrometer with electrospray ionization in both positive and negative modes. Optima grade solvents were used in all LC-MS experiments. LC-MS separations were achieved with an Alltech Altima C18 column (2.1×100 mm, 3 µm) applying a gradient mobile phase of 40∶60 to 95∶5 acetonitrile∶water with 0.01% acetic acid throughout. PGA₂ was detected in fractions by matching chromatographic retention times and MS fragmentation patterns with those obtained for pure synthetic PGA₂. For fractions in which PGA₂ was detected, the negative-mode ESI-MS selected ion recording for *m/z* 333.3, corresponding to \[M-H\]⁻ of PGA₂, was integrated and compared to a standard curve for PGA₂ at six concentrations (r² = 0.94). The concentration of PGA₂ in each HPLC fraction was determined by interpolation of this standard curve data. # Supporting Information We thank the staff of the Perry Institute for Marine Science; Carly Gaebe, Terry Rioux and Ann Tarrant for their assistance with animal collection. We also thank Dexter Morin for his assistance with GST separations. [^1]: Conceived and designed the experiments: KW JK MH. Performed the experiments: KW ALL. Analyzed the data: KW ALL MH. Contributed reagents/materials/analysis tools: KW JK MH. Wrote the paper: KW. [^2]: Current address: Centre for Marine BioInnovation, University of New South Wales, Sydney, New South Wales, Australia [^3]: Current address: Scripps Institution of Oceanography, University of California San Diego, La Jolla, California, United States of America [^4]: The authors have declared that no competing interests exist.

# Introduction Sulfotransferase (SULT) enzymes catalyze the sulfate conjugation of a broad range of substrates and play an important role in metabolism of endogenous and exogenous compounds including thyroid and steroid hormones, neurotransmitters, drugs and procarcinogens. There are many isoforms of the *SULT*s supergene family, each with different amino acid sequence identity and substrate specificity. SULT1A1 is an important member of the sulfotransferase family involving in the pathogenic process of various cancers. The *SULT1A1* gene is located on chromosome 16p12.1–p11.2. Previous study indicated that exon 7 of the *SULT1A1* gene contained a G to A transition at codon 213 (rs9282861) that causes an Arg to His amino acid substitution. Some studies have shown that this genetic polymorphism leads to a decrease in enzymatic activity of SULT1A1 and the sulfonation efficiency thus associating with susceptibility to several cancers. Although the specific role of *SULT1A1* Arg213His polymorphism in carcinogenesis has been investigated in numerous case- control studies, the results have been inconclusive, even conflictive. In order to give a comprehensive and precise result, we performed this meta-analysis study to analyze the association between this polymorphism and cancer risk. # Materials and Methods ## Identification of eligible studies The meta-analysis was conducted following the criteria of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). In this study, we did an exhaustive literature search on studies that examined the association of the *SULT1A1* gene polymorphisms with cancer risks. All eligible studies were identified by searching the following databases: PubMed, Web of Knowledge and China National Knowledge Infrastructure (CNKI, <http://www.cnki.net/>). The following terms were utilized: “sulfotransferase, *SULT* or *SULT1A1*”, “polymorphism, variation, variant or mutation” and “cancer or carcinoma”. In the CNKI database, we searched with these corresponding key words in Chinese characters. Included studies should meet the following criteria: (1) evaluating the association between *SULT1A1* Arg213His polymorphism and cancer risk; (2) study designed as case-control; (3) sufficient data available to estimate an odd ratio (OR) with its 95% confidence interval (95% CI). ## Data extraction Two investigators extracted data independently and reached consensus on the following characteristics of the selected studies: first author's name, the year of publication, ethnicity of the study population, matching criteria, number of participants, genotype distribution and control source. ## Statistical analysis Hardy-Weinberg equilibrium was assessed by Chi-square test. Crude odd ratio (OR) and 95% confidence interval (CI) were used to estimate the association between *SULT1A1* polymorphism and cancer susceptibility under the dominant model (Arg/His+His/His vs. Arg/Arg), recessive model (His/His vs. Arg/Arg₊Arg/His), homozygous model (His/His vs. Arg/Arg), heterozygous model (His/Arg vs. Arg/Arg) and allelic model (His vs. Arg). The heterogeneity among the studies was evaluated by Q-test and *I²* value ranging from 0% to 100% to describe the percentage of between-study variation caused by heterogeneity. P value for the Q-test less than 0.10 indicates existing heterogeneity among studies. And then the pooled OR was measured by a random effect model (the DerSimonian-Laird method). Otherwise, a fixed effect model (the Mantel-Haenszel method) was chosen. Subgroup analyses were performed according to cancer type (breast cancer, colorectal cancer, urothelial cancer, prostate cancer, lung cancer, upper aero digestive tract (UADT) cancer, ovarian cancer and gastric cancer), ethnicity (Caucasian, East Asian, Indian and African) and source of controls (hospital based and population based). When heterogeneity was detected, a multivariable meta-regression analysis including cancer type, ethnicity, control source and year of publication to explore potential source of heterogeneity and sensitivity analysis were performed. The potential publication bias was estimated using Egger's linear regression test by visual inspection of the funnel plot. P_\<0.05 was considered statistically significant, and all P values were two-sided. Analyses were performed using the software Review Manager 5.3 (Cochrane Collaboration), R software ([www.r-project.org](http://www.r-project.org)) and STATA 12.0 software (StataCrop). # Results ## Characteristics of eligible studies The flow diagram of literature search was given in. A total of 91 studies focusing the association between the *SULT1A1* Arg213His polymorphism and cancer risks were identified. 25 of them were ruled out because of unavailable data or repeated data. Thus, the allele and genotype frequencies of the *SULT1A1* Arg213His polymorphism were extracted from 66 articles. However, 18 articles didn't meet with Hardy-Weinberg equilibrium and were abandoned. As a result, 53 studies of 48 articles, involving 16733 cases and 23334 controls were included in the pooled analyses. The characteristics of studies included in the current meta-analysis are shown in. Among these studies, 13 were conducted for breast cancer, 10 for colorectal cancer, 7 for urothelial cancer, 5 for prostate cancer, 5 for lung cancer, 5 for UADT (upper aero digestive tract) cancer, 3 for ovarian cancer, 2 for gastric cancer and 1 for myeloid leukemia, multiple myeloma, and endometrial cancer, respectively. By ethnics, there were 27 studies of Caucasians, 11 studies of East Asians, 4 studies of Indians, 2 studies of Africans and 9 studies of mixed ethnics. By source of controls, 16 studies were population-based, 17 studies were hospital-based and 20 studies were not clear. ## Overall Analysis showed the results of overall analysis and the subgroup analysis. The analyses on the full data set indicated a significant association of the *SULT1A1* Arg213His polymorphism with cancer risk: heterozygous (OR = 1.09, 95% CI = 1.01–1.19, P = 0.035), homozygous (OR = 1.20, 95% CI = 1.04–1.39, P = 0.014), dominant (OR = 1.12, 95% CI = 1.03–1.22, P  =  0.008), recessive (OR = 1.16, 95% CI = 1.02–1.32, P = 0.027) and allelic model (OR = 1.11, 95% CI = 1.04–1.20, P = 0.003), with high heterogeneity among studies (*I²* = 63.1%, 62.6%, 68.5%, 58.3% and 73.7%, respectively, all P\<0.001). ## Subgroup Analyses We analyzed the association in cancer type subgroup. *SULT1A1* Arg213His polymorphism can increase cancer risks in the following cancer types: breast cancer (homozygous model: OR = 1.37, 95% CI = 1.01–1.87, P = 0.045; dominant model: OR = 1.18, 95% CI = 1.00–1.40, P = 0.050 and allelic model: OR = 1.15, 95% CI = 1.00–1.32, P = 0.044); UADT cancer (heterozygous model: OR = 1.62, 95% CI = 1.11–2.35, P = 0.012; dominant model: OR = 1.63, 95% CI = 1.13–2.35, P = 0.009 and allelic model: OR = 1.52, 95% CI = 1.10–2.11, P = 0.012). Forest plots of breast cancer risk and UADT cancer risk were shown in and separately. Analyzed by ethnicity, a moderately increased risk was observed in Caucasians (homozygous model: OR = 1.20, 95% CI = 1.01–1.43, P = 0.035 and allelic model: OR = 1.10, 95% CI = 1.01–1.19, P = 0.019) and Indians (recessive model: OR = 1.93, 95% CI = 1.22–3.07, P = 0.005). No significant association was found in other ethnicities in any model. By control source, significant association was observed in hospital based study, but not the population based study. ## Meta-regression analysis To find potential source of heterogeneity, multivariable meta-regression analyses were conducted in total group and subgroups including cancer type, ethnicity, control source and publication year. In the breast cancer subgroup, ethnicity (heterozygous model, P = 0.027; recessive model, P = 0.020) and publication year (heterozygous model, P = 0.019; recessive model, P = 0.012) are significant sources of heterogeneity. Other variables don't affect heterogeneity. ## Sensitivity analysis The sensitivity analysis was constructed by repeating the meta-analysis sequentially removing each study. In the recessive model, two studies, were found to affect the pooled OR and the heterogeneity when removed. The study conducted by Khvostova was focused on breast cancer and Sun's study was focused on colorectal cancer among Caucasians, so further sensitivity analyses were conducted in total data set and breast cancer, colorectal cancer and Caucasian subgroups after removing the two studies. In total group, the heterogeneity was significantly decreased (*I²* = 58.2, 42.2, 63.5, 33.1 and 66.4, respectively). In the subgroup sensitivity analyses, removing the two studies can significantly decrease the heterogeneity among studies, most *I²* values less than 50%. And this polymorphism didn't show any obvious correlation with breast cancer risk. At last, we conducted the sensitivity analyses on the remaining studies and the result was stable. ## Publication bias Funnel plots and Egger's test were carried out to assess publication bias. The shapes of funnel plots indicated no obvious asymmetry. Egger's test found no publication bias in the heterozygous (P = 0.074); homozygous (P = 0.146); dominant (P = 0.076); recessive (P = 0.282) and allelic model (P = 0.081). # Discussion SULT1A1 enzyme encoded by *SULT1A1* gene plays an important role in xenobiotic metabolism. The Arg213His polymorphism, the most widely studied polymorphism within *SULT1A1* gene, can reduce enzyme activity and thermostability, and consequently results in an individual's susceptibility to cancer. There have been a few meta-analyses focusing on this mutation and cancer risk. However, most of these analyses were conducted before the year 2012 and a new meta-analysis is needed to give a comprehensive conclusion due to the increasing data of case-control studies. This present meta-analysis, including 16733 cases and 23334 controls from 53 case-control studies, explored the association between the *SULT1A1* Arg213His polymorphism and cancer risk. This is the largest scale meta-analysis so far. Our results suggested that the *SULT1A1* Arg213His was associated with UADT cancer risk. As the upper aero digestive tract is exposed to numerous potential carcinogens such as phenolic xenobiotics, polycyclic aromatic hydrocarbons and heterocyclic aromatic amines contained in cigarette smoking, environmental pollutants and some food, this result manifests that the mutation within *SULT1A1* causes the low SULT1A1 activity and is associated with high susceptibility to cancers related with environment. In the sensitivity analyses, the study conducted by Khvostova influences the pooled estimates and the heterogeneity most in breast cancer subgroup. And after removing this study, the significant association between *SULT1A1* Arg213His and breast cancer risk became null. We further checked data from Khvostova and observed the percentage of wild homozygous genotype in Khvostova's study was obviously lower than that in other studies thus causing great heterogeneity. At last a robust result was achieved and failed to reveal significant association in breast cancer subgroup. This result is similar to Wang, Lee and Jiang, but they found a positive association of this polymorphism with breast cancer susceptibility among Asians. While in our meta-analysis, we only recruited one paper focused on breast cancer among Asians because other papers on Asians deviate from *HWE* and were excluded. This is a limitation of this meta-analysis and more independent case-control studies conducted on Asians are needed to conclude a more comprehensive result. In the ethnic subgroup analysis, we found that the genotype distributions of the SNP site are different in ethnic groups. When calculating the percentage of alleles in every ethnic, we found that His allele in Asians (9.58%) is significantly less than in Caucasians (35.2%). Different ethnicities may have different genetic backgrounds, thus causing different genotype frequencies in Asian and other ethnic groups which may influence cancer susceptibility. Li and Kotnis have conducted meta-analyses focused on environment-related cancers, such as tobacco-related cancers and found cancer risk could be modulated by interaction between genetic variants and environmental factors. As exposed environmental factors are different according to cancer types, for example smoking leads to lung cancer, while the intake of meat influences breast cancer and colorectal cancer, and our analysis took many kinds of cancer into account, we decided not to include environmental factors. Moreover, the definitions of exposed environmental factors were not consistent in the studies, which could cause great heterogeneity. Our estimates were based on crude OR values, not adjusted OR values, which may yield inaccurate calculation. There were several sources bringing in heterogeneity, such as study design, age and sex distribution, and ethnicity. Meta-regression analysis was conducted to find source of heterogeneity. In the breast cancer subgroup, publication year could cause great heterogeneity and further attention was paid to years. We found all the recruited studies were carried out before 2005 or after 2010, and there were no studies between 2006 and 2009. The His allele was 29.6% in the studies before 2005 and 33.0% after 2010, which was significantly different (P = 0.02). This may be caused by the different study population, and needs more case-control studies to illustrate. In conclusion, our meta-analysis suggests that the *SULT1A1* Arg213His polymorphism may contribute UADT cancer risk. As the result was calculated through sampling statics and statistical difference is not the same as clinical difference, the result can be used for clinical reference, not for clinical diagnosis of cancer. Further detailed investigation with larger number of worldwide participants is needed to clarify the role of this polymorphism in cancer risk. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: XLY MHW WPW. Performed the experiments: JJX YBZ YHZ PZ. Analyzed the data: YBZ LXF. Contributed reagents/materials/analysis tools: JJX JGW FYS. Wrote the paper: JJX YBZ VKK.

# Introduction **T**elomeres consist of long tandem arrays of TTAGGG repeats, bound by proteins, placed at the end of linear chromosomes, which are involved in several essential biological functions., These non-coding telomeric repeats represent a buffer zone preventing the adjacent coding region of the genome from erosion. In normal human cells, telomeres decrease by some 5–20 repeats with every cell division. Therefore telomere shortening limits the number of times a cell can divide. Hence, they can regulate the onset of replicative senescence in the somatic cells.– In human cells, several pathways regulating telomeres length have been identified. The most important is regulated by telomerase, that catalyzes extension of 5′-ends of the lagging DNA strand by adding TTAGGG repeats onto the telomeres using its intrinsic RNA as template for reverse transcription. Two major subunits of the human telomerase core complex have been identified, namely h-TERC and h-TERT. The former serves as a template for telomeres elongation; instead, the latter subunit (h-TERT) contains a reverse transcriptase domain that catalyzes this reaction. The length and structure of telomeres are also controlled by a variety of proteins. Collectively, these telomeric proteins protect telomere integrity and function, connect DNA damage/repair network with the controls of cellular senescence, monitor telomere homeostasis and modify the access of telomerase to telomeres. The two major proteins are the duplex TTAGGG repeat-binding factors 1 and 2 (TRF1 and TRF 2) that are localized at telomeres. These proteins play a key role in the maintenance of telomere function and structure modifying telomerase activity.– Recent evidence shows that TRF1 interacts with other telomere-binding molecules. TRF1 accepts adenosine diphosphate (ADP)-ribosylation catalyzed by the tankyrase-poli-ADP-ribose polymerase (TANKs- PARP) complex. The ADP-ribosylation of TRF1 reduces its ability to bind telomeric DNA, allowing telomerase to elongate telomeres and extending the cellular life span.– The alteration of telomere length homeostasis affects telomere structure and leads to genomic instability by generating chromosome end-to end fusion and chromosomal abnormalities. It has been demonstrated that telomeres shortening could initiate successive events, such as aberrant fusion or recombination of the end of chromosomes, genomic instability, loss of cell growth control, and finally cancer development., The phenomenon of telomeres alteration during carcinogenesis and cancer progression is well known and established at the molecular level.– Nevertheless, studies focused on the analysis of telomere dysfunction in astroglial brain tumors are missing. The present study was designed to investigate the expression levels of a panel of genes controlling the length and structure of telomeres in human astroglial brain tumors with different grade of malignancy (WHO Grade 2–4). We analyzed telomeres length, telomerase activity and the expression levels of TRF1, TRF2, h-TERT and TANKS-PARP complex in tumor samples obtained *in vivo*, investigating the presence of a specific genes expression profile during the different stages of tumor progression, from low grade astrocytomas to glioblastoma. # Materials and Methods ## Ethics Statement The manuscript has been submitted to the Ethics Committee on February 21, 2011. The above mentioned Committee issued a formal written waiver for the need of ethics approval. All patients signed a written informed consent for the purpose of publication of clinical data, according to the internal regulation. Results were analyzed anonymously. ## Patient population Tumors samples, histologically verified as grade 2–4 astrocytomas, were obtained in adult patients who underwent craniotomy for microsurgical tumor resection. All tumors were located in the supratentorial compartment. Only patients who had undergone gross total resection (more than 95% of the tumor volume) were eligible for the study. Summary of demographic and clinical data are reported in. Samples obtained from single or multiple stereotactic biopsies were not included in the present study. We carefully excluded tumors containing components that were suspicious of oligodendroglioma. No case of recurrent tumors and no patient who underwent radio and/or chemotherapy before surgery were employed in the present study. ## Tissue Samples All tumor tissue samples were obtained from resection specimens, within 15 minutes from surgical tissue removal. Specimens were taken from viable areas of tumor, avoiding areas of gross necrosis. Three to seven anatomically separate areas of tumor tissue were sampled from each resection specimen, according to the volume of resected tissue available. Tumor samples were placed in cryovials and immediately flash-frozen in liquid nitrogen in the operating room and stored at −70°C. Tissue samples adjacent to the frozen tissue, as well as additional tissue submitted *in toto* from the resection specimens, were both used for histological typing and grading according to WHO criteria. Three samples of normal brain tissue were used as controls. Non-neoplastic brain tissue samples were derived from the temporal lobes of patients surgically treated for temporal lobe epilepsy, histologically verified as normal cortex and white matter. ## Telomere length analysis Terminal restriction fragment (TRF) length measurements in tumor specimens and in normal samples were obtained by using the *Telo* TTAGGG telomere length assay kit (Roche Diagnostics, Milan, Italy), according to the manufacturer's recommendations. The intensity of the hybridization was evaluated by densitometric analysis with Quantity One software (Bio-Rad Laboratories, Hercules, CA) and mean TRF length of a sample was estimated according to the formula as described by Harley CB at al. ## Telomerase activity Assay Telomerase activity was measured by a telomere repeat amplification protocol (TRAP) assay using TeloTAGGG Telomerase PCR ELISA plus® (Roche, Mannheim, Germany) according to the manufacturer's recommendations. The relative telomerase activity (RTA) within different samples was calculated using the following formula: RTA = (*A*_sample/*A*_sample,IS)/(*A*<sub >TS8</sub>/*A*_TS8,IS)×100%, where *A*_sample  =  absorbance of sample; *A*_sample,IS  =  absorbance of internal standard of sample; *A*_TS8  =  absorbance of control template; and *A*_TS8,IS  =  absorbance of internal standard of control template. ## Extraction of total RNA and Real-Time Quantitative PCR Total RNA was extracted from each specimen using TRIizol reagent and purified with RNA Purification kit (Rneasy Mini Kit clumns- Qiagen). The quality and quantity were checked respectively on agarose gel and spectrophotomety. Three µg of total RNA from each sample was reverse-transcribed by Archive kit (Applied Biosystems Milan, Italy). Generated cDNA was used as template for real time quantitative PCR analysis using gene expression products according to the manufacturers recommendations (Applied Biosystems). All reactions were performed with a 7300 Sequence Detection System apparatus (Applied Biosystems) to measure and compare the mRNA level of TERF-1, TERF-2, TNKS, PARP1, h-TERT, and β-actin (as an endogenous control). Relative quantification (RQ) for these genes was expressed as fold variation over control, and calculated by the ΔΔCt method, using normal brain tissue (control) as calibrators. ## Proteins extraction, electrophoresis (SDS-PAGE) and immunoblotting Frozen tumor tissues (∼50 mg) was harvested by homogenization with a Potter homogenizer in a 15 volumes ice-cold triple detergent lysis buffer. Immunoblots were probed with goat polyclonal antibody anti TRF1 and TANKS, mouse monoclonal antibody TRF2, PARP1 and β-actin (Santa Cruz Biotechnology Inc., Santa Cruz, California, USA). Following incubation with primary antibody (1∶200) at RT for 2 hours, blots were incubated with a secondary antibody: mouse- anti- goat and rabbit – anti-mouse IgG, (1∶1000; DAKO) conjugated to peroxidase at RT for 1 hour. Enhanced chemiluminescence reagents were used to visualize immunolabeling on Kodak Biomax ML chemiluminescent film. (ECL, Amersham biosciences, Little Chalfont, Buckingamshire, UK). ## Quantification of telomeric proteins Semi-quantitative evaluation of protein levels detected by immunblotting was performed by computer-assisted densitometric scanning (AlphaImager 4.2 Digital Imaging System, Italy). Different time of exposure were used for each blot (15–25 seconds), and longer exposures were performed in an attempt to detect very low levels of proteins in normal brain tissue. Data were acquired as integrated densitometric values and expressed as percentages of the densitometric levels obtained on scans from normal brain tissue used as control visualized on the same blot (ADU - Arbitrary Densitometric Unit). We used two different negative controls for each blot. ## Statistical Data Analysis Statistical analysis was accomplished using the unpaired Student t-test to compare the expression levels of mRNA of TRF1, TRF2, h-TERT, Tankyrase, PARP1 as quantified on real time RT-PCR and telomere length on southern blotting. The Spearman nonparametric correlation test was used to assess the correlation among mRNA expression levels and the nominal variables (WHO grade). Computer software programs (INSTAT \[version 3.0\] and PRISM \[version 4.0\]; GRAPHPAD, San Diego, CA and MedCalc \[version 7.2.1.0\]) were used to perform the data analysis. A probability value less than 0.05 was considered statistically significant. All values where expressed as mean ± standard deviation. All tests are two-tailed; values are expressed as mean ± standard deviation. # Results ## Telomere length in normal brain and astroglial brain tumors Telomere length varied in astroglial brain tumors with different grade of malignancy. Telomere length was 7.4±0.23 Kbs in LGGs; 6.9±0.14 Kbs in AAs; 9.2±0.5 Kbs in GBMs; 10.93±2.35 Kbs in NBT. Elongated telomere was frequently found in GBMs as compared with both AAs and LGGs (p\<.001). No statistical differences in telomere length between AAs and LGGs were observed. Interestingly, telomeres in tumor specimens were always shorter as compared with NBT. ## Telomerase activity and h-TERT expression Telomerase activity (TA) in NBT was weak or undetectable using the TeloTAGGG Telomerase PCR ELISA plus. Telomerase activity was 3.9-fold in LGAs, 15.75-fold in AAs, and 51-fold in GBMs as compared to normal control values. Telomerase activity significantly correlates with h-TERT mRNA expression and WHO grade. h-TERT mRNA expression was 1.71±0.90 in LGAs, 8.45±8.49 in AAs, 14.8±12.46 in GBMs. The expression levels in GBMs resulted statistically higher as compared with those in LGAs (P = .009). The expressions in AAs differed significantly as compared with that in LGAs (P = .04). A correlation was found between h-TERT and WHO grading (P\<.001). Conversely, no statistical differences in telomere length between h-TERT and Telomere length were observed. A correlation was also found between expressions of hTERT mRNA and PARP-1 mRNA (P = .008), and Tankyrase mRNA (P = .058). ## Telomere-associated proteins ### Down-regulation of TRF1 and 2 and TANKS-PARP up-regulation occurs along with malignant progression in astroglial brain tumours Summary of biomolecular data are reported in. TRF1 mRNA expression was 3.54±3.65 in LGAs, 0.30±0.34 in AAs; 0.19±0.22 in GBMs. The expression levels in LGA were statistically different as compared with those in GBMs (P = .007). The expression in LGAs was significantly higher than that in AAs (P = .009) ( A). An inverse correlation was found between expressions of TRF1 and WHO grade (P = .006). The TRF2 mRNA expression was 0.51±0.22 in LGAs, 0.38±0.43 in AAs; 0.16±0.19 in GBMs. The expression levels in LGAs resulted statistically higher as compared with those in GBMs (P = .008) ( B). An inverse correlation was found between expressions of TRF2 and WHO grade (P = .008). The Tankyrase mRNA expression was 1.28±0.58 in LGAs, 4.46±6.23 in AAs, 1.08±1.35 in GBMs. The expression levels in AAs resulted statistically higher as compared with those in LGAs (P = .009) and GBM (P\<.001) ( C). A correlation was found between expressions of Tankyrase mRNA and that of both hTERT mRNA (P = .03), and TRF-2 mRNA (P = .03). The PARP-1 mRNA expression was 0.97±0.59 in LGAs, 17.4±5.69 in AAs, 11.5±15.5 in GBMs. The expression levels in AAs resulted statistically higher as compared with those in LGAs (P\<.001). The expression in GBMs showed a tendency toward the higher expression as compared with LGAs (P = .07) ( D). Western blot analysis of proteins confirmed previous results. ### Possible clinical implications of telomere length, Telomerase activity and Telomere-associated proteins A statistically significant correlation between telomere length and WHO grade was observed (P\<.001). Conversely, no statistical differences between telomere length and both age and KPS, were found. A correlation was found between h-TERT and WHO grading (P\<.001). No statistical differences between h-TERT and Telomere length were observed. An inverse correlation was found between expression of TRF1 and WHO grade (P = .006), and between expression of TRF2 and WHO grade (P = .008). # Discussion Our results suggest that factors controlling telomere length are expressed at variable level in astroglial brain tumors with different grade of malignancy when compared with normal brain tissue. They also suggest the presence of a specific expression profile for different tumor grade, with telomere length depending on the balance of expression levels of different genes involved in the control of telomere maintenance. In details, up-regulation of TRF1 and 2, and shorter telomere featured LGGs, suggesting a pivotal role of these telomeric proteins in the early stage of cell immortalization. A down-regulation of TRF1 and 2, and up-regulation of both telomerase and TANK-PARP1 mainly observed in AAs and GBMs, may play a role in malignant progression of astroglial tumors toward higher malignancy levels. In order to better understand the role of telomere dysfunction, including elongation or attrition, in carcinogenesis and tumor progression, we measured telomere length in astroglial brain tumor with different grade of malignancy (WHO Grade 2–4). Terminal restriction fragment measurement showed that telomere length was reduced in astroglial tumors as compared with NBT. In GBM however, telomeres were often longer than AAs and LGGs. Telomere length is generally reduced in human tumors.– The relevance that this shortening plays in carcinogenesis has been extensively studied using knockout mice., – In the absence of genome checkpoint functions, telomere dysfunction caused by telomere shortening accelerates genomic instability, facilitating cancer initiation and progression. On the other hand, the telomere length tended to increase along with malignancy of tumors.– Gertler et al., showed that in colorectal tumors increased TRF correlated with higher tumor stage, decreased overall survival, and in a multivariate analysis, TRF was an independent prognostic factor. Our findings support the hypothesis that hypervariability of telomere length, already described in other human cancers, probably depends on the different stages of carcinogenesis and tumor progression. Hence, to ascertain this hypothesis, we evaluated, for the first time in gliomas, telomerase activity and the expression levels of the main genes involved in telomere maintenances including h-TERT, TRF1 and TRF2, and TANKs-PARP, in tumor tissues obtained during surgery and their paired normal tissue. In cancer cells telomere length depends on the balance between the loss of telomeric repeats during DNA replication and the elongation of telomeric repeats mediated by telomerase. In most normal human somatic cells, telomerase activity is weak or undetectable. As in almost all tumors, malignant brain tumors are associated to higher telomerase activity than benign tumors, such as neurinomas, meningiomas or normal brain tissue. Increased telomerase expression has been also associated with higher proliferative index, tumor grading, age, vascular and endothelial proliferation, poor outcome.– In our investigation, both telomerase activity and h-TERT mRNA levels varied in astroglial brain tumors and both were significantly correlated with tumor grading. On the other hand, no statistical correlation between h-TERT and telomere length was observed. Similar results have also been reported in other studies, suggesting that a telomerase-independent mechanism for the regulation of telomere lengthening possibly exist in these telomerase-negative tumors. Recent studies indicate that telomere-associated proteins can regulate telomerase accessibility in either positive or negative ways, suggesting a role in telomere maintenance. The two major telomere- binding proteins are TRF1 and TRF2. Both may function individually or by interacting with other binding proteins, such as tankyrase, TIN2, hRap1, Mre11/Rad50/Nbs1 DNA repair complex, Ku86., Recent in vitro studies indicate that, overexpression of TRF1 in tetracycline-responsive human fibrosarcoma cell line HTCC75 resulted in a gradual decline in telomere length and it has been reported that the forced tethering of a large number of TRF1 molecules to a single telomere induce a significant shortening of telomere length. These findings suggest that in human cancer cells an upregulation of TRF1 may results in a progressive telomere shortening. Our results documented a differential telomere-associated proteins expression in astroglial brain tumors of different grade compared with normal brain tissue. An inverse correlation between TRF1 and 2, and WHO grade was also found. Several studies, dealing with correlation of TRFs expression and telomere length in human cancers, also confirm our results. Oh et al. documented that up regulation of telomere-binding proteins, TRF1, TRF2, and TIN2 is related to telomere shortening during human multistep hepatocarcinogenesis. These findings support the hypothesis that an up regulation of the telomeric binding proteins, inhibiting telomerase, results in a progressive telomere shortening and may play a role in immortalization of cancer cells. Our findings reveled a weak TA In low grade tumors, TRF1 and TRF2 mRNA is up-regulated, while TA was weak and h-TERT, tankyrase, and PARP resulted absent or minimally expressed. These data, also, confirm our previous observation that glioma cells express higher levels of TRF1 as compared with normal brain tissue., As expected, telomere length in such tumors was shorter as compared with normal brain tissue. In high grade tumors h-TERT, PARP and tankyrase are over-expressed with an increased telomerase activity. These findings, are consistent with the evidence of longer telomeres in those tumors. # Conclusions Our findings, are consistent with the hypothesis that in astroglial brain tumors, up-regulation of TRF1 and TRF2 may occur in the early stages of carcinogenesis when telomerase has not yet been activated or is down-regulated by TRF1. This stage is characterized by short telomeres, genomic instability, low proliferative rate and prolonged life span, that are typical biological behaviour of LGGs. Later, telomere dysfunction caused by telomere shortening accelerates genomic instability, facilitating cancer initiation and progression. At this stage, an up-regulation of PARP-Tankyrase complex and telomerase activation may occurs. The ADP-ribosylation of TRFs, mediated by PARP1, reduces its ability to bind to telomeric DNA, allowing telomerase to elongate progressively telomeres and extending cellular life span. Down-regulation of TRF1 and TRF2, increasing telomerase activity, persistent over-expression of PARP-TANKs and elongated telomere are typical features of GBMs. Elucidation of additional telomerase components and associated proteins will certainly contribute to further investigations of the effect of telomerase in telomeres elongation, telomere length maintenance, oncogenesis, and new unidentified cellular functions. As far as we know, no other studies regarding these issues in astroglial brain tumors are available in the literature, so further studies should be performed to better understand the pathways involved in the telomeres length maintenance and, consequently, in the process of carcinogenesis and malignant progression of human astroglial brain tumors. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: DLT. Performed the experiments: DLT AC MA SR MGDP CT. Analyzed the data: DLT AC FFA SC CA SR. Contributed reagents/materials/analysis tools: MA SR MGDP CT CA. Wrote the paper: DLT AG AC.

# Introduction Accurate assessment of respiratory muscle strength is clinically important in patients with neuromuscular disorders (NMD) or unexplained dyspnea. Measurement of maximum inspiratory pressure (MIP) at the mouth is the most commonly employed test to evaluate inspiratory muscle function, as it is non-invasive and relatively convenient. However, particularly in patients with NMD, MIP suffers from the possible occurrence of falsely low values due to difficulties in maintaining an effective mouth seal or sustaining a maximal inspiratory effort. For these reasons, sniff nasal inspiratory pressure (SNIP) has been used as an alternative non-invasive test of inspiratory muscle function which is easier to perform for many NMD patients. It can be employed to monitor inspiratory muscle strength over time in NMD, and a normal SNIP can also effectively rule out inspiratory weakness in individuals with spuriously low MIP values. In amyotrophic lateral sclerosis (ALS), the SNIP has been reported as the best prognostic indicator. The SNIP measurement entails brief maximal sniff efforts by the patient during simultaneous intranasal pressure recordings within a nostril that is sealed by a snugly fitting plug containing the pressure-sensing probe. The sniff maneuver has long been used in the assessment of diaphragm function. It was initially described for radiological assessment of unilateral diaphragm weakness. Subsequently, sniffs were found to be a representative approximation of phrenic stimulation in studies of diaphragm contraction. The sniff maneuver is now commonly used when diaphragm strength is being assessed by measuring transdiaphragmatic pressure (Pdi). The SNIP, in turn, has been devised as a less invasive alternative. In the original description of SNIP measurements, the contralateral nostril remained unobstructed or open (henceforth referred to as SNIP_OP). Under these conditions, a reliable SNIP_OP value presumably requires inspiratory collapse of the contralateral nasal valve in order to allow for quasi-equilibration of intrathoracic and nasal cavity pressures. The SNIP can also be measured as a static maneuver with the contralateral nostril closed (SNIP_CL), as reported in a much smaller number of studies. Although they appear to be used and reported interchangeably in the literature, it is unclear if SNIP_OP and SNIP_CL in fact produce the same results. Theoretically, they could be very similar in healthy subjects, but individuals with inspiratory muscle weakness may be unable to generate sufficient negative inspiratory pressure to collapse the nasal valve when performing SNIP_OP. Therefore, SNIP_OP might poorly reflect the actual negative intrathoracic pressure values and thus provide inaccurate information about the true level of inspiratory muscle strength in some NMD patients. In the present study, we sought to determine whether there are any systematic or clinically significant differences between values of SNIP_OP and SNIP_CL in patients with known NMD as well as in patients without clinical evidence of respiratory dysfunction. Our primary hypothesis was that values of SNIP_CL would be significantly higher than SNIP_OP in NMD patients. The secondary hypothesis was that in patients with a reduced MIP value, SNIP_CL would give results within a normal range (suggesting an absence of respiratory muscle weakness) more often than SNIP_OP. # Materials & methods ## Study subjects The two groups of study subjects consisted of: 1) NMD patients recruited from a home non-invasive ventilation program, and 2) a control group comprised of individuals with obstructive sleep apnea (OSA) and no known NMD or significant lung disease who were also participants in a Pompe disease screening study. The NMD diagnosis was obtained from the medical record, as made by a clinical neurologist. Two patients in the initial control group were identified as having a NMD and thus transferred into the NMD group. In addition, several control group patients (n = 8 with asthma or chronic obstructive pulmonary disease (COPD), n = 5 with morbid obesity, n = 7 with other) had abnormal spirometry (FEV1 or FVC \< 80% of predicted, or FEV1/FVC ratio \<70%) and were thus excluded. Subjects were recruited between September 2013 and November 2014, and provided their written informed consent. Participants, none of whom were hospitalized at the time of testing, took their usual medications without modification. All study subjects underwent spirometry and respiratory muscle strength measurements as outlined below, all in the sitting position, and all within a single testing session between 10 am and 2 pm. The study was approved by the institutional Research Ethics Board of the McGill University Health Center (13-379-BMB). ## SNIP measurements Both SNIP_OP and SNIP_CL were performed using a commercially available device (MicroRPM, VIASYS Healthcare, Hochberg, Germany) with disposable nasal probes. Factory-set calibration of the device was verified using a manometer. The nostril that appeared most patent clinically was chosen for insertion of the nasal probe and the appropriate nasal probe size was verified by ensuring the absence of air leak during sniffs. Without a prior training period, the patient was asked to perform short, sharp sniffs of maximal intensity from functional residual capacity (FRC) in the sitting position with the mouth closed. Normal breathing was allowed between trials. At least 10 trials were done in total: five sniffs with contralateral nostril open (SNIP_OP), and five with the contralateral nostril closed (SNIP_CL). Half of the participants performed SNIP_OP first, whereas the reverse order was used in the other half, to account for any potential learning or order effect. The highest value for each SNIP method is reported for each individual. A single research assistant performed all testing. ## Standard PFT measurements Spirometry was performed (Jaeger FlowScreen V2.6.0, Carefusion Corp, San Diego, CA) to determine forced expiratory volume in one second (FEV1), forced vital capacity (FVC), and peak expiratory flow (PEF) according to ATS guidelines and established reference values. Supine spirometry was subsequently similarly performed. Wheelchair-bound participants who could not easily transfer did not have supine measurements, unless their wheelchair tilted to at least a 30 degree recline. MIP was measured through a flanged mouthpiece from residual volume (RV). The highest of at least three consistent values was recorded as recommended. Reference values were taken from Vincken et al.. Individuals with values reaching the upper saturation limit of the MIP manometer (≥150 cmH₂O) were excluded from the analysis (n = 5 from the control group) to avoid a ceiling effect which could introduce error into the analyses. ## Statistical analysis Unpaired t-tests were used for comparisons of baseline characteristics between NMD and control groups. Descriptive statistics are presented as mean and standard deviation (SD), unless specified otherwise. Normality of outcomes data was tested using the Shapiro-Wilk test. Control data were normally distributed but not NMD data. Therefore, we used ANOVA to compare the SNIP_OP, SNIP_CL and MIP measurements in controls, and (nonparametric) Friedman ANOVA for the NMD group. Intraclass correlation coefficients (ICC) were performed for combinations of SNIP_OP, SNIP_CL and MIP within groups. Scatter plots and Bland-Altman plots were generated, and bias was defined as the mean of the differences between two measurement values. Limits of agreement were calculated as bias +/- (1.96 x SD for the difference). The Fisher exact test was used to compare counts. Statistical significance is defined as p\<0.05. Analyses were done using SAS software, version 9.3. ### Power calculation Our sample size was based on a detectable difference between SNIP_CL and SNIP_OP of 10 cmH₂O, which we considered the minimum that would be relevant, and assumed a normal distribution. For a sample size of 50 patients (in each group separately), using a paired t-test and conservative estimate for the standard deviation of the difference of 20 cmH₂O, we would have a power of 93% to detect a difference of 10 cmH₂O with type I error of 0.05. # Results shows demographic and PFT data for the 52 NMD patients and 52 control subjects included in the study. The two groups did not differ with respect to age, although the control group tended to include more females and had a higher average body mass index. NMD patients demonstrated mild to moderate reductions in spirometric values, which were significantly lower than the control group (73.1% vs. 98.6% of predicted for FVC, p\<0.001). shows MIP, SNIP_OP and SNIP_CL values in the control and NMD groups. As expected, all values were significantly lower in NMD compared with control subjects (p\<0.001). Neither age nor sex correlated with SNIP_CL values in the control and NMD groups, whereas age was weakly correlated with SNIP_OP (r = 0.295, p = 0.03) in the NMD group only. The mean SNIP_OP value was significantly lower than SNIP_CL and MIP in both groups. Results were identical irrespective of the order in which SNIP_OP and SNIP_CL were performed (results not shown). Scatter plots demonstrating the relationships between these parameters in individual patients are shown in. To assess agreement between measurements, ICC was calculated between SNIP_OP, SNIP_CL and MIP. Agreement was poorer in the control subjects than in the NMD group for all combinations of measures. In both groups the highest agreement was for SNIP_OP vs. SNIP_CL, and SNIP_CL was in better agreement with MIP than SNIP_OP. Agreement and bias were further assessed using Bland-Altman plots. These plots indicate that SNIP_CL is greater than SNIP_OP for the majority of subjects, with a mean bias of -15.04 in NMD and -19.9 in controls. Moreover, the bias between SNIP_CL and MIP was substantially lower (consistent with better agreement) than between SNIP_OP and MIP. This was true for both NMD and control groups, although the limits of agreement are narrower (less scatter) for NMD compared with control subjects. Visual inspection of the plots also suggests less scatter at lower values, particularly in NMD patients. To assess whether SNIP_CL might be more useful clinically than SNIP_OP to rule out inspiratory muscle weakness in subjects with reduced MIP, we determined how often SNIP_OP or SNIP_CL were higher than MIP in individuals with a low MIP value, as determined using three different thresholds for MIP. For MIP \< 80% of predicted, SNIP_CL was higher than MIP more frequently than SNIP_OP in NMD patients (40% vs. 14%, p = 0.03). For MIP \< 60% of predicted, SNIP_CL was also more often higher than MIP as compared to SNIP_OP in NMD patients (48% vs. 10%, p = 0.02). In control subjects, MIP was \< 80% of predicted in 4 subjects and \< 60% of predicted in 1 subject. The latter subject had both SNIP_OP and SNIP_CL higher than MIP, while another control subject had a SNIP_CL (but not SNIP_OP) higher than MIP. Finally, we assessed subjects with MIP \<80 cmH₂O, selected because this represents a threshold value above which clinically significant inspiratory muscle weakness is considered to be highly unlikely. SNIP_CL and SNIP_OP were higher than MIP in 44% and 14% of 36 NMD patients with MIP falling below this threshold, respectively (p = 0.003). In controls, this occurred in 69% vs. 31% of 13 subjects, respectively (p = 0.12). Lastly, we evaluated how often SNIP_OP or SNIP_CL values fell within the normal range in subjects with reduced MIP. The recommended lower limit of normal (LLN) for SNIP is 70 cmH₂O for males, and 60 cmH₂O for females. In NMD patients with MIP\< 80 cmH₂O, 1 subject had a SNIP_CL value \> LLN, whereas this did not occur for SNIP_OP in any NMD patient. In controls with MIP \< 80 cmH₂O, 3 subjects had both SNIP_OP and SNIP_CL values \> LLN, while 4 had only SNIP_CL \> LLN and none had only SNIP_OP \> LLN. # Discussion Although MIP is the most widely used test of inspiratory muscle strength in standard clinical practice, it is clear from previous work that the use of a single test such as MIP tends to overdiagnose weakness. SNIP has thus been recommended as a complementary test to help address this issue, particularly in NMD patients. However, SNIP_OP and SNIP_CL have been utilized in a seemingly interchangeable fashion by different investigators to assess inspiratory muscle strength. Moreover, very few studies in the literature have actually reported on the use of SNIP_CL to evaluate inspiratory muscle function. To our knowledge, this is the first study to formally compare the two methods of SNIP measurement in NMD patients as well as control subjects with normal inspiratory muscle strength. Our main findings are that SNIP_CL values are systematically greater than SNIP_OP in both NMD and controls, and that the level of agreement with MIP is also superior for SNIP_CL in comparison to SNIP_OP. Therefore, in patients with a low MIP value, SNIP_CL appears to be a more useful test than SNIP_OP for excluding inspiratory muscle weakness. Both MIP and SNIP have a learning effect and are operator-dependent, but several aspects of SNIP may be more advantageous in NMD patients. The SNIP requires only a short burst of maximal inspiratory muscle contraction, whereas the MIP involves sustaining a maximal inspiratory effort for at least 1 second. This more prolonged effort required for MIP may be difficult for some patients, resulting in falsely low values. Furthermore, in principle SNIP can be performed in individuals who are unable to maintain a tight lip seal around a mouthpiece, which is frequently the case in NMD. The maneuver required for SNIP is also generally regarded as more natural and easier to explain to patients. In keeping with the above, SNIP was reported to be more predictive of outcomes than MIP in ALS and Guillain-Barré syndrome. SNIP_OP has been reported to be higher than MIP in some studies, and this appears to be more prevalent in those individuals with the least amount of weakness. Conversely, Hart et al. found, in a group of NMD patients, that MIP was greater than SNIP_OP (4.8 cmH2O bias with both tests performed from FRC), and SNIP_OP was lower as a proportion of MIP in those patients with the most severe impairment. We speculate that the above findings are at least partly explained by an inability of very weak patients to generate a sufficiently negative inspiratory pressure to collapse the nasal valve within the open nostril during SNIP_OP measurements. In contrast, by occluding the nostril during SNIP_CL, the measurement becomes a static one such that pressures are more readily equilibrated throughout the airways. It is interesting to note that SNIP_OP (but not SNIP_CL) was also lower than MIP in the control subjects of our study, suggesting that additional factors other than weakness are involved in the better equilibration of pressures achieved with SNIP_CL. One potential factor could be the presence of airflow obstruction at the lower airway level, although this appears unlikely since we excluded individuals with abnormal spirometry in our control group. However, obstruction could occur at the upper airway (e.g., nasal) level, which we did not assess. A possibility also exists that the pattern and/or level of inspiratory muscle recruitment differs between SNIP_CL and SNIP_OP. This might also help to explain a closer correlation between SNIP_CL and MIP, since the two are similar in being "static" in nature compared to the more "dynamic" SNIP_OP. The specific maneuver itself, i.e., sniff vs. Mueller, is also an important element in determining inspiratory muscle recruitment, with brief sniffs generally producing higher values of diaphragm activation and transdiaphragmatic pressure than the inspiratory maneuver employed for MIP. However, SNIP_OP may conversely generate lower pressures than static maneuvers due to shortening of inspiratory muscles and the attendant pressure-velocity relationship. In a clinical context, these sources of variability are difficult to ascertain, but the tests are not interchangeable and should be viewed as complementary. ## Critique of methods It should be noted that our study design contains several elements which accurately reflect the routine clinical evaluation of NMD patients but may also introduce increased variability in the measurements. For example, NMD patients were comprised of a heterogeneous group of diagnoses with different levels of weakness. In this regard, Terzi et al. previously reported much wider limits of agreement between SNIP_OP and MIP in myotonic dystrophy than in Duchenne muscular dystrophy. In addition, our control group was a clinical one rather than being composed of entirely healthy volunteers, although it should be emphasized that all control group subjects had normal spirometry. The MIP was initiated from RV as per standard clinical practice and American Thoracic Society recommendations, whereas SNIP was measured at FRC according to the original description of the technique. These lung volume differences would be expected to result in a small (less than 10 cmH₂O) change in inspiratory force generation, which is quite consistent with the average magnitude of MIP minus SNIP_CL differences found in our study. Given that the study subjects did not undergo any prior training period, one possible limitation of the current study might be insufficient learning of the procedure. However, this appears unlikely since results were similar regardless of which test was performed first. Finally, as noted above we did not objectively measure nasal resistance, which can also affect SNIP reliability. It is important to emphasize that although SNIP test result variability may have been increased by one or more of the above factors, our findings are likely more generalizable to real world clinical practice for the very same reasons. In addition, since technical measurement errors often underestimate but are very unlikely to overestimate muscle strength, respiratory muscle pressure generation is primarily used as a “rule out” test for muscle weakness. Accordingly, our study suggests that the use of SNIP_CL in this manner may help to prevent clinical misclassification of certain patients who might otherwise be considered as having significant inspiratory muscle weakness based on low values for either MIP or SNIP_OP. # Conclusions The SNIP_CL maneuver produces values which are systematically higher than SNIP_OP and therefore likely represents a more useful test for ruling out inspiratory muscle weakness. Accordingly, we propose that whenever MIP is low or cannot be performed, SNIP_CL should be used to obtain further information on inspiratory muscle strength. Clearly, the use of different tests of respiratory muscle strength should be considered complementary in nature as previously suggested by others. # Supporting information We would like to thank Pei Zhi Li for her work on data analysis and figures. [^1]: Basil Petrof has received an investigator-initiated grant and speaker fees from Sanofi Genzyme Inc. This does not alter our adherence to PLoS One policies on sharing of data and materials. [^2]: **Conceptualization:** BP MK. **Data curation:** FN MK. **Formal analysis:** MK FN. **Funding acquisition:** BP. **Investigation:** BP. **Methodology:** BP MK. **Project administration:** BP MK. **Resources:** MK BP. **Supervision:** MK BP. **Validation:** MK FN. **Visualization:** MK BP. **Writing – original draft:** MK. **Writing – review & editing:** BP MK.

# Introduction Plant communities are known to be shaped by competition for resources such as light, space, nutrients, and water. Differences in species' abilities to capture or alter these resources across space and time can drive competitive hierarchies, successional dynamics, community diversity, and invasions. One of the best known attempts to describe these complex interactions with a simple set of rules is R\* resource competition theory. In this theory, Tilman posits that plant species differ in their ability to deplete a limiting resource, and the species that maintains a positive growth rate at the lowest level of this resource will emerge as the competitive dominant. Following from Tilman, R\* is the equilibrium concentration of resources necessary for the consumer (plant) species to maintain a stable population. Although seemingly simplistic, this theory has successfully predicted the outcome of plant competition in various ecosystems, including California grasslands, alpine meadows, and Midwestern prairies. Part of R\* theory's success stems from providing clear predictions and testable hypotheses for ecologists. For example, it is relatively simple to quantify differences in resource depletion among species by planting monocultures and measuring resource concentrations after biomass stabilizes (R\*). Determining whether R\* values predict competitive hierarchies and structure communities can be accomplished by comparing these monoculture R\* values with the relative abundance of those species in mixtures (i.e. after competition). A negative relationship between species' R\* and their relative biomass in mixture supports R\* theory, since the species depleting resources to the lowest levels are the competitive dominants. An open question is, what drives species-specific R\* values in the first place? Differences in species R\* levels for a soil nutrient may be due to differences in plant uptake, differences in how plants affect microbial nutrient cycling and sequestration, or both. However, the relative contribution of plants and microbes to the observed resource depletion remains largely untested. R\* theory in terrestrial ecosystems has largely been tested in perennial communities or perennial communities subjected to annual invaders,. In such systems it is difficult to disentangle over multi-year timescales the relative contributions to R\* of plant uptake, microbial nutrient cycling, and the indirect effects of plant traits such as litter quality and other plant-soil feedbacks. Plants are often assumed to be the major drivers of R\*. For example, in systems limited by nitrogen (N), depletion has largely been assumed to be due to plant N uptake and sequestration in plant tissues coupled with litter-mediated feedbacks. Plant species can also alter microbial community composition, size, and activity, and thus alter available soil N concentrations through differences in plant traits such as quality of root exudates or root and leaf litter. These effects of plant species on microbial N cycling rates can occur over short time scales. Carbon-rich root exudates from grasses, for example, have been shown to increase microbial respiration and N turnover in rhizosphere soils within 24 hours, resulting in higher N uptake by grasses. However, microbes are also strong competitors for N in soils, given their high surface area. Microbes are integral in the cycling of N in soils by breaking down large organic N molecules into smaller compounds (e.g., breaking proteins into amino acids and ammonium), and converting ammonium into nitrate. While microbial biomass is typically not large, it turns over rapidly. As microbial biomass turns over, N from microbial “necromass” can become less available to plants and microbes as it is incorporated into soil organic matter and is protected from decomposition by a range of chemical, physical, and biological mechanisms. To test the extent to which plant or microbial controls on N availability underlie the R\* theory of plant competition, we established an experiment in a California grassland ecosystem. We planted annual grass and forb species in monoculture and mixture plots and followed them over one growing season. Our focus on an establishing annual system is very different than the majority of work examining R\* theory, most of which has been done over multiple years in perennial systems and does not explicitly examine the extent to which plants or microbes are driving the patterns. Although it has been previously established that plant-microbial feedbacks mediated by litter are associated with species' R\* after several years in perennial systems, this has not been studied mechanistically (i.e., by explicitly examining the plant and microbial N pools and fluxes). Moreover, in the first year of community establishment in an annual system, such feedbacks are unlikely to be important drivers. Thus, examining an annual system over a single growing season allows for a detailed examination of short-term impacts of plants on microbial activity, and microbial activity on plant growth and nutrient acquisition, all in the relative absence of litter mediated feedbacks. As such, this study represents a first step in describing the mechanisms that drive patterns of resource competition in R\* theory. We have previously reported that the available soil N levels in monocultures in this experiment were correlated with the competitive hierarchy among plant species, and thus, that R\* for N (and therefore, competition for nitrogen) strongly influences community dynamics in this system. The importance of competition for N is not surprising, given that grasslands in this locality have been shown previously to be N-limited. Here we examine the degree to which plant and microbial controls drive N levels in soils, and directly or indirectly influence plant-plant competitive dynamics. A range of different relationships could emerge between plants, microbes, and R\*. If differences in R\* are driven primarily by plant N uptake, we would expect to see a strong negative relationship between plant N uptake and dissolved inorganic N (DIN), our metric of R\*. Furthermore, we would expect no relationship between microbial N cycling and DIN. However, if differences in R\* are driven primarily by changes in microbial N cycling rates (*e.g.*, net N mineralization), we would expect that microbial N cycling would be positively correlated to DIN, while plant N uptake would have no relationship with DIN. If differences are driven by both differences in plant uptake and in microbial N cycling, we would expect a pattern somewhere in between the plant dominated and microbe dominated patterns with a slight positive correlation between microbial N cycling and DIN, and a slight negative relationship between plant N uptake and DIN. While we expected that both direct plant uptake of DIN and microbial N cycling would contribute to R\*, we were interested in where on the continuum of plant to microbial control this annual ecosystem would fall. Across the different monoculture and bare plots, we measured: 1) pool sizes of total N in plants; 2) DIN at the end of the growing season (our measure of R\*) and over time; 3) microbial pools and process rates (microbial N, microbial biomass, net N mineralization, gross NO₃⁻ production, and nitrification potential). To tease out drivers we then examined relationships in bare and monoculture plots between DIN and: plant biomass; microbial biomass and biomass N; and microbial process rates. # Materials and Methods ## Site and Experimental Design Experimental plots were established in a pasture/grassland with permission from the owner, the Midland School, in Los Olivos, California. The climate is Mediterranean, with hot, dry summers and cool, wet winters and an average annual rainfall of 550 mm. The land had not been plowed or tilled since 1940, but these soils were not undisturbed; gopher activity causes extensive physical turnover of surface soils in California grasslands, and has been estimated to give complete turnover of surface soils every 3–5 years. Plots have also been grazed by cattle annually, presumably also causing soil disturbance. The vegetation is a mixture of annual and perennial herbaceous forbs and grasses growing with occasional oak trees. Soils are typic Argixerolls, with gravelly fine sandy loam texture. Study species included three native annual forbs (*Amsinkia menziesii, Calandrinia ciliata, Clarkia purpurea*), three native annual grasses (*Muhlenbergia microsperma, Vulpia microstachys,* and *Vulpia octoflora*), and six exotic annual grasses (*Avena barbata, Bromus hordeaceous, Hordeum murinum, Lamarkia aurea, Polypogon monspielensis* and *Vulpia myorus*). All species were present in nearby grasslands (Stan Harpole, personal communication) and thus our study plants reflect a realistic representation of current California annual species. All data were collected in winter, spring, and summer of 2006. Five ‘blocks’ were established within the grassland and were separated by 50 to 500 m. Vegetation was cleared by spraying with Roundup in fall 2005 and roto- tilling with a tractor two weeks after plants had died. Within each block, fourteen 0.64 m² plots were established separated by 1 m buffers, and plots were randomly assigned treatments (species monocultures, mixtures, or bare plots). There was one full replicate of the experimental design within each block, consisting of one bare, one mixture (with all 12 species), and 12 monoculture plots, giving a total of five replicate plots for each plot type between the five blocks. Each block was fenced to exclude cattle grazing (see for further details). Data from mixture plots was used to establish the R\* relationship in HilleRisLambers et al., by relating relative biomass in mixture to DIN in monoculture plots. Because we were trying to determine what drives DIN concentrations in the monoculture plots used to test R\* theory, we focused on monoculture plots only for this study. To establish plots, we collected seeds locally (*A. barbata, B. hordeaceous, H. murinum, L. aurea*, and *V. myorus*) or used seed from local seed companies (*A. menziesii, C. ciliata, C. purpurea, M. microsperma, P. monspielensis, V. microstachys*, and *V. octoflora*). We added a total of 15 grams of seed/m² to each plot, with that weight being divided equally among the species in mixture plots. Seeds were added at the beginning of the growing season (late November 2005) and plots were watered with a volume equivalent to 3 inches of rainfall to encourage germination and establishment. We weeded plots twice to remove nontarget species, once soon after germination and once midway through the growing season. Seeding amounts in monoculture and mixture plots were sufficiently high to generate plots with little bare ground visible for all but two species (JHRL, personal observation), including *M. microsperma*, which failed to germinate. *M. microsperma* thus served as a type of bare plot, and may be considered as a more appropriate control because it had an initial N input from seed, similar to other planted plots. We refer to it as bare (seeded) in the rest of the manuscript and figures. ## Soil N Pools and Dynamics We collected all soils immediately after plant biomass harvest. We sampled soil by extracting two 5 cm diameter, 10 cm depth soil cores from each plot. From each of the five replicate plots for each species, duplicate cores were sieved (2 mm mesh size) prior to soil and microbial analyses (see *Microbial N Pools and Process rates*, below). Dissolved inorganic N (DIN hereafter) was quantified by extracting ammonium and nitrate (NH₄⁺ and NO₃⁻) with 2M KCl which were then analyzed on a Lachat flow injection autoanalyzer (Lachat Instruments; Loveland, USA), while soil moisture was determined gravimetrically after six days drying at 60°C. DIN was our measure of R\*. Total soil N pool (which includes DIN, as well as organic N) was measured on dried soils after they were ground in a ball mill (Wig-L-Bug amalgamator, Crescent Dental, Lyons IL), after which Carbon (C) and N content were measured on a Carlo Erba NA 1500 CHN analyzer (Fisons Instruments, Beverly, MA). Soil bulk density was measured in June by collecting five soil cores randomly placed within plots in each block (4 cm diameter, 10 cm deep). These were returned to the lab, dried for 48 h at 60°C, rocks were removed, and remaining soil weighed. Block-specific bulk densities were used to convert all soil measures to g m⁻² unit values to enable comparison of N budgets between soils, plants, and microbes. Outcomes of our statistical analyses on soil N pools did not differ given different N measurement units (i.e., g N m⁻² versus g N kg⁻¹ soil). To determine whether our R\* measure (DIN pools in May) was representative of N dynamics over the growing season, we used resin stakes (PRS Probes; Western Ag Innovations, Saskatoon, Canada) to track inorganic N levels in soils throughout the experiment. Plastic stakes with ion exchange resin surfaces were placed in 0–5 cm soils in mid-month February 2006, and switched each month thereafter until mid-month June 2006. Thus the May sample date represents resin-available N sampled between mid-April and mid-May. Excess soil was washed off of probes with DI water. Probes were then extracted in 0.5 M HCl, and extracts were analyzed for NH₄⁺ and NO₃⁻ as described above. Regenerated stakes were placed into the same slots in the soil as previous stakes. ## Plant N Pools Plant biomass was harvested in mid-May for all species; this timing was chosen based on close monitoring of the diverse phenologies of the species in the experiment. At this time, all species had reached peak biomass and set seed, but had not senesced. Aboveground plant biomass was sub-sampled in each plot by clipping all vegetation within one 10×50 cm quadrat. Roots were collected from duplicate soil cores collected within the clip plots by sieving soils (see above in *Soil N Pools and Dynamics*) in each of the five replicate plots for each vegetation type. All plant material was dried at 60°C for six days and weighed. Plant material was ground with a Wiley mill (Thomas Scientific, Philadelphia, PA) followed by a ball mill. Plant samples were then analyzed for C and N on the CHN analyzer. Total plant N (g N/m²) in monoculture plots was then quantified as (aboveground biomass/m² × shoot % N/100) + (belowground biomass/m² × root % N/100). ## Microbial N Pools and Process Rates Prior to examining microbial biomass and N cycling, sieved soils were adjusted to 35% water holding capacity (WHC) to eliminate the confounding effects of variable water content on process rates. Soils were allowed to equilibrate for 7 days to allow the effects of sieving and drying/rewetting to pass, after which soils were weighed out for all analyses. Lab replicates were not included given that soils were sieved, there were five field replicates per treatment, and past work using these same techniques in soils from this area showed tight agreement among lab replicates. To examine how microbial biomass N relates to R\*, we used a chloroform slurry extraction to examine the flush of N released from microbial cell lysis and extraction. Briefly, 25 mL of 0.5 M K₂SO₄ was added to each glass tube containing a 4 g soil sample. To this, 0.5 mL of EtOH-free chloroform was added. Tubes were sealed with PTFE lined caps, and shaken for 4 h at 150 rev min⁻¹ on an orbital shaker. Tubes were allowed to settle for 10 minutes allowing the bulk of the chloroform and soil slurry to separate. The top 10 mL of extract was filtered through a Pall A/E glass fiber filter (Pall Corporation, Port Washington, NY, USA), then bubbled with air for 20–30 minutes to remove any residual chloroform. Chloroform slurries were compared to slurries extracted without chloroform. To test the influence of microbial abundance on R\* we used substrate induced respiration (SIR), an index of microbial biomass. Briefly, 10 mL of a 12 g L⁻¹ Difco Yeast Extract (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) was added to 4 g soil. Soils were capped and shaken for four hours with 1 mL headspace CO₂ subsamples withdrawn at 20 minutes, two hours, and four hours. Gas samples were analyzed on a LI-6262 infrared gas analyzer (LI-COR, Lincoln, NE, USA). The SIR biomass was then calculated by applying a linear fit to the change in CO₂ over time (average r² = 0.99). Because microbial biomass turnover rates are faster than those of annual plants, microbes may have driven inorganic N availability in soils via process rates that are not reflected in microbial biomass pools. Thus, we also quantified potential net N mineralization, gross NO₃⁻ production, and nitrification potential. Net N mineralization is the net accumulation of extractable NH₄⁺ and NO₃⁻ in soil, and represents the balance between microbial production and consumption. Gross nitrification separates microbial production and consumption of NO₃⁻, and offers insight into nitrifier abundance and ambient substrate (NH₄⁺) availability. In contrast to gross rates, nitrification potential assays add NH₄⁺ in excess and thus rates are driven by—and serve as an index of—the abundance and potential activity of ammonia oxidizers. Net accumulation of inorganic N is associated with conditions where heterotrophic microbes are not limited by N availability, while high gross rates of nitrification and high nitrification potential are associated with high availability of NH₄⁺ and NO₂⁻ to autotrophic nitrifiers. For net N mineralization, two sets of 4 g samples were used, with one set extracted on the first day of incubation for one hour with 25 ml of 0.5 M K₂SO₄, while the other set was incubated for 60 days at 20°C prior to extraction. Net N mineralization was then calculated as the difference in extractable NH₄⁺ and NO₃⁻ in soils between the initial and final time points. These rates can be seen as potential rates because they were held at optimal moisture and temperature for microbial activity. Our method has the advantage of controlling for microclimate variability that may affect mineralization rates, but is not expected to perfectly match field rates. To quantify gross nitrification, we took duplicate 4 g samples of soil at 35% WHC and added 0.25 ml of 25.5 mg N/L (98% enriched ¹⁵N-NO₃⁻) stirring with the pipette tip after addition. Samples were then extracted for one hour with 25 ml of 0.5 M K₂SO₄, with one sample extracted at 15 min, and the other sample at 24 h. After NO₃⁻ was measured, extracts were prepared for isotope ratio mass spectrometry. First, NH₄⁺ was converted to NH₃ with MgO and driven out of solution. Then, NO₃⁻ was converted to NH₃ by Devarda's Alloy in the presence of MgO, and the NH₃ was then captured on acidified filter disks in teflon packets using the methods of Sørensen and Jensen. Filter packs were removed and placed in a desiccator with DrieRite and concentrated H₂SO₄ to dry filters and trap any free NH₃. Filters were folded into tin capsules, and ¹⁵N enrichment was measured at UCSB's Marine Science Institute Analytical Lab using a Thermo- Finnigan MAT Delta+ Advantage (Thermo Fisher GmBH, Dreieich, Germany). Gross nitrification rates were calculated following Hart et al.. We aimed to measure gross N mineralization in addition to gross nitrification, however NH₄⁺ pools in soils were too low, and consumption too high, such that it precluded quantifying gross N mineralization. Nitrification potential was measured by adding 35 ml of a stock solution containing 8 mL 0.2 M K₂HPO₄, 1 ml 0.2 M KH₂PO₄, 5 ml 0.2 M (NH₄)₂SO₄, and 20 ml of 1 M NaClO₃ (added to block the activity of nitrite oxidizers;). The rate of appearance of nitrite was measured at three time points over 4 h using a modified Griess- Illosvay reagent, with 1 ml of color reagent added to 4 ml of sample. ## Statistical Analyses We asked what the patterns in DIN were over the course of the growing season under our different monoculture types (i.e. plant community composition, time) by fitting the resin N data with five linear generalized mixed effects models (with block as random effect in all), including: a null model; a model with only species identity as explanatory variable; a model with only time as the explanatory variable; a model with both time and species identity as explanatory variables; and a model with the main effects of time and species identity as well as their interaction. Specifying block as a random effect with time as an explanatory variable is analogous to a repeated measures ANOVA. The log of resin N values was used as our response variable to ensure normality. We used AIC's to determine the best-fitting of the five models, and the appropriate likelihood ratio tests to determine the significance of explanatory variables retained in the best fitting model. We also used resin-N data to test whether our one time measure of soil DIN represented soil dynamics across the season, by correlating species-specific cumulative resin N values for each species (over the entire growing season) to species-specific DIN values. We used Kendall's tau rather than Pearson's r for this correlation because both DIN values were non-normally distributed. We also asked whether DIN values were greater in plots without any vegetation, as we would expect if plants influence available soil N. We did this using a generalized mixed effects model with DIN (log transformed) as the response variable and the categorical explanatory variable ‘vegetated’ vs. ‘bare’. We then asked how species-specific DIN related to N pools and fluxes. First we used 1- way ANOVAs with species as a fixed effect and block as a random effect to ask whether plant and microbial parameters were different between species. We then quantified the correlation between DIN and either: total plant N; Microbial N; Microbial biomass (substrate induced respiration, SIR); Net N mineralization; Nitrification potential; and Gross nitrification (we use the term “microbial biomass/process rates” to encompass these measurements of biomass or process rates hereafter). Kendall's tau was used to assess the relationship between these variables since many were non-normally distributed. If plant uptake was driving N dynamics, we expected to find a negative relationship between DIN and plant N, and no clear relationship between DIN and other N pools and fluxes. By contrast, if microbial processes were driving R\*, then we would expect to find positive correlations between microbial biomass/process rates and DIN (our proxy for R\*), but no clear relationship between plant biomass N and DIN. If a combination of microbial cycling and plant uptake were driving DIN, we would expect a modest positive correlation between microbial biomass/process rates and DIN, and a modest negative correlation between plant biomass N and DIN. While examining the longer term effects of litter mediated plant-soil feedbacks is beyond the scope of this study, we used linear regression to examine the relationship between plant C:N and plant biomass to provide some insight into potential relationships. A positive correlation between biomass and litter C:N would reinforce the short-term R\* patterns, with plants that reach high biomass and draw N levels to the lowest levels further depleting soil N with C rich and N poor litter. A negative correlation could suggest that litter-feedbacks might drive the system in the opposite direction, with plants that achieve lower biomass and have a higher R\* potentially slowing decomposition over longer time scales with high C:N litter. Finally, we asked how plant productivity (as measured by total plant biomass per m²) related to total available N pools (summed over soils, microbes and plant biomass). We used Kendall's tau for these correlations. All analyses were performed using R version 2.13.1. # Results ## N Pools and Dynamics In examining N pools in this annual grassland, total soil N accounted for 154±4 g N/m². Microbial biomass N and DIN pools (NO₃⁻+NH₄⁺) comprised a small fraction of soil N pools, with each accounting for about 0.3% of the total soil N pool. In contrast to microbes, plants were a much larger N pool, about ten- fold higher than microbial biomass N. Our indicator of growing season N dynamics, resin N, varied by species over time for all monoculture plots. A model including species identity, time, and their interaction gave the best fit, as indicated by AIC values (data not shown). Resin available N increased in May in all plots, with a pronounced increase in both seeded bare plots as well as plots with low plant biomass (*Vulpia octoflora*, Vo; *Hordeum murinum*, Hm; and *Calandrinia ciliata*, Cc). In contrast, this response was tempered in monoculture plots with high plant biomass (e.g., *Clarkia purpurea*, Cp). The five-month sum of resin-available N was significantly correlated with DIN pools measured in May. This shows that our R\* measure (May DIN pools) was correlated with the soil N dynamics through the growing season. The temporal trends in the resin N data showed that plots with high plant biomass maintained lower DIN. Consistent with these data, we found that vegetated plots had lower DIN (R\*) values than bare plots. ## Plant N Uptake Monoculture plant biomass N was significantly and negatively correlated with soil DIN, and gave a similar trend with resin-available N. Monoculture plots with high plant N (e.g., *Clarkia purpurea*) consistently exhibited lower levels of resin-available N than bare plots. When comparing N pools of plants and microbes within monoculture plots, even for those plants that took up very little N (*e.g., C. ciliata*), plant biomass was still a larger N pool than microbial biomass. This difference became greater with increasing aboveground plant biomass. We found that plant species differed in their uptake of N into plant biomass. Differences in plant biomass N were related to differences in total plant biomass yielding a strong positive correlation between the two (r² = 0.81, p\<0.001, n = 54), whereas plant % N showed a slight negative correlation with N plant biomass N (r² = 0.10, p = 0.02, n = 54). Plant species that obtained higher biomass in monoculture plots also had higher C:N ratios (C:N aboveground biomass versus biomass in monoculture: r² = 0.67, p = 0.007, data not shown). ## Microbial N Uptake and Process Rates Microbial biomass N and SIR biomass did not differ in soils across plant monoculture plots, and had no relationship with DIN. Similarly, there was no relationship between DIN and our indicator of microbial N cycling rates, net N mineralization. Our two metrics of the conversion of NH₄⁺ to NO₃⁻, nitrification potential and gross nitrate production, did show significant positive correlations with DIN. # Discussion While a number of studies have shown that a plant species' depletion of soil available N can be predictive of its competitive ability in mixture, the extent to which plants or microbes drive the N depletion has remained largely unexplored. Set in the context of an N-limited annual grassland ecosystem, our study suggests that plant N uptake was more important than species-specific effects on microbial processes in driving plant species' differences in their R\*, and thus uptake determined their competitive dominance in the system. Three lines of evidence support this interpretation: 1) plant N and soil DIN were negatively correlated across the monocultures of different plant species ; 2) microbial N was not correlated with soil DIN; and 3) net N mineralization was not correlated with soil DIN. ## Evidence that Plants Wear the Pants The negative correlation between plant N and soil DIN suggests that plant uptake was largely responsible for the observed species level differences in R\* at the end of the growing season. Additionally, throughout the growing season most of the monocultures—especially those species that achieved high biomass—maintained lower levels of resin-available N than bare plots, where N levels were primarily a function of microbial processes. We found that neither microbial biomass N nor net N mineralization was correlated with soil DIN at the end of the growing season This suggests that microbes were not driving differences in soil DIN via direct uptake and immobilization of available N into microbial biomass, and that microbial processing of organic N was not driving differences in DIN levels across plant species. It has been suggested in the literature that, while microbes do not drive available N via direct uptake, they do regulate N via processing and plants take up the “leftovers”. If this were the case in the present study net N mineralization would be expected to positively correlate with plant N uptake, which is a relationship we did not find in our system (net N mineralization versus plant N uptake: r² = 0.05, p = 0.47). In fact, nitrification potential and gross nitrification were the only measured microbial parameters that were correlated with DIN. Nitrification is a dissimilatory process and thus only alters N form, not DIN pool size. Since nitrifier abundance (nitrification potential) and process rates (gross nitrification) correlated to substrate pools (DIN), we hypothesize that they were limited in their activity by the size of their substrate pools of NH₄⁺ and NO₂⁻. This is consistent with the current understanding of these chemoautotrophic organisms, which are generally considered poorer competitors for soil N than heterotrophic microbes or plant roots. In contrast, the heterotrophic microbes which drive N mineralization and immobilization were likely not limited by N availability. Rather, they functioned similarly across monoculture plots regardless of N uptake by plants and DIN concentrations. This is consistent with low soil C:N values, which averaged 7.3 (±0.1), indicating that heterotrophic microbes were more limited by C than N. C-limitation in turn may explain why heterotrophic microbes did not seem to drive final DIN levels. ## Evidence of plant-microbial interactions Our data suggest that plant N uptake drove DIN pool sizes in this annual system, and thus drove R\*. This is consistent with ¹⁵N tracer experiments which postulate that plants immobilize a greater amount of N into plant biomass over an entire growing season, and thus drive the longer-term patterns of N fate in ecosystems. However, these same studies make it clear that microbes can outcompete plants for available soil N in the short term, with some of the best evidence for this coming from California grasslands. Our goal was to test the mechanisms underlying R\* by determining the role of plant uptake as compared to changes in microbial biomass and microbially mediated N cycling rates. Understanding such mechanisms is particularly important in such ecosystems where competition outcomes among annuals in the first year of community assembly are likely to be important for long-term community trajectories. By constraining our study to one year in an annual system, we explicitly removed the confounding effect of plant-microbe feedbacks as mediated by litter, which are more likely to develop over several years. Short-term competition stands to be particularly important in these California grasslands, where disturbance plays a large role in exotic annual grass invasion, and the dominant competitors are annual. While our data suggest that plant N uptake drove soil DIN pools, another story emerges when the focus is shifted from looking at comparisons of individual pool sizes and fluxes to looking at the total N in plants, microbes, and DIN for each set of monoculture plots. Examining the sum of N in plants, microbes, and DIN reveals that there were larger pools of biological and/or bioavailable N in plots with higher plant biomass. This suggests that microbes released more plant available N over the course of the growing season in those plots with higher plant biomass and lower R\*, even though our measure of microbial N turnover, net N mineralization, did not differ between plots. Comparing planted vs. bare plots, there was more N in these pools (plant N, microbial N, and DIN) in planted (5.4±0.3 g N/m²) than bare plots (1.3±0.7 g N/m²; F_1,62 = 26.7, p\<0.001) and total N in these pools increased with increasing plant biomass and decreasing R\*. Thus, while plant N uptake was working in a manner consistent with our conceptual models, microbial N cycling may have actually been responding in the opposite direction of our predictions. In other words, microbial N cycling in the field may have been negatively, rather than uncorrelated or positively correlated with DIN; in plots with lower R\* and higher plant N uptake, microbes may have generated more plant available N than in plots with higher R\* and lower plant N uptake. It is possible that this pattern—where the sum of plant N, microbial N, and DIN was higher in plots with lower R\*—could be explained by large losses of available N from bare plots due to denitrification or leaching. Several lines of evidence suggest that these explanations are unlikely. Denitrification depends on labile C, which would have led to higher potential for N losses due to denitrification in planted plots with roots exuding labile C as opposed to those that were bare or had sparse cover. Thus denitrification would be expected to yield more loss of N with increasing biomass, which would lead to a pattern opposite to that observed in the total N pools. While leaching losses would have been higher in plots with lower biomass due to decreased evapotranspiration, we suggest it is unlikely to have been the main driver of the observed patterns. Over 94% of the annual rainfall had fallen by mid-April, prior to peak plant growth and strong differences in monoculture soil resin available N. The two remaining rainfall events were small enough that they would not have caused significant leaching. While we cannot rule out denitrification and leaching, we hypothesize that plant ‘priming’ of microbes led to differences in microbial N cycling in the field, which in turn altered plant uptake of DIN, and thus the total distribution of N in different ecosystem compartments. Priming is the stimulation of microbial activity via labile C-rich root exudates. The low C:N ratio of these soils makes it likely that microorganisms were limited more by C than by N. Plant roots releasing labile C into rhizosphere soil may have led to localized increases in microbial activity leading to higher N availability in rhizosphere soil. Because we did not separate rhizosphere from bulk soil, localized increases in microbial activity, such as net N mineralization, were less likely to be observed. Priming has been shown to increase plant N uptake in various ecosystems, – including California grasslands, where it was shown to increase gross N mineralization rates by ten-fold. Thus plant-microbe interactions may have increased N cycling in this current experiment in plots with lower R\*, but our measures of N cycling were not sensitive to those differences given several factors, including: the localized nature of this interaction; the removal of roots and fresh C inputs which would have been the drivers of this interaction; and the short preincubation following sieving and adjustment of water content. While our results were consistent with plants driving patterns in R\* through uptake of N, and showed no evidence of microbial processes or biomass driving R\*, when ecosystem N pools are taken as a whole those plots that accumulated more N in microbial, plant, and DIN pools likely had increased cycling and availability of N to plants. This suggests that even over the course of one growing season, plant-microbial interactions (*e.g.*, priming) may be important to ecosystem N cycling and ultimately, may influence plant competition. It remains an open question as to whether these patterns would persist over multiple growing seasons, and what kind of impact litter mediated feedbacks would have in this annual system. In our experiment, plant species that obtained higher biomass in monoculture plots also had lower quality biomass (higher C:N in aboveground biomass), suggesting that the patterns observed in this study would be reinforced over multiple years by litter mediated feedbacks. It is likely that monoculture plots of annual species with high N use efficiency and low N litter would depress N mineralization, helping drive low R\* and the magnitude of this effect could increase with time as feedbacks become better established. Thus for annuals growing in continuous monoculture, we would expect to see similar patterns to perennial systems, where the persistence of plants allows for strong litter feedbacks that help determine plant species' R\* for nitrogen. We thank the Midland School (Los Olivos, CA) for letting us perform this experiment on their property. Ben Munger and Rick Skillin provided invaluable help in setting up the experiment, and Peter Adler, Anna Oliver, Lilly Hayden and Rebecca Harris assisted in data collection. We also thank Carla D'Antonio, Steve Perakis, Josh Schimel, Alexia Kelley, Bonnie McGill, Marissa Lee, Erin Mordecai, Louie Yang, and three anonymous reviewers. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SGY BPC JHRL JML. Performed the experiments: BPC SGY JHRL. Analyzed the data: JHRL SGY BPC. Contributed reagents/materials/analysis tools: SGY BPC JHRL JML. Wrote the paper: SGY BPC. [^3]: Current address: Pacific Island Ecosystem Research Center, United States Geological Survey, Kilauea Field Station, Hawaii National Park, Hawaii, United States of America [^4]: Current address: Department of Biology, Duke University, Durham, North Carolina, United States of America [^5]: Current address: Institute of Integrative Biology, Eidgenössische Technische Hochschule Zurich, Zurich, Switzerland [^6]: Current address: Biology Department, University of Washington, Seattle, Washington, United States of America

# Introduction Delirium is a frequently observed complication in patients in an intensive care unit (ICU), that has been associated with long-term cognitive impairment, prolonged length of ICU stay and with increased health care costs. The pathophysiology of delirium is complex and heterogeneous. Metabolic disorders such as hypo- and hyperglycemia have been identified as risk factors for delirium onset, but extensive research is lacking. To improve patient related outcomes, identification of modifiable factors in delirium need to be further explored. Tight glucose control has been implemented as regular care in critically ill patients to reduce extreme glucose deviations as hypo- and hyperglycemia, glucose variability and to decrease the mean glucose concentration with decreased mortality risk as result. However, the optimal blood glucose range in tight glucose control is controversial. Intensive glucose control (glucose target between 4.5–6.0 mmol/l (81.0–108.1 mg/dL)) has been shown to increase mortality rates compared to conventional glucose control (glucose target ≤ 10.0 mmol/l (180.1 mg/dL)). The occurrence of hypoglycemia during intensive glucose control may be responsible for this increased risk of death. Furthermore, it has been reported that the mortality rate after hyperglycemia is higher in non- diabetic patients compared to diabetic patients, due to adaptive mechanisms to chronic hyperglycemia in patients with diabetes. Glucose variability has been associated with higher mortality risk in critically ill patients. A gold standard for measurement of blood glucose fluctuations is lacking. Glucose fluctuations were frequently reported as glucose variability and refer for example to mean glucose concentration, mean absolute glucose (MAG) change, standard deviation (SD) or hypo- and hyperglycemia. Delirium and glucose variability have both been associated with negative outcomes, but their mutual relation has been poorly studied. Higher glucose values have been reported in critically ill patients with hyperactive delirium compared to critically ill patients with non-hyperactive delirium. Given that delirium results from acute illness, it is plausible that this acute illness may increase activity of hypothalamic-pituitary-adrenal axis leading to increased cortisol release and subsequent decreased peripheral insulin sensitivity, thereby contributing to hyperglycemia. It is unclear whether glucose variability is higher during delirium within the window of glucose control during ICU admission. The aim of this study was to determine whether estimates of glucose variability are altered during delirium in critically ill patients with and without diabetes in the ICU. # Materials and methods ## Setting, study design and population Data were used from a prospective cohort study conducted in the 32-bed mixed ICU of the University Medical Centre Utrecht (UMCU), the Netherlands. All patients hospitalized for longer than 24 hours on the ICU in the period from January 2011 to June 2013 were included in this study, except in the case of neurological illness, if delirium assessment was impossible or patients were unable to speak Dutch or English. The local Institutional Review Board waived the need for informed consent in this non-interventional investigation (IRB 010/056/c and 12/421/c) and approved further research with the anonymous data. The mental status of all ICU patients was daily classified by the research team as ‘delirious’, ‘awake and non-delirious’ or ‘comatose’ using a 5-step validated algorithm (interobserver agreement, 0.94–0.97; sensitivity, 0.75; and specificity, 0.85). This multistep algorithm incorporates a review by a research nurse of all Confusion Assessment Method for the ICU (CAM-ICU) assessments conducted by the bedside nurses, whether delirium treatment was initiated and a meticulous chart review for the presence of documented terms clinically associated with delirium. When delirium could not be ruled in or out using this procedure, the research nurse conducted an additional CAM-ICU assessment. Delirium episodes were recorded and delirium subtype was classified using the 3 hourly registered RASS scores (10 point scale ranging -5 (comatose) to +4 (heavily agitated)). A delirium episode ended if a patient had a classification of ‘awake and non-delirious’ or a classification of ‘comatose’ for at least two days. For this study, patients with delirious and non-delirious observation days were selected from the study cohort. In case of one delirious episode during ICU stay all observation days were included until ICU discharge. In case of more than one delirious episode, observation days until the start day of the second delirious episode were included for that patient. Patients were excluded if there was no glucose value available during a delirious episode or during a non-delirious episode. Observation days were excluded from the study if there were no glucose values available or if the observation day was classified as ‘comatose’. ## Data collection Trained, assigned physicians collected data (baseline and per day) from all ICU patients including demographic data, (chronic) co-morbidities and medication use, ICU admission characteristics, daily physiological measurements and vital signs, and therapeutic interventions. Diabetes was marked present if noted in the medical record or if patients used insulin and/ or oral antidiabetic drugs at ICU admission. Current alcohol intake was marked as positive if patients used more than three units of alcohol per day, as documented in the medical records or history. Current smoking was marked as positive if smoking was written in the medical records or history. Planned admissions were those admissions which could be postponed for at least 12 hours without adverse consequences. The Acute Physiology and Chronic Health Evaluation (APACHE) IV classification was used to determine the admission diagnosis, severity of disease, and infection at ICU admission. The extent of chronic comorbidities were measured with Charlson Comorbidity Index (CCI). The Sequential Organ Failure Assessment (SOFA) score without central nervous system component was used daily to classify severity of disease. The presence of severe sepsis or septic shock was classified using international sepsis definitions at the time of study. During the study period, a glucose regulation protocol was used to maintain the target glucose concentration during ICU admission between 5.0 and 8.0 mmol/l (90.1–144.1 mg/dL), except in those ICU-patients with a low risk of prolonged hyperglycemia such as per- and postoperative patients with one bolus injection of dexamethasone. Continuous insulin infusion was initiated in patients with diabetes and in ICU-patients with a (drug-induced) glucose concentration \> 8.0 mmol/l. Glucose levels were measured on fixed time points between 0.5–4 hours after the last glucose measurement (details are described in the glucose regulation protocol) from blood samples obtained from an arterial catheter using BeckmanCoulter AU5800 (Beckman Coulter Inc., Brea CA, USA) or if arterial catheter was absent by finger stick using Precision Xceed Pro (Abbott, Abbott Park, USA). Glucose levels were automatically stored in the electronic patient data management system (EPDMS, MetaVision, version 5.45, iMDsoft). Medication use (drug, dose, route and time of administration including total parenteral nutrition) and glucose measurements (concentration and time of measurement) were retrieved from the EPDMS and added to the prospectively collected data. Continuous infusions, such as insulin, were recorded in the EPDMS, including end date and time of administration. A change in infusion rate resulted in a new medication record. If a continuous infusion covered more than one day, the dose per day was calculated using the ratio between infusion times of both days. Energy intake was defined as the sum of daily caloric intake from continuous infusion of glucose, total parenteral or enteral nutrition, and high caloric medication, such as propofol. ## Outcome The primary outcome was the within-patient difference in glucose variability during delirious and non-delirious observation days. Glucose variability was measured each observation day, expressed by the following five measures: 1. mean glucose concentration (mmol/l) 2. SD of all glucose levels (mmol/l) 3. MAG change, defined as the mean absolute glucose change per hour (mmol/l/hour). To calculate the MAG, all absolute changes in blood glucose levels were added up and were divided by the time between first and last glucose levels (in hours). 4. Daily delta, defined as the difference of daily maximum and daily minimum glucose concentration (mmol/l) 5. Hypo- and hyperglycemia. Hypoglycemia was defined as a glucose concentration \<3.5 mmol/l (63.1 mg/dL) and severe hypoglycemia was defined as a glucose concentration \<2.2 mmol/l (39.6 mg/dL). Hyperglycemia was defined as glucose concentration \> 8.0 mmol/l (144.1 mg/dL) and severe hyperglycemia as glucose concentration \> 11.0 mmol/l (198.2 mg/dL). ## Data analyses Patient and observation day characteristics were reported as numbers with percentages in the case of nominal data and means with SD or median with interquartile range (IQR) in the case of continuous data. Continuous data were compared using Student independent sample t tests when the data was normally distributed; otherwise the Mann-Whitney U test was used. Chi-square tests were used to compare nominal data. Characteristics of delirious and non-delirious days in non-diabetic and diabetic patients were compared in a multilevel technique using linear mixed-effects models for continuous characteristics and generalized mixed-effects models with logit link function for dichotomous characteristics. Statistical significance was considered at p-value \<0.05, when appropriate 95% bootstrap percentile confidence intervals (CIs) were expressed. Two-stage bootstrap resampling procedure with ‘patient’ as cluster variable was used for obtaining CI’s and p-values from 1000 replications. In the case of one glucose concentration per day the mean glucose concentration, SD and the difference of daily maximum and minimum could not be calculated. The MAG change was calculated if there were more than two glucose levels per day available. Hyperglycemia and hypoglycemia were described as dichotomous outcome per observation day, but glucose values were analysed individually. Linear mixed-effects models and generalized mixed-effects models with logit link function were used as multilevel techniques to test whether delirium was associated with increased glucose variability. The effects were expressed as regression coefficients or odds ratios, both with bootstrap 95% CIs. Covariates were included in the model as fixed effects, when possible as time dependent covariate. The use of medication was classified dichotomous per day. All models included random effects for ‘patient’. The degree of glucose variability depends on diabetic status, therefore separate models were developed for patients without and with diabetes. The adjusted models always included the following covariates; age, gender, total dose of insulin (bolus injection and continuous infusion) in the 30 minutes before glucose measurement or total dose of insulin per day and energy infusion in the 30 minutes before glucose measurement or energy infusion per day. Confounders were selected based on p-values (\< 0.05) and effect sizes. The following variables were tested as potential confounders: age, gender, body mass index (BMI), current alcohol intake, current smoking, admission type, planned admission, confirmed infection, APACHE IV-score, CCI, SOFA-scores, support of mechanical ventilation, presence of severe sepsis or septic shock, number of observation day, length of stay (LOS) at ICU, the use of antipsychotic drugs, norepinephrine, corticosteroids, clonidine, ACE-inhibitors, cyclosporine or tacrolimus, beta-blockers and beta-agonists. All statistical analyses were carried out with R version 3.2.3 with package ‘lme4’ (R Foundation for Statistical Computing, Vienna, Austria). # Results During the study period, 2669 patients were admitted to the ICU and of whom 1557 patients were excluded. Delirium was diagnosed in 535 patients. Of those patients, 125 patients were excluded: 88 (16.4%) patients because they had only delirious or comatose observation days during their ICU admission and 37 (6.9%) patients because of the absence of glucose values during delirious or non- delirious observation days. Therefore, the final population consisted of 410 patients with 1233 delirious and 1775 non-delirious observation days. Patient characteristics are shown in. Diabetic patients were on average older, had a higher BMI and had a higher APACHE-IV score compared to non-diabetic patients. The number of delirious days was higher in diabetic patients compared to non-diabetic patients. Diabetic patients had a higher maximum glucose concentration in the first twenty-four hours of ICU-stay than patients without diabetes. shows the characteristics of delirious and non-delirious days in non-diabetic and diabetic patients. During delirious days, diabetic and non-diabetic patients had more often insulin infusions, had more insulin rate adjustments, and had a higher average of numbers of glucose measurements in comparison with non- delirious days. In total 19,962 glucose levels were collected. Estimates of glucose variability are presented per observation day in Tables and. In the unadjusted models, delirium was associated with a higher MAG change (β:0.038; 95% CI:0.017–0.061; p = 0.001) and increased daily delta (β:0.325; 95% CI:0.134–0.494; p = 0.001) in patients without diabetes. After adjustments for potential confounders, the association was not maintained in non diabetic patients using the same definitions for glucose variability (MAG change; β adj.:0.021; 95% CI:-0.004–0.043; p = 0.076 and daily delta β adj.:0.100; 95% CI:-0.096–0.282 p = 0.287). Delirium was positively associated with hypoglycemia in diabetic patients (OR adj.: 2.78; 95% CI: 1.71–6.32, p = 0.005), but not in non-diabetic patients (OR adj.: 1.16; 95% CI: 0.58–2.28, p = 0.689). Generalized mixed- effects models with logit link function were not performed for the association between delirium and severe hypoglycemia as the number of glucose levels below 2.2 mmol/l was insufficient. We found similar results for glucose variability when all delirious and non- delirious days during ICU stay were analysed compared to the observation days of the first episode, or when consecutive episodes (delirious and non-delirious episodes) were analysed (data not shown). # Discussion In this cohort of ICU patients, mean and SD of glucose concentrations, MAG change, daily delta and the risk of hyperglycemia were unaltered during delirious days compared to non-delirious days in non-diabetic and diabetic patients. Furthermore, we demonstrate that in diabetic patients delirium was associated with hypoglycemia. The association was even stronger after adjustment for several confounding factors. This association was not found for non-diabetic patients. Little is published about the mutual relationship between glucose levels and delirium. It has been reported that mean glucose levels did not differ between patients with delirium and without delirium within non-critically ill older patients. Although we conducted our study in an ICU cohort with critically ill patients, our results are in concordance with their study. In the ICU setting, one study has been conducted reporting higher mean glucose levels in patients with hyperactive delirium compared to patients with non-hyperactive delirium. In our study, we were not able to identify any hyperactive delirium. This may be related to the use of sedatives. Additionally, our study was designed to compare mean glucose concentrations during delirious and non-delirious days per individual. In concordance with our results, tight glucose control has been linked to increased hypoglycaemia rates and increased delirium rates. (Insulin- induced) hypoglycaemia affect brain function. One of the strengths of our study is that we were able to conduct our study in one of the largest high quality cohorts with ICU patients with different delirium episodes. In addition, we had extensive information on glucose measurements, allowing us to assess subtle and detailed changes in glucose levels over time, both in diabetic and non-diabetic patients. In particular, this detailed information enabled us to look at various definitions of glucose variability. Furthermore, we were able to look at within- patient patterns (comparing delirious and non-delirious days in each individual), which minimizes the risk of confounding. Finally, we were able to control for various potential confounders in a time dependent manner, such as glucose-influencing drugs including insulin, norepinephrine, corticosteroids and energy infusion. However, this study has some limitations. The generalizability is possibly limited as this study was performed as monocenter study at a university hospital. Selection bias could have occurred because patients and observation days without glucose measurements were excluded. Despite our rich information on glucose levels, a potential limitation is the possibility that peaks and nadirs in blood glucose levels have been missed as glucose levels were not measured continuously. We considered this misclassification as non-differential as this misclassification occurred at random during delirious and non-delirious days. Due to the multiple testing, it remains a possibility that the association between delirium and hypoglycemia was based on a type I error, despite the stronger positive association after adjustment for confounders. Unmeasured confounding may have occurred as there could have been unmeasured confounding covariates. The measures of glucose variability could depend on the number of glucose determinations. Especially, the MAG-change is sensitive for higher frequency of measurement. We consider this as less important because observation days were compared, but not whole ICU stays. Furthermore, we adjusted for disease severity and insulin infusion which indirectly correct for the frequency of measurement. For the number of glucose measurements has not been adjusted because this indices can been seen as glucose variability measure. Hypoglycemia at the ICU has been associated with increased mortality independent of diabetic status. For this reason, our findings suggest that in clinical practice blood glucose levels should be monitored more often during delirium in critically ill patients with diabetes to avoid hypoglycemia. More research is needed to explore the impact of our findings concerning diabetic patients on ICU outcome and determine whether any causality consists between delirium and glucose variability. # Conclusions Mean glucose concentration, its SD, MAG change, daily delta and the risk of hyperglycemia were not significantly altered during delirium in non-diabetic and diabetic ICU patients. Delirium in critically ill patients with diabetes was associated with hypoglycemia. This association was not found for non-diabetic ICU patients. Our findings suggest that glucose levels should be monitored more closely in diabetic patients during delirium at the ICU to prevent hypoglycemia. # Supporting information The authors thank W. Plasma, DVM., Department of Intensive care Medicine, University Medical Centre Utrecht, Utrecht, the Netherlands for his assistance in data acquisition. [^1]: The authors have declared that no competing interests exist.

# Introduction Blood transfusion is a crucial life saving therapy to many who have experienced road accidents, maternal hemorrhage, anemia, different surgical procedures and a number of other medical and surgical conditions. Blood comes from blood donors defined as “persons who donates either whole blood or blood products for transfusion” who provide a global estimate of 112.5 million blood donations yearly. About half of all donations come from developing countries where more than 80% of world’s population lives. World Health Organization (WHO) further provide estimations of nearly nine times greater average blood donations rates in developed countries when compared to developing countries, equivalent to 4.6 donations per 1000 people in developing countries compared to 33.1 donations recorded in developed countries. This brings greater restrain to blood donation needs among the under five year old, who are majority users of blood in developing countries when compared to the needs in developed countries where elderly are the majority users. Persons donating blood may be voluntary non-remunerated blood donors or replacement donors as required by a member of their own family or community. WHO recommends voluntary non-remunerated blood donors over replacement donors due to the degree of blood safety from the two groups. Although a person can voluntarily decide to donate blood, they may be disqualified from donating blood due to reasons pertaining to the donors’ safety and/or recipient safety, which is simply referred to as donor deferral. Deferral may be temporal postponement or permanent exclusion from donating blood due to being suspected or confirmed of having an infectious disease, hematological disease, or any other medical condition that will either influence the safety of blood or affect donors’ own health. However, the prevalence of blood donor deferrals varies widely, and examples that follow substantiate the variations. In Asia, blood donor deferrals differ from one locality to another and different studies report the prevalence that ranges from 4.6 to 30%. Observation in different countries within Europe shows that the prevalence of blood donor deferrals is slightly lower that from Asia. In Africa, the prevalence of blood donor deferrals seems to be comparable to that of middle income countries. For example, the prevalence is 10.8% in Ivory Coast, and 7% in Zimbabwe. These studies confirm that blood donor deferral is an issue in all countries including Tanzania however, the prevalence is not known. Regardless of the prevalence of deferral, we may generally note that it is an issue of concern to most blood transfusion centers in the world; it affects both low income and high income countries and leads to inadequate blood for transfusion due to lack of eligible blood donors. Blood donor deferral is a sad experience to the donor and the blood bank as whole. Analysis of the causes of blood deferrals show that the causes do differ from one country to the other, calling for a need of setting specific analysis of causes based on the donor selection criteria. However, low haemoglobin has remained to be the main cause of temporal deferral in many places. For example, in Turkey the leading cause of donor deferral is low haemoglobin 20.7% followed by common cold and/or sore throat 17.7%, high risk sexual activities 6.7%, hypertension 5.6% and polycythaemia 2.8%. Similar findings on the role of hemoglobin of deferral has been reported in Netherlands and Asian countries. However, risk factors related to infections of Human Immunodeficiency Virus (HIV) and Hepatitis B have been reported to be the leading cause of permanent blood deferrals in other studies. Understanding causes of blood donor deferrals are important in instituting appropriate preventive strategies towards conditions identified, including appropriate referral systems for clinical care. According to WHO annual blood estimates formula, Tanzania needs 508,000 units of blood annually, but the amount of blood collected in 2016 was less than two- fifths of actual needs (196,735 units). About 71% of the collected blood is used during child birth and by children under five years of age. To make sure that there is constant supply of blood for transfusion in Tanzania, the Tanzania national blood transfusion services (NBTS) was established in 2004. This was followed by establishment of centralized system through the establishment of seven zonal centers. As the part of NBTS policy, guidelines to ensure high standards of blood safety, screening and testing for transfusion transmissible infections (TTIs) is done and include infections such as Human Immunodeficiency Virus (HIV), Hepatitis B Virus (HBV), Hepatitis C virus (HCV) and Syphilis. Screening and deferring clients with medical conditions like fever, low hemoglobin, unstable blood pressure and social reasons like risk behaviors such as sex workers and drug users is done. Although there has been a system to screen and document causes of deferral in Tanzania, no studies have presented a comprehensive analysis on the causes of blood donor deferrals in the country. Information on deferral causes for blood transfusion is important when considering future interventions that aim at increasing blood donation and in the prevention of diseases and conditions associated with blood. Increasing demand of blood for transfusion calls for a need to maximize enrollment to blood donation program and increase the number of those who ultimately donate blood. This study was therefore aimed at estimating the prevalence of blood donor deferrals and to identify the causes of deferrals in the northern zone blood centre in Tanzania. # Materials and methods ## Study design and setting This was a cross-sectional study where information from all participants who presented for blood donation at the northern zone blood bank from January 1^st to December 30^th 2016 were retrospectively analyzed between March and June 2017. The northern zone blood bank center is one of the six centers in Tanzania assigned to collect blood from all donors in four regions of Kilimanjaro, Arusha, Tanga and Manyara. The center was established in 2005 and is located in Moshi municipality about four (4) kilometers from Moshi town, just proximal but within the compound of KCMC Hospital–a consultant hospital in the zone. This center deals with blood donor recruitment, blood collection, screening and distribution of blood to all health facilities in the area. The goal of northern zone blood transfusion center was to collect 15,846 blood units in 2016, however it managed to collect 7,163 (45%) blood units. The center serves a total population of about 6,949, 880 people from the four regions, which is 16% of the Tanzanian population. To achieve the goal the center had to collect blood from voluntary non-remunerated blood donors as well as from replacement donors within the zone. The procedures for donor intake is as follows: The clients presents for blood donation at the center where they are received at reception and registered. This is followed by counseling provided by trained donor counselors. This counseling process goes in hand with checking on the eligibility for blood donation using a standardized donor questionnaire. The questionnaire inquires about the socio- demographic characteristics of participants such as age, marital status, occupation and address; the type of donor–either voluntary or replacement; and general health check of the donor in terms of diseases and risks for acquiring transmissible infections such as HIV and HBV. Voluntary non-remunerated blood donor was defined as a person who gives blood on his or her own free will and receives no payment, either in the form of cash or in kind. After counseling, prospective blood donors then receive measurements for weight, blood pressure and hemoglobin estimation. The facility then does testing on blood samples for HIV, Hepatitis B, Hepatitis C and Syphilis. Clients who do not fulfill the eligibility criteria for donation are then deferred either temporarily or permanently depending on reason for the deferral. Blood donor deferral in this study was therefore defined as the temporally postponement or permanent exclusion from donating blood by a person suspected of having an infection, disease or any other medical condition that will either influence the safety of blood or affect donors’ own health. Blood donation is done after the client has signed informed consent. Regarding testing for TTIs, the Tanzanian NBTS tests four TTI markers which are HIV, Hepatitis B, hepatitis C and Syphilis. These tests are done using ELISA antigen–Antibody combination with test kits shown in. For each marker, appropriate screening test (Screening Test \# 1) is done with three possible results; either negative, positive or grey zone. Those with negative test results are recorded as negative and can be recalled for next donation. The blood unit with positive or grey zone test results for the initial screening test is discarded and subjected to a duplicate repeat (Screening test \# 1 in Duplicate Repeat) and then supplemental test (Supplemental test \# 2). The donor with the blood unit testing positive after this duplicate repeat are deferred permanently. ## Study population and data collection We were provided by the NBTS with information from all clients who presented for blood transfusion at the northern zone blood bank center from January to December 2016. This study extracted data from all registered clients with complete information on donor type, decision to donate or defer and clearly indicated reason for deferral among deferred clients. Participants excluded from this analysis were those who decided not to proceed with the assessment for qualifying to donate blood before being administered with donor questionnaire. Information extracted from the database included those from donor questionnaire and from laboratory where the results for TTIs and other estimations such as of hemoglobin are documented. Information also included the final verdict of donate or defer. ## Statistical analysis Statistical Package for Social Studies (SPSS) program version 20.0 was used to analyze the data. Descriptive statistics was used to summarize data where frequency tables and cross tabulation were made while describing the data in numbers and percentages. The association between donor types; voluntary and replacement; and between deferrals and accepted donors, are made using Pearson Chi-square test for categorical variables. Results of the difference were considered statistically significant when p-value was less than 0.05. ## Ethical consideration The ethical approval for carrying out this study was obtained from the Kilimanjaro Christian Medical University College Research and Ethical Committee (CREC) and the National Blood Transfusion Services Ethical Committee. The letter to request permission to do the research and to access the database for the registered blood donors was granted by the National Blood Transfusion Services. The head of laboratory records, head of blood donors and head of records were informed about the study. The information obtained was used for research purpose only. Confidentiality of participant information was observed during the entire period of the study. # Results Out of all clients 16537 presented for blood donation in northern blood transfusion centre, 14377 (87.0%) met inclusion criteria. Reasons for exclusion were incomplete data for the status of deferral, donation and reasons for deferral of participants. ## Demographic characteristics of participants Demographic characteristics of all participants are presented in below. The majority of participant were males, who formed more than three-quarters 11377 ~~(~~79.1%) of all participants. More than two third of all participants 9719 (67.6%) were aged between15 and 30 years followed by 3377 (23.5%) who were aged 31 to 45 years. Almost half of the participants 6671 (46.4%) came from Kilimanjaro region where the center for blood transfusion in the northern zone is located and nearly a quarter 3408 (23.7%) came from a nearby region of Arusha. Almost nine out of every ten participants 12775 (88.9%) who presented for blood donation were voluntary non remunerated blood donors. Family replacement donors were significantly more likely to be females, with increasing age above 31 years and come from nearby regions of Arusha and Kilimanjaro (P value \<0.01). ## Prevalence of blood donor deferrals Out of 14377 participants who presented for blood donation in 2016, 1829 (12.7%) were deferred. About 15.6% of all men were deferred and 12.0% of all female were deferred. Blood donor deferrals were also categorized according to age where participants aged 46–65 years had higher deferral rate (27.6%) when compared to those who were in the younger age groups of 18 to 30 years (12.1%) and between 31 and 45 years (17.6%). Replacement donors had higher deferral rate 15.8% as compared to voluntary non-remunerated blood donors who had 12.3%. Deferral was significantly more likely among females, increasing age, came from Tanga or surrounding region of Kilimanjaro and/or a family replacement donor (P value \<0.01). ## Causes of blood donor deferrals Detailed analysis of permanent and temporary causes of blood donor deferrals is shown in below. The general picture show that infections contribute to 62% of all deferral causes in this study. Looking at temporary causes, slightly more than half of all blood donor deferrals 927 (50.7%) were due to temporary causes. Of all temporary causes, low haemoglobin was the leading cause; contributed one- fifth (21.1%) of all causes of blood donor deferrals and the majority of participants with low haemoglobin were females 224 (58.0%). Syphilis was the second leading cause of temporary blood donor deferral, carrying about one-tenth 171 (9.3%) of the remaining causes. Majority of those who were deferred because of syphilis were males 149 (87.1%). However, the prevalence of syphilis in the study population was 1.2%. Of all permanent deferrals causes, hepatitis B was the leading cause. It contributed more than a quarter (29.6%) of all deferrals. About 90% of those deferred due to hepatitis B were males. HIV was the second leading cause of permanent deferrals accounting for 13.3% of all deferrals. Of all participants who were deferred because of HIV, 81.1% were males. The prevalence of HBV, HIV and HCV in this study was 3.8, 1.7 and 0.1% respectively. # Discussion This study aimed to determine the prevalence and causes of blood donor deferrals among clients presenting for blood donation in northern zone Tanzania. The prevalence of blood donor deferrals in our study was found to be 12.7%. Analysis of deferral causes indicated that slightly more than half of all deferred clients were due to temporary causes. Infections were the leading cause of deferral in this study by 62%; with Hepatitis B being the leading single cause followed by low haemoglobin and HIV. The majority of those who were deferred were replacement donor when compared to voluntary non-remunerated donors. Exploring on the background characteristics of the study subjects, the majority of participants were males, with a male to female ratio of 3.8:1. Other studies have indicated very high male to female ratio. Having very high male to female ratio, as compared to our study has implication to the study results as some of deferral causes are more prevalent in either males or females. For example, having more males in the study decreased the deferral rates in some studies since females are more deferred than males due to low haemoglobin. The present study also indicated that voluntary non-remunerated blood donors were nearly 90% of all the blood donors with remaining proportion being replacement donors. Other studies on blood donor deferrals have found a reverse composition where replacement donors are more than voluntary non-remunerated blood donor. Voluntary non remunerated blood donors are safer than replacement donor. However, the reported deferral rates by these studies are within similar range with what has been reported in the present study. The characteristics of participants enrolled in these studies therefore influence the deferral rate, calling for further exploration on other factors. This might be due to differences in geographical location where these studies have been done, difference in levels of infections and/or awareness towards blood donor eligibility among populations where these studies were conducted. In the present study the prevalence of blood donor deferrals was 12.7%. To the best of our knowledge there is no similar study that have been done in Tanzania thus no other study in the country to be compared with the present study. The deferral rate reported in our study is comparable with what has been reported in other studies done elsewhere. While some studies have shown congruent prevalence results with the present study, some have reported slightlyr lower or higher prevalence rates of deferral. The studies which reported congruent prevalence results used the same study design and tools, and were from low income countries. Most of these studies used donor screening questionnaire together with TTIs result from laboratory and had employed retrospective study design. In that regard, all the studies presenting comparable results with the present study assessed the pre-donation and post-donation deferrals. Of the studies that reported lower rates of deferral, two studies used information on TTI’s similar to the methodology of the present study. Similarly, two of the studies which reported higher rates of deferral than what is reported in the present study included results of TTIs in their analysis. It can therefore be concluded that the variation in deferral rate presented in these studies when compared to the present study might be explained by the difference in donor eligibility criteria, level of awareness of the population from where donors come on blood donor eligibility criteria and social economic status due to difference in geographical location. In middle and high income countries the level of awareness on blood donor criteria is high, have good nutrition status and low prevalence of infectious diseases which results in the prevalence of deferrals to be low. More strict blood donor eligibility criteria and the level of keenness of blood donor counselors on exploring risk behaviors to clients may lead to the increased prevalence of deferrals. The differences in social- economic status between the countries also contribute to differences in the prevalence of deferrals. For example, with high prevalence of low hemoglobin in low income countries when compared to high income countries, low income countries tend to have high prevalence rates of deferrals. The common cause of blood donor deferrals in our study was infections, which are measured by the positivity of TTIs when screening, accounted for two third of all deferrals. TTIs that are measured in northern zone blood transfusion center are HBV, HCV, HIV, and Syphilis. These results correspond to the nature of low income countries where there is a double burden of infectious diseases and non communicable diseases. Prevalence of infectious and non communicable disease plus malnutrition is higher when compared to the high income countries. The present study further found that HBV was the leading cause of permanent deferrals followed by HIV. Other studies elsewhere have indicates similar trend. Our study has found that HBV is more prevalent among blood donors (3.8%) than HIV (1.7). Even in the general population of Tanzanians, the prevalence of HBV (6.0%) is higher than that of HIV(4.7%). It is important noting here that clients testing positive for HIV and HBV will never donate blood, reducing the pool of potential donors in the general population. However there is low level of awareness on HBV, its mode of transmission, its causative agent and its consequences among the general population when compared to HIV in the country. This might be the reason as to why the prevalence of HBV is high among the clients who presented for blood donation in northern zone blood transfusion center. If those positive on HBV and HIV are refereed to appropriate services then Blood Transfusion centers will have identified the diseases and referred to appropriate care, improving the integration with other health care services. The Health Sector Strategic Plan for the health sector and the national policy on health emphasizes the need for continuous care and highly recommends referrals to appropriate care. However, there is a need to balance between the risks for test seeking behavior among blood donors which may attract high risk persons to donate blood and increase the likelihood of window period donations. The best practice is improving community based initiatives for screening of infections coupled with preventive messages, while linking those tested positive for treatment in nearby centers. The present study has shown that low hemoglobin was the leading temporary cause of blood donor deferrals and the second leading cause of all deferrals followed by Syphilis. This study implies that low hemoglobin is prevalent in northern zone. Similar finding as in our study was reported in Turkey, Netherlands, Asian countries and in other parts of the world. The low Hb is caused by parasitic infections like hookworm and poor nutrition status like low consumption of iron containing diet, Vitamin B12 and folic acid. It seems that most of the participants don’t know about their health status as Hb estimation is a simple and cheap measure of which at least many could have known before presenting at the blood transfusion center. Some of those with low Hb could have known their status before they presented themselves for blood donation. This may mean that the level of understanding on blood donation eligibility criteria is very low among population in the northern zone. From public health perspective, low hemoglobin is amenable to address in the short term and depending on the root cause, re-entry programs to donation are important among clients deferred because of low Hb levels. The rate of blood donor deferrals in the present study was higher among replacemenr donors than in voluntary non remunerated blood donors. Similar findings have been documented by Stokx et al and by Meinia and Sawhney where replacement donors were deferred more than voluntary non remunerated blood donors. This imply that voluntary non remunerated blood donors are relatively safer than replacemnt donors, inline with the WHO recommendation. Regarding the proportion of type of donors, there has been ethusiasm among blood transfusion centers in order to comply with WHO recommendation which requre 100% of all blood to be donated from voluntary non remunerated blood donors. Some of blood donation settings have been reported to falsify the donor status from replacement donors to voluntary donors. We may not know if this may have happened in our setting where 90% of the participant were voluntary non remunerated blood donors. When interpreting the result of this study, it is important to cosider some of the study limitations. A considerable number of participants had incomplete information and were excluded from this study. Due to having incomplete information, we were not able to compare with those we included in the study. Also the study could not capture the information from self defferals blood donors (who included below 18 years and underweight 50kg), who were not registered into blood donor database at transfusion centre. The potential donors who declined interview and were not recorded may constitute special risk factors different from those who accepted interview. We may therefore not know their influence to the results presented in study. Regardless of these limitation, the present study had several strengths. The study has used a very large sample size. The study also present the combined information from blood donor screening questionnaire and TTIs results thus studying both pre-donation and postdonation deferral causes at the same time. # Conclusion A siginificant proportion of blood donor deferrals has been reported in the present study, that accounted for the inadequate supply of blood for transfusion. Transfusion transmissible infections are the leading cause of permanent blood doonor deferrals where HBV and HIV forms the highest proportion of infections. Low Hb is a leading cause for temporary deferal. There is a need to improve blood donor recruitment plans by increasing awareness of the people on blood donation and the causes for deferrals. Mass education on HBV is important so as to increase awareness of the population and ultimate prevention of HBV. The authors would like to acknowledge staff from the northern zone blood transfussion center for their cooperation and support during conducting this study. Special appreciations are registered to Dr Efesper Nkya, the program manager of National Blood Transfusion Services, for providing permission to conduct this study. We are also grateful to Dr Sia Msuya, the Director of Institute of Public Health and KCMUCollege for her support throughout the process of developing this work. [^1]: The authors have declared that no competing interests exist.

# Introduction Global cerebral ischemia/reperfusion (GCIR), a syndrome characterized by the rapid interruption of cerebral blood flow, occurs in patients who suffer from cardiac arrest, shock or complex cardiac surgery and is usually accompanied by a broad range of neuronal death in the brain\[–\]. Many of these patients suffer from various degrees of memory loss and learning dysfunction, suggesting an impairment in the hippocampus, which is the primary region of the brain that controls the formation of memories and learned behaviors. Currently, counteracting ischemia-induced cognitive impairment is challenging, and effective strategies that can attenuate the negative effects of ischemia are insufficient. The increased neurogenesis, especially in the hippocampus of adults after GCIR, has been reported to be a compensatory adaptive response to brain injury that could counteract the negative effects of cell death and cognitive dysfunction. Sleep is necessary for health and survival. After decades of research into the function the sleep-wake cycle, a large body of evidence strongly indicates that sleep deprivation (SD) has complex detrimental consequences on rodents and humans, though the mechanisms underlying these negative effects remain largely unclear\[–\]. However, growing evidence also suggests that short-term sleep deprivation (SD) with a duration of less than 48 h produces positive effects. For example, previous studies suggest that 6–12 h of short-term SD prior to cerebral ischemia produces neuroprotective effects by attenuating inflammatory responses and glial reactions in the rat hippocampus\[–\]. Moreover, a recent study showed that 24 h of short-term SD immediately following traumatic brain injury (TBI) reduces morphological damage and enhances recovery in rats. Hence, we suggest that the effect of sleep deprivation may largely depend on the time window and duration of SD. More specially, emerging studies have also reported that 12 h of short-term SD can promote neurogenesis in the hippocampus of normal rats. Based on these findings, we hypothesized that short-term SD may improve cognitive function in an ischemic model through neurogenesis- induced neuronal regeneration. Similarly, the ideal durations and time window of SD is significant. Brain-derived neurotrophic factor (BDNF) has recently been shown to be a homeostatic regulator of sleep. BDNF is also known to increase neurogenesis, neurite sprouting and other processes related to the general enhancement of hippocampal function in normal and ischemic rodents. Some reports have demonstrated that short-term SD increases BDNF expression in the hippocampus in normal rats. We speculate that the BDNF pathway mediates the neurogenesis induced by short-term SD, thereby ameliorating cognitive function in global ischemia. In light of this, the present study aimed to determine whether different durations of short-term SD could stimulate hippocampal neurogenesis and ameliorate the impaired cognitive functions induced via GCIR in rats in an appropriate time window and whether this self-repair is related to the hippocampal BDNF pathway. # Materials and Methods ## Animal groups One hundred male Sprague—Dawley rats (weighing 230±20 g), were purchased from the Experimental Animal Center of Chongqing Medical University. They were housed in groups of five in polycarbonate cages under a 12:12 h light:dark cycle (lights on from 9:00 to 21:00) with unlimited access to food and water at an ambient temperature of 21±1°C and a relative humidity of 40–50%. All experiments were approved by the Chongqing Medical University Institutional Lab Animal Care and Use Committee and were in accordance with the National Institutes of Health guidelines. The rats were maintained in the institutional animal facilities for at least 2 weeks and then randomly assigned to 3 groups: the Sham group (n = 20), the GCIR group (n = 20), and the short-term SD group. Based on the different durations of SD, the short-term SD group was randomly divided into three subgroups: the GCIR+6hSD\*3d-treated group (n = 20) rats were subjected to GCIR followed by sleep deprivation for 6 h/d continuously for 3 d; the GCIR+12hSD-treated group (n = 20), rats were subjected to GCIR followed by sleep deprivation for 12 h; and the GCIR+12hSD\*3d-treated group (n = 20) rats were subjected to GCIR followed by sleep deprivation treatment for 12 h/d continuously for 3 d. The rats were sleep-deprived starting at 48 h following GCIR. ## Global ischemia reperfusion model establishment Global cerebral ischemia/reperfusion (GCIR) modeling was performed in a manner similar to a previous method with slight modifications. Briefly, GCIR was induced via ligation of the bilateral common carotid artery combined with hemorrhagic hypotension in rats. The rats underwent fasting for 12 hours prior to surgery. After anesthesia was induced (3.5% chloral hydrate, 1 ml/100 g, i.p.), the bilateral common carotid arteries were slightly isolated, and the right jugular vein was cannulated with tubing connected to a heparinized syringe. Blood (2.5 ml/100 g) was slowly withdrawn from the right jugular vein until the volume reached standard requirements, at which time the bilateral common carotid arteries were temporarily occluded using artery clamps for 20 min; the extracted blood was then slowly reinfused, and the catheters were withdrawn. The rats in the sham group were subjected to the same operation described above, with the exception of the bilateral carotid artery occlusion and hemospasia from the right jugular vein. We discussed the success and exclusion criteria of the GCIR model in our previous report. ## Sleep deprivation procedures Sleep deprivation was conducted 48 h following GCIR using the modified multiple platform method, as described previously, which does not involve immobility stress or forced activity, resulting in less interference from other factors\[, \]. Sleep deprivation was initiated at 09:00AM during the rest circadian phase of the rats (light phase: 09:00AM-9:00PM). Briefly, groups of 8 rats were placed in water tanks (75×34×17 cm) containing 8 small circular platforms (6.5 cm in diameter). The surfaces of all platforms were 1 cm above the water level. The rats fall into the water if they lose muscle tonus, forcing them to climb back onto the platform, thus being awakened. Additionally, the rats in the sham and GCIR groups were placed in an identical apparatus that was equipped with a larger platform (18 cm in diameter) to permit sleep, serving as a control for the small platform to filter out the effects of nonspecific stressors. All of the rats were placed on the platform for 10 min twice per day beginning one week prior to the experiment to adapt to the homemade device. Moreover, food and water were provided ad libitum throughout the study, and the water in the tanks was changed daily. ## BrdU Incorporation BrdU (B5002, Sigma Aldrich, Munich, Germany) was dissolved at a concentration of 10 mg/ml in fresh isotonic sterile saline prepared just prior to use and the solution was injected in a volume of 5 ml/kg/d of the body weight. And BrdU was injected four times with a 2 h interval in the 13th day and 27th day after GCIR. Rats were sacrificed 24h after the last BrdU injection. ## Morris Water Maze Test The Morris water maze (MWM) procedure was employed to assess spatial learning and memory function of the rats 7 days after the beginning of SD procedure and included 5 days of spatial acquisition and 1 day of a probe trial. The MWM (Institute of Materia Medica, Chinese Academy of Medica Sciences, Beijing, China) was equipped with a diameter of 150 cm, height of 50 cm, water depth of 40 cm, and temperature of 24±1°C. A platform (10 cm in diameter) was submerged 1 cm below the surface of the water and placed in the middle of the same quadrant throughout the training phase. During the learning process (1–5 d), the rats (n = 6 per group) were subjected to four consecutive trials per day with intervals of 5 min. In each trial, an individual rat was placed into the pool and permitted to search for the submerged platform for 90 s. If a rat failed to locate the platform within 90 s, it would be gently guided to the platform, and the escape latency was recorded as 90 s. The mean escape latency of 4 trials was noted as the daily result of learning ability for the animal. On the 6th day of the test, each rat was placed into the pool after the platform had been previously removed and then allowed to explore the pool for 90 s. The frequency with which each rat passed the hidden platform and the resident time that each rat spent in the target quadrant were noted as the result of the spatial memory function. ## Preparation of paraffin-embedded tissue Four rats were selected in each group for histopathological observation respectively at the 14 d and 28 d timepoints after GCIR. Rats were anesthetized with 3.5% chloral hydrate (1 ml/100 g, i.p.) and rapidly perfused with ice-cold saline (approximately 200 ml. Then, rats were perfused with 250 ml of 4% paraformaldehyde until the liver became pale and the limbs and neck became straight and stiff. Finally, paraffin sections of coronal slices in the hippocampal dentate gyrus (DG) were cut with a slice thickness of 4μm. ## Preparation of fresh tissue On the 7th day after GCIR, the rats (n = 6) were decapitated, and their hippocampi were rapidly removed on ice, quickly frozen in liquid nitrogen and stored at -80°C in a refrigerator until ELISA. ## Immunohistochemistry and immunofluorescence analysis Briefly, brains embedded in paraffin were used to examine BrdU incorporation 14 d after GCIR to assess the proliferation of newly generated cells. The sections were immersed in 3% H2O2 for 30 min at 37°C to block endogenous peroxidase activity. After being washed in PBS, the sections were blocked with 5% BSA for 30 min, incubated with a primary mouse monoclonal antibody against BrdU (1:50,B2631,Sigma)overnight at 4°C, and then incubated with biotinylated goat anti-mouse IgG(1:100,Zhongshan Inc) for 60 min. Immunoreactivity was detected with 0.05% diaminobenzidine (DAB) containing 0.03% H2O2 for 5 min. To further determine the cell lineage of the proliferating cells, we examined the expression of BrdU and the neuronal marker neuron-specific enolase (NSE) 28 d after GCIR. After pretreatment with 2 N HCl, the sections prepared as described above were incubated overnight at 4°C in anti-BrdU (1:50, B2631, Sigma) and rabbit anti-NSE antibodies (1:300; Cell Signaling Technology, Danvers, MA) and then incubated for 1 h at room temperature in Alexa Fluor 488 goat anti-mouse IgG(1:500, Invitrogen, Carlsbad, CA) and Alexa Fluor 594 goat anti-rabbit IgG (1:500, Invitrogen, Carlsbad, CA) for 1 h at room temperature. The sections were washed with PBS-T 3×10 min followed by 2× 5 min washes with PBS and 2×1 min washes with water; they were then mounted with water-based mounting medium containing anti-fading agents (Biomeda, Fischer Scientific, Pittsburgh, PA). ## Image Analysis The sections were examined by light microscopy (BX51, Olympus, Tokyo, Japan). Cells were counted, in a blinded manner, within defined regions of interest in the inner edge of the granule cell layer (GCL) of the DG. Cell counting in hippocampal DG was performed in the ipsilateral DG using a series of coronal sections between 10.0 mm and 10.6 mm from the front of the brain. In total, six 4-μm coronal sections were examined per animal (n = 4 per group), spaced 100μm apart, under high-power (40×objective) using Image-Pro Plus 6.0 for Windows (Media Cybernetics, MD, USA). In each section, measurements were made in ten regions of interest. The planar area enclosed by each region was 50μm×50μm. The average ratio of BrdU-positive cells in the proportion of the total number of cells and the average ratio of BrdU/NSE-positive cells in the proportion of BrdU-positive cells in the DG per section were respectively calculated. ## BDNF ELISA The hippocampal tissue was placed in Eppendorf tubes and immersed in isopentane cooled by dry ice. This process did not take longer than 3 minutes. These samples were homogenized with a Dounce homogenizer (Kontes-7 ml, Vineland, NJ, USA) in ice-cold homogenate buffer solution. The ELISA procedure was performed according to the manufacturer’s instructions (BOSTER, China) using a microplate reader (Bio-Rad, Richmond, CA, USA). ## Statistical analysis The results were expressed as the means ± standard deviation of the means. Two- factor multi-level analysis of repeated measures was used for the comparison of the water maze escape latencies. Statistical comparisons of differences between groups for different interventions were performed using one-way ANOVA. All statistical analyses were performed with the Statistical Package for the Social Sciences (SPSS 10.0) software, and p\<0.05 was considered statistically significant. # Results ## Effect of short-term sleep deprivation on spatial learning and memory ability in rats with GCIR The Morris water maze results showed that rats in all the five groups exhibited a rapid reduction in their escape latencies to find the platform over the five training days. Compared with the S group, the rats subjected to GCIR showed a prolonged escape latency, which implies that global ischemia significantly impaired spatial learning ability, and this impairment occurred from training day 1 onward (p\<0.05). Surprisingly, there was no apparent difference between the escape latency of rats in the GCIR+12hSD group and the GCIR group. And the rats in the GCIR+6hSD\*3d group just spent less escape latency than the rats in the GCIR group on day 3 (p\<0.05). However, the escape latency in the GCIR+12hSD\*3d group was significantly shorter than which in the GCIR group, and the decrease trend substantially occurred in training day 1 onward (p\<0.05), which indicates that the 12hSD\*3d-treated method can improve the spatial learning ability of rats with global cerebral ischemia/reperfusion-induced injury. During the probe test, the platform was removed, then the resident time and frequency with which the rats crossed the target quadrant was recorded. Relative to the S group, the rats with GCIR showed less crossing times and resident time in the test (p\<0.01). However, the rats in all short-term SD groups (including GCIR+6hSD\*3d, GCIR+12hSD, and GCIR+12hSD\*3d) spent significant longer and made substantial more frequent cross-platform movements in the original platform quadrant than the rats in the GCIR group (respectively p\<0.05, p\<0.01, p\<0.01). This indicates that the rats in the S and all short-term SD groups exhibited better spatial memory regarding the original platform location. ## Effect of short-term sleep deprivation on cell proliferation in the dentate gyrus (DG) of the hippocampus in rats 14 d after GCIR BrdU-positive cells were showed brown or black irregular granular nuclear markers distributed along the hippocampal dentate gyrus (DG). The results showed that a small number of BrdU-positive cells were observed in the hippocampal DG in the S group. Relative to the S group, the number of BrdU-positive cells was substantially increased in the GCIR group (P\<0.01). While the number of BrdU- positive cells had no significant difference in the GCIR+6hSD\*3d group and GCIR+12h SD group compared with GCIR group. Furthermore, the number of BrdU- positive cells in GCIR+12hSD\*3d group was largely higher than that in the GCIR group(P\<0.05). ## Effect of short-term sleep deprivation on cell differentiation in the hippocampal dentate gyrus in rats 28 d after GCIR Four weeks after GCIR, BrdU/NSE double fluorescence-positive cells were rarely found in the S group. However, the number of double-positive cells was significantly increased in the GCIR group compared to the S group (P\<0.01). The BrdU/NSE double-positive number in the GCIR+6hSD\*3d group had no difference compared to the GCIR group. Moreover, a higher number of BrdU/NSE-positive cells was observed in short-term SD groups (including the GCIR+12hSD group and GCIR+12hSD\*3d group) compared with the GCIR group (P\<0.05). ## The expression of hippocampal BDNF in rats 7 d after GCIR The expression of hippocampal BDNF in all short-term SD groups (including GCIR+6hSD\*3d, GCIR+12hSD, and GCIR+12hSD\*3d) was significantly increased relative to the GCIR group (P\<0.05), and the most substantial increase was observed in the GCIR+12hSD\*3d group. # Discussion In this study, we examined the beneficial effects of different durations of short-term SD on learning, memory function and neurogenesis in the hippocampus of rats exposed to GCIR. We found that after GCIR, (1) short-term SD could ameliorate the impairments in learning and memory, (2) short-term SD could promote the proliferation and differentiation of newly generated cells in the hippocampal DG, (3) the BDNF signaling pathway participated in the stimulation of neurogenesis, and (4) the GCIR+12hSD\*3d-treated method may be the most appropriate way to implementation of sleep intervention in the rat with GCIR. As we know the brain is highly sensitive to ischemia, and reperfusion exacerbates ischemic damage. The hippocampal neurons, particularly the pyramidal neurons in the CA1 region, are known to be the most sensitive to the deleterious effects of cerebral ischemia/reperfusion. And as the hippocampus is an important region for learning and memory in mammals, the majority of ischemia survivors suffer from various cognitive dysfunctions. Using the current model, our study showed that global ischemia significantly compromised spatial learning and memory function. In the present study, the rats in the GCIR+12hSD\*3d group showed less escape latency in the five training days relative to the GCIR group, suggesting an improvement in spatial learning. And all groups of rats treated with short-term SD in our experiment spent significant longer and made substantial more frequent cross-platform movements in the original platform quadrant than the rats underwent GCIR alone in the probe test, suggesting an attenuation of memory dysfunction. In total, the group of rats subjected to 12 h SD for 3 d consistently showed the greatest recovery of cognitive function. Similar results have also been reported in the literature; for instance, Moldovan et al. found that rats with occlusion of the middle cerebral artery (MCAO) that were subjected to pretreatment with 6 h of SD showed better learning and memory performance during the first week of recovery; this was also consistent with reduced signs of morphological damage. Additionally, Martinez-Vargas showed that 24 h of SD after a TBI had a neuroprotective effect by reducing morphological damage and enhancing recovery in rats. The mechanism of how short-term SD promotes neurological recovery from brain injury is largely unclear. Some previous reports suggested that sleep deprivation prior to transient GCIR could attenuate hippocampal damage by attenuating inflammatory responses or glial scar formation. In the present study, we induced sleep deprivation intervention 48 h following GCIR, and focused on observing the neurogenesis effects of different durations of short-term SD. A considerable degree of neurogenesis has been reported to occur following cerebral ischemia in the adult mammalian brain. The regional correlation of neurogenesis in the hippocampal DG has also been studied in various disorders that cause memory impairment. Global ischemia stimulates cell proliferation in the hippocampal DG, which peaks at 1–2 weeks after ischemia. Many newly generated cells die in the weeks following ischemia, but the minority of cells that survive will differentiate into neurons approximately 4 weeks later. In our study, compared with the GCIR group, a larger number of BrdU-positive cells was observed 14 days after GCIR in the hippocampal DG of the rats treated with short-term SD.NSE, a neuronal marker, can further determine the cell lineage of the newly generated cells. We also observed an increased number of BrdU/NSE- positive cells in the short-term SD groups, especially in the GCIR+12hSD group and GCIR+12hSD\*3d group, 28 days after GCIR compared with the GCIR group. These results indicated that short-term SD not only promoted the proliferation of newly generated cells but also stimulated the proliferating cells to differentiate into neurons in the hippocampal DG. However, the most effective proliferation and differentiation occurred in the GCIR rats with 12 h SD for 3 successive days. Our results are in agreement with previous studies indicating that sleep deprivation for relatively short periods(\<48 h), such as 12 h, significantly stimulates neurogenesis in the hippocampus of normal rats by enhancing cell proliferation and the survival of newly generated cells. Based on these data, it may be tentatively concluded that delayed short-term SD (after 48 h) facilitates cell repair in the hippocampus through the stimulation of neurogenesis via endogenous progenitor cells to improve behavioral recovery and the most appropriate way is 12 h SD for 3 successive days. In our experiment, the ELISA results showed that the BDNF level of hippocampal tissues was significantly increased after different periods of short-term SD 7 d following GCIR rats, which is in line with earlier reports suggesting that 12 h of short-term SD could increase the expression of hippocampal BDNF. Numerous evidences have also indicated that BDNF is required for basal neurogenesis in the hippocampal DG. The knockdown of BDNF by RNA interference and lentiviral-mediated gene silencing in the DG reduces hippocampal neurogenesis. Therefore, BDNF signaling is one possible mechanism for short- term SD-induced neurogenesis, and further research is needed to clarify the down-stream signaling mechanisms. # Conclusion The current study clearly demonstrates that 48 h delayed short-term SD, especially the GCIR+12hSD\*3d-treated method, is effective in promoting neurogenesis in the hippocampal DG and facilitating cognitive recovery following GCIR in the adult rat. Additionally, the underlying mechanism of short-term SD in neurogenesis may be mediated by the BDNF signaling pathway. These results suggest that short-term SD in an appropriate way may be a potential strategy for the clinical treatment of ischemic injury in brain tissues. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: OC RL JY. Performed the experiments: OC RL LZ LY BY JW. Analyzed the data: OC RL LZ. Contributed reagents/materials/analysis tools: OC RL LZ LY BY JW BC. Wrote the paper: OC RL JY. [^3]: ‡ These authors are co-first authors on this work.

# Introduction The giant clam *Tridacna crocea*, inhabiting many Indo-Pacific coral reef communities, is remarkable for its fantastic mantle and unique ability to bore fully into coral rock. Similar to other tridacnids, there is a symbiotic relationship between *T*. *crocea* and symbiotic dinoflagellates, which are commonly called zooxanthellae. Symbionts of clams are intercellular, and the zooxanthellae live within a branched, tubular structure permeating the mantle;, [the](http://bbs.pigai.org/t45156-1-1.html) symbionts produce nutrients via photosynthesis to aid in the host’s autotrophy, while they gain essential nutrients for growth and metabolism from the host. Because of the characteristics of giant clams, they are effective ecosystem engineers playing multiple roles in coral reefs. Nevertheless, because of environmental and anthropogenic disturbances, such as over-harvesting and habitat destruction, giant clam populations have been depleted, and their densities have been insufficient for self-replenishing and maintaining their populations, which might eventually lead to their extinction. In recent years, the breeding of giant clams has attracted increasing attention worldwide. However, diseases caused by bacterial pathogens have been reported to result in losses in aquaculture populations of these commercially important shellfish. Pathogenic vibrios can affect larval stages of cultured bivalves and are also involved in diseases of juveniles and adults. *Vibrio coralliilyticus* was originally known as a temperature-dependent etiological agent involved in coral bleaching; it targets the mucus of the coral host through the use of dimethylsulfoniopropionate as a chemotaxic and chemokinetic agent, and its use of extracellular proteases results in coral tissue lysis and symbiont density decrease. Lately, it was demonstrated to cause mortalities in larval oysters. Since *V*. *coralliilyticus* is phylogenetically related to *Vibrio tubiashii*, many marine isolates of *V*. *coralliilyticus* from shellfish were misidentified as *V*. *tubiashii*. Therefore, previous reports of *V*. *tubiashii* in bivalve shellfish aquaculture on the west coast of North America were possibly caused by *V*. *coralliilyticus*. It has been reported that gross pathological changes to the velum and cilia occurred in diseased *Crassostrea gigas* larvae. Moreover, *V*. *coralliilyticus* has also been associated with outbreaks of vibriosis in several other bivalve species, including Eastern oysters (*Crassostrea virginica*), the European flat oyster (*Ostrea edulis*), the great scallop (*Pecten maximus*), the Atlantic bay scallop (*Argopecten irradians*) and New Zealand green-lipped mussels (*Perna canaliculus*). Consequently, *V*. *coralliilyticus* is a significant pathogen for bivalves, contributing to dramatic losses in mollusks worldwide. Bivalves mostly rely on an innate immune system composed of cellular and humoral components. Hemocytes are thought to be responsible for the main activities by which bivalves respond to infectious agents, and those are phagocytosis, encapsulation and nacrezation. The first action of the bivalve’s immune system when challenged by microbial pathogens is the recognition of these foreign organisms. This is achieved by expressing various pattern recognition receptors (PRRs), which sense diverse pathogen-associated molecular patterns (PAMPs). Recognition of PAMPs by PRRs activates intracellular signaling pathways that culminate in the induction of inflammatory cytokines, chemokines, interferons (IFNs) and the upregulation of costimulatory molecules. Toll-like receptors (TLRs) are particularly essential members of PRRs, and they are responsible for the recognition of PAMPs and the activation of downstream signaling adaptors. Among these adaptors, myeloid differentiation factor 88 (MyD88), which mediates the activation of the TLRs, except for TLR3, was originally identified as a curial and conserved signaling proteins. MyD88 interacts with IL-1 receptor- associated kinase (IRAK), and IRAK associates with TNF receptor-associated factor 6 (TRAF6), which subsequently activates the NF-кB pathway. In addition, the TLR pathway not only activates inflammation and phagocytosis but also regulates the induction of apoptosis. Apoptosis is important in the immune system and plays significant roles in the control of the immune response, the removal of immune cells recognizing self-antigens, and cytotoxic killing. Pathogens can be eliminated by the ability of hemocytes to recognize foreign targets and induce apoptosis. Similarly, *V*. *coralliilyticus* exposure can induce significant changes in the host TLR pathway and apoptotic systems. Since the first TLR was identified in *Drosophila melanogaster*, large members of TLR family have been recently investigated in marine bivalves, such as *Chlamys farreri*, *Chlamys nobilis*, *Mizuhopecten yessoensis*, *C*. *gigas*, *C*. *virginica*, *Mytilus edulis* and *Hyriopsis cumingii*. Among them, the innate immune regulation of TLR genes in bivalves has been reported in *C*. *farrer*i (*Cf*Toll-1) and *C*. *nobilis* (*Cn*TLR-1), respectively, both of which might be involved in the immune response against pathogen invasion. In addition, 23 TLRs were identified and arranged in 4 clusters according to extra-cellular LRR domain content in *Mytilus galloprovincialis*. Furthermore, there were 83 TLR genes in the genome of *C*. *gigas*, 19 of which had different responses to *Vibrio* infection. Despite the fact that *T*. *crocea* has high ecological value and is under the stress of populations, there is limited information available about its immune molecule response mechanisms. In the present study, we injected adult *T*. *crocea* with *V*. *coralliilyticus* to investigate the individual and cellular responses of giant clams. Furthermore, high-throughput sequencing was applied to analyze the differentially expressed genes in *T*. *crocea* hemocytes at 0 h, 6 h, 12 h, and 24 h after *V*. *coralliilyticus* challenge. Afterwards, using KEGG pathway enrichment analysis, some molecular mechanisms of response and candidate genes involved in *V*. *coralliilyticus* infection were identified. Therefore, we explored the possible sensing patterns for *V*. *coralliilyticus* in the innate immune system of *T*. *crocea*. The results provide insights for a better understanding of *V*. *coralliilyticus* pathogenicity and the future development of disease prevention strategies. # Materials and methods ## Ethics statement All *T*. *crocea* used in the present study were bred in our laboratory (Hainan Tropical Marine Biology Research Station), Chinese Academy of Sciences, Sanya, China. No specific permits were required for clams sample collection or described sampling. The location was not privately-owned or protected, and the field studies did not involve any endangered or protected species. ## Animals, tissue, embryonic development collection and challenge experiment Adult *T*. *crocea* (average 6.7–10.8 cm shell length) were obtained from Sanya, Hainan Province, China and maintained in tanks filled with natural seawater (temperature: 28±1°C; salinity: 33). Cultures were illuminated with metal halide bulbs from 6:00 am to 7:00 pm every day for two weeks prior to use. To analyze the tissue distribution of *Tc*MyD88, the following tissues were collected from six healthy adult *T*. *crocea*: siphonal mantle, pedal mantle, inner mantle, byssus gland, pedis, heart, gill and hemocytes. For analysis of the developmental expression patterns of *Tc*MyD88, samples of 0, 0.5, 1, 4, 12, 16, 24, 48, 72, 96, and 120 h-developing embryos were also obtained. For the bacterial challenge, an in vivo infection experiment was performed. Healthy giant clams were injected into the adductor muscle either 100 μL of 2×PBS or 100 μL of live *V*. *coralliilyticus* (1.0 ×10⁸ CFU/mL) suspended in 2×PBS. Specifically, purified bacterial inoculum were centrifuged for 10 min to collect bacterial pullets and were washed 3 times with PBS and re- suspended in 2×PBS to a concentration of OD600nm = 1.0. *V*. *coralliilyticus* (CAIM616) was purchased from the Marine Culture Collection of China and was cultured in Zobell marine broth 2216 (Difco Laboratories) at 28 °C overnight. After injection, the giant clams were returned to tanks full of seawater at 28 °C for subsequent treatment or sampling. If the valves of a giant clam were not closed and the mantle did not react after stimulation, it was considered dead. Mortality in each tank was assessed visually and counted every 6 h after injection. Hemocytes were collected at scheduled intervals (0, 3, 6, 12, 24, and 36 h post injection) from both the challenged and control groups. Among them, hemocytes taken at 0, 6, 12, and 24 h post challenge were stored in liquid nitrogen for transcriptome analysis. Five individuals were randomly sampled from each group at every time point after injection to obtain biological replicates. ## Flow cytometric analysis of apoptosis Hemocytes were harvested 6 h after *V*. *coralliilyticus* injection and were then resuspended in 100 μL of binding buffer containing 5 μL of Annexin V-FITC and 5 μL of propidium iodide, provided in an apoptosis detection kit (Vazyme, A211); the mix was incubated at room temperature for 10 min in the dark. Finally, another 400 μL of binding buffer was added to the solution and applied to a Guava^® easyCyte^™ (Millipore). At least 10,000 cells were obtained to analyze the population by Flow Jo v10.0 software. ## De novo assembly and gene annotation Based on the sequencing by synthesis (SBS) technique, cDNA libraries were sequenced using an Illumina Hiseq high-throughput metering platform to obtain raw data. The raw data were processed to discard the dirty reads and low-quality sequences. After filtering, the remaining reads were called “Clean Reads”. The unigenes were de novo assembled using Trinity. For gene annotation analysis, the assembled transcripts were scanned against NR (NCBI nonredundant protein sequences), Swiss-Prot databases and KOG (euKaryotic Ortholog Groups) using diamond with E-values at 1.0×10⁻⁵ (E-values of less than 1.0×10⁻⁵ were considered significant). NT (NCBI nucleotide sequences) and PFAM (protein family) analyses were performed using the NCBI BLAST + v2.2.28 and HMMER 3.0 programs, respectively. In addition, the unigenes were also classified according to the GO (Gene Ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) databases. ## Differential expression analysis Read counts were estimated by mapping clean reads to unigenes using Bowtie2, and they were calculated according to the comparison result with RSEM. The expression abundance of the corresponding Unigene was expressed by the FPKM (expected number of fragments per kilobase of transcript sequence per millions of base pairs sequenced) value. We used the adjusted p-value to detect differentially expressed genes (DEGs). When the adjusted p-value was less than 0.05 and had a greater than two-fold change (absolute value of log2 ratio≥1), the gene was considered differentially expressed in a given library. Significantly enriched terms were obtained by mapping DEGs to the corresponding KEGG pathways. ## RNA extraction and cDNA synthesis Total RNA (50mg) was isolated with TRIzol Reagent (Invitrogen, 15596–026) according to the manufacturer’s protocol. RNA was dissolved in DEPC-treated water, and the integrity of the RNA was assessed by electrophoresis with a 1.0% agarose gel, and then the concentration and purity were examined at 260/230 and 260/280 absorbance ratios. Purified RNA was diluted to 1 mg/ml to synthesize first-strand cDNA using a Primer Script^™ First Strand cDNA Synthesis kit (TAKARA Bio Inc. Japan). The cDNA was used as the template for amplifying gene sequences and analyzing their expression. All primers used in this study were designed with Primer Premier 5.0 and are shown in. ## Cloning the full-length cDNA of *Tc*MyD88 A search of the transcriptome data of *T*. *crocea* revealed a TIR contig homologous to the MyD88 gene of *T*. *crocea*. The intermediate fragment sequences of *Tc*MyD88 were obtained by polymerase chain reaction (PCR). Then, gene-specific primers were designed to amplify the unknown 5’ and 3’ ends of *Tc*MyD88 cDNA using rapid amplification of cDNA ends (RACE). For *Tc*MyD88 3’ sequencing, the primer pairs Takara3P/*Tc*MyD88-F1 and Takara3NP/*Tc*MyD88-F2 were employed for primary PCR and nested PCR, respectively. Similarly, the 5’ end of the *Tc*MyD88 gene was obtained by nested PCR using Takara5P/*Tc*MyD88-R1 and Takara5NP/*Tc*MyD88-R2. Full-length cDNA sequences were obtained by combining intermediate fragment sequences and 3’ and 5’ end sequences. ## Sequence and phylogenetic analyses The cDNA sequences and deduced amino acid sequences of *Tc*MyD88 were analyzed using the BLAST program (<http://blast.ncbi.nlm.nih.gov/Blast.cgi>) and the Expert Protein Analysis System (<http://expasy.org/>). The nucleotide and protein sequences were analyzed using BLASTN and BLASTX, respectively. The molecular weights and the theoretical isoelectric points were calculated using the Compute pI/Mw tool (<http://web.expasy.org/compute_pi/>). A structural analysis of proteins was performed using the Simple Modular Architecture Research Tool (SMART) program (<http://smart.embl-heidelberg.de/>). Multiple protein sequence alignments were performed using Bio-edit software via the Clustal W method. A phylogenetic tree was constructed using the MEGA5.0 software based on the alignment of the complete amino acid sequences with the neighbor- joining method and 1000 bootstrap replicates. ## Plasmid construction, cell culture and transfection The *Tc*MyD88 ORF was amplified from *T*. *crocea* cDNA by PCR using specific primers designed by CE Design V1.03. The target PCR products were subcloned into eukaryotic expression vectors, pcDNA3.1-HIS (Promega, USA) and pEGFP-N1 (Promega, USA), by homologous recombination using a Vazyme^™ one step cloning kit. The constructed recombinant plasmids were digested with enzymes, and the inserted fragments of each clone were sequenced. All of the plasmids used for transfection were extracted from overnight bacterial cultures using a HiPure Plasmid EF Micro kit (Magen) according to the manufacturer’s protocol. HEK293T cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco) supplemented with 10% heat-inactivated fetal bovine serum (FBS) and antibiotics (100 mg/L streptomycin and 105 U/L penicillin, Gibco) at 37 °C with 5% CO₂, and they were subcultured every three days. Transfections were performed using ViaFect reagent (Promega, USA) according to the manufacturer’s instructions. ## Subcellular localization and dual-luciferase reporter assays For subcellular localization analysis, HEK293T cells were seeded into a 6-well cell plate, cultured for 24 h and then transfected with a pEGFP-N1-*Tc*MyD88 mix. Forty-eight hours after transfection, cells were washed once with 1×PBS (pH 7.4) and then were fixed with 4% paraformaldehyde for 10 min, which was followed by staining with 6-diamidino-2-phenyl-indole (DAPI) (1 mg/ml) for 5 min. Finally, the cells transfected with fluorescent vectors were directly observed by fluorescence microscopy. For dual-luciferase reporter assays, cells were plated in a 48-well cell plate, cultured until they reached 40–50% confluency and then were transfected with pRL-TK (20 ng/well), NF-κB reporter (200 ng/well) and a target plasmid (0, 100, 200 or 400 ng/well). The pRL-TK vector and NF-κB vector (Promega, USA) were used as internal controls. After 48 h of transfection, the luciferase activity of total cell lysates was measured using a luciferase reporter assay system (Promega, USA). Renilla luciferase activity was expressed as the fold stimulation relative to that of the empty vector transfected cells. The values are expressed as the mean relative stimulation for a representative experiment from four separate experiments, and each experiment was performed in duplicate. ## Quantitative real-time PCR analysis qRT-PCR reactions were conducted using 2 × Real Star Green Power Mixture (GenStar, A311) and a LightCycler^® 480 II (Roche, Switzerland) according to the manufacturer’s protocol. All the template cDNA were diluted to 200ng/μl. PCR conditions were as follows: 95 °C for 10 min, followed by 40 cycles of amplification at 95 °C for 15 s, 55 °C for 30 s and 72 °C for 30 s. At the end of each qPCR, a melting curve analysis was performed to confirm the specificity of the PCR products. The data from each experiment are expressed relative to the expression levels of the β-actin gene to normalize expression levels between the samples. All of the samples were analyzed in triplicate, and the expression values were calculated with the 2^-ΔΔCT method. ## Statistical analysis Data processing and statistical analyses were performed using GraphPad Prism v5.0.1. All data are presented as the means ± S.D. Comparisons between two groups of samples were performed with Student’s t-tests, and comparisons among more than two groups were performed with one-way ANOVA followed by Turkey’s or Dunn’s post hoc test. Values \*p \< 0.05, \*\*p \< 0.01 and \*\*\*p \< 0.001 were considered significant. # Results ## The mortality and apoptosis after vibrio challenge To investigate the possible effects of a *V*. *coralliilyticus* challenge on *T*. *crocea*, challenge experiment was performed. After injection with *V*. *coralliilyticus*, the siphonal mantle of *T*. *crocea* contracted inward and collapsed. Bacterial challenge directly affected adult survival; dead individuals appeared at 6 h post infection, and 45.56% and 56.78% mortality rates were observed in the two challenged groups, while no death was observed in the control group after 36 h. Flow cytometry was performed to detect the apoptosis of hemocytes. The results showed that the apoptosis rate of the hemocytes dramatically increased 1.8-fold (including early and late apoptotic cells) in the *V*. *coralliilyticus*-injected *T*. *crocea* compared with that of the control group, which clearly implied that *V*. *coralliilyticus* could cause apoptosis in *T*. *crocea* hemocytes. ## Transcriptome sequencing and assembly of transcripts The transcripts of *T*. *crocea* injected with *V*. *coralliilyticus* were assembled to analyze the molecular response mechanism of *T*. *crocea* during *V*. *coralliilyticus* infection. After filtering dirty reads from the raw reads, 46525916, 44448554, 49647448 and 53853546 clean reads were obtained from hemocytes in the *V*. *coralliilyticus* challenged group at 0 h, 6 h, 12 h, and 24 h, respectively. Sequencing reads were found to be of high quality, with Q30% values of 92.97%, 92.33%, 91.89% and 93.03%, respectively. The GC content of nucleotides was approximately 38%. For all four sequencing libraries, the percentages of reads that could be matched with assembled reference sequences were higher than 64.80%. All sequencing reads were stored into the Short Read Archive (SRA) database and available with the accession PRJNA596951. ## Annotation and function analysis of unigenes To achieve protein identification and gene annotation, transcripts were compared with data from the NR, NT, KO, Swiss-Prot, PFAM, GO and KOG databases. A total of 195651 unigenes (ranging from 301 to 25533 bp) were generated with an N50 size of 810 bp. All sequences from 75468 unigenes were annotated in at least one database, with 28385 annotated unigenes (14.50%) having significant matches with sequences in the NR. We also assembled the species distribution of the unigenes by aligning sequences against the NR database to learn the sequence similarity of *T*. *crocea* with other species. A total of 62.5% of the unigenes matched with sequences from five top-hit species, *M*. *yessoensis* (25.4%), *C*. *gigas* (14.3%), *C*. *virginica* (14.0%), *Lottia gigantea* (5.2%) and *Lingula anatina* (3.6%), all of which are mollusks. GO classification is a unified gene functional classification system. The analysis of annotated transcripts demonstrated that 36361 matched unigenes (18.58%) were divided into 26, 20, 10 entries for three categories, biological processes (BP), cellular component (CC) and molecular function (MF), respectively. The unigenes in “binding,” “catalytic activity,” “signaling,” “response to stimulus,” “membrane,” which were potentially related to immune existed in higher percentages compared to counterparts at sub-categories at the whole transcriptome reference. They played important roles during the defense process after challenge. The results provided a comprehensive view for screening candidate genes related to immune and defense mechanism. In addition, we also used KEGG to identify the probable functional status of assembled transcripts. A total of 12912 unigenes (6.59%) were assigned to five main categories containing 232 KEGG pathways. Remarkably, immune system (558) was the well-represented term in the KEGG organismal systems category. Signal transduction (1455) was the most enriched terms in both environmental information processing and the whole transcriptome reference, suggesting the dominant position of various signal transduction in processing the stimulation of environment. In the metabolism category, abundant unigenes were found in the amino acid metabolism (510) sub-category. These results indicated that these unigenes may play a role in mediating *V*. *coralliilyticus* exposure and associated impacts. ## Identification of differentially expressed genes To identify the DEGs involved in *V*. *coralliilyticus* infection, a comparison of the relative transcript abundance for each unigene was performed. We compared the expression levels of each unigene for 6 h, 12 h, and 24 h with 0 h, obtaining 1941, 2172 and 1116 DEGs, respectively. By taking the union of three sets of DEGs, we ended up with 3446 DEGs. Subsequently, KEGG pathway enrichment analysis was used to classify the DEGs and highlight biological associations. As expected, the results emphasized the immune system, where the Toll-like receptor signaling pathway and the NF-кB signaling pathway were enriched in DEGs. Furthermore, some DEGs were associated with apoptotic signaling, such as the TNF signaling pathway. Meanwhile, we selected some of the DEGs related to innate immunity and constructed a heat map in which 39 genes were upregulated and 16 genes were downregulated. The upregulated genes were mainly TLR pathway-related genes, immune effector molecules (which included inflammatory cytokines interleukin-1 (IL-1), IL-17 and tumor necrosis factor (TNF)), antimicrobial peptides (β defensin and phage lysozyme 2), and apoptosis-related genes (such as inhibitor of apoptosis proteins 1 (IAP1) and the anti-apoptotic genes of the Bcl-2 family). For the TLR pathway, one TLR homolog, the adaptor molecule MyD88, TRAF6, kinase IRAK4 and two transcription factors IкB-α and NF-кB were all significantly upregulated after 6 h and 12 h of infection. The downregulated genes included members of the lectin family and the complement pathway, including perlucin, galectin, and C-type lectin. The results revealed that the MyD88-dependent pathway was activated after *V*. *coralliilyticus* challenge, which promoted the release of immune effector molecules such as IL-1 and TNF, and indirectly affected the expression of some apoptosis-related genes. ## Cloning and sequence analyses of *Tc*MyD88 *Tc*MyD88 was dramatically enriched in multiple KEGG pathways and occupied an essential position in the TLR pathway. The full-length sequence of *Tc*MyD88 (MN829819) is 1853 bp, containing a 105 bp 5’-untranslated region (UTR), a 140 bp 3’-UTR and an open reading frame (ORF) of 1608 bp. The ORF encodes a putative protein of 535 amino acids, with an estimated molecular mass of 60.430 kDa and a theoretical isoelectric point (PI) of 5.04. The protein structure of *Tc*MyD88 was predicted and analyzed using the SMART program. Typical Myd88 domains comprising a DEATH domain (residues 24–123) and a TIR domain (residues 179–316) were identified in the *Tc*MyD88 protein. To clarify the evolutionary relationship among variable MyD88 proteins, multiple sequence alignments were performed with the deduced amino acid sequences of MyD88. Compared with other species, *Tc*MyD88 displayed a high degree of homology and conservation, especially in the TIR domain. Moreover, a phylogenetic tree was also constructed using MEGA5.0 software using the neighbor-joining method. The deduced phylogeny of MyD88 revealed two major clusters, one for mollusks and the other for fish and mammals. *Tc*MyD88 initially clustered with bivalves, such as *Ruditapes philippinarum*, *H*. *cumingii*, *C*. *virginica*, and *C*. *giga*, to form a single branch. In conclusion, the evolution of *Tc*MyD88 is relatively well conserved in bivalves, and it is distantly related to that of vertebrates and other invertebrates. ## Expression pattern variation of *Tc*MyD88 in tridacna The tissue distributions of *Tc*MyD88 mRNA were detected by qRT-PCR. According to the results, *Tc*MyD88 was widely expressed in all examined tissues, including the siphonal mantle, pedal mantle, inner mantle, byssus gland, pedis, heart, gill and hemocytes. With the change in the position of mantles, the expression level of *Tc*MyD88 varied, while it was relatively highly expressed in the siphonal mantle. Furthermore, it was expressed predominantly in the gills, followed by the siphonal mantle and hemocytes, whereas the pedis, heart and byssus gland contained low levels. The expression level of *Tc*MyD88 was approximately 33.7-fold higher in gills than it was in hearts. The mRNA expression pattern of *Tc*MyD88 was also observed throughout developmental stages. As shown in the figure, the expression of *Tc*MyD88 maintained a stable, low level from 0 h to 72 h. At 72 h, there was a sudden increase, and the level continuously increased after 72 h. ## Subcellular localization of *Tc*MyD88 The subcellular localization of *Tc*MyD88 was examined by transient transfection of the *Tc*MyD88-GFP plasmid into HEK293T cells. Imaging of the GFP-tagged *Tc*MyD88 revealed that *Tc*MyD88 was distributed mainly in the cytoplasm, whereas the GFP protein was found in both the cytoplasm and the nucleus. ## Dual-luciferase reporter assays To determine whether *Tc*MyD88 could modulate NF-кB transcriptional activity, dual-luciferase reporter assays were performed in HEK293T cells. As shown in, *Tc*MyD88 can activate the NF-κB responsive reporter, and the overexpression significantly increased the activation in a dose-dependent manner from 100 ng to 400 ng. The most marked increase was approximately 10.9-fold (P \< 0.01) over what was observed in cells transfected with pCDNA3.1 alone. These results implied that *Tc*MyD88 could potentially be involved in the NF-кB signaling pathway. ## The responses of TLR pathway-related genes in *T*. *crocea* to *V*. *coralliilyticus* challenge On account of the DEG analysis showing that the TLR pathway was activated, we proceeded to analyze pathway-related genes by qRT-PCR. For the TLR pathway genes, significant differences between control and infected *T*. *crocea* individuals were observed for TLR, MyD88, IRAK4, TRAF6 and IкB-α. Upon *V*. *coralliilyticus* challenge, the level of TLR transcripts was significantly upregulated at 6 h (22.9-fold; p \< 0.01) and then returned to the basal level at 36 h. With *V*. *coralliilyticus* stimulation, the expression of *Tc*MyD88 initially significantly increased at 3 h (4.8-fold; p \< 0.01) and reached the highest expression levels at 6 h (9.6-fold; p \< 0.01), and then it declined at 12 h. During the *V*. *coralliilyticus* challenge, the expression level of IRAK4 was upregulated at 3 h and maintained a high level throughout the whole stage, with the highest value at 6 h (4.0-fold; p \< 0.01); then it returned to the basal level at 36 h. With exposure to *V*. *coralliilyticus*, TRAF6 expression was significantly upregulated by 3.6-fold (p \< 0.01) at 6 h, and then it peaked with a 2.8-fold increase at 12 h (p \< 0.001). After *V*. *coralliilyticus* infection, the expression of IкB-α rapidly increased at 6 h (3.9-fold; p \< 0.001) and then dropped to 1.5-fold at 12 h (p \< 0.05) before nearing the control level at 36 h. The expression of cytokine-related genes was also affected by infection. In response to *V*. *coralliilyticus* challenge, IL-17 exhibited an increase at 3 h and 6 h by 6.3-fold (p \< 0.001) and 10.5-fold (p \< 0.001), respectively. There was also a significant change in apoptosis- related gene IAP1. During *V*. *coralliilyticus* infection, IAP1 mRNA was increased at 3 h (2.0-fold; p \< 0.01), and it dramatically increased at 6 h with a 3.0-fold change (p \< 0.001). In summary, the qRT-PCR analyses greatly confirmed the direction of changes determined by the transcriptome analysis. Preliminary indications are that the expression of TLR pathway-related genes changed after *V*. *coralliilyticus* challenge, and the downstream genes were activated, which promoted the release of inflammatory and led to the activation of a series of signals. # Discussion In a global warming scenario, an increase in the seawater temperature could promote the potential disease outbreaks associated with *V*. *coralliilyticus* in mollusks. In the present study, we demonstrated that *V*. *coralliilyticus* could cause acute mortalities of adult *T*. *crocea* when previous work was mainly focused on mortalities of shellfish larvae caused by *V*. *coralliilyticus*. After injection with *V*. *coralliilyticus*, the siphonal mantle of *T*. *crocea* contracted inward and collapsed, while the siphonal mantle of healthy *T*. *crocea* extended outward. In previous studies, *V*. *coralliilyticus* infection resulted in the loss of *C*. *gigas* larval motility due to the gross pathological changes to the velum and cilia. Histopathology indicated that the route of infection by *V*. *coralliilyticus* was the digestive system in Greenshell^™ mussel larvae. Furthermore, vacuolation of the tissues of the digestive tract similar to oyster *C*. *virginica* larvae and necrotic tissue were observed. Moreover, the results of this study suggest that *V*. *coralliilyticus* could trigger apoptosis in *T*. *crocea* hemocytes. Diverse metalloprotease and effector genes like the pore forming toxin hlyA were identified in the genome of *V*. *coralliilyticus* and expressed proteases were also detected in the secretome, which caused mortality in *Drosophila* and *Artemia* and may be involved in the infection of *T*. *crocea*. It has been reported that hemolysins play a role in inducing apoptosis. In addition, *C*. *gigas* larvae infected by *V*. *coralliilyticus* showed higher activities of catalase and superoxide dismutase, two key enzymes implicated in antioxidant defense, indicating their activation after pathogen stimulation; further, PO activity was significantly increased in challenged mussel larvae. While sequencing the genomes of *V*. *coralliilyticus* isolates has indicated a varying repertoire of potential virulence factors that may function independently or in concert to induce pathogenicity, hemolysin and extracellular protease activities are thought to play important roles during pathogenesis in oysters. It was found that lytic enzymes produced by aquaculture pathogens include haemolysins and proteases. Extracellular metalloproteases facilitate bacterial invasion and the infection process, acting to enhance tissue permeability and leading to necrotic tissue damage and cytotoxicity in the host. The discovery in our study filled the research blank of rapid lethality caused by *V*. *coralliilyticus* in adult bivalves. Furthermore, we dissected this phenomenon from a mechanistic molecular level. To shed light on the molecular response mechanism of *T*. *crocea* during *V*. *coralliilyticus* infection, we used transcriptome sequencing analysis and other relevant techniques to analyze the *T*. *crocea* hemocytes at 0 h, 6 h, 12 h and 24 h after *V*. *coralliilyticus* challenge. *T*. *crocea* employs the innate immune response as the sole defense mechanism against pathogen infection, such as pathogen recognition and apoptosis systems. Our results revealed that the total number of unigenes and DEGs was 195651 and 3446, respectively. More details were uncovered by KEGG pathway enrichment analysis, where DEGs were found to be significantly enriched in immune-related signaling pathways, such as the TLR signaling pathway and the NF-κB pathway, and some were associated with apoptotic pathways, such as the TNF signaling pathway. These results were consistent with the quantitative validation, where it was preliminarily demonstrated that the expression of TLR pathway-related genes changed after *V*. *coralliilyticus* challenge, and downstream genes were activated, which promoted the release of inflammatory factors such as IL-17 and led to the activation of a series of signals. Taken together, these results coincided with previous work in Pacific oyster larvae and adult *Nematostella vectensis* in which the TLR-to-NF- κB pathway was activated during pathogenic conditions. However, this activation reaction is not absolute, as shown here; some downregulated genes, such as the partially lectin family, are components of the complement system. The complement system enables one of the major innate immune mechanisms, which has the ability to remove microbes and attack pathogens, and the existence of a potential multicomponent complement system has been identified in shellfish. When *C*. *gigas* was infected with *V*. *splendidus*, suppressed genes may have helped the bacteria escape the giant clam response, and afterward, the bacteria were able to establish the pathogenic intracellular and intravesicular forms mediated by localization that enhanced bacterial protection from an innate immune attack. Furthermore, *Tc*MyD88 of the TLR pathway was identified by KEGG pathway enrichment analysis. MyD88 has been demonstrated to be a key adapter protein in TLR signal transduction that triggers downstream cascades involved in innate immunity in organisms ranging from mammals to mollusks, that includes human, fish, Drosophila, and *C*. *gigas*. The MyD88-dependent pathway leads to the activation of NF-κB and the expression of proinflammatory genes, such as TNF and IL-1. *Tc*MyD88 was found to have two conserved domains, the death domain and the TIR domain. Furthermore, the expression profile of *Tc*MyD88 and the activation of NF-кB both revealed that *Tc*MyD88 plays a crucial role in the regulation of the *T*. *crocea* immune system. These results verified the existence and significance of a MyD88-dependent signaling pathway in *T*. *crocea*. Acting as an intermediate receptor for signal transduction, MyD88 participates in the transmission of multiple signaling pathways. TNF expression and apoptosis have been reported to share the same signal transduction molecule, MyD88, in human myelomonocytic cells. Meanwhile, the TLR pathway and other signaling pathways may increase the activation of MyD88. Therefore, *Tc*MyD88 may serve as a link among different immune regulatory mechanisms against *V*. *coralliilyticus* infection. In conclusion, we discovered that *V*. *coralliilyticus* could cause acute mortality of adult *T*. *crocea* at 28°C, which is the first evidence of the rapid lethality of *V*. *coralliilyticus* in adult bivalves at natural and agricultural temperatures, since previous work was mainly focused on mortality of shellfish larvae. Moreover, transcriptomic analyses of the differences in molecular mechanisms between healthy and infected giant clams were obtained, and abundant differentially expressed immune-related genes and signaling pathways were identified, which drew our attention to the TLR pathway. Following quantitative validation and functional analysis, the results suggested that adult *T*. *crocea* could initiate the innate immune response through the TLR pathway against Vibrio infection, where changes in TLR pathway-related gene expression promoted the release of inflammatory factors such as IL-17, leading to the activation of a series of signals driving activities such as apoptosis. Despite the fact that *V*. *coralliilyticus* appears to be a global bivalve pathogen, limited information about its pathogenicity, infection mechanism and disease mitigation is available. Further studies are needed on the immune defense mechanisms of adult giant clams. These studies will be conducive to the development of health management in aquaculture. # Supporting information We would like to acknowledge the contributions of several other people to this work. We would like to thank Chuanjie Fu, Kunna Liu and Yunqing Li for their helps with samples collection. We appreciate the sequencing services of Novogene (Beijing, China). 10.1371/journal.pone.0231399.r001 Decision Letter 0 Fugmann Sebastian D. Academic Editor 2020 Sebastian D. Fugmann This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 22 Jan 2020 PONE-D-19-35528 Mechanistic molecular responses of the giant clam Tridacna crocea to Vibrio coralliilyticus challenge PLOS ONE Dear Dr zhiming, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. 1\) Reviewer \#1 raised numerous concerns about experimental details.  This should be addressed by making appropriate changes to the Material & Methods section and to the text of the Results / Figure legends where appropriate.  2\) The concerns about the apoptosis assay in Fig. 1 need to be addressed.  While the percentages of cells in the quadrants change upon infection, there is no actual "cell population" coming up which in annexin V positive (and hence early apoptotic) - it appears like a general no specific increase in fluorescence signal.  Also the is a considerable number of PI+ cells at t=0 h raising concerns about a large number of cells dying during the cell extraction and purification procedure.  Also these cells should likely be PI+ but Annexin V negative; thus there is an apparent issue with the compensation between the fluorescence channels in this experiment ( take a look a the sample plots of the vendor of the kit used: <http://www.vazymebiotech.com/products_detail/productId=84.html>). One important positive control in this context would be a treatment of the cell with an apoptosis inducing agent (e.g. etoposide). 3\) The assembled transcriptome dataset should also be made publicly available such that a consistent gene annotation for this organism can be established for future references.  4\) The labeling in heat map shown in Figure 3 is not really helpful - instead of only referring to Cluster XXXX YYYYY, common names should be utilized that give a clear hint on the function of the genes being differentially expressed.  5\) The description of the results regarding the GO/ KEGG classification (lines 311-322) is currently meaningless, e.g. it is quite obvious (even without knowing anything about the data) that the majority of genes fall in to classes like "cellular processes", "metabolic processes", "cell", and "cell part".  What else should they fall in? I strongly urge the author to use their biological "common sense" to omit this non-useful information (that likely was just copid as the highest ranking scores from a bioinformatics software produces) and rather include information that actually provides some biological insight into the immune system of their model organism.   We would appreciate receiving your revised manuscript by Mar 07 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Sebastian D. Fugmann, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> 2\. In your Methods section, please provide additional information regarding the permits you obtained for collection of Tridacna crocea. Please ensure you have included the full name of the authority that approved the animal collection and, if no permits or approvals were required, a brief statement explaining why. 3\. PLOS requires an ORCID iD for the corresponding author in Editorial Manager on papers submitted after December 6th, 2016. Please ensure that you have an ORCID iD and that it is validated in Editorial Manager. To do this, go to ‘Update my Information’ (in the upper left-hand corner of the main menu), and click on the Fetch/Validate link next to the ORCID field. This will take you to the ORCID site and allow you to create a new iD or authenticate a pre-existing iD in Editorial Manager. Please see the following video for instructions on linking an ORCID iD to your Editorial Manager account: <https://www.youtube.com/watch?v=_xcclfuvtxQ> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The manuscript presented here is of interest, well-written and presents new data. My main concern is about the functional demonstration of apoptosis induction by V. coralliilyticus. First, I am not confident in the chosen flow cytometry areas. Can I see your dead and apoptosis-induced controls? For me, PI+ cells should be separated at 102. With such area, I am quite afraid about your control hemocyte mortality (more than 10%?) Secondly, to my opinion, one unique technic (on one sampling point) is not enough to demonstrate apoptosis. Can’t you perform TUNEL assay, caspase dosage, TEM, …? In introduction section, I would appreciate to find more information on the huge diversity of TLR in marine bivalves. Material and method sections lack information \- How many individuals were challenges? \- Did you check bacterial suspension purity and concentration? \- Did you analyze moribund animals to ensure V. coralliilyticus ‘imputability’ in mortality? \- What is the difference in the hemolymph sampling explained L138-139 and 141-142 ? \- If hemocytes sampled at 0, 3, 6, 12, 24 and 36h, can we see flow cytometry results on all sampling points? \- Were the RT-QPCR analyses (L440) and RNA-seq analyses performed on the same biological samples \- L150. Triplicate = technical replicates and not biological replicates if I correctly understand what you mean \- Can you precise the volume/weight of tissue for RNA extraction L177 \- For RNA, 230 nm is also informative (L180) \- For qPCR, were the cDNA used diluted or pure? Can you precise it L241? In discussion section, I would recommend to authors to be more careful on the potential bacterial virulence factors that could induce the immune response measured (L485-500). There is a diversity of hemolysins for instance. Even if some hemolysins were described in other models as playing a role in apoptosis, you could not say that ‘some hemolysins \[…\] in the secretome of V. cora \[…\] may be responsible of this phenomena’ (L487). Are they expressed in vivo ? and which hemolysin are we talking about? Except by performing dual-RNAseq, you should be more moderate in this part of your discussion. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0231399.r002 Author response to Decision Letter 0 3 Mar 2020 PONE-D-19-35528 Reviewer \#1 1\) Reviewer \#1 raised numerous concerns about experimental details. This should be addressed by making appropriate changes to the Material & Methods section and to the text of the Results / Figure legends where appropriate. Response: We have appropriately changed the Material & Methods section as the reviewer suggested. We totally used 330 individuals to complete the experiment. Each experiment had biological repeats. Some imprecise statements have been corrected in the revised MS (please see line151-152, 287, 193). And the insufficient information has been improved and supplemented in the revised MS (please see line142-144, 190, 256). Through the reculture of bacteria from the moribund animals hemocytes and no deaths occurring after PBS and Vibrio alginolyticus injection, we can ensure V. coralliilyticus ‘imputability’ in mortality. Meanwhile, based on previous studies and our data, we chose the time point, 6 h to perform flow cytometry in consideration of the scarcity of experimental samples. 2\) The concerns about the apoptosis assay in Fig. 1 need to be addressed. While the percentages of cells in the quadrants change upon infection, there is no actual "cell population" coming up which in annexin V positive (and hence early apoptotic) - it appears like a general no specific increase in fluorescence signal. Also the is a considerable number of PI+ cells at t=0 h raising concerns about a large number of cells dying during the cell extraction and purification procedure. Also these cells should likely be PI+ but Annexin V negative; thus there is an apparent issue with the compensation between the fluorescence channels in this experiment ( take a look a the sample plots of the vendor of the kit used: <http://www.vazymebiotech.com/products_detail/productId=84.html>). One important positive control in this context would be a treatment of the cell with an apoptosis inducing agent (e.g. etoposide). Response: We have repeated the experiment of apoptosis as the reviewer suggested and submitted a new version of Fig 1, and we have compensated for the fluorescence channels appropriately according to the manufacturer’s instruction for both this time and before. Specific experimental operations and data processing were referred to Lin’s and Qin’s articles, where the early apoptosis of hemocytes in marine invertebrates may not be accompanied by the new “cell population” coming up \[1-3\]. Moreover, due to the specificity of species and the difficulty of experimental operation, some cells died during the previous experimental treatment. However, the control group and the experimental group were performed simultaneously, with the same protocol and statistical methods, which making the FACS results comparable and statistically significant differences credible. The Annexin-positive cells in the experimental group was significantly higher (P\<0.05) than that in the control group, although there was no new “cell population” coming up, which indicates V. coralliilyticus infection could induce hemocytes apoptosis (please see the details below). 3\) The assembled transcriptome dataset should also be made publicly available such that a consistent gene annotation for this organism can be established for future references. Response: We will release the assembled transcriptome dataset, as soon as the article is accepted. 4\) The labeling in heat map shown in Figure 3 is not really helpful - instead of only referring to Cluster XXXX YYYYY, common names should be utilized that give a clear hint on the function of the genes being differentially expressed. Response: Thank you for reviewer’s helpful comments. We have revised the labeling in the heat map and submitted the new version of Fig 3. 5\) The description of the results regarding the GO/ KEGG classification (lines 311-322) is currently meaningless, e.g. it is quite obvious (even without knowing anything about the data) that the majority of genes fall in to classes like "cellular processes", "metabolic processes", "cell", and "cell part". What else should they fall in? I strongly urge the author to use their biological "common sense" to omit this non-useful information (that likely was just copid as the highest ranking scores from a bioinformatics software produces) and rather include information that actually provides some biological insight into the immune system of their model organism. Response: Thank you for reviewer’s valuable comments and we changed the description of the results as reviewer suggested (please check the MS line 320-326, 336-342 and below). Revised: GO classification is a unified gene functional classification system. The analysis of annotated transcripts demonstrated that 36361 matched unigenes (18.58%) were divided into 26, 20, 10 entries for three categories, biological processes (BP), cellular component (CC) and molecular function (MF), respectively. The unigenes in “binding,” “catalytic activity,” “signaling,” “response to stimulus,” “membrane,” which were potentially related to immune existed in higher percentages compared to counterparts at sub-categories at the whole transcriptome reference. They played important roles during the defense process after challenge. The results provided a comprehensive view for screening candidate genes related to immune and defense mechanism (S2 Fig). In addition, we also used KEGG to identify the probable functional status of assembled transcripts. A total of 12912 unigenes (6.59%) were assigned to five main categories containing 232 KEGG pathways. Remarkably, immune system (558) was the well-represented term in the KEGG organismal systems category. Signal transduction (1455) was the most enriched terms in both environmental information processing and the whole transcriptome reference, suggesting the dominant position of various signal transduction in processing the stimulation of environment. In the metabolism category, abundant unigenes were found in the amino acid metabolism (510) sub-category. These results indicated that these unigenes may play a role in mediating V. coralliilyticus exposure and associated impacts (S3 Fig). Reviewer \#2 Reviewer \#1: The manuscript presented here is of interest, well-written and presents new data. 1\) My main concern is about the functional demonstration of apoptosis induction by V. coralliilyticus. First, I am not confident in the chosen flow cytometry areas. Can I see your dead and apoptosis-induced controls? For me, PI+ cells should be separated at 102. With such area, I am quite afraid about your control hemocyte mortality (more than 10%?) Secondly, to my opinion, one unique technic (on one sampling point) is not enough to demonstrate apoptosis. Can’t you perform TUNEL assay, caspase dosage, TEM, …? Response: Firstly, we have repeated the experiment of apoptosis as reviewer suggested and submitted the new version of Fig 1, where the control hemocyte (PBS injection group) mortality were all less than 10% (please see the details below). Secondly, RNA-seq and RT-QPCR have confirmed the change of apoptosis-related genes expression. Moreover, it was at the same time point as flow cytometry, which further proves the occurrence of apoptosis. 2\) In introduction section, I would appreciate to find more information on the huge diversity of TLR in marine bivalves. Response: Thank you for reviewer’s valuable comments and we supplemented more information on the huge diversity of TLR in marine bivalves in the introduction (please check the MS line 108-117 and below). Revised: Since the first TLR was identified in Drosophila melanogaster, large members of TLR family have been recently investigated in marine bivalves, such as Chlamys farreri, Chlamys nobilis, Mizuhopecten yessoensis, C. gigas, C. virginica, Mytilus edulis and Hyriopsis cumingii \[30-36\]. Among them, the innate immune regulation of TLR genes in bivalves has been reported in C. farreri (CfToll-1) and C. nobilis (CnTLR-1), respectively, both of which might be involved in the immune response against pathogen invasion \[31, 37\]. In addition, 23 TLRs were identified and arranged in 4 clusters according to extra- cellular LRR domain content in Mytilus galloprovincialis \[38\]. Furthermore, there were 83 TLR genes in the genome of C. gigas, 19 of which had different responses to Vibrio infection \[39\]. 3\) Material and method sections lack information \- How many individuals were challenges? Response: 330 individuals were challenged. 40 individuals were performed for flow cytometry considering individual differences and death. 180 were used to investigate mortality where the experiment was repeated three times, with 20 individuals in each group tested every time. Another 110 individuals were for RNA-seq and RT-QPCR, where 5 individuals were randomly sampled from each group at every time point and 50 of them were used to make up for the loss of death. \- Did you check bacterial suspension purity and concentration? Response: We have supplemented the information in the revised MS (please see line 142-144). Specifically, for bacterial challenge, purified bacterial inoculum were centrifuged for 10 min to collect bacterial pullets and were washed 3 times with PBS and re-suspended in 2×PBS to a concentration of OD600nm = 1.0. \- Did you analyze moribund animals to ensure V. coralliilyticus ‘imputability’ in mortality? Response: Hemocytes were collected from moribund animals and then homogenized and diluted in PBS and spread on Thiosulfate Citrate Bile Salts Sucrose (TCBS) Agar plates, followed by overnight culture. Through the successful reculture of bacteria, we can ensure V. coralliilyticus ‘imputability’ in mortality (please see the details below). Meanwhile, when giant clams were injected with an equal amount of PBS and Vibrio alginolyticus under the same conditions, there was no observable death, which also demonstrated that V. coralliilyticus should be the main cause of mortality in T. crocea. \- What is the difference in the hemolymph sampling explained L138-139 and 141-142 ? Response: We have corrected the misunderstanding statement in the revised MS (please see line151-152). There is no difference in the hemolymph sampling explained L138-139 and 141-142. Revised: Hemocytes were collected at scheduled intervals (0, 3, 6, 12, 24, and 36 h post injection) from both the challenged and control groups. Among them, hemocytes taken at 0, 6, 12, and 24 h post challenge were stored in liquid nitrogen for transcriptome analysis. \- If hemocytes sampled at 0, 3, 6, 12, 24 and 36h, can we see flow cytometry results on all sampling points? Response: It has been reported that after 6 h post-injection of V. coralliilyticus, haemocyte cell concentrations in haemolymph of infected and non-infected mussels were similar. However, the proportion of viable haemocytes in haemolymph of infected mussels was substantially lower than that in non- infected mussels \[4\]. Hence, we speculated that 6 h might be a critical time point. Meanwhile, in terms of phenotype, dead individuals appeared at 6 h post infection. Furthermore, in terms of mechanism, the expression of apoptosis- related genes like IAP1 was dramatically increased at 6 h, which both RNA-seq and RT-QPCR confirmed. Based on these, it is reliable that flow cytometry was performed at 6 h post infection. On the other hand, giant clams are rare and difficult to collect. Taking into account the loss of death, this experiment needs a large number of samples, which should be considered very carefully and ecological friendly. \- Were the RT-QPCR analyses (L440) and RNA-seq analyses performed on the same biological samples Response: The RT-QPCR analyses and RNA-seq analyses were performed on the same biological samples. Each sample was divided into two parts, one for RNA-seq, the other for RT-QPCR. \- L150. Triplicate = technical replicates and not biological replicates if I correctly understand what you mean Response: We have corrected the misunderstanding statement in the revised MS (please see line287). We performed biological replicates and n=3. \- Can you precise the volume/weight of tissue for RNA extraction L177 Response: We have supplemented the information in the revised MS (please see line190). We sampled 50mg of each tissue for RNA extraction. \- For RNA, 230 nm is also informative (L180) Response: We have corrected the unspecific statement in the revised MS (please see line193). Revised: The concentration and purity were examined at 260/230 and 260/280 absorbance ratios. \- For qPCR, were the cDNA used diluted or pure? Can you precise it L241? Response: We have supplemented the information in the revised MS (please see line256). For qPCR, all the template cDNA were diluted to 200ng/μl. 4\) In discussion section, I would recommend to authors to be more careful on the potential bacterial virulence factors that could induce the immune response measured (L485-500). There is a diversity of hemolysins for instance. Even if some hemolysins were described in other models as playing a role in apoptosis, you could not say that ‘some hemolysins \[…\] in the secretome of V. cora \[…\] may be responsible of this phenomena’ (L487). Are they expressed in vivo ? and which hemolysin are we talking about? Except by performing dual-RNAseq, you should be more moderate in this part of your discussion. Response: Thank you for reviewer’s valuable comments and we have corrected the related description in the discussion (please check the MS line 510-514 and below). Revised: Diverse metalloprotease and effector genes like the pore forming toxin hlyA were identified in the genome of V. coralliilyticus and expressed proteases were also detected in the secretome, which caused mortality in Drosophila and Artemia and may be involved in the infection of T. crocea. References 1\. Lin Y, Mao F, Zhang X, Xu D, He Z, Li J, et al. TRAF6 suppresses the apoptosis of hemocytes by activating pellino in Crassostrea hongkongensis. Dev Comp Immunol. 2020;103:103501. doi: 10.1016/j.dci.2019.103501. PubMed PMID: 31634519. 2\. Qin Y, Zhang Y, Li X, Noor Z, Li J, Zhou Z, et al. Characterization and functional analysis of a caspase 3 gene: Evidence that ChCas 3 participates in the regulation of apoptosis in Crassostrea hongkongensis. Fish Shellfish Immunol. 2020;98:122-9. doi: 10.1016/j.fsi.2020.01.007. PubMed PMID: 31917320. 3\. Wang X, Wang M, Xu J, Jia Z, Liu Z, Wang L, et al. Soluble adenylyl cyclase mediates mitochondrial pathway of apoptosis and ATP metabolism in oyster Crassostrea gigas exposed to elevated CO2. Fish & Shellfish Immunology. 2017;66:140-7. doi: 10.1016/j.fsi.2017.05.002. 4\. Nguyen TV, Alfaro AC, Young T, Ravi S, Merien F. Metabolomics Study of Immune Responses of New Zealand Greenshell™ Mussels (Perna canaliculus) Infected with Pathogenic Vibrio sp. Marine Biotechnology. 2018;20(3):396-409. doi: 10.1007/s10126-018-9804-x. 10.1371/journal.pone.0231399.r003 Decision Letter 1 Fugmann Sebastian D. Academic Editor 2020 Sebastian D. Fugmann This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 24 Mar 2020 Mechanistic molecular responses of the giant clam Tridacna crocea to Vibrio coralliilyticus challenge PONE-D-19-35528R1 Dear Dr. zhiming, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Sebastian D. Fugmann, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No 10.1371/journal.pone.0231399.r004 Acceptance letter Fugmann Sebastian D. Academic Editor 2020 Sebastian D. Fugmann This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 Mar 2020 PONE-D-19-35528R1 Mechanistic molecular responses of the giant clam *Tridacna crocea* to *Vibrio coralliilyticus* challenge Dear Dr. zhiming: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Sebastian D. Fugmann Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Family planning is the a technique for either limiting number of children, or want to delay their next birth \[Spacing\] after having unprotected sexual intercourse. Informed choice of family planning is defined as when a woman chooses a family planning method after receiving information about the possible side effects, what to do in case of side effects happen, and possible alternative methods by healthcare providers. If there are no contraindications, the choice of family planning (FP) methods is ultimately at the decision of the user. Clients are entitled to make voluntary, informed decisions about family planning services based on options, knowledge, and comprehension. Ethiopian national family planning guideline appreciate health care professionals to provide informed decision making as the best strategy to quality family planning services. In sub-Saharan Africa countries, the level of informed choice of family planning is not satisfactory. Potential African countries are providing family planning services without informing possible side effects. The level of informed choice of family planning is very low in Burundi \[35%\], Niger \[38%\], Gambia \[42%\], Benin \[47%\] and Mali (48%). In Ethiopia, the level of informed decision of family planning is poor (25%) compared to other African countries. Recent research has confirmed that the extent of informed choice for family planning methods is inconsistent with the place of residency in Ethiopia. Women residing in urban areas were 1.4 times more likely to be informed about family planning methods than women living in rural slums. However, the factors responsible for the disparity of uninformed choice between the two groups (urban and rural) are not examined. Education about the side effects of family planning and other alternative methods [plays a significant role](https://www.classicthesaurus.com/play_an_essential_role/synonyms) in the uptake of long-acting methods. The low use of long-acting family planning methods may be related to uninformed family planning choices. To our knowledge, and after a thorough review of literatures, previous studies in Ethiopia have investigated the magnitude and socio-demographic factors associated with uninformed family planning choices. However, none of these studies examined the extent of urban-rural inequalities. Moreover, none of these studies identified the factors contributing to the urban-rural gap in the informed family planning decision. In addition to that, this study determined the spatial distribution of uninformed choices of family planning in Ethiopia and identified areas exhibiting elevated aggregation of uninformed choices (popularly known as hotspots\], which facilitate the formulation of interventions tailored to specific locations. Therefore, the objective of this study is to identify and quantify the factors that contribute to the observed urban-rural gap in informed family planning decisions. Besides, information obtained from Georeferenced datasets could help program developers and local decision-makers develop location-specific strategies to address the problem of uninformed family planning choices, rather than indiscriminately providing services to the entire part of the country. # Methods and materials ## Study design and period This study was conducted using information from the fourth cross-sectional demographic health survey of Ethiopian (EDHS) communities, which was carried out between January 18 and June 27, 2016. ## Study area Ethiopia is the second-most populous country in Africa. The country’s overall population was expected to reach roughly 103,000,000 by 2021, according to the Population and Housing Census (PHC) projection from 2007. According to the 2007 Census, the majority of people lived in rural regions (83.6%), there were 4.7 people per family on average, and 47% of all women were between the ages of 15 and 49. ## Population ### Source population All reproductive age women in Ethiopia. ### Study population Women aged 15–49 years who were modern family planning users during the data collection period. ### Dependent variable The outcome variable for this study is informed choice of family planning for the decomposition analysis and uninformed family planning choice for the spatial analysis. ### Independent variables According to our extensive literature reviews, maternal age, maternal education (none, primary, secondary, higher), marital status, occupation, wealth status (poorest, poor, middle, rich, and richest), sources of contraception (private, non-governmental organizations (NGO), pharmacy, other), marital status **(**never married/ever married), family planning message on mobile phone (got messages/ not receive messages), Working status (working/ no working), visit a health facility in the last 12 months, frequency of reading newspapers or magazines (not at all, less than once a week, at least once a week), frequency of listening to the radio (not at all, less than once a week, at least once a week), and Frequency of watching television (Not at all, Less than once a week, At least once a week) were considered as independent variables for this study. ### Operational definitions *Informed choice of family planning*. women of reproductive age who used contraceptive methods but were not informed of side effects, what to do if they experienced the side effects or problems, and who were not informed of other contraceptive methods that could be used, were labeled as “uninformed choice” with code 1, otherwise as” informed choice” with code 0. *Media exposure*. created by combining whether a respondent reads the newspaper, listens to the radio, and watches television. If the respondent was exposed to at least one of the three media this is labeled “exposed” and coded” 1”, otherwise “not exposed” is coded “0”. ### Data source We used the Kids’ record (KR) dataset of the 2016 EDHS to further analyze. ### Data collection tools and procedures The website of the DHS measure ([http://www.dhsprogram.com](http://www.dhsprogram.com/)) was employed to effectuate registration for access to the 2016 Ethiopian DHS Datasets and Global Positioning System (GPS) data, with requisite permission having been duly procured in order to gain access to the requested tools. Accordingly, all required data were downloaded from the Demographic and Health Surveys Program website. A global positioning system was used to collect the geographic coordinates of each cluster. ### Data quality control The data collectors of this survey documented that questionnaires were pre- tested in all three local languages (Amharic, Afaan Oromo, and Tigrigna) to ensure that questions were clear and understandable to respondents. ### Sample size and sampling procedures A two-stage, stratified, clustered sampling procedure was implemented across all 11 (now expanded to 13) geographic administrative areas comprising 9 regions and 2 city administrations. The first strata included a total of 645 enumeration areas (EAs) that were selected proportionally to the EAs size of the nine geographical regions and two administrative cities. In the second strata, every eleven administrative divisions were further sub-grouped into urban and rural residents, yielding a total of 21 sampling strata. From each cluster, 28 households were chosen using an equal probability technique. This study ultimately includes a weighted sample of 3511 participants. ### Data management Data were processed, reviewed, sorted, and recoded using STATA/SE version 16.0. To account for the effects of the survey’s complex sampling design or the hierarchical nature of the EDHS dataset, to restore survey representativeness, and to obtain reliable statistical estimates, the data were weighted via applying the STATA command "svyset." This command was prepended to each analysis in this study. Arc GIS version 10.8 software was used to visualize the spatial distribution and locate hotspot areas (clusters).To measure the deviation of the spatial arrangement of the uninformed choice from randomness, the global spatial autocorrelation (Global Moran’s I) was calculated. A positive value of Moran’s I represent positive spatial autocorrelation (cluster together), whereas a negative value indicates dispersed arrangement. ### Logit-based multivariate decomposition analysis To find factors that contributed to the gap of informed choice of FP between urban and rural residents, multivariate decomposition analysis was calibrated. This analysis utilizes the output from the logistic regression model to assign the observed change in informed choice of FP rate between urban and rural through subdividing into components (i.e., due to endowment and coefficients). For logistic regression, the Logit or log-odd of difference between urban and rural is taken as: Logit (rural)-Logit (urban) = F (X urban\*rural)–(F Xurban\*βrural) = \[(F(X urban\*β2rural)-F (Xurban\*βrural)/E)\] + \[(F(X urban\*βrural)-F (Xurban\*βrural)/C)\]. X indicates the outcome variable i.e., informed choice of FP Beta (β) indicates that, regression coefficient of each selected independent variables The E component refers to change in informed choice of FP imputable to differences in endowments or characteristics. The C component refers to a change in informed choice of FP imputable to differences in coefficients or effects. During decomposition analysis the categories urban and rural were recorded as “0” and”1” respectively. Percent contributions with a 95% confidence interval (CI) of coefficients and a p-value \<0.05 were reported. Candidate variables for multivariable decomposition regression were selected according to the p-value from bivariable decomposition regression. Those variables with p-value less than 0.2 in the bivariable regression were moved to multivariable decomposition regression analysis. ### Spatial scan statistical analysis Bernoulli model and purely spatial Kulldorff’s scan statistical analysis was deployed to detect clusters (areas with concentrated numbers of women making uninformed choices FP). If the spatial pattern of uninformed choice of FP is randomly distributed across space, it is not doubtful that the development of site-specific strategies will be effective in curbing the high prevalence of uninformed choice. Only areas with a high rate of uninformed choices of FP were applied to determine the geographical location of statistically significant clusters using SaTScanTM v10.0.1 software. We used the Bernoulli model because the data were binary (uninformed or informed choice). Women who were uninformed about the choice of FP were considered cases (1), whereas those who were informed were considered non-cases (0). The case file (1), non-case file (0), and coordinate files (latitude and longitude) were imported into SaTScan ^TM software to determine the location of significant clusters. The maximum size of the scan window was scaled according to the percentage of the total population at risk. The maximum radius of the circle was adjusted to be less than 100 km. This is done to facilitate the development of intervention strategies, where clusters with optimal radius are easier to manage than large clusters. To avoid overlooking very small and very large clusters, the maximum geographic cluster size was adjusted to \< 50% of the population at risk as an upper limit. The most likely (primary), cluster was determined using p-value and likelihood ratio tests. The cluster with the highest likelihood ratio represented the most likely cluster, and the remaining clusters with statistically significant log- likelihood ratios (LLR) were designed as possible secondary possible clusters. The relative risk (RR) parameter of the uninformed choice of FP in each cluster was calculated to estimate the risk of uninformed choice within the cluster areas. ## Ethics approval and consent to participate As the authors of this manuscript did not involve collection of data from the participants, consent to participate and ethical approval is not required. But the authors have granted to access the dataset of 2016 EDHS through registering to [www.dhsprogram.com](http://www.dhsprogram.com/) website. The data granted from this website have no personal identifying information (anonymous) and the participants’ confidentiality and privacy issue is not an inconsequential in this scenario. All the methods were performed in accordance with the relevant guidelines and regulations of DHS measures. # Results A total of 3511 (weighted) modern family planning users were included in this study. The median age of the study participants was 28 years (IQR = from 23 to 34 years). As shown in the Figure (below), only 40.7% \[95% confidence interval: 39.12% to 42.4%\] of current contraceptive users were informed about FP choices. Additionally, the magnitude of informed choice was 12 percent higher among urban residents than rural residents. The greater number of study participants were from rural slums (76.5%). More than 60% of the study participants were from Oromia and Amhara regions. Of all participants, 1838 (52.3%) were orthodox Christian followers, accompanied by Protestants (25%). ## The spatial arrangement of the uninformed choice of FP The spatial arrangement of the uninformed choice of FP in Ethiopia was not evenly distributed. The null hypothesis of global spatial autocorrelation assumes that the uninformed choice of FP was randomly distributed across the study area. The positive z-score, positive Moran’s index, and statistically significant p-value from prove that the spatial arrangement of uninformed choice was clustered. The corresponding p-value of less than 0.01 on the right side of the figure can be interpreted to mean that the probability of randomness of the observed spatial pattern (i.e., clustering) of uninformed choice is less than 1%. shows hotspot areas (areas with an overwhelming number of women who have uninformed FP choices) among contraceptive users in Ethiopia. the primary cluster (shaded in red) (LLR = 34.8, p-value\<0.001) focused primarily on the southern part of the Amhara region, which included East & West Gojjam, South Gondar, and south wollo zones and was centred at the latitude of 710.984556°N, and longitude of 38.044450°E with a radius of 97 km. ## Geographic clustering of uninformed choice . ## Decomposition analysis Variables with a p-value of less than 0.2 from the bivariable decomposition analysis were picked as the contender variable for the multivariable decomposition analysis. The composition factors of region and religion were not included in the multivariable composition analysis because of having a p-value greater than 0.2 at bi-variable decomposition analysis. The multivariable decomposition regression models revealed that 74% of the changes in informed choice of FP could be attributed to changes in population characteristics (population dynamics). This can be interpreted to mean that if household characteristics related to the explanatory variables had been equalized between urban and rural residents, the Urban-rural gap of informed choice of FP would have declined by 74%. However, the change due to the coefficients is not statistically significant and the detailed result of the decomposition of each variable is not included in the table. **Non-linear decomposition of informed choice of FP.** From the multivariable non-linear decomposition analysis, the age group between 35 and 49 years, receiving the last FP at private health facilities, and being visited health facilities in the last year were the gap tightening variables (negative percentage contribution). Conversely, the age group between 25 and 34 years, lack of access to FP-related mobile short message service (SMS) and listening to the radio were variables that increased the difference in informed choice of FP between these two groups. As can be seen from the table (above), the difference between urban and rural areas is mainly explained by the location of FP supply (i.e., private facilities), followed by listening to radio programs. In addition, the age of current users of FP also had a strong influence on the urban-rural gap in informed choice. To illustrate, if the composition (characteristics) of where FP is offered (in private facilities) were identical in urban and rural areas, the difference in informed choice of FP in Ethiopia would have increased by 16 percent. Similarly, if the composition (characteristics) of women aged 35–49 had remained the same for urban and rural participants, the difference in the extent of informed choice would have increased by 9.3%. Oppositely, the gap in informed choice of FP between rural and urban residents would have been narrowed by 9%, if the composition of women aged between 25 and 34 years was the same between rural and urban in 2016. Besides, the gap would have been narrowed by 11 percent, if the composition of radio listeners stayed the same between the two groups. # Discussion This study found that the spatial arrangement of uninformed choice of FP is clustered. Moreover, variables like woman’s age, place of FP offer, media exposure, access to mobile SMS, and visit to health facilities are among the contributing compositional variables to the gap of informed FP choice between urban and rural residents. The current study concentrated on the quantitative analysis of the informed choice of FP data from a spatial vantage point using cluster analysis. Using the local spatial statistics (SaTScan), the spatial arrangement of uninformed choice was concentrated in east & west Gojjam, south Gondar and wollo zones of the Amhara region. These provinces were also identified with the poor practice of health communication on family planning. Moreover, earlier research also documented that informed choice of FP varied across regions and places of residency. A possible explanation for that inequality might be, women residing in urban areas are highly likely to get in touch with well trained and experienced health professionals who can nourish theme information related to the choice of FP. Tailoring plans and programs based on geographic locations would be more effective and resource-saving instead of random health service provisions to all areas of the country. Moreover, women living in rural provinces were laggards compared to their urban counterparts in having informed choices of FP. Those results are consistent with another similar prior study in the context of India. These results are likely to be related to rural women who might have less opportunity to access well trained health professionals who give adequate information about methods of choice. Rural health facilities in Ethiopia has less knowledgeable and less autonomous in their choice of health care services. It would be interesting to focus on rural residing women in offering information related to the choice of family planning. Interestingly, getting family planning services at private health facilities is found to narrow the gap in informed choice between rural and urban residents. This finding was in agreement with other studies. The reason for this could be the client flow at private health facilities are lower than at public health facilities so that informed choice of family planning can be practically achieved in private health facilities. Moreover, since private health facilities are commercial, they could have a good client handling approach including delivery of detailed information about the service and treatment they are providing to their clients. It is better to expand private health facilities to rural areas as equal as the public health facilities to have informed choice services of FP. Compared to the age group of 25 to 34 years being age group between 35 and 49 years is found decrease the gap in informed choice of family planning between rural and urban residents. Other studies also reported similar findings. The possible reason behind this could be elder mothers from rural areas may also seek more information about the family planning method that could be delivered to them because they might experience side effects and discomforts in their prior utilization of family planning methods. Unsurprisingly, visit to health facilities in the last 12 months were found to narrow the urban to rural gap in the informed choice of family planning, and it was consistent with another study done in Ethiopia. This could be due to the fact that rural resident women might remember their informed choice of family planning if they visit health facilities recently but rural resident women who had not visited health facilities recently may forget about whether they were offered informed choice of family planning services. On the hand, urban women are more likely to remember the informed choice of family planning services as they are more educated and exposed to media than their rural counterparts. Additionally, information flow and access are increasing over time which could make family planning providers in rural health facilities to be more aware of the right of being informed during the choice of their family planning method. This study showed that Access to FP related mobile SMS significantly widened the urban-rural gap in the informed choice of family planning. This finding was supported by other studies. The possible justification behind this could be related to the fact that the availability and utilization of mobile phones were limited in rural areas. Besides, rural resident women are less educated and could not read and understand messages sent to them via SMS. This study demonstrated that the frequency of listening to the radio was a significant variable that widen the urban-rural gap in the informed choice of family planning methods. The possible reason might be there is better access and coverage of mass media including Radio in urban areas because of improved energy access. Additionally, urban women are more literate to understand and internalize the information streamed through radio programs than their rural counterparts. ## Strengths and limitations In this study, nationally representative data were used, allowing for better generalizability. The design effect due to the hierarchical nature of the samples was accounted for by using the STATA command "svyset" for each descriptive and analytical analysis. But, as this article is analyzed using secondary data, important behavioral and other socio-demographic variables might have declined model performance. The EDHS survey relies on respondents’ self- reports, which may be subject to recall bias Furthermore, some women may be embarrassed to disclose all the details (self-report) concerning FP because it is a sensitive subject. Conclusion In this study, visit to private health facilities, women’s age group between 25 to 39 and visit to health facilities in the last 12 months are factors that significantly widen urban -rural gap of family planning. On the other hand, factors such as Access to FP related mobile SMS and frequency of listening to radio were found to narrow the urban -rural gap of informed choice of family planning. These findings suggest that to narrow residential inequalities of informed choice, tailoring interventional plans based on age, health seeking behavior and family planning promotion via media are effective strategies. The spatial irregularity of not informed choice of family planning identified in this research needs further research to identify the factors behind these geographic inequalities using geographic weighted regression. The authors acknowledge MEASURE DHS for permitting us to access and download the Ethiopian 2016 DHS datasets. EDHS Ethiopian Demographic health survey LLR Log-Likelihood Ratio FP Family Planning NGO Non-Governmental Organization GPS Global Positioning System RR Relative Risk IQR Inter-Quartile Range EA *enumeration areas* SNNP Southern Nations, Nationalities, and People and SMS Short Message Service 10.1371/journal.pone.0289099.r001 Decision Letter 0 Simegn Wudneh Academic Editor 2023 Wudneh Simegn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 13 Apr 2023 PONE-D-23-01109Urban-rural inequalities and spatial arrangement of informed choice of FP in Ethiopia: further analysis of 2016 Ethiopian demographic health survey.PLOS ONE Dear Dr. Tareke, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by May 28 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Wudneh Simegn Belay, MSc Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. 3\. Please amend either the abstract on the online submission form (via Edit Submission) or the abstract in the manuscript so that they are identical. 4\. We note that Figure 3 in your submission contain map images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: \(1\) You may seek permission from the original copyright holder of Figure 3 to publish the content specifically under the CC BY 4.0 license.   We recommend that you contact the original copyright holder with the Content Permission Form (<http://journals.plos.org/plosone/s/file?id=7c09/content- permission-form.pdf>) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (<http://creativecommons.org/licenses/by/4.0/>). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an ""Other"" file with your submission. In the figure caption of the copyrighted figure, please include the following text: “Reprinted from \[ref\] under a CC BY license, with permission from \[name of publisher\], original copyright \[original copyright year\].” \(2\) If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. The following resources for replacing copyrighted map figures may be helpful: USGS National Map Viewer (public domain): <http://viewer.nationalmap.gov/viewer/> The Gateway to Astronaut Photography of Earth (public domain): <http://eol.jsc.nasa.gov/sseop/clickmap/> Maps at the CIA (public domain): <https://www.cia.gov/library/publications/the- world-factbook/index.html> and <https://www.cia.gov/library/publications/cia- maps-publications/index.html> NASA Earth Observatory (public domain): <http://earthobservatory.nasa.gov/> Landsat: <http://landsat.visibleearth.nasa.gov/> USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): <http://eros.usgs.gov/#> Natural Earth (public domain): <http://www.naturalearthdata.com/> 5\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting- information. ](http://journals.plos.org/plosone/s/supporting-information. ) \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Review Comments to the Author Using 2016 EDHS data the authors assessed urban-rural inequalities and spatial arrangement of informed choice of family planning in Ethiopia. The paper reported that spatial configuration of not informed choice was clustered and magnitude of informed choice was 12 percent higher in urban residents compared to rural residents. Title: It is not recommended to use abbreviation on the title: so better you write full form. Abstract: correct editorial error such as not capitalizing the first letter of new sentence and capitalizing the middle word ‘’Descriptive’’. The same editorial issues are there under discussion. This sentence is incomplete and what authors want to say is not clear -More specifically, place of FP offers i.e., private health facilities (-16%), listening to the radio programs (+12%), age 35 to 49 years (-9.3%) and 25 to 34 years (+9.2%) (p\<0.05. Check grammar for the following sentence: In this study, visit private health facilities, women’s age group between 25 to 39 and visit to health facilities in the last 12 months factors that significantly widen urban - rural gap of family planning. You said: Besides, the spatial arrangement of not informed choice of family planning is not regular. To express spatial distribution the term random/non- random vs clustered are better rather than using not regular. Introduction: Check grammar issue in the following sentence: •Informed choice of family planning means that when women choose a family planning methods, all available information regarding side effects, what to in case of side effects, and possible alternative methods •Write full form followed by abbreviations in bracket first time abbreviations appear in manuscript Check grammar issue in the following sentence: •Moreover, the current study identified the spatial distribution and identified hotspots areas where women who had uninformed family planning choices. Methods: •Even if the authors used secondary data some sections of methods were absent. Please add subsections on study area, design, population and data source •Editorial issue: don’t capitalize letters in the middle of the sentence “Visit” Results: •In figure 1 title mention time component, you can use Year EDHS was released. Similarly for table 1 title is incomplete address where and when component. •Under the following section the authors wrongly cited Figure 1 replace it with Figure 2 because there are no p value in Figure 1 “The spatial arrangement of the uninformed choice of FP” •Similarly in the paragraph under figure 2, omit figure 2 (wrongly cited) Discussion •Check grammar issue in the following sentence: In this study, being in the age group 35 to 49 years was significantly narrowed the urban and rural gap in the informed choice of family planning where as being in the age group 25 to 34 found to widen the urban -rural gap of family planning compared to age group 15 to 24 year. •Try to include strength and limitations of this study Conclusions •The first sentence is unnecessary Declarations You specified only single abbreviations but there are a lot of abbreviations inside document. So include all abbreviations-SNNP, NGO, FP…… References Some references are too old (16=1996, 24=2007, 28=2001). Try to use recent publications Reviewer \#2: Dear editors. Thank you very much for this opportunity to review this paper. This paper assessed urban-rural disparities and spatial distribution of informed choice of FP in Ethiopia, which is crucial for plociy makers and program managers to identify intervention on inequality. The following are my suggestions for the improvement of this manuscript: General comments The authors don't include page numbers and lines in the document, making it difficult to give suggestions and comments by page and line Make sure to leave space before each bracket or parenthesis There is typo and spelling error thoughout the document Introduction. The introduction part must be rewritten orked and organised in a logical way. The aouthors must highlights what lloks the condition in Africa, then in Ethiopia. Incorporate national family planning strategy and initiatives, as well as health service accessibility and distribution by residence. Describe the efforts and gaps in providing access to quality family planning service to everyone who needs it. It should be discussed that how much govt has spent on family planning and benefit in terms of preventing unintended pregnancy and highrsik fertility behaviors FP: to be defined first Method EDHS is used two stage stratified study design so there is a need for adjustment of cluster and weight while doing multivariate analysis. Need to elaborate the design of the study and whether any type of stratification has been carried out? In addition, the authors did a decomposition analysis but did not elaborate on it in the method section. Which residence coded as “0” or “1”? it need more details Independent variables Be consistent, for some variable athors mentioned the coatergy but form sone not yet Results Present the frequency of Table 1 by residence, then the total and weighted frequencies are enough. Decomposition analysis The authors stated that "the composition factors of region, religion, and place of residence were not included in the multivariable composition analysis because they had a p-value greater than 0.2 at the bivariable decomposition analysis". How could residence be an explanatory variable here? It was an urban-rural disparity study, hence the grouping variable in this case is residence. In table one, I didn't notice a variable “place for the FP offer” to be classified as government or private, but in the decomposed table, I did????? Correct the FF typoerror “in the f informed choice” “urbanand rural” “by 9.3%.. Oppositely” Discusion It may be helpful to have a short summary of important findings in the first paragraph. The authors did not discuss limitations of the study such as the self-reported nature of the outcome informed choice of FP. It is surprising that the composition of private health facilities had a narrowing effect, which indicates the composition of private health facilities was higher in rural areas, which is far from reality. Most private health facilities in Ethiopia are concentrated in urban areas. Conclusion State the the policy implications of this study based on the finding. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0289099.r002 Author response to Decision Letter 0 13 May 2023 Background: Ethiopia has made satisfactory progress in improving maternal and child health over the past two decades. The introduction of family planning through informed choice is one of the main strategies to improve maternal and child health. However, this positive progress may have masked the significant urban-rural disparities in informed choice for family planning. Objective: To identify factor contributing to observed urban-rural disparities and to determine the spatial distribution of informed family planning choices in Ethiopia. Methods: The study used information from 3,511 (1) women currently using contraceptives (rural-2685 and urban-826)c from the most recent Ethiopian demographic health survey cross-sectional data. Spatial and descriptive, bivariable, and multivariable logit-based decomposition analysis methods were used. Results: The spatial configuration of uninformed choice was clustered. The primary cluster (LLR=34.8, p-value\<0.001) was located at the southern portion of Amhara region that covers east & west Gojjam, south Gondar and south Wollo administrative zones. The magnitude of informed choice was 12 percent higher in urban residents compared to rural residents. Urban-rural gap was attributed to variations in characteristics (74%). Place of family planning offer i.e., private health facility, being aged between 35 and 49 years, and having visited to health facility in the last 1 year are found decrease the urban-rural gap of informed family planning choice by 15%, 9% and 5% respectively. Conversely, being aged between 25 and 34 years, being a listener to radio has increased the gap by 9% and 12% respectively. Conclusion: The variables being private health facility visitors, being aged between 25 and 39 years and having visited health facilities in the last one year are found to increase the gap of informed family planning choices between urban and rural residents Besides, the spatial distribution of uninformed family planning choices is non-random. 10.1371/journal.pone.0289099.r003 Decision Letter 1 Simegn Wudneh Academic Editor 2023 Wudneh Simegn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 5 Jul 2023 PONE-D-23-01109R1Urban-rural inequalities and spatial arrangement of informed choice of Family planning in Ethiopia: further analysis of 2016 Ethiopian demographic health survey.PLOS ONE Dear Dr. Tareke, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Aug 19 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Wudneh Simegn, MSc Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. Additional Editor Comment: Author’s contributions form must be corrected as per the journal guideline. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: Dear Editor, thank you for allowing me to review “Urban-rural inequalities and spatial arrangement of informed choice of Family planning in Ethiopia: further analysis of 2016 Ethiopian demographic health survey”. After I reviewed the revised manuscript, I got that the author/authors had corrected all the raised issue. However, see carefully the age group in the abstract section particularly at result and conclusion section (35 to 49 VS 25 to 39). \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0289099.r004 Author response to Decision Letter 1 6 Jul 2023 Authors’ response to reviews Title: Urban-rural inequalities and spatial arrangement of informed choice of family planning in Ethiopia: further analysis of 2016 Ethiopian demographic health survey. Authors: Abiyu Abadi Tareke (<abiyu20010@gmail.com>) Bayley Adane Takele , Mohammedjud Hassen Ahmed Masresha Derese Tegegne Habitu Birhan Eshetu Version: 2 Date: May 13, 2023 Point by point response for editors/reviewers’ comments Manuscript number: PONE-D-23-01109 Dear editor/reviewer: Dear all, We express our profound appreciation for the insightful and productive feedback that you have provided. Your invaluable comments have significantly enriched the quality of the manuscript, and have greatly augmented our expertise in the realm of scientific paper writing. The authors have diligently taken into account each of the comments and queries raised by the editors and reviewers, and have responded to them in a targeted manner. Our comprehensive point-by-point rejoinders to all the comments and questions can be found in the subsequent pages. In addition, an accompanying supplementary document has been enclosed, which showcases the modifications made in detail, using the track changes feature. We also made some change to fix grammatical error in some paragraphs. Review Comments to the Author Reviewer’s comment: see carefully the age group in the abstract section particularly at result and conclusion section (35 to 49 VS 25 to 39). Authors’ response: dear reviewer we are here because of your insight and deep review to this manuscript. Sorry for this typographical error. After checking the result written in the regression table, we changed the phrase “being aged between 25 and 39 years” to “being aged between 35 and 49 years”. 10.1371/journal.pone.0289099.r005 Decision Letter 2 Simegn Wudneh Academic Editor 2023 Wudneh Simegn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Jul 2023 Urban-rural inequalities and spatial arrangement of informed choice of Family planning in Ethiopia: further analysis of 2016 Ethiopian demographic health survey. PONE-D-23-01109R2 Dear Dr. Tareke, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Wudneh Simegn, MSc Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0289099.r006 Acceptance letter Simegn Wudneh Academic Editor 2023 Wudneh Simegn This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Aug 2023 PONE-D-23-01109R2 Urban-rural inequalities and spatial arrangement of informed choice of Family planning in Ethiopia: further analysis of 2016 Ethiopian demographic health survey. Dear Dr. Tareke: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Wudneh Simegn Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Heat shock protein 70 (Hsp70) family members are present in all organisms and are the most highly conserved heat shock protein family. Hsp70 proteins function as molecular chaperones, and are involved in various cellular processes such as protein folding and assembly of nascent polypeptides, refolding aggregated proteins, protein translocation across membranes, protein degradation and controlling the activity of regulatory proteins. Allosteric coupling between the N-terminal ATPase domain and the C-terminal substrate binding domain is essential for the function of Hsp70 and is mediated via the inter-domain linker. ATP binding and hydrolysis leads to conformational changes in the two domains and this regulates substrate affinity. The functions of Hsp70 are regulated by co-chaperones, such as J-proteins (also called Hsp40 or DnaJ), and by GrpE-like nucleotide exchange factors. Due to their crucial role in proteostasis, Hsp70 members are present in almost all cellular compartments. The function of the mitochondria is critically dependent on mitochondrial Hsp70 (mtHsp70) and it plays a major role in the translocation of nuclear encoded proteins across the mitochondrial membranes and folding of proteins in the matrix. Ssc1, Ssq1 and Ssc3/Ecm10 are the Hsp70 proteins found in the mitochondria of yeast, with Ssc1 being the major Hsp70. Protein folding by Ssc1 in the matrix is regulated by Mdj1 (the only Type I J-protein in the mitochondria) which delivers substrate and stimulates the ATPase activity. Unexpectedly, mtHsp70 has the propensity to self-aggregate and it requires an additional essential regulator, Hep1 (<u>H</u>sp70 <u>e</u>scort <u>p</u>rotein), to maintain its functional state. Hep1 orthologues are conserved in many eukaryote species including protozoa, but not in prokaryotes. Only Ssc1 and Ssq1 produce aggregation-prone conformers of the ATPase domain that bind to Hep1. Yeast cells deprived of Hep1/Zim17/Tim15 accumulate insoluble mtHsp70 and generally exhibit mitochondrial defects similar to those observed upon mtHsp70 deletion. Hep1 is a zinc-finger protein with one tetracysteine motif that is part of the zinc finger domain. The binding of zinc ions to Hep1 in the mitochondria is critical for its function. Hep1 only binds to nucleotide-free mtHsp70 and is released upon nucleotide binding. The N-terminal ATPase domain of mtHsp70 in association with the interdomain linker is prone to aggregation, while the ATPase domain and C-terminal substrate binding domain are both soluble when expressed separately. The inter-domain linker attached to the ATPase domain is the minimal binding entity required by Hep1 to keep mtHsp70 soluble and active. The *P*. *falciparum* genome encodes six Hsp70 orthologues, and only PfHsp70-3 has been predicted to localise in the mitochondria. Plasmodial Hsp70 proteins and their interactions with co-chaperones have received attention as potential avenues for drug targeting as they play an integral part in the survival and pathology of the parasite. Selective modulation by small molecules of *P*. *falciparum* Hsp70 proteins has been demonstrated, as well the modulation of the Hsp70/J-protein partnership. Little is known about the role of PfHsp70-3 in the parasite, and a protein translocation model was previously described by van Dooren et al.. More recently, an overview of the proposed roles of PfHsp70-3 and its co-chaperones in the mitochondria was presented by Njunge et al.. The putative co-chaperones of PfHsp70-3 which form part of the import machinery are Tim44 (PF11_0265), GrpE (PF11_0258) and PfPam18 (PF07_0103). Other proposed co- chaperones of PfHsp70-3 involved in protein refolding in the matrix are Pfj1 and PFF1415c but these have not been experimentally validated. PfHsp70-3 is an orthologue of Ssc1 and is predicted to localize in the parasite’s mitochondrion and play a central role in the translocation of proteins into the mitochondria and their subsequent folding in the matrix. Our results show that PfHsp70-3 is indeed insoluble when heterologously produced in *E*. *coli* cells. We identified a putative Hep1 orthologue in *P*. *falciparum*, and provide experimental evidence that PfHep1 prevents the self-aggregation of PfHsp70-3 that is required for structural and functional activities. We also examined the abilities of PfHsp70-3 and PfHep1 to function as holdases and suppress protein aggregation, and demonstrate that the zinc ion in the zinc finger domain is essential for stabilising the structure of PfHep1. # Materials and Methods ## Primary structure sequence analysis and homology modelling of the zinc-binding domain (ZBD) of PfHep1 The protein domain mapping for PfHep1 was conducted using the online programs SMART 7 (Simple Modular Architecture Research Tool; <http://smart.embl- heidelberg.de/>;), and Prosite (<http://prosite.expasy.org/>;). In order to predict the subcellular localisation of PfHep1 and PfHsp70-3 a number of online programs that included NucPred (<http://www.sbc.su.se/~maccallr/nucpred/cgi- bin/single.cgi>;), MitoPROT (<http://ihg.gsf.de/ihg/mitoprot.html>;), MultiLoc (<http://abi.inf.uni-tuebingen.de/Services/MultiLoc>;, SignalP version 4.1 (<http://www.cbs.dtu.dk/services/SignalP/>;), and WoLF PSORT (<http://www.genscript.com/wolf-psort.html>.;) were used. The primary amino acid sequence of PfHep1 and other selected well-characterised Hep1 orthologues were obtained from PlasmoDB v4.4 (<http://plasmodb.org/plasmo/>;), and the NCBI database. Alignment was conducted using MAFFT (<http://www.ebi.ac.uk/Tools/msa/mafft/>;). The zinc-binding domain structure of PfHep1 was modelled using the online Swiss Model server. The solution structure of Tim15c (yHep1) solved by NMR (PBD accession number 2EZZ) was used as the template. The model was rendered using PyMol. ## Construction of *E*. *coli* expression plasmids encoding PfHsp70-3 and PfHep1 *E*. *coli* codon-optimized versions of the Hsp70-3 coding sequence (PlasmoDB accession number: PF3D7_1134000) (41 – 622aa) and PfHep1 coding sequence (PlasmoDB accession number: PF3D7_1420300) (15–302 aa), both lacking the mitochondrial signal peptides, were synthesized by the GenScript Corporation (Piscataway, New Jersey, USA) and inserted into a pQE30 expression vector (Qiagen, Germany) to create the pQE30-PfHsp70-3 and pQE30-PfHep1, while the pACYCDuet1 expression vector was used to create pACYCDuet1-PfHep1. ## Expression and purification of PfHep1 *E*. *coli* M15 (pREP4) cells were transformed with pQE30-PfHep1 and grown at 37°C in 2x YT medium supplemented with 100 μg/ml ampicillin and 50μg/ml kanamycin and grown to mid-logarithmic phase (A₆₀₀ 0.4–0.6). Protein production was induced by addition of 0.1 mM IPTG. Cells were harvested prior to induction and at hourly intervals post induction for 5 hours and overnight. Cells were harvested by centrifugation (13000×g; 2 min) and re-suspended in PBS buffer. Protein production levels were evaluated using SDS-PAGE and Western blot analysis. For recombinant protein purification, cells were harvested at the fourth hour post induction and the harvested cells were re-suspended in lysis buffer (10 mM Tris-HCl, pH 7.5, 300 mM NaCl, 10 mM imidazole, 1 mM PMSF, 1 mg/ml lysozyme) and frozen at -80°C overnight. The cells were then thawed on ice and sonicated at 4°C in the presence of 3% sarcosyl (Sigma-Aldrich, Germany). The lysed cells were centrifuged (16000×g, 40 min, 4°C) and the soluble supernatant fractions were mixed with cOmplete His-tag purification resin (Roche, Germany) and allowed to bind overnight at 4°C with gentle agitation. The resin was then pelleted by centrifugation (5000×*g*; 2 min) to remove unbound proteins and washed three times using wash buffer (100 mM Tris-HCl, pH 7.5, 300 mM NaCl, 50 mM imidazole) to remove non-specific contaminants. The bound protein was eluted by re-suspending the resin in elution buffer (10 mM Tris-HCl, pH 7.5, 300 mM NaCl, 750 mM imidazole). The eluted protein was extensively dialysed using SnakeSkin dialysis tubing (Pierce-MWCO 10,000; Thermo Scientific, USA) into dialysis buffer (10 mM Tris, pH 7.5, 100 mM NaCl, 0.5 mM DTT, 10% (v/v) glycerol, 50 mM KCl, 2 mM MgCl₂), and concentrated against PEG 20000 (Merck, Germany). The efficacy of the purification process was assessed using SDS-PAGE and western analysis using mouse monoclonal anti-His primary antibody and HRP-conjugated goat anti-mouse IgG secondary antibody (Santa Cruz Biotechnology, USA). Chemiluminescence-based protein detection was achieved using the Clarity^TMWestern ECL blotting kit (Bio-Rad, USA) as per the manufacturer’s instructions, and captured with a Chemidoc chemiluminescence imaging system (Bio-Rad, USA). The protein concentration for the purified proteins was quantified using the Bradford’s assay (Sigma-Aldrich, USA) with BSA as the standard. A fraction of the purified PfHep1 protein was dialysed extensively against buffer (50 mM Tris-HCl pH 7.5, 300 mM NaCl, 1 mM DTT, and 1 mM PMSF) containing 200 mM EDTA. The protein was then further dialysed in buffer without EDTA. ## Co-expression and co-production of PfHsp70-3 and PfHep1 in *E*. *coli* Co-production of PfHep1 and PfHsp70-3 in *E*. *coli* was conducted in order to functionally assess the escort activity of PfHep1. *E*. *coli* BL21 (DE3) cells were transformed with pQE30-PfHsp70-3 in the presence of pACYCDuet1 (empty vector) or pACYCDuet1-PfHep1 and grown at 37°C in 2x YT broth supplemented with 100 μg/ml ampicillin and 34 μg/ml chloramphenicol to mid-logarithmic phase, and protein production was induced by adding 0.1 mM IPTG. Cells were harvested prior to induction and at hourly intervals post induction for 5 hours and overnight. The harvested cells were centrifuged (13000×g; 2 min) and re-suspended in PBS buffer. Protein production levels of PfHsp70-3 in the presence or absence of PfHep1 were evaluated using SDS-PAGE and western analysis. The procedure for purification of the co-expressed PfHep1 and PfHsp70-3 was carried out as described for PfHep1 except that cells were harvested 5 hours post induction. For subsequent *in vitro* analyses, gel filtration was employed in order to separate the co-produced proteins. The dialysed protein solution was filtered through 0.2 μm filters and loaded into a HiPrep™ 16/60 Sephacryl™ S-200 HR column driven by an ÄKTA fast-protein liquid chromatography system (GE Healthcare, Biosciences, UK). The proteins were eluted at a flow rate of 0.5 ml/ min and 1 ml elution fractions were collected and analysed using SDS-PAGE and western analysis. The eluted proteins were concentrated against polyethylene glycol (PEG) 20000 (Merck, Germany). ## Suppression of thermally-induced PfHsp70-3 aggregation by PfHep1 An evaluation of the ability of PfHep1 to suppress the thermally-induced aggregation of PfHsp70-3 was adapted and modified from Dores-Silva et al.. The suppression of PfHsp70-3 aggregation by PfHep1 was monitored by light scattering at 360 nm for 30 min at 50°C in assay buffer (50 mM Tris-HCl, 100 mM NaCl; pH 7.4). For this assay, 0.8 μM PfHsp70-3 was used with stoichiometric concentrations of PfHep1 (dialysed in the presence and absence of EDTA). An evaluation of PfHep1 and EDTA-treated PfHep1 (E-PfHep1) to self-aggregate under the assay conditions was also conducted. Each assay was conducted in triplicate and three independent experiments on independent batches of proteins were conducted. Absorbance was plotted as percentage of PfHsp70-3 aggregation subsequent to normalizing against assays with PfHsp70-3 alone. ## Suppression of thermally-induced MDH aggregation by PfHsp70-3 and PfHep1 An evaluation of the ability of PfHep1 and PfHsp70-3 to suppress the thermally- induced aggregation of MDH was adapted from Burger et al.. Varying concentrations of PfHsp70-3 (0.25–1 μM), PfHep1 (0.25–1 μM) and combinations of these proteins in assay buffer (50 mM Tris-HCl, 100 mM NaCl; pH 7.5) were used to assess the abilities of these proteins to suppress the aggregation of 0.72 μM MDH by monitoring light scattering at 360 nm for 30 min at 45°C. PfHep1 alone did not self-aggregate under the assay conditions (data not shown). Absorbance was plotted as percent MDH aggregation over 30 min subsequent to normalizing against assays with MDH alone. Each assay was conducted in triplicate and three independent experiments on independent batches of proteins were conducted. ## Suppression of thermally induced citrate synthase aggregation by PfHsp70-3 and PfHep1 The ability of PfHep1 and PfHsp70-3 to suppress thermally induced aggregation of citrate synthase was adapted and modified from Lee et al.. Different concentrations of PfHsp70-3 (0.25 and 1 μM), PfHep1 (0.25–1 μM) and a combination of these proteins in an assay buffer (100 mM HEPES-KOH, pH 7.5) together with 0.15 μM citrate synthase from porcine heart (Sigma-Aldrich) were used. Suppression of aggregation was determined by monitoring light scattering at 320 nm for 30 min at 45°C Absorbance was plotted as percent over citrate synthase aggregation subsequent to normalizing against assays with citrate synthase alone. Each assay was conducted in triplicate and three independent experiments on independent batches of proteins were conducted. # Results and Discussion ## The functional domain of PfHep1 is conserved Based on the observation that mitochondrial members of the plant zinc ribbon (ZR) protein family show sequence similarities to Hep1 from yeast and humans, ZR and Hep proteins were classified as part of a family consisting of five subgroups: ZR1, ZR2, ZR3, Hep1 and Hep2. The classification was based on sequence identity and sub-cellular localization of individual proteins within the cell. The ZR subfamily refers to plant zinc finger proteins, ZR1 and ZR2 being plastidic and ZR3 residing within the mitochondria. The Hep subfamily refers to non-plant zinc finger proteins with Hep1 being mitochondrial and Hep2 being plastidic. The putative mitochondrial targeting sequence and the zinc finger domain (zf-DNL) of the hypothetical zinc finger protein from *P*. *falciparum* (PF3D7_1420300) was identified, and based on its sequence identity and predicted subcellular localization, it was denoted as PfHep1. Hep and ZR proteins have been observed to play roles in suppressing the self- aggregation of the respective mtHsp70 orthologues in humans, yeast, Leishmania, green algae and in *Arabidopsis thaliana*. The primary structure of full-length PfHep1 was aligned with its orthologues and the putative mitochondrial or chloroplast signal peptides are highlighted for each sequence. The PfHep1 sequence is asparagine-rich, with a continuous stretch from residues 128–166. In fact approximately a quarter of all amino residues found in PfHep1 are asparagines. The asparagines repeats are characteristic of the *P*. *falciparum* proteome and are often absent from heat shock proteins. PfHep1 is also larger than its orthologues with 302 amino acid residues. Whilst the region upstream of the zf-DNL is longer in PfHep1 than its orthologues, the C-terminal region is slightly longer in both the hsHep1 and ScHep1 sequences. There is lack of sequence conservation outside of the zf-DNL domain. No functional domains, with the exception of the zf-DNL, were identified in the primary sequence of PfHep1. However, the generation of truncation mutants in LbHep1 indicated that the region upstream of the zf-DNL contributes to enhancing the solubility of LbmtHsp70. The highest sequence identity was 38% between PfHep1 and CrHep2. This is not surprising as the apicoplast, non-photosynthetic plastid, in *P*. *falciparum* is of algal origin. There was approximately 25% sequence identity between PfHep1 and the remaining orthologues. Despite the low overall sequence identities between Hep proteins, the zinc-finger motifs (CXXC) that are part of the zf-DNL were found to be conserved in all of the sequences including PfHep1 and are separated by 21 amino acids. Yeast mutants harbouring either a C75S or C100S mutation in the tetracysteine motifs of Hep1 were found to be incapable of rescuing growth defects in cells lacking Hep1. Human Hep1 stimulated the ATPase activity of mtHsp70 but this was not observed in yeast. Mutations of the key residues R81, H107, and D111 in human Hep1 decreased the binding affinities for HSPA9. Furthermore H107 played a critical role in stimulating the ATPase activity of HSPA9. Interestingly two of these residues are conserved in PfHep1 (H255 and D259), while R81 is replaced with K229. In yeast Hep1/Zim17, residues R106, H107 and D111have been shown to play a critical role in interactions with mtHsp70. Nuclear magnetic resonance (NMR) structural elucidations have shown that Zim17 has an L-shape with the two zing- finger motifs located at the end of the L sandwiched by two anti-parallel beta- sheets. The zf-DNL of yeast Hep1 was used as a template to model the zf-DNL of PfHep1 and the overall structure of PfHep1 resembled that of yeast Hep1 with the cysteine residues found at the ends of two anti-parallel beta-sheets. ## Aggregation of PfHsp70-3 is prevented by PfHep1 co-expression The coding sequences of PfHsp70-3 and PfHep1 were inserted into pQE30 expression vectors without the mitochondrial signal sequences. PfHep1 was also inserted into pACYCDuet1 for the purposes of co-expression with PfHsp70-3. Induction of PfHsp70-3 with PfHep1 in *E*. *coli* cells was monitored over 5 hours and after overnight growth; western analysis revealed that both proteins were produced with lower levels of PfHep1 (data not shown). Examining the effect of PfHep1 co- expression with PfHsp70-3 revealed that it substantially enhanced the solubility of PfHsp70-3 (lane 5). PfHsp70-3 alone was virtually insoluble, as seen by the prominence of the protein in the insoluble pellet (lane 3) compared to the soluble fraction (lane 2). The solubility of PfHsp70-3 facilitated its native purification using nickel affinity chromatography without the need for denaturants such as urea. Some PfHsp70-3 was removed during the wash steps (lanes 3–5) and a low concentration of PfHep1 (35 kDa) co-eluted with PfHsp70-3 (70.4 kDa) in elutions 1 and 2, however PfHep1 was not detected in elution 3 (lanes 6–8). Human Hep1 displayed the features of a Type I J-protein and suppressed the aggregation of rhodanese. To determine if PfHep1 performed a similar aggregation suppression role, the recombinant protein was purified after heterologous expression in *E*. *coli* \[pQE30-PfHep1\] cells. PfHep1 was initially found to be insoluble (lanes 2 and 3), such that purification required the addition of the ionic detergent sarcosyl to the lysis buffer to solubilise the protein, prior to purification by nickel-affinity chromatography. Some PfHep1 was removed during the wash steps (lanes 3–5) and PfHep1 was successfully eluted (lanes 6–8). To our knowledge, the orthologues of PfHep1 are soluble and can be purified under native conditions. The insolubility of PfHep1 is probably due to the challenges associated with the heterologous expression of soluble and functional plasmodial proteins. Codon bias due to the AT-rich genome of *P*. *falciparum* can lead to poor protein expression levels in *E*. *coli*. However, codon optimisation does not ensure expression as a large scale screen involving the heterologous expression of numerous *P*. *falciparum* genes revealed that a quarter of the codon optimised proteins remained insoluble. The reason for this failure to express in a soluble form is unclear but could be attributable to the presence of long repeats of asparagine residues, resulting in a propensity of these proteins to form insoluble aggregates when heterologously expressed. ## PfHep1 prevents thermally induced aggregation of PfHsp70-3 PfHsp70-3 aggregated at 50°C which enabled us to assess the ability of PfHep1 to suppress its thermally induced aggregation. PfHep1 suppressed the aggregation of PfHsp70-3 in a dose-dependent manner, with equimolar concentrations displaying complete suppression of aggregation of PfHsp70-3. Similar results were obtained for *Leishmania braziliensis* Hep 1 (LbHep1), which suppressed the thermal aggregation of *L*. *braziliensis* mtHsp70 (LbmtHsp70). PfHep1 alone did not aggregate, while EDTA-treated PfHep1 (E-PfHep1; zinc ions removed) did aggregate under the assay conditions. E-PfHep1 failed to prevent the thermal aggregation of PfHsp70-3, and an additive effect was observed as the percentage aggregation was due to aggregated PfHsp70-3 and aggregated E-PfHep1. EDTA has been shown to destabilize the structural integrity of LbHep1 probably by chelating zinc ions that in turn weakens the secondary and tertiary structure of the protein. E-PfHep1 aggregated at 50°C and this was probably due to disruption of the structure after the removal of zinc ion. Zinc ions were found to be essential for maintaining the overall secondary structure of yeast Hep1 and LbHep1. ## PfHep1 did not prevent thermally induced aggregation of malate dehydrogenate or citrate synthase To determine if the prevention of aggregation by PfHep1 is specific to PfHsp70-3, it was necessary to assess its ability to suppress the aggregation of proteins that are known to aggregate thermally, such as MDH and citrate synthase. The addition of varying concentrations of PfHep1 resulted in less than 10% aggregation suppression of MDH and citrate synthase, while PfHsp70-3 suppressed the aggregation of MDH and citrate synthase in a dose-dependent manner. PfHsp70-3 did not aggregate under the assay conditions, neither did PfHep1 (data not shown). PfHep1 did not enhance the ability of PfHsp70-3 to suppress aggregation of either MDH or citrate synthase. Similarly, LbHep1 could not prevent the aggregation of two model proteins, MDH and Luc, but was specific for LbmtHsp70. # Conclusion We have shown for the first time that PfHep1 is required for maintaining the solubility and thereby the activity of PfHsp70-3. Our study has indicated that PfHep1 functions as a specialised co-factor that facilitates the folding of PfHsp70-3. PfHep1 is larger than its orthologues and there is no sequence conservation outside of the zf-DNL. The zinc binding domain of PfHep1 was predicted to contain many of the conserved amino acid residues important for its interaction with PfHsp70-3. Not surprisingly, PfHsp70-3 was insoluble when heterologously expressed in *E*. *coli* cells, as many eukaryotic mtHsp70s are insoluble. We did not anticipate that PfHep1 would be insoluble and this may be due to plasmodial proteins being notoriously difficult to express heterologously in a soluble form, even after codon-optimization. The co-expression of PfHep1 enhanced the expression and solubility of PfHsp70-3. The co-expression of PfHep1 and PfHsp70-3 facilitated the production of soluble and functional PfHsp70-3 that enables further biochemical characterisation of this chaperone in future studies. At high temperatures, beyond that of the human host, PfHep1 suppressed the aggregation of PfHsp70-3 but not other aggregation-prone proteins, such as MDH or citrate synthase. Furthermore, PfHep1 did not enhance the aggregation suppression activities of PfHsp70-3. Taken together, these data suggest that PfHep1 is not a co-chaperone of PfHsp70-3, but rather a specific co-factor to prevent its self-aggregation. To our knowledge, human Hep1 is the only Hep protein that has been demonstrated to display the features of a J-protein as it stimulated the ATPase activity of mtHsp70 and functioned as a holdase by binding unfolded proteins such as rhodanese. The sequence alignment revealed that the region downstream of the zf- DNL or C-terminal sub-domain, is longer in both humans and yeast with a lack of sequence identity between the two species. There is evidence to suggest that the C-terminal sub-domain of human Hep1 is responsible for regulating the activity of zf-DNL and conferring co-chaperone activity. An examination of the direct functional effects of Zim17 on mtHsp70 in the cell indicated a novel role of Zim17 in assisting its interaction with client proteins in a J co-chaperone- dependent manner. There is also the proposition that Zim17 functions as a “fractured” J-protein that provides a zinc finger domain to Type III J-proteins for substrate binding. However, this remains to be experimentally elucidated. The mechanism of action of human and yeast Hep1 appears to be different from that of other orthologues, and further studies are required to understand this differentiation. A greater understanding of Hep proteins from different organisms is required to determine whether or not they have the properties of *bona fide* co-chaperones. Thus future mechanistic studies on PfHep1-PfHsp70-3 could include determination of the ability of PfHep1 to stimulate the ATPase activity of PfHsp70-3 and identification of potential J protein co-chaperones in the mitochondrial matrix. In view of the fact that LbmtHsp70 plays a key role in the adaptation of the parasite in the host, disrupting the interaction between LbHep1 and LbmtHsp70 is a potential target for new therapies. Likewise, abrogating the specific partnership between PfHep1 and PfHsp70-3 may also be a target for new antimalarials. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AB GLB HH DON. Performed the experiments: DON LAMV SJB. Analyzed the data: AB DON GLB HH. Contributed reagents/materials/analysis tools: AB HH. Wrote the paper: DON SJB AB.

# Introduction ## Contemporary society and trends in adolescent mental health The secular trends in adolescent mental health are, in many Western countries, unclear due to changes in recognition, diagnosis, and how adolescents perceive their health, and this perception appears to be deteriorating among adolescent girls. In Norway, as in other Nordic countries, reports of mental health issues have increased in recent cohorts of adolescents. The apparent increase in Nordic adolescents affected by depressive symptoms and suffering from mental health disorders must be considered in the context of social and cultural changes that have occurred during the 21st century. Essential aspects of the past year’s societal changes could have affected people’s mental health status through psychological factors. These modern trends include (I) rising income inequality, (II) changes in family consultations and dynamics, and (III) particularly among adolescents, the growing use of modern online technology such as social networking sites, which have created a significant arena for social comparisons that could be psychologically harmful. Furthermore, secondary and tertiary students commonly report high levels of academic-related stress, and feelings of anxiety linked to schoolwork are common, particularly among girls, across OECD countries. The social factors are presented at different levels of society: individual, family, community, and national. ## Adolescence, social gradient in mental health, and family socioeconomic position (SEP) The social gradient in mental health, as in health in general, is a widespread phenomenon, and it has been well documented worldwide. Socioeconomic inequalities affecting health emerge early on in life and occur across the life course. Among adolescents, several studies have demonstrated a higher prevalence of mental health issues with decreasing family SEP calculated using the parents’ income, education. and occupational levels. Low affluence may affect mental health both directly and indirectly through a variety of mechanisms. Growing up in families with limited socioeconomic resources may affect both access to resources (e.g., time with parents, leisure activities, books and learning materials, housing conditions) and psychosocial conditions among the parents (e.g., stress, conflicts between parents, psychological difficulties,), which together affect the children’s socio-emotional development. According to the relative deprivation theory, it is also crucial how adolescents perceive their situations relative to others. Being unable to afford goods and activities that are considered affordable to most can be detrimental to mental health, particularly in adolescence when peer influences are intense as the material display of social status—how symbols of wealth are consumed and displayed (i.e., symbolic capital)—may be as important as the income itself. Adolescence, particularly late adolescence, may well be the period of life with the highest social equalization in health as adolescents earn independence from their parents but are still in the process of determining their own socioeconomic potential in terms of education, income, and job opportunities. However, several studies demonstrate an association between subjective SEP and adolescent health outcomes, particularly regarding mental health, although the health inequalities are considered less clear and consistent than those for adults. Lack of consistent findings may be due to geographical differences, methodological factors, use of different measurements of SEP and mental health, or the possibility that SEP may have different effects on various dimensions of mental health and at different stages of adolescence. Measuring adolescents’ SEP is an important methodological challenge in examining health disparities among young people. School-aged children and youth are likely to still be attending school and living with their parents, and consequently they lack their own social and economic status. Material conditions in the family can also be measured by the Family Affluence Scale (FAS) and used as an alternative to uncertain estimates of household income, parental education level, and parents’ occupational status/position. Inequalities in self-reported health and psychosomatic symptoms in adolescents have been revealed using FAS. However, Amone-P’Olak et al. reported that adolescents in the low SEP category are at risk of mental health problems, but family SEP accounts for a relatively small proportion (≤ 5%) of the total variance in mental health. Another aspect to consider is that objective SEP relates differently to adolescent health than does subjective social status (SSS), and they are not equivalent indicators of the same construct. It may be that SEP differences in adolescent health relate more closely to psychosocial processes than to material inequality. ## Socioeconomic environment, psychological distress, and social comparison Previous studies have emphasized different contextual domains that are likely to affect mental health such as the structure and capacity of health care resources in a community, local economic conditions, income inequality, social disruption, and social capital. Lantz, Pritchard define socioeconomic environment “as a place with geographically defined boundaries that also has economic, educational, social, cultural, and political characteristics,” and it can be measured within different units of geography (e.g., census tract, school district, zip code, municipality, county, country). In addition to family- related SEP, the socioeconomic environment shapes resources, opportunities, and exposures that can both directly and indirectly influence health outcomes. For example, in comparison to young people from wealthier neighborhoods, those from areas with high levels of poverty and distress tend to have higher levels of psychological distress and are at higher risk of suffering physical issues; however, neighborhood deprivation is not necessarily a causal risk factor for poor health nor for damaging health behaviors. Previous research suggests that people living in more equal communities in terms of material standards report better physical and mental health than do their peers in less equal communities. However, it is not a straightforward process to determine which socioeconomic conditions and community circumstances are considered beneficial for adolescents’ mental health and which conditions have an adverse impact. From a social interaction perspective, one might expect that adolescents living in high socioeconomic communities achieve health benefits from living in an environment that promotes an active and healthy lifestyle, having access to high-quality health services when needed, and surrounded by people with hope in life and faith in the future who facilitate (positive) learning, play, and growth. Moreover, it could be argued that living in a safe neighborhood with high materialistic standards has a positive psychological effect. Alternatively (and worth considering even though the literature is sparse), adolescents residing in high-level socioeconomic communities may be prone to creating harmful stress for themselves as a result of being exposed to social comparison and pressure to achieve. Being surrounded by ambitious and competitive fellow students from well-educated families may, for example, produce high academic stress due to educational expectations and pressure for academic achievement as well as difficulties with social comparison making them feel inferior and worthless. Other school-based social status dimensions, such as being attractive and sporty, have also been associated with well-being and psychological distress, and these might be reinforced in communities with high socioeconomic profiles. A weakness of many previous studies is that family SEP is not seen in relation to the socioeconomic standards of the local community. The potential interplay between the effects of structural, family, and community resources on adolescents’ depressive symptoms and well-being may manifest in various ways. For example, in line with the contextual amplification hypothesis, we might assume that the detrimental influence of familial factors and adverse community conditions, including concentration of poverty, noxious environment, weak degree of social integration, and collective socialization, reinforce one another. From another perspective, we might assume a buffer effect brought about by the beneficial resources at the family level which protect against adverse influence on the community level, or vice versa. However, an alternative explanation may be that the beneficial influence of family social resources on adolescents’ depressive symptoms decrease under adverse community conditions. Wickrama, Bryant call this “moderation of a contextual dissipation” and argue that, instead of buffering, the negative influence of family social resources on adolescents’ depressive symptoms level off under adverse community conditions. Several psychosocial factors could increase adolescents’ vulnerability to symptoms of depression and anxiety. Perceiving oneself as having low social rank compared to others has been demonstrated to be consistently linked to a higher degree of depressive symptoms. Wetherall et al. stated in their review article that “although markers of SES are consistently associated with depressive symptoms (e.g., measures of social rank may have a stronger association as they tap more psychosocial constructs than the objective indicators (Marmot and Wilkinson, 2001).” Adolescents’ perceptions of social status may be two- dimensionally rooted. One dimension is class identification based on parental SEP and familial placement in society: adolescents in high-level socioeconomic communities might live in families with relatively low or moderate socioeconomic resources; these adolescents are particularly vulnerable to the adverse effects of social comparison. Another dimension is the adolescents’ sense of personal standing compared to peers/schoolmates when it comes to, among other things, school performance and popularity in their school community. As adolescents age, they undergo a process of cognitive maturation that may increase their self- conception and ability to place themselves on a social status ladder. In Norway, young adults without high school diplomas face a higher risk of receiving a medically based disability pension (where the leading cause is mental illness) before turning 40-years-old if they reside in a municipality with a high socioeconomic profile. ## Study aims As income and educational inequalities in Norway are rising, socioeconomic disparities in adolescents’ mental health may be rising concurrently. Furthermore, within a community, the average educational level should be considered as a proper proxy measure of the socioeconomic profile of the area, considering local economic conditions and other social community benefits. On the other hand, there may be psychological mechanisms associated with well- educated communities such as competitive-oriented environments, higher expectations, and a rush for status that also impact (in a positive or negative way) on adolescents’ mental health. Despite this, few studies have explored the interactions between individual-level family affluence and community-level socioeconomic profile. This study aims to explore, in the national context of Norway, how the municipal socioeconomic indicators *education level* and *income inequality* affect anxiety and depressive symptom scores among students in their later teenage years, as well as to understand to what extent these associations are conditioned by family affluence. This study asked the following three research questions: (I) Do high school students (in general) achieve psychological benefits by living in a municipality with a high average education level? (II) Does the level of income inequality in a municipality affect the psychological symptom load among students? (III) Is this hypothetical municipal effect on high school students dependent on family affluence level? # Methods ## Design and data sources ### The Ungdata survey The studies involving human participants were reviewed and approved by the Norwegian Centre for Research Data (NSD), <https://www.nsd.no/en/>. Informed consent to participate in this study was obtained from the students, as well as from legal guardian/next of kin if the student was under 16 years of age. This cross-sectional study is based on questionnaire data collected in five waves (2014–2018) of the Ungdata survey. Ungdata is a quality assured and standardized system for local questionnaire surveys aimed at adolescents attending high school in Norway. The Welfare Research Institute NOVA (at OsloMet) is, together with Norway’s seven regional drug and alcohol competence centers (KoRus), responsible for conducting the survey. Participation in the survey is voluntary and based on the students’ informed consent. The survey covers different aspects of the students’ lives encompassing a wide range of thematic areas, and it is an important source of information on young people’s health and well-being, both at the municipal and national levels. The surveys take place during school hours and are carried out electronically. The response rate varies between surveys, schools, and school years. The overall response rate among senior high school students was 66% for surveys conducted in 2014–2016 and 69% for surveys conducted in 2016–2018. According to Bakken, the data are considered nationally representative during a three-year period. Our data are based on participants in the Ungdata survey during the period 2014 to 2018, and we thus consider that our study material in general gives a representative picture of Norwegian youths. (See Frøyland for a detailed description of the content and theoretical framework of the Ungdata survey.) ### Municipality state reporting (KOSTRA) We combined the individual data with census information on individuals’ home municipalities, sourced from Municipality-State-Reporting (KOSTRA) database administered by Statistics Norway (SSB). KOSTRA is a national information system that contains management information on municipal key activities including, demography, economy and social services development. ## Study population Our study sample consists of students who voluntarily undergo three years of education at level 3 in the International Standard Classification of Education (ISCED), which is the final stage of general and vocational secondary education. In Norway, students generally begin ISCED level 3 at age 16 and complete it the year they turn nineteen. Programs classified at ISCED level 3 may, for example, be referred to as “upper secondary education” or “(senior) high school”; in the present paper, we use the term “high school.” In this study, all high school students completing the version of the Ungdata survey containing questions on psychological distress, anxiety, and depression in the period 2014–2018 were included (N = 144,239). We excluded individuals with missing information on gender (n = 6,072), school year (n = 955), family affluence (n = 687), and municipal residential identifiers (n = 31,923). The final dataset with complete municipal identifiers contained 104,602 individuals. The study sample was further reduced in the parametric estimations due to individuals’ missing information on mean scores of psychological distress (n = 8,095), symptoms of depression (n = 7,142), and symptoms of anxiety (n = 7,459). We found no statistical difference between included and excluded individuals for mean symptom scores of psychological distress (1.91 vs. 1.91), depression (2.16 vs. 2.15), or anxiety (1.55 vs. 1.55). There was a noted higher share of first- year students (54% versus 45%, p\<0.001) for excluded compared to included groups. The fact that first-year high school students are overrepresented in the study population should be mentioned and may partly be explained by the fact that some high schools only offered the youngest students the opportunity to participate in the surveys. Moreover, it is more difficult to carry out surveys among the oldest students due to exams and the fact that many of the students are apprentices. A highly relevant explanation is also that the dropout rate in high school increases with increasing age of students. We should therefore be aware that the data, especially towards the end of high school, consists of a more selected group of students compared to lower grade levels. ## Assessment of variables ### Psychological distress, and depressive and anxiety symptoms Our dependent variables are adolescent’s psychological distress, and depressive and anxiety symptoms. Psychological distress was measured using the 10-item Hopkins Symptom Checklist, consisting of two subscales: a depression dimension (six items that constitute the “Depressive Mood Inventory”) and an anxiety dimension (four items). In addition, there is a total mean score (10 items). The students reported how often they were bothered by each of the following symptoms during the past week (the six first items relate to the depression dimension of the scale): “felt that everything is a struggle”; “had sleep problems”; “felt unhappy, sad, or depressed”; “felt hopeless about the future”; “felt stiff or tense”; “worried too much about things”; “suddenly felt scared for no reason”; “felt constant fear or anxiety”; “been nervous or felt uneasy”; and “felt worthless”. Each item was answered on a four-point scale ranging from “not at all” (1) to “very much” (4). Separate measures for psychological distress (total mean) and depressive/anxiety symptoms were constructed by adding up the scores (1 to 4) on all the items covering each dimension (10 items in total: 6 for depression and 4 for anxiety) and dividing the total by the number of completed items, given responses to at least half the statements for each scale. The resulting mean symptom scale scores—one for psychological distress, one for symptoms of depression, and one for symptoms of anxiety—ranged from 1 to 4. In addition to the mean score, a validated cut-off score of ≥1.85 was used to identify students reporting moderate to high symptom loads related to overall psychological distress, depression, and anxiety. ### Socioeconomic position (SEP) The socioeconomic position (SEP) of the adolescents was measured using a collective measure of SEP developed by Bakken et al. which includes, in addition to four questions from FAS II, information on parental education levels and the number of books in the home. The students answered the following four questions retrieved from FAS II: “Does your family have a car?”; “Do you have your own bedroom?”; “How many times have you travelled somewhere on holiday with your family over the past year?”; and “How many computers or tablet computers does your family have?” FAS II has been validated alongside other measures of adolescents’ SEP and compared to measures in which adolescents report their parents’ income, occupations, and education levels, and the scale has been found to have better criterion validity and less susceptibility to non-response bias. The goal with the collective measure of SEP is to capture three dimensions of a family’s socioeconomic position—parents’ education level, number of books in the home, and the family’s affluence level—and combine these into a collective measure of the family’s socioeconomic status. Parents’ level of education, number of books in the home, and family’s affluence level each have some clear limitations as measures of a family’s socio-economic situation. A collective index based on these three dimensions will probably provide a more robust and valid measure of family socioeconomic status. Consideration of anonymity is the main reason that the Ungdata surveys do not include questions about parents’ occupations or incomes. A critical review of each question included in the collective affluence measure, as well as detailed information on how the measure is developed, may be found in Bakken et al.. We calculated a mean sum score, ranging from 0 to 3, for each study participant, and then the total study sample was split into three equally sized groups ordered by increasing affluence level from low to high (low, medium, and high). ### Municipal sociodemographic characteristics In our descriptive analysis, we included municipal education level, median gross household income, income inequality, unemployment rate, disability pension rate, immigration rate, divorce rate, and life expectancy. The municipal education level was defined as the percentage of municipal inhabitants aged 15–80 years who completed tertiary education. In the descriptive analyses, the municipalities were grouped into quartiles where Quartile 1 is the 25% of municipalities with the lowest education level; Quartile 2 is the second lowest education level group; Quartile 3 is the group of municipalities with the second highest education level, and Quartile 4 is the 25% of municipalities with the highest education level among the adult population. The Gini coefficient, the most commonly used measure of income inequality, was used to estimate income inequality within municipalities. We used the census Gini coefficient (only available in the period 2014–2017) calculated for each municipality by Statistic Norway. The Gini coefficient varies between 0 and 1, where 0 corresponds to so-called “perfect income equality,” meaning that every household has the same income and wealth, and 1 corresponds to perfect income inequality, meaning that one household possesses the population’s entire income and wealth. According to the OECD, in 2017 Norway was ranked as the sixth country with the least income inequality, with a Gini coefficient of 0.262. Statistics Norway excludes students and uses a different method when adjusting for large households in their calculations of the Gini coefficient, which in 2017 was 0.252. The municipality characteristics enter the analyses as continuous (percentages or median) census variables. In our parametric estimations, we only included municipal education level and income inequality. ### Covariates The individual variables in Ungdata used in this study are *gender*, *school year* (proxy for age and categorized as follows: first, second, or third year of high school), and *survey cycle* (survey conducted in 2014, 2015, 2016, 2017, or 2018). These variables were included because earlier studies show strong associations with adolescents’ mental health. Introducing other individual or contextual sociodemographic variables was considered problematic in the modelling in regards to problems with potential statistical over-adjustment due to uncertainty whether the covariates represent real cofounding issues; or rather act as mediators, colliders or have bi-directionally roles. ## Statistical approach First, descriptive analyses of percentages on individual variables and municipal sociodemographic variables, by municipal education level were tested by chi square tests and analysis of variance (ANOVA). Second, we investigated the relevance of the residential context as well as the association between family affluence, municipal education level/income inequality, and psychological distress, depression, and anxiety symptoms among high school students and tested the hypothetical interactions using multilevel models. Linear multilevel models with individuals (level 1, n = 97,460) nested within municipalities (level 2, n = 156) were estimated. A two-level random intercept model was fitted using maximum likelihood estimation to distinguish the individual and municipality sources of variation in adolescents’ mental health. We modelled the prediction of adolescents’ mental health in five steps. First, we estimated an intercept model, only including the random intercept, to determine the impact of the municipality context on adolescents’ mental health. Second, we included the individual and family level variables (gender and school year) and family affluence. Third, as shown in, we extended the random intercept model for the relationship between family affluence and adolescents’ mental health to allow family affluence effect to vary across municipalities. A likelihood ratio test (LR test) was used to compare the random intercept and the random slope models’ goodness of fit. In the final steps, we included the municipal education level (main effect model) and the interaction terms of family socioeconomic status with municipal education level (interaction model) for all three mental health outcomes: psychological distress, depressive symptoms, and anxiety symptoms. Similarly, in municipal income inequality as well as the interaction terms of family affluence with municipal income inequality were included. In addition, supplementary analyses exploring the association between municipal socioeconomic conditions and adolescent risk of moderate-to-high psychological symptoms (main effect model) as well as analysis of the hypothetical interactions with family affluence (interaction model) were performed; the results are presented in – Tables. Estimates for fixed effects are reported as coefficients with 95% confidence intervals (CI). To quantify the influence of municipality of residence on adolescents’ mental health, we computed the intraclass correlation coefficients (ICCs) for each outcome. The ICC expresses the correlation in the outcomes (i.e., psychological distress, depressive and anxiety symptoms) between two individuals randomly selected from the same municipality: the larger the ICC, the stronger the clustering of psychological distress within the municipality and the larger the general contextual effect of the municipality. The Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) were used as measures of goodness of fit for our models. The model parameters were estimated by a general linear model (GLM) mixed effects method using Stata/MP software (version 13). # Results ## Characteristics of the study population presents descriptive information regarding the individual and municipality contextual variables among high school students in Norway, both in total and by municipal education level. In all, 50.4% females and 49.6% males participated in this study. More first-year students (44.6%) compared to second- (33.7%) and third-year (21.7%) students completed the questionnaire. Each family affluence group was (approximately) equally represented in the study population, with about one-third of the students each in high (36.3%), medium (30.8%), and low (32.9%) affluence groups. The mean family affluence score was 1.9, and, as expected, mean affluence score as well as share of high affluence students increased with increasing municipal education (p\<0.001). Inequalities in adolescents’ mental health were observed both between family affluence groups and municipal education groups in all three mental health domains, with increasing symptom load associated with decreasing affluence and education level. Mean psychological distress, depressive, and anxiety symptoms were higher (p\<0.001) in low affluence adolescents (1.95, 2.19, 1.58, respectively) compared with their peers with medium (1.91, 2.16, 1.54, respectively) and high (1.87, 2.11, 1.51, respectively) affluence levels (results not shown in table). The prevalence of students with moderate-to-high symptom loads showed similar patterns. Furthermore, mean psychological distress (1.8 to 2.0, p\<0.001), depressive (2.1 to 2.2, p\<0.001), and anxiety (1.5 to 1.6, p\<0.001) symptoms increased with increasing municipal education level. Similarly, the prevalence of students with moderate-to-high symptom loads increased with increasing municipal education level in all three mental health domains (psychological distress: 43% to 51%, depressive symptoms: 53% to 61%, and anxiety symptoms: 23% to 28%). We noted differences in other municipal characteristics across municipal education levels when divided into quartiles. The mean share of municipal inhabitants who have completed tertiary education increased from 22% in municipalities with the lowest education levels to 51% in the most educated municipalities. Income inequality (measured by the Gini coefficient) increased incrementally from the 25% least educated municipalities up to the municipal group in the top 25% education level. The most educated municipality (75^th percentile) had a larger population (528,217 vs. 169,047 inhabitants), higher median household income (648,476 NOK vs. 642,806 NOK), a lower disability pension rate (5.2% vs. 8.7%), a higher immigration rate (30% vs. 18.4%), and a slightly lower divorce rate (10% vs. 10.6%) compared to the average of the municipalities. The unemployment rate was 0.2 lower in the municipal group with the 25% lowest education level of the other groups (1.6 vs. 1.8). The life expectancy was 0.6 years higher in the municipal group “upper middle education level” than in the other groups (81.8 years vs. 81.2 years). ## Individual- and municipal-level variation in adolescents’ psychological distress In the first step of our parametric estimations, an intercept model containing only the second random intercept was estimated. We found that the ICCs for psychological distress, depressive symptoms, and anxiety symptoms among adolescents were 0.017, 0.019, and 0.010, respectively. In other words, our estimations suggest that about 2% of the variation in the students’ psychological distress and depressive symptoms could be attributed to differences between municipalities. However, only 1% of the variation in students’ anxiety symptoms could be attributed to differences between municipalities. shows the individual and family covariates of the three mental health outcomes: psychological distress (Model 1), depressive symptoms (Model 2), and anxiety symptoms (Model 3). Family affluence was negatively associated with each of the outcomes, with decreasing symptom scores accompanying increasing affluence levels. Being female and a third-year student was correlated with higher psychological symptoms. The interaction terms with gender and school year are negative and statistically significant, indicating that the positive association between females and psychological distress and depressive/anxiety symptoms decreases over time. In, we extend the random intercept models for the relationship between family affluence and a) psychological distress, b) depressive symptoms, and c) anxiety symptoms to allow the impact of family affluence to vary across municipalities. The two-level random intercept model, which is nested in the random slope model, is rejected at the 5% significance level (using a likelihood ratio test), suggesting that the impact of family affluence on adolescents’ anxiety and depressive symptoms does vary between municipalities. In relation to the municipal socioeconomic variables, we found that municipal education level is associated with higher psychological distress and depressive/anxiety symptoms among high school students in Norway. However, the interaction models suggest that these associations are also conditioned by the family affluence level of the students ( and). demonstrates that the predicted depressive and anxiety symptoms increase with increasing municipal education level among all high school students. Notably, among adolescents at the medium affluence level, the adverse effect of high municipal education level is statistically steeper than for students with high and low family affluence levels. For example, for students that live in medium affluence families, the predictive depressive symptom score is 2.12 in a municipality where 26% of the inhabitants have completed a tertiary degree, compared to 2.24 in municipalities with a tertiary education level of 46%. The same impact of municipal education and family affluence and their interactions were found on the predicted probability of moderate-to-high depressive and anxiety symptom complaints. Students from families of medium affluence who lived in one of the Norwegian municipalities with the lowest percentage of residents with tertiary education (for example, 22%) had a 53% chance of experiencing depressive symptoms. Students from families of medium affluence living in municipalities where the tertiary education level was (as high as) 50% had a predicted likelihood of 63% for moderate-to-high depressive symptom complaints. The effects of municipal income inequality and its interaction with family affluence on psychological symptoms were also tested, with similar results as with municipal education level. (main effect models) shows that psychological distress and depressive/anxiety symptoms among high school students increase with increasing income inequality. Notably, when adding the interaction terms with family affluence, the associations with municipal income inequality and psychological symptom load is most significant for high school students with medium affluence levels (See). # Discussion ## Key findings This national representative study exploring the impact of interactions between family affluence and municipal socioeconomic characteristics on psychological distress among Norwegian high school students has produced findings that should be highlighted. Overall, our results indicate substantial psychological symptom loads among 16–18-year-old students in Norway. Inequalities in adolescents’ mental health between family affluence groups were evident, with increasing symptom loads accompanying decreasing affluence levels. Females, particularly those attending the third year of high school, showed higher psychological symptom scores than males, mostly evidencing moderate-to-high psychological symptom loads. The average psychological symptom load and prevalence of students suffering from moderate-to-high symptom loads increased slightly with increasing education levels and income inequalities in their residential municipalities; this applies to adolescents in all three family affluence groups: low, medium, and high. However, our parametric estimations suggest that these municipal socioeconomic characteristics appear to have the highest impact on the mental health of adolescents living in families at the medium affluence level. This study deepens our understanding of how family affluence interacts with residential contexts to promote or inhibit young people’s emotional development. ## Education level in municipality and other socioeconomic conditions The first research question this study aimed to answer was whether high school students in general achieve psychological benefits by living in a municipality with a high average education level. Our parametric estimations suggest that the predicted mean of depressive and anxiety symptoms among high school students to some degree increases in line with higher proportion of municipal residents with tertiary education. Although there are several plausible reasons why living in a high socioeconomic environment with a high proportion of well-educated inhabitants positively affects young people’s health and well-being, the findings in the present study are contrary. In our study, the Norwegian municipalities with high proportions of well- educated people were typically characterized by high household incomes and low disability pension rates, which have been used in comparison with lower educated municipalities as proxy measures of local economic and neighborhood conditions in multilevel analyses of health status. It is argued in the literature that education creates social benefits to society beyond private returns through, among other factors, civic participation, enhanced political behavior, and lower crime rates. In accordance with these assumptions, we found that the education level in a municipality was positively associated with the voting rate, whereas it was inversely associated with the reported crime rate. Thus, the education level in a municipality might be a contextual indicator not merely of the local economic conditions, but also of social determinants of mental health such as level of social support and community social capital. Despite that, we found a positive association between municipality education level and adolescents’ psychological distress scores. This may indicate that contextual domains typically related to community infrastructure and material living conditions are either not of crucial relevance when considering differences in adolescents’ psychological distress between Norwegian municipalities, or that there are psychological mechanisms that mask or balance out the beneficial effects stemming from structural and materialistic factors. Elgar et al. indicated in their study of 1,371 adolescents in seven European countries that socioeconomic inequality in health is more closely related to psychosocial processes than to material inequality. Living in a community with a high share of the adolescents in socioeconomic advantageous families may, however, trigger some types of psychological stress. For example, expectations can be an important source of stress, and adolescents living in well-educated communities might perceive that the people they are surrounded by demand and expect much from them, including their parents, other family members, neighbors, classmates, teachers, training partners, and team leaders, as well as society as a whole. Furthermore, during adolescence the sense of social position is developing, and this psychological mechanism might also influence adolescents’ health and well-being; perhaps this source of stress is particularly prevalent in communities with high degrees of symbolic capital. The evolutionarily based social rank theory (SRT) accounts for the inferiority and submissiveness that is typical in depression. It might be that well-educated communities are characterized by underlying factors that increase the pursuit of social status and the chances of adolescents’ feelings of inferiority. Societies that accentuate individual competitiveness will be particularly vulnerable to the emotionally disturbing effects of social comparison. Furthermore, there are several other underlying contextual features that potentially explain why the average depression and anxiety scores among our study sample rise with increased municipal education level. It should be taken into consideration that the immigration rate was nearly three times higher in the most educated municipalities compared with the least educated municipalities. Moreover, major differences were observed regarding the municipalities’ per capita numbers. The least educated municipalities had the smallest population size; in contrast, municipalities where the 60^th percentile of adult inhabitants had a tertiary education had the highest populations. Differences in psychological distress between settlement types have been shown in several other studies, but the results are mixed. Rural and urban communities differ in environmental factors such as culture, socioeconomics, and access to healthy diets and health care that may influence adolescents’ functioning and well-being. In many Western countries, urban residents have a higher likelihood of suffering from psychological distress than rural residents. However, across the degree of centrality, small geographical variations in anxiety and depressive symptoms were reported in a previous study from Norway. Although we found an effect of education level at the municipal level, it is important to note that we found small variations between municipalities in adolescents’ depressive and anxiety symptom scores which is in line with previous Nordic population studies exploring geographical variations and rural- urban differences in mental health, This might be because Norway has a high- level welfare state which benefits all Norwegians no matter where in the country they live. Norway is considered an egalitarian welfare country with generally low income inequality (e.g., the Gini coefficient was 0.26 in 2017, compared to a Gini coefficient of 0.32 across OECD countries and it imposes national regulation on social and health services. The Nordic countries are leaders in promoting health through public policy action. In 2018, 92.3% of high school students attended public schools. ## Income inequality The second research question in this study was whether the level of income inequality in a municipality affects the psychological symptom load among students. In line with other population-based studies, our findings linked income inequality with poor mental health outcomes. According to Wilkinson, Pickett, unequal societies become dominated by status competition and class differentiation, and consequently they experience more health disadvantages. The prominent and well-accepted phenomenon of income inequality/health association is explained by hypotheses operating at different ecological levels, from the individual up to the national. Rising income inequality in a society can lead to a weakening of collective social capital and social integration, thereby affecting the residents’ health. A second possibility, which overlaps with the social capital mechanism, is that income inequality reinforces the negative health effects of social comparison through a more prominent status hierarchy in the local community. According to the status anxiety hypothesis, feelings of social defeat and inferiority produce stress reactions that lead to poorer health. However, the contextual mechanism behind the income inequality/health association phenomena could also be interpreted without psychological explanations, but rather based on distribution of material resources in a society. The neo-materialist hypothesis argues that material conditions per se and public and social infrastructure influence individual well-being and health. A review study analyzing data from 27 European countries found the most support for the status anxiety hypothesis and the social capital hypothesis, and the author suggests that the mechanisms through which income inequality influences mental well-being vary depending on the wealth of the country. Similarly, a review of the inequality-depression relationship by Patel et al. suggests that the main mechanisms at the local level are both the status anxiety hypothesis and the social capital hypothesis. At the individual level, the inequality- depression association is likely to be primarily mediated through psychological stress. The heterogeneity of study findings across populations on the effect of income inequality on depressive symptoms reflects the complexity of mechanisms and pathways. The size and geographical unit of analysis matters when considering the link between income inequality and mental health because different geographical levels have different meanings and do not relate to the same contextual characteristics. A study from The Netherlands, where income inequality is almost as low as in Norway, found an association between municipal income inequality and psychological distress (measured with the Kessler Psychological Distress Scale) in the adult population. Our study found that estimated income inequality had more effect on depressive symptoms than on anxiety symptoms. Notably, in the present study, well-educated municipalities were positively associated with higher income inequality. This can be seen from an urban-rural perspective. Both level of education and average income characterize municipalities within a relatively large city. In geographic areas where there are many knowledge-based jobs, many well-educated people will reside, and this is reflected in the average income of the citizens. However, there is still a need for low-income jobs in these areas, which apparently creates a connection between education level and income inequality in Norwegian municipalities. ## Family affluence, and the case of medium affluent students The third and final research question we raise in the present study is whether the degree of association between socioeconomic municipal characteristics and adolescents’ psychological distress depends on family affluence level. We found that the psychological disadvantages resulting from increasing municipal education levels and income inequalities were most significant among adolescents living in families at the medium affluence level. Our parametric estimates suggest that as the municipal education level and income inequalities increase, the depressive and anxiety symptom mean scores in families of low and medium affluence converge, while the differences between those in families of medium and high affluence are slightly increased. It is somewhat puzzling why the students in families of medium affluence (measured at the national level in Norway) seem to be most affected by an increase in municipal education level. This may partly be explained by the fact that students in families of medium affluence residing in the least educated municipalities have proportionally more peers at lower family affluence levels than themselves compared to students of medium affluence living in the most highly educated municipalities in Norway. Consequently, students in families of medium affluence may have more to lose than affluent students in terms of socioeconomic rank by residing in highly educated municipalities. A more open question asks why the difference in average psychological symptom scores between students of low and medium affluence gradually becomes smaller with increasing socioeconomic level. One speculation is that students of medium affluence might be (somewhat) more concerned about their social status than students of low affluence. In a study of adolescents aged 14–17 years from six European cities, the socioeconomic inequalities in adolescent health were more closely related to adolescents’ perceptions of relative family SEP than to objective indicators of material family affluence. Thus, our findings give some support to the assumption that late adolescents’ psychological symptom loads depend not only on their families’ affluence levels, but also on the socioeconomic position that level of affluence confers in the social hierarchy (see Subramanian, Kawachi for a brief review of linking income inequality, relative income, and relative rank to health). In general, our likelihood estimates show the highest psychological distress among students in families with the lowest family affluence, while the most affluent families had the fewest students suffering from anxiety and depressive symptoms. By using the FAS, Elgar et al. demonstrated socioeconomic differences in adolescents’ mental health in teenagers aged 11–15 years from high-income countries. Our study indicates that the late period of adolescence (i.e., 16–18 years) is not exempt from a socioeconomic family gradient of mental health. Given that people use conventional socioeconomic criteria to assign themselves subjective status, we may assume that family affluence is involved in adolescents’ assessments of their subjective SEP relative to their peers. Adolescents who perceive themselves to be at a lower level in the socioeconomic hierarchy compared to their peers are at higher risk of poor well-being and health outcomes. ## Strengths and limitations A main strength of the study is its relatively large and nationally representative sample, with explanatory factors at the individual and family levels linked to population-based municipal socioeconomic factors from national administrative databases. Moreover, family affluence level and adolescents’ symptoms of depression were evaluated in a standardized manner using validated measures. There are, however, several limitations of this study that should be noted. First, the use of a cross-sectional design limits the interpretation of our findings as we cannot draw inferences regarding the direction of the observed associations or the risk factor status of explanatory variables. Nor can our results reveal what may be the underlying causes of the associations we found. Second, although overall response rates were relatively high, approximately 30% of students were absent from the surveys, which inevitably poses the possibility of non-response bias due to illness and truancy. Third, while the items from the Depressive Mood Inventory have been validated in clinical studies, the present combination of anxiety items has not been validated. Thus, the reliability of the anxiety measure is uncertain, and the use of exclusively validated instruments would have strengthened the study findings. Fourth, the various high schools and municipalities included in the study sample differ in each survey/study year. Fifth, using municipality as an area-level when investigating how community environments may be related to health outcomes poses methodological challenges. Other geographical areas and social contexts (e.g., neighborhoods, school districts, peer groups, and other types of communities that facilitate social interaction) may be more relevant than municipalities for adolescents’ well-being and mental health outcomes. Moreover, the socio-demographic distribution of the population varies between Norwegian municipalities, and it is important to take this into account when interpreting the results of this study. However, presenting and interpreting estimates of effect measures for secondary risk factors (confounders and modifiers of the exposure effect measure) from a single statistical model may lead to several interpretative difficulties and may be misleading. Sixth and finally, an ideal study examining contextual effects on various health outcomes should include explanatory variables at multiple levels and allow the levels (i.e., contexts) to change over time. Family background, neighborhood of residence, and schools the children attended before high school are three highly important levels, and adjustment for them would constitute a strong advancement of the study. Ideally, this should be performed using a multiple-membership, cross-classified, multilevel analysis that allows the levels (municipality, neighborhood, school, and families) to change over time. However, our data do not include information about school, neighborhood, or particular household the child attended before entering secondary school, and thus this analytic framework cannot be used in this case. ## Interpretations and conclusion Living in a community with a high proportion of well-educated people is associated with several benefits. One possible benefit is the opportunity to learn and imitate healthy behaviors. Another benefit may be the sharing and receiving of social capital as well as the materialistic facilities in these types of communities. Despite these potential benefits of living in a well- educated society, our study demonstrates that there must be other factors at play which apparently are even more important for psychological well-being in the high-income and egalitarian welfare state of Norway with its national regulation of social benefits and health services. Thus, there may be psychological mechanisms associated with well-educated communities that have negative effects on adolescents’ mental health. The Norwegian municipalities with high proportions of well-educated people were typically characterized by an increase in income inequality compared to less educated municipalities. Living in competitive-oriented environments with high degrees of demands and expectations in several social arenas (such as at home, school, and sports activities, and when socializing with friends both in the virtual world and when physically present) might generate “status anxiety” and daily stress. We suggest that this type of disadvantageous environment is amplified by increased education level and income inequality in the Norwegian municipalities. Although we identified a modest degree of association between adolescents’ socioeconomic circumstances and psychological distress in Norway, the findings have potentially important implications for population health and society at large which should be considered when developing community planning and policy in high-income countries facing social changes and within-region inequalities in socioeconomic status. # Supporting information The Ungdata surveys were conducted by the Norwegian Social Research (NOVA) institute in cooperation with regional centers for drug rehabilitation (KoRus). We wish to thank them for their cooperation and for conducting the data collection. 10.1371/journal.pone.0254033.r001 Decision Letter 0 Yang Xiaozhao Yousef Academic Editor 2021 Xiaozhao Yousef Yang This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Apr 2021 PONE-D-21-02793 Psychological distress in late adolescence: The role of inequalities in family affluence and municipal socioeconomic characteristics in Norway PLOS ONE Dear Dr. Haugan, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jun 03 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Xiaozhao Yousef Yang, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <u><https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf></u> and <u><https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_t itle_authors_affiliations.pdf></u> 2\. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For information on unacceptable data access restrictions, please see <http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data- access-restrictions>. In your revised cover letter, please address the following prompts: a\) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially identifying or sensitive patient information) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent. b\) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. Please see <http://www.bmj.com/content/340/bmj.c181.long> for guidelines on how to de- identify and prepare clinical data for publication. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data- availability#loc-recommended-repositories>. We will update your Data Availability statement on your behalf to reflect the information you provide. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: I think this manuscript titled: "Psychological distress in late adolescence: The role of inequalities in family affluence and municipal socioeconomic characteristics in Norway" is of great interest and, in general, is well organized and well written. The topic is exact and relevant in the actual situation where mental health is more than ever valued. Besides, it brings to the reader new insights about the variables to consider in terms of youth, mental health, and social contexts and their associated features (family, municipalities...). Although focused on psychological distress in adolescence, it highlights several psycho-social aspects underlying a new vision and understanding about youth mental health. I think this psycho-social focus on mental health is one of the significant innovative contributions of the study. The outcome is a study that contributes to deep analysis and understanding of a psychological issue but going out from a psychologically perspective and thus becoming a comprehensive tool for researchers and practitioners. Also of relevance is the large and nationally representative sample used in the study. Still, I have a few comments on strengths, and also suggestions about possible changes to improve the article. Following the structure of the manuscript, they are: 1\. INTRODUCTION Very well organized and clear, straight to the point. The authors did a good summary of fundamental literature to support the study's aim and the problem. Although being targeted in Norway, the information is sufficiently valuable and adequate to understand other countries through this particular social organization's lens. Even related to just one country, the N of participants is significantly large, making this a robust study with broad implications and applications. The introduction in subsections makes it easy to follow and incorporate main conceptual aspects/issues and variables to consider. Study aims and research questions are clear and emerging from the introduction's contents. 2\. METHODS \- Sample and data collection The authors present a lot of pertinent information about participants and the data collection. According to this, some more aspects could be pointed out in a more explicitly way (e.g using subsections/subtitles), for example, if there were any eligibility criteria to participate or if there was a differentiation between the clinical and normative population, or if both were integrated into the study or not. The sentence (see lines from 216-218): "We found no significant differences between the study sample and excluded individuals in terms of mean symptom scores of psychological distress (1.91 vs. 1.91), depression (2.16 vs. 2.15), or anxiety (1.55 vs. 1.55)" is not clear, please add the word "respectively" at the end of the sentence. Besides, giving some information/characteristics about excluded individuals could help understand reported values and compare excluded participants and the study sample. \- Measures This section requires a more attentive organization in terms of titles and subtitles. Some doubts emerged: are the titles of "Measures" and "Individual and family level" and "Municipality level" at the same level, or all these points are about measures and so "Individual and family level" and "Municipality level" should be inside the section "Measures" as are the first subtitle: The outcome variables: psychological distress, symptoms of depression and anxiety"? Please review this organization of contents inside the subsection "Measures", even if authors considered different measures under each level. Besides, when reporting specific scales, I suggest informing the particular version used because some scales have several different versions (with, for example, different number of items); and/or if there were some changes in the original ones to serve specific purposes in this study. This means that more information on measurement instruments should be provided. \- Statistical approach I think the proposed statistical analysis is according to the study's aims and data. Besides this section, the authors also report some other specific analysis inside other parts of the manuscript (e.g., family and municipality levels have references to particular statistics). I suggest the authors write in this section of the "statistical approach" that some other information on specific statistics will be (or have already been) shown in specific other parts of the article. 3\. RESULTS I think this is a strong part of the manuscript. According to study aims and methods, the authors obtained several types of results. Tables and Figures are very elucidative, clear and readable, aggregating significant results and improving readers' facility to follow all the information. 4\. DISCUSSION In the discussion, the authors summarize the main findings, showing how the results support the conclusions. The authors highlighted the leading and innovative findings produced by the study in an excellent way, supporting and interpreting them clearly, through well-supported literature and previous research. Although this study focus in a specific country, the new and unexpected results and conclusions can inspire and support other studies on youth mental health in general. Still, I ask if authors could explicitly re- organize some of the research questions' findings. These questions defined at the beginning (in the introduction) should be explicitly again in the discussion in order to benefit the manuscript (just as a suggestion). The authors also report the study's strengths and limitations, and they also give suggestions to overcome some of these limitations in future studies. this is a very relevant part in the discussion. The list of references is complete and diverse, with very recent references supporting the topics discussed along with the article. In general, I think this is a relevant study demonstrating significant advances in adolescent mental health. I hope authors will reflect on comments made and improve the manuscript following suggestions on few aspects. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0254033.r002 Author response to Decision Letter 0 27 May 2021 Dear editor and reviewer, In the revised manuscript we have endeavored to meet PLOS ONE’s style requirements and made some changes based on comments from reviewer. Response to the reviewer. Thank you for constructive suggestions for improvements in the method sections. In the revised manuscript under “Methods” we have made the following revisions: 1\. Sample and data collection \- renamed the "sample and data collection" into "Design and data sources" and included the following subtitles "The ungdata survey". "Municiplaity state reporting (KOSTRA)" and "study population", please see line 183-235. \- rewritten the following sentences (line 216-219): "We found no statistical difference between included and excluded individuals for mean symptom scores of psychological distress (1.91 vs. 1.91), depression (2.16 vs. 2.15), or anxiety (1.55 vs. 1.55). There was a noted higher share of first-year students (54 % versus 45 %, p\<0.001) for excluded compared to included groups. 2\. Measures \- Rewritten the section into "assessment of variables" with the following subtitles: " Psychological distress, and depressive and anxiety symptoms", " Family affluence as a proxy for socioeconomic position (SEP)", " Municipal sociodemographic characteristics " and "Covariates" \- Revised the sections “Psychological distress, and depressive and anxiety symptoms” and “Socioeconomic position” – with clearer and more information of the particular scales used. 3\. Statistical approach \- We included the following sentence in the first part of the statistical approach section (line 321-323): "First, descriptive analyses of percentages on individual variables and municipal sociodemographic variables, by municipal education level were tested by chi square tests and analysis of variance (ANOVA)". We are also very grateful for the reviewer's advice to reorganize parts of the “Discussion” chapter. In the revised version of the manuscript, we have highlighted our three research questions in separate sections. In each of these sections, we first answer the respective research question by stating our related main finding. Data Availability statement The data and materials from the Ungdata-surveys are closed and stored in a national database administered by Norwegian Social Research (NOVA). The present study and analysis of the Ungdata were approved by the Norwegian Centre for Research Data (NSD). Norwegian legislation prohibits deposition of these data to open archives. The data are freely available for research purposes upon application. Details about the application process to NSD can be found at: <https://nsd.no/nsddata/serier/ungdata_eng.html>. Yours sincerely, Tommy Haugan 10.1371/journal.pone.0254033.r003 Decision Letter 1 Yang Xiaozhao Yousef Academic Editor 2021 Xiaozhao Yousef Yang This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 18 Jun 2021 Psychological distress in late adolescence: The role of inequalities in family affluence and municipal socioeconomic characteristics in Norway PONE-D-21-02793R1 Dear Dr. Haugan, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Xiaozhao Yousef Yang, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0254033.r004 Acceptance letter Yang Xiaozhao Yousef Academic Editor 2021 Xiaozhao Yousef Yang This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 23 Jun 2021 PONE-D-21-02793R1 Psychological distress in late adolescence: The role of inequalities in family affluence and municipal socioeconomic characteristics in Norway Dear Dr. Haugan: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Xiaozhao Yousef Yang Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Controlled release of drugs into the epidermis often results in large amounts of the drug remaining at the delivery site, whereas much smaller amounts of the drug enter the body system. Although some transdermal delivery systems can be efficient in supplying drugs that have a systemic effect, they are not practical for controlling the delivery of drugs when the final target is the skin itself. Thus, it would be useful to develop a delivery system that could maximize the period of drug deposit, either in the epidermis or the dermis, while minimizing its transdermal penetration into the body. Microsponge has been increasingly investigated to reach the aim in recent years. Microsponges are a polymeric delivery system consisting of porous microspheres, and they may enhance the rate of dissolution of poorly water-soluble drugs by entrapping such drugs in microsponge pores. These pores are very small, and the drug can be reduced to microscopic particles in these pores, which significantly increase the surface area of the drug and increase the rate of solubility. Hence, this system may improve the efficacy and bioavailability of some poorly soluble drugs. Microsponges are also designed to deliver active ingredients efficiently at minimum dose and to enhance stability, reduce side-effects, and modify drug release. Thus, the microsponge delivery system may be a promising drug-delivery system in treating skin disease. Paeonol is one of the main active components from the root bark of *Paeonia suffruticosa* (the tree peony), which has been widely used in Asia and Europe. Paeonol has potential for the treatment of neurodegenerative diseases in humans by alleviating morphological damage , increasing neuron viability, and reducing cerebral infarction. Paeonol also possesses anti-atherogenic and anti-arrhythmic activities, and is widely used in cardiovascular diseases. Paeonol also has anti-tumor and anti-inflammatory activity, as well as the ability to inhibit melanin. Recently, it has been reported that paeonol has anti-anaphylactic activity through regulating histamine and tumor necrosis factor-α, and these effects could be exploited in treating eczema. However, paeonol is hydrophobic and has a low aqueous solubility, with an oil/water partition coefficient of 2.21. Consequently, it is likely that it will be unable to penetrate the stratum corneum, and using enhancers may cause the drug to reach the blood too easily, but it is not desirable for the side effect. We know that, for any drug, including paeonol, to be effective in treating eczema, there must be a sufficient concentration of the drug in the epidermis. It is likely that paeonol would not be able to have an effect in treating skin disease, not because of the inefficacy of the drug itself, but because the drug cannot reach its site of action. Furthermore, increasing the amount of paeonol in the plasma might induce production of drug-metabolizing enzymes by the liver, which would affect the metabolism of other drugs. Therefore, formulation of a drug-delivery system to increase the rate of solubility of paeonol and its deposition in the epidermis, as well as to reduce its systemic action in patients with eczema, is of interest. Previous authors have developed paeonol-loaded liposomes to increase paeonol bioavailability in skin tissues. However, liposomes are very expensive and difficult to formulate in the final product and their manufacturing processes are very complex, thus it is difficult to carry out large-scale production with liposomes. <u>Rui-guang et al</u> did not report whether the prepared liposomes were stable over a long period at room temperature. Complexing paeonol with β-cyclodextrin was shown to increase its solubility, but this formulation could not control the release rate of the active agent by itself. Recent studies have shown that microsponge formulations were able to deliver drugs to the colon and then release the drug while retained in the colonic lumen, indicating that this formulation could be a new approach for colon-specific drug delivery,. In studies of a microsponge intra-dermal drug-delivery system, researchers used mupirocin, hydroxyzine hydrochloride, and benzoyl peroxide as the model drugs to evaluate the characteristics of this formulation, and showed that the main mechanism of drug release was diffusion and that the microsponge formulation could enhance the rate of drug release,. Generally, microsponges were prepared mainly by a quasi-emulsion solvent-diffusion method and free-radical suspension polymerization. However, free-radical suspension polymerization requires irradiation, high temperature, or catalysis to activate the monomers, and thus the manufacturing processes are very complex. In our project, the authors used an emulsion solvent-diffusion method to prepare the paeonol microsponges. Generally, the microsponges were stable over a wide range of pH values (1 to 11) and temperature (up to 130°C), and were compatible with most vehicles and ingredients. Because of the small pore diameter of the microsponges, bacteria cannot penetrate inside and reach their contents, thus, unlike liposomes and β-cyclodextrin, microsponges do not require any preservative compounds. It has been shown that some microsponge formulations can significantly maximize the time that the active ingredient remains in the skin, while minimizing its penetration through the dermis into the body. In addition, the controlled release of the drug from the microsponge formulation into the epidermis means that the drug remains primarily localized, with only a restricted amount entering the systemic circulation, and thus can be used as a means of controlling side-effects. Therefore, development of a microsponge delivery system for paeonol could improve drug efficacy in the local region and reduce side-effects. In this study, we prepared paeonol microsponges using an emulsion solvent- diffusion method. Scanning electron microscopy (SEM), production yield, loading efficiency, and particle size distribution were used to evaluate the characteristic of the paeonol microsponges thus formed. The microsponges were then incorporated into a cream base and we carried out *in vitro* and *in vivo* drug-release studies. # Materials and Methods ## Materials Paeonol was purchased from Sigma Co. Ltd. (Tianjin, China). Ethyl cellulose-M70, polyvinyl alcohol (PVA) 1788, white beeswax, and stearic acid were obtained from Aladdin Chemistry Co. Ltd. (Shanghai, China). Triethanolamine and liquid paraffin were obtained from Tianjin Fuyu Chemical Reagents (Tianjin, China). All other chemicals and solvents were analytical grade. ## Animals Female Wistar rats weighing about 200 g and female nude mice aged 6 to 7 weeks were purchased from the Center of Experimental Animals, Southern Medical University (Guangzhou, China). All animal experiments were performed according to the guidelines of the Experimental Animal Ethics Committee of Southern Medical University. And the protocol was approved by the Experimental Animal Ethics Committee of Southern Medical University (Permit Number for rats: 4402101707, Permit Number for nude mice: 4402102090). All surgery was performed under 10% chloral hydrate anesthesia, and all efforts were made to minimize suffering. ## Preparation of paeonol microsponges Preparation was optimized based on production yield, loading efficiency and sorting coefficient (data not shown). The preparation procedure was as follows. The paeonol microsponges were prepared using a quasi-emulsion solvent-diffusion method with external and internal phases. To prepare the internal phase, paeonol 7 g and ethylcellulose 1 g were dissolved in dichloromethane 20 ml. In this procedure, dichloromethane was an effective solvent for dissolving both the drug and the polymer. The external phase, which contained the emulsifying agent PVA 3 g dissolved in 100 ml of distilled water, was placed in the vessel, and stirred with a propeller-type agitator at 1000 rpm, and then the internal phase was gradually added into the stirring external phase. The mixture was then stirred at 1000 rpm for 4 h at room temperature to remove the dichloromethane from the reaction flask. After that, the formed microsponges were filtered through filter paper with a pore size of 0.45 µm (Millipore, Maidstone, Kent, UK), washed with distilled water, and dried at room temperature. ## Analytical system The quantitative determination of paeonol in microsponges was carried out using reverse-phase high-performance liquid chromatography (HPLC) using a chromatograph equipped with a quaternary pump (both model series 1100; Agilent Technologies Inc., Wilmington, DE, USA), on-line vacuum degasser, autosampler, column temperature controller, diode array, and multiple-wavelength detectors, along with an analytical workstation. The chromatographic separations were performed using an HPLC column 250×4.6 mm with a particle size of 5 µm (Syncronis C18; Thermo Fisher Scientific Inc., Rockford, IL, USA). A mixture of methanol and distilled water (65∶45) was used as the mobile phase. The filtered mobile phase was pumped at a flow rate of 1 ml/min, and the wavelength was set at 276 nm. All the determinations were performed at 25°C. The retention time for paeonol was found to be about 8.3 min, and the total run time was 10 min. A good linear relationship was found between the peak areas for various concentrations, from 0.004 mg/ml to 0.02 mg/ml (R² = 0.9996). External standardization by peak area was used for quantitative determination of the paeonol microsponges. The developed method had good precision (0.114%) and accuracy (2.76%). Each determination was calculated in triplicate, and the mean of the values were reported. ## Characterization and evaluation of microsponge formulations ### Scanning electron microscopy The morphology and appearance of the microsponges were studied using SEM with a system (H-3000N, Japan) operating at 10 kV. The samples were dusted onto double- sided tape on a metal stub and coated with gold/palladium alloy under vacuum. The obtained photograph was recorded at ×400 magnification. ### Determination of production yield The production yield of the microparticles was calculated according to the following equation:where M_ms is the final weight of the microsponges obtained, and M_rm is the initial weight of the raw materials (polymer and drug). All the experiments were performed in triplicate and the mean of the values were reported. ### Determination of drug content and loading efficiency The quantitative determination of paeonol in microparticles was carried out using HPLC as described above. The drug-loaded microsponges were weighed accurately and treated with ultrasonic waves for 30 min with methanol as the extraction reagent. Before injection into the HPLC, the samples were filtered through a nylon membrane filter (0.45 µm). The actual drug content and loading efficiency were calculated according to the following equation :where M_act is the actual quantity of paeonol in the weighed quantity of microparticles, M_ms is the weighed quantity of the microsponges, and M_the is the theoretical amount of paeonol in the microsponges. ### Particle size distribution analysis The particle size distribution of the microsponges was determined by a laser light-scattering technique (Mastersizer 2000; Malvern Instuments Ltd., Malvern, Worcestershire, UK). Before measurement, samples were dispersed in distilled water. The particle size range was set to 0.02 to 2000 µm, and the particle refractive index was set to 1.520. Sorting coefficient (σ) was calculated to evaluate particle uniformity, and particle size distribution by volume of the paeonol microsponge was calculated internally. ## Preparation of paeonol microsponge creams Creams were prepared using a standard reverse-emulsification method, using an oil phase and an aqueous phase in order to carry out convenient administration of the microsponges to the skin. The aqueous phase, contained 3.7 g triethanolamine, and the oil phase consisted of 7 g stearic acid, 3.5 g liquid paraffin, and 1 g white beeswax. Both phases were heated to 65°C in a water bath. The aqueous phase was then added dropwise into the oil phase while the mixture was stirred using a magnetic stirrer, while being allowed to cool to room temperature, thus forming a cream. Once the cream was complete, we added either paeonol alone or the paeonol-containing microsponges to the cream, and thoroughly mixed the compound to ensure a homogenous preparation. As a comparison, a saturated solution of paeonol, using normal saline as solvent, was also prepared. In our study, the content of paeonol in the microsponge cream or paeonol cream was set at 50 mg/g to keep the cream stable (data not shown), while it reached 505.52 µg/ml in the saturated solution. ## *In vitro* drug-release studies The *in vitro* release studies were carried out using a recirculating water bath and three Franz diffusion cells, with a receptor compartment volume of 15 mL and an effective area of 3.14 cm². The permeation membrane was mouse skin, obtained from female nude mice aged 6–7 weeks. After the mouse was euthanized, the whole skin was excised, the subcutaneous fat was carefully removed with forceps, and then the skin was washed with normal saline and examined for integrity. The skin was clamped between the donor and the receptor chambers. The receptor chamber was filled with normal saline and set at 37°C, and then the solution in the receptor chambers was stirred continuously at 300 rpm. The respective formulation (paeonol cream 1.0 g, paeonol microsponge cream 1.0 g, or saturated aqueous solution 1 ml) was gently placed in the donor chamber. At 15 min, 30 min, 45 min, 1 h, 1.5 h, 2 h, 3 h, 4 h, 5 h, 6 h, 7 h, 8 h, 9 h, 10 h, 11 h, and 12 h, samples of 1 ml were withdrawn from the receiver compartment and replaced immediately with an equal volume of normal saline kept at 37°C. The collected samples were then analyzed by HPLC. The cumulative curve was plotted of the total amount of paeonol that permeated at each time interval *vs.* time. The release kinetics of the paeonol cream and paeonol microsponge cream were calculated, and their release patterns were analyzed using different mathematical modes. ## *Ex vivo* drug-deposition studies For determination of the amount of drug deposited in the skin, we used Franz diffusion cells as described above for the *in vitro* release studies. Skin obtained from female nude mice was clamped between the donor and the receptor chambers, with the respective formulations (paeonol cream 1.0 g, paeonol microsponge cream 1.0 g, or saturated aqueous solution 1 ml) placed gently in the donor chamber. The permeation membrane was dismantled after 4, 8, 12 and 24 h, respectively. The remaining cream outside the skin was carefully removed, and then the skin was weighed and cleaned with 10 ml distilled water each time (five times in total). The treated skin was then cut into small pieces, and the drug in the skin was extracted by homogenization with 5 ml methanol. Finally, the tissue samples were centrifuged at 10,000 rpm for 30 min. The supernatants were collected and analyzed using HPLC. ## *In vivo* microdialysis Female Wistar rats weighing about 200 g were anesthetized with 10% chloral hydrate 0.35 ml/100 g, given as an intraperitoneal injection with supplementary injections of half the dose every 90 min if needed. For intra-dermal microdialysis, the abdominal fur of the rats was carefully shaved, and then the skin was incised over the dermis, followed by intra-dermal insertion of an introducer, assembled by inserting a stainless needle into the tubing. After placing the tubing into the dermis, the needle was withdrawn, followed by insertion of a microdialysis probe (CMA 20 Elite, 10 mm in membrane length, 20 kDa cut-off; CMA Microdialysis AB, Solna, Sweden) and then the tubing removed. After placement, the probe was perfused with normal saline using a syringe pump (CMA 402; CMA Microdialysis AB, Solna, Sweden) at a constant flow rate of 5 µl/min. While the rat was still under anesthetized condition, the jugular vein was isolated. For plasma microdialysis, the microdialysis probe described above was placed in the jugular vein, and perfused with normal saline by the microdialysis pump at a constant flow rate of 5 µl/min. The system was equilibrated for 1 h, and then 1.0 g paeonol cream or paeonol microsponge cream was applied to an area of skin 3.14 cm² in size. The intra-dermal and plasma dialysates were collected for HPLC analysis, using a refrigerated fraction collector (MAB 85; Microbiotech/se AB, Stockholm, Sweden) every 20 min for 12 h. In order to obtain absolute tissue concentrations from dialysate concentrations, *in vivo* relative recovery (R*_{in vivo}*) was carried out using retrodialysis prior to the microdialysis studies. In this experiment, one microdialysis probe was inserted into the dermis as indicated above, while the other probe was inserted into the jugular vein. Probes were perfused with 6.15 µg/ml paeonol at a constant flow rate of 5 µl/min, and the respective dialysates were collected at 20-min intervals and analyzed using HPLC. Recovery was determined from six consecutive dialysis samples per probe. R*_{in vivo}* was calculated as:where C_d is the drug concentration in the dialysate, and C_p is the initial perfusate concentration of paeonol. Prior to the PK analysis, concentrations for all dialysate samples were corrected using the following equation:where C_dermis/blood is the unbound drug concentration in the intradermal fluid or plasma. ## Statistical analysis Data in all experiments are presented as means ± SD. Statistical differences were tested by one-way ANOVA and the independent samples *t*-test. *P*\<0.05 was considered significant. # Results ## Characterization of the microsponge formulation The morphology of the microsponges was studied by SEM. The typical shape and surface characteristics of the microsphere are shown in. The microsponges were finely spherical and uniform in shape, and porous in nature, with no drug crystals on the surface. The production yield, actual drug content, and loading efficiency was calculated according to to. The production yield was 72.20±3.59% (n = 3), the actual drug content was 77.40±1.06% (n = 3) and the loading efficiency of the paeonol microsponge was 55.90±3.27% (n = 3). Using the laser light-scattering technique, a particle size distribution map by volume of the paeonol microsponges was determined, which showed that the specific area, surface diameter and diameter by volume of the particles were 0.65 m²/g, 9.5 µm and 22.4 µm, respectively. The particle size distribution of d(0.1), d(0.5) and d(0.9) were 6.3 µm, 15.8 µm and 37.4 µm respectively, and the sorting coefficient was 0.569. ## *In vitro* release studies The *in vitro* release profiles of paeonol cream, paeonol microsponge cream, and saturated aqueous solution are shown in. In order to have a better comparison between the release profiles of the three formulations, the slopes (flux) of the linear portion of the release profiles were calculated, as was the steady state flux (Jss), based on the cumulative amount of drug permeated per unit area plotted against time, and the estimated as steady state flux (J_ss). Based on these results, the *in vitro* release profile of paeonol microsponge cream could be best expressed by zero order kinetics, as the cumulative drug release *vs.* time were found to be linear (R² = 0.997). ## *Ex vivo* drug-deposition studies The amount of paeonol deposited in the skin from different formulations at different time intervals was determined by HPLC. As shown in, the amount of paeonol deposited in the skin was much higher for the paeonol microsponge cream than for the paeonol cream, especially at 4 h (0.5675±0.0394 mg/cm² *vs.* 0.3194±0.0091 mg/cm²) and 24 h (1.3627±0.0699 mg/cm² *vs.* 0.9988±0.0801 mg/cm²). Therefore, the highest amount of paeonol deposited by the microsponge cream at the same dose was found after 24 h, indicating that the microsponge cream improved drug residence in the skin. ## *In vivo* microdialysis The average percentage recovery of paeonol by the retrodialysis method over 120 min was 52.44±3.67% (n = 6) in the intradermal fluid and 51.70±3.75% (n = 6) in plasma. The dialysis membrane showed steady loss of paeonol for 120 min in the subcutaneous liquid and in plasma through the CMA-20 probe. The corrected dialysate concentrations of paeonol in plasma and intradermal fluid after topical application of paeonol cream and paeonol microsponge cream over the time interval of 12 h are shown in. The pharmacokinetic parameters for unbound paeonol were calculated. As shown in, the paeonol absorption was significantly higher with the paeonol microsponge cream than with the paeonol cream, which is in accordance with *in vitro* drug-release studies. Likewise, in the intradermal fluid, the area under the curve (AUC) for concentration versus time (AUC_0∼t) value was much higher for paeonol microsponge cream (2396±258.2 µg/ml/min) than for paeonol cream (1587±308.0 µg/ml/min). Maximum time (T_max) was 220 min for paeonol microsponge cream and 480 min for paeonol cream, while t_1/2 for paeonol microsponge cream (935.1 min) was almost twice that of the paeonol cream (548.6 min). By contrast, the AUC_0∼t and C_max values of the paeonol microsponge cream were less than half that of the paeonol cream (270.1±14.10 µg/ml/min *vs.* 656.7±153.2 µg/ml/min and 1.145±0.2126/ml *vs.* 0.6359±0.04150 µg/ml, respectively). # Discussion ## Characterization of microsponge formulations SEM images showed that the paeonol microsponges were porous and had a spherical shape. The pores were caused by the diffusion of the solvent from the surface of the microsponges. Thus, the volume of dichloromethane has a key role to play in the preparation of microsponges. ## *In vitro* release studies It is considered that microsponge formulations are too large to pass through the stratum corneum, hence they would be expected to remain on the skin surface, gradually releasing their contents over time. However, our results showed that preparing paeonol in microsponges could increase the permeation rate of the drug, compared with paeonol cream and saturated aqueous solution. This showed that the particle size has a significant effect on the drug release, and probably accounts for the discrepancy between our study and that of Jelvehgari et al, who used a pore size of about 300–400 µm, which is more than 10-fold larger than that used in our study. Our results indicate that microsponge formulations are able to increase the permeation rate of some liposolubility drugs formulated as small particles. Furthermore, the *in vitro* release profiles shown in indicated that paeonol microsponge cream was able to show a sustained release for up to 12 h compared with paeonol cream. The mechanism of drug release from microsponges may be associated with its porous surface, as this enables easy penetration of the release media and accessibility to the entrapped drug molecule. As the release media penetrated into the porous of the microsponge, the drug was then dissolved into it, thus drug released from the microsponge. As the release media first gained access to the surface of the microsponge and then gradually into the internal, the drug release measured over the first few hours was due to the presence of non-encapsulated paeonol on the surface of the microsponge, then as the release media gained access to the pores, there was release of the drug entrapped in the pores, resulting in sustained drug release for up to 12 h. ## *Ex vivo* drug-deposition studies Effective topical drug therapy requires a sufficient amount of drug to be taken up into the skin over a particular period of time to allow maximal pharmacological activity. Thus, the larger amount of drug deposited in the skin from paeonol microsponge cream indicates greater drug bioavailability in the topical area, which is in agreement with an earlier report. The higher drug retention by the paeonol microsponges for all time points may be partly explained by the occlusive effect, as the microparticles produced a film on the skin surface, which reduced transepidermal water loss. And then increased the hydration state of the stratum corneum, and finally leading to increased drug penetration into the skin. Moreover, the high lipophilicity of the drug microsponge formulation prevented drug diffusion from the skin into the receiver fluid, thus maintaining efficacious local drug levels for a long period of time. ## *In vivo* microdialysis As shown in, the observed T_max for paeonol microsponge cream was 220 min, compared with 480 min for paeonol cream. This may be due to the microsponge formulation producing an occlusive layer on the skin surface and thus reduced the transepidermal water loss and leading to high drug penetration, which was in accordance with the *in vitro* release studies. The average t_1/2 for the paeonol microsponge cream and paeonol cream was 935.1 min and 548.6 min, respectively, thus the microsponge formulation showed good distribution into the dermis. As shown in, the drug concentration of paeonol microsponge cream was very stable compared with that of paeonol cream, showing that the microsponge could act as a drug reservoir within the upper layers of the stratum corneum, allowing the drug to be released in a controlled fashion. For treating skin disease, the drug (in this case, paeonol) should stay in the treated areas and there should be no absorption to other areas. The AUC measured by microdialysis represents the total amount of drug that penetrated through the stratum corneum, into the epidermis and finally into the dermis after 12 h. Therefore the higher values of AUC_{0∼12 h} for paeonol microsponge cream in the subcutaneous liquid showed that the microsponge formulations could deliver significant amounts of paeonol to the dermis, thus leading to high drug bioavailability, but with high inter-individual variability. However, the lower values of AUC_{0∼12 h} and C_max for paeonol microsponge cream in plasma showed that the amount of drug absorbed into the plasma was not significant, thus this should result in reduced side-effects. This reduced absorption may be due to the high lipophilicity of the drug formulation preventing drug diffusion from the skin into the interstitial fluid. The results indicated that preparing paeonol as a microsponge formulation may be an effective drug-delivery system for skin targeting in treating skin disease. Major limitations of the microdialysis method are the low recoveries of molecules with large molecular weights (\>20 kDa) and high lipophilicity or high protein binding of some drugs. To determine the appropriate perfusate for paeonol, we also measured the average recovery and protein binding. The hydrophilicity (2.21 of partition coefficient), low molecular weight (166.18), high recovery (52.44% in intradermal fluid and 51.70% in plasma) and low protein binding (33.01±5.75) of the drug, showed microdialysis method was particularly appropriate for paeonol with saline as perfusate. There are controversial results regarding the influence of the actual position of the probe on recovery. The different probe depth into the dermis may influence the drug concentration in intradermal fluid. All the probes were inserted into nearly the same depth by the same person and the histological analysis was used to confirm that. And Esther et al. has reported that nearly consistent depths can be achieved if the probes are inserted by the same trained person. However the exact value in the dermis depth was not available. But in a skin penetration study of salicylic compounds, there was no significant correlation between probe depth and drug concentration in the range from 0.7 to 1.1 mm. The same conclusion had been found in the report of Muller et al. and Hegemann et al.. And Esther had also reported that there would be no significant correlation between probe depth and drug concentration when the variation in probe depth was very small. # Conclusions In this study, we first found that the microsponge delivery system could not only increase paeonol permeation rate but also minimize transdermal penetration of the drug into the body, which should increase drug bioavailability at the level of the skin and reduce side-effects when treating skin disease. Additionally, microsponges were able to improve the drug residence in skin and allowed sustained drug release for up to 12 h, resulting in a long active time for the drug in topical treatment of the skin. These properties indicate that a microsponge delivery system could be a useful strategy for a new generation of pharmaceutical and cosmetic treatments. However, the level of drug distribution in the stratum corneum, epidermis, and dermis and the mechanism of drug penetration is still unknown, thus further studies into these aspects are needed. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: QL SSL. Performed the experiments: SSL XJ BZ ZGL XLL TZ. Analyzed the data: SSL XJ. Contributed reagents/materials/analysis tools: LDW LL. Wrote the paper: SSL GFL.

# Introduction In domains ranging from the economy to national security, large-scale decisions often involve judgments about the machinations of a group agent, such as a terrorist organization, government, or corporation. Sometimes, judgments about a group agent simply reduce to judgments about one or more of its individual members (for example, thinking about whether or not a *country* is hiding nuclear weapons may primarily involve consideration of that country's *leader*). However, people also sometimes appear to make judgments about a group by treating it as an entity in and of itself. Individuals assign moral blame and punishment to whole organizations, interpret laws by looking for the ‘intentions’ of the legislature, may get into financial trouble by reasoning about the ‘mind’ of the market, and, in a recent decision by the United States Supreme Court, extended rights typically granted to individuals to a corporation as a whole. Although an abundance of research has investigated the effects of group membership on how people perceive and reason about the minds of individual people (for recent reviews, see, less is known about how perceivers reason about the ‘mind’ of a group agent itself. To investigate this question, the present work uses a combination of behavioral and fMRI approaches to examine the extent to which understanding the ‘mind’ of the group as a whole shares important properties and processes with understanding the minds of individuals. Specifically, we ask (1) to what extent people sometimes reason about the beliefs and intentions of a group agent separately from those of the groups’ members and (2) to what extent brain regions associated with understanding individuals also support understanding group agents. In order to predict or understand the behavior of a single individual, perceivers often appeal to that individual’s *mental states* (i.e., his or her thoughts, beliefs, intentions, desires, and feelings). This capacity to ascribe mental states to others—that is, to *mentalize*, or engage *theory-of-mind*, —reveals itself in the words perceivers use when talking about other people. For example, we can say that Dick *thought* he was aiming for a partridge and never *intended* to shoot his friend. Words like *think, believe, feel, intend, want*, and *plan* all refer to the inner contents of other minds, allowing perceivers to speak about the purported underlying causes of others’ behavior even as they diverge from that behavior itself. In turn, inferences about these internal causes guide moral decisions about how others should be treated, including the extent to which they deserve praise or punishment. Over the past two decades, an abundance of neuroimaging research has linked mentalizing or theory-of-mind to a consistent set of brain regions, including the medial prefrontal cortex (MPFC), temporo-parietal junction (TPJ), and precuneus/posterior cingulate, sometimes collectively called the ‘theory-of-mind network’. Using carefully controlled tasks that aim to isolate theory-of-mind, these regions show preferential engagement when people are thinking about humans versus other entities – and when people are thinking about humans' minds versus their other aspects, such as their physical attributes,. Although much of this evidence has been correlational, recent work using TMS has demonstrated a causal role for the Right TPJ (RTPJ) in the use of mental state information for moral judgment, and research on individuals with damage to MPFC and TPJ has demonstrated a role for those regions in the ability to make inferences about others' mental states. Intriguingly, mental state words pervade perceivers' statements not only about individuals but also about groups. In recent news reports, we learn that “Apple thinks carefully about its entire product lineup”, that “Apple wants owners to sell their old iPhones back to the company for a discount on a new phone”, and that “Apple intends to work with record labels to identify and promote up and coming artists”. In cases like these, people apply words normally associated with the psychological states of an individual person—words like ‘thinks’, ‘wants’, and ‘intends’—to a corporation as a whole. These same expressions can also be applied to other sorts of group agents. People talk about what a government agency ‘intends’, what a religious organization ‘thinks’, or what a sports team ‘loves’ or ‘hates’. Indeed, archival studies show that people speak about groups using mental state words spontaneously, even outside the context of an experiment, and cross-cultural studies document the use of mental state words in descriptions of groups not only in the West, but also in East Asian cultures. Does the use of such language indicate that people understand governments and other organizations by attributing mental states to a group? Critically, there are two different senses in which one might think about ‘groups’ and, accordingly, two different senses in which one might investigate the processes perceivers use to understand groups. On one hand, one could think about a ‘group’ as referring to the *members* of groups. If each group member is a human being, then the group is simply a collection of human beings. A first sense in which one might investigate how perceivers understand groups, then, is to investigate how people understand collections of human beings. On the other hand, one could think about a ‘group’ as referring to a *group agent*. A group agent itself is not merely a collection of separate human beings but, instead, an entity with whatever sort of status attaches itself to corporations, nations, and sports teams. Thus, a second sense in which one might investigate how perceivers understand groups is to investigate how people understand not collections of individuals, but group agents. An example highlights the distinction between a group in the sense of a collection of individuals and a group in the sense of a group agent. Consider the sentence “The employees and stockholders of Acme Corp. are all in debt.” This sentence says something about the financial condition of various individual human beings while making no claims about the financial condition of the corporation with which they are associated. In other words, the sentence ascribes a property to the members without ascribing that property to the group agent itself. By contrast, consider the sentence, “Acme Corp. is in debt.” This sentence says something about the financial condition of a corporation, but it makes no claims at all about the financial condition of any individual human beings. (The corporation itself could be in debt even if all of the employees and stockholders were in excellent financial shape.) Thus, this sentence ascribes a property to a group agent without ascribing that same property to any of the members. Existing work already provides some evidence for the claim thinking about groups in the first sense—i.e., thinking about collections of human beings—shares properties and processes with thinking about individual people. Behaviorally, the vast literatures on stereotypes and intergroup relations show that people are willing to ascribe psychological attributes to whole collections of others,, and studies indicate that some of the same principles that apply to the ascription of properties to individual agents also appear in the ascription of properties to whole collections of agents. Moreover, a recent neuroimaging study observed activation in brain regions associated with theory-of-mind—MPFC, TPJ, and precuneus—when participants evaluated the applicability of certain preferences both to individual people and to collections of individuals, compared to a non-mental control condition. Taken together, these behavioral and neuroimaging studies provide support for the view that people can ascribe psychological attributes not only to individual human beings but also to collections of human beings, and that they may use similar processes to do so (even if the outcomes of those processes may sometimes differ). Yet studies like these still leave open the question of how people understand groups in the second sense—i.e., how they understand group agents. As we saw above, people can ascribe a non-mental property to all of the members of a group agent without ascribing that property to the group agent itself (“All of the employees and stockholders are in debt”). Similarly, perhaps people can ascribe a mental property (i.e., a mental state) to all of the members of a group without in any way ascribing these states to the group agent itself (“The employees and stockholders all love Jeopardy!”). We have also seen that people can ascribe a non-mental property to a group without ascribing that property to the individual members (“Acme Corp. is in debt.”). Similarly, perhaps people can ascribe mental states to a group agent without ascribing that state to any of the members. Indeed, recent research suggests that the more people perceive a ‘group mind’, the less they tend to perceive the minds of the members of that group. With this in mind, the current studies investigate how perceivers understand group agents by examining the extent to which understanding group agents shares important properties and processes with understanding individuals. Experiment 1 examines behaviorally the extent to which people ascribe mental states to group agents over and above attributions of mental states to their individual members. Experiment 2 uses fMRI to investigate the extent to which understanding and predicting the behavior of group agents recruits brain regions associated with understanding and predicting the behavior of individuals—i.e., brain regions associated with theory of mind. # Experiment 1: Ascriptions to group agents vs. ascriptions to group members When people use sentences that appear to ascribe mental states to a group agent, are they actually ascribing something to the group agent, or are they merely attributing something to the group's members? For example, consider the sentence, “United Food Corp. believes that the new policy is morally unacceptable.” At least on the surface, this sentence appears to attribute a mental state (the belief that the policy is morally unacceptable) to a group agent (United Food Corp). However, it is possible that this is just a linguistic shortcut, and that when people use or hear sentences like this one, they are really attributing mental states to the members of the group, not to the group itself. Existing research demonstrates that people sometimes do use sentences that appear to attribute a property to a group when referring to its members, specifically when the members of the group have the particular property in their roles as group members. For example, if each member of the Sigma Chi fraternity gets drunk, and if each of them does so in his role as a Sigma Chi member, people tend to agree with the sentence, “The Sigma Chi fraternity got drunk”. This sentence appears on the surface to be ascribing a property to the fraternity itself—the actual organization— but is in fact just a shorthand way of ascribing a property to the individual members in their roles as members. In Experiment 1, we examine whether apparent mental state attributions to group agents can involve attributions of a property to a group agent itself, or whether they reduce to attributions to individual group members. To the extent that perceivers genuinely attribute a property to the group agent itself, attributions to group agents should sometimes diverge from attributions to the members of those groups. That is, we should observe (a) cases in which perceivers attribute a mental state to all of the members of the group without attributing that state to the group agent itself and (b) cases in which perceivers attribute a mental state to the group agent without attributing that state to any of the group's members. In contrast, to the extent that apparent attributions to group agents are merely shorthand for attributions to the group members, participants should not attribute properties to the group agent that they do not also attribute to the members of the group. Thus, finding that individuals attribute mental states to a group agent without attributing that state to any of the group's members would be the most unambiguous evidence that perceivers can apply mental states to group agents themselves. ## Method ### Participants 116 Yale students and faculty (33% female; age range 18-54, mean age 21 years) were recruited outside a dining hall to fill out a questionnaire for payment. ### Ethics statement This study was approved by the Institutional Review Board at Yale University. All participants provided written informed consent. ### Materials and Procedure This experiment used a 2 (mental state: individual-only or group-only) × 3 (question: any member, each member, group) design in which target was manipulated within-subject and question type was manipulated between subjects. Each participant received eight vignettes in counterbalanced order. Four vignettes were designed in such a way that it would be logically possible to ascribe a particular mental state to each of the individuals in the group without ascribing that state to the group itself (*Individual-only* condition). For example, one vignette described an organization devoted to fighting the death penalty. All of the members of this anti-death penalty organization are also interested in antebellum American history, so they decide to form a separate organization, with exactly the same members, called the Shady Grove Antebellum Historical Society (SGAHS), which meets to discuss historical questions. If participants are willing to ascribe a mental state to all of the individual members without ascribing that mental state to the group as a whole, participants should report that all of the members of SGAHS want to fight the death penalty but that the SGAHS itself *does not* want to fight the death penalty. On the other hand, to the extent that attributions to a group simply reduce to the attributions made to the individual members, participants should report that SGAHS *does* want to fight the death penalty. The other four vignettes were designed such that that it would be logically possible to ascribe a mental state to the group itself without ascribing that state to any of the individual members (*Group-only* condition). For example, one vignette described a large organization that was commissioned to build a space shuttle. Some members of the organization put together the software, others build the exterior, still others are in charge of the fuel, and so forth. But there is no single person who works on every aspect of the project. To the extent that people are willing to ascribe a property to a group agent over and above its members, participants should say that the organization knows how to build a space shuttle, but the individual members do not. In another vignette, a Community Association needs to choose music for an upcoming event. Some members really want to play punk music and can't stand classical, others really want to play classical music but strongly dislike punk, so in the end, the Association selects a third option: classic rock. If people are willing to attribute properties to group agents over and above their members, participants should say that the Community Association itself preferred playing classic rock but that none of the individual members shared this preference. On the other hand, to the extent that attributions to the group simply reduce to the attributions made to the individual members, participants should report either that most or all of the individual members prefer playing classic rock or that the group itself does not prefer playing classic rock. For full texts of the vignettes, see. Each participant was randomly assigned to one of three question conditions: ‘any member,’ ‘each member,’ or ‘group.’ Participants in the ‘any member’ condition received after each vignette a question about whether any individual member of the group had a particular mental state (‘Do any of the members of the Community Association prefer the idea of playing classic rock to the idea of playing every other type of music?’). Participants in the ‘each member’ condition were asked whether *each* member had the relevant state (‘Do each of the individual members of the Community Association prefer…?’). Finally, participants in the ‘group’ condition received questions about whether the group itself had the relevant state (‘Does the Community Association prefer…?’). Each question was answered on a scale from 1 (‘No’) to 7 (‘Yes’). ## Results Two participants failed to complete all items of the questionnaire. We calculated the mean response to ‘group’, ‘any member’, and ‘each member’ questions in the ‘Members Only’ vignettes and the ‘Group Only’ vignettes for the remaining participants. To the extent that participants attributed purported mental states to group agents themselves, we should observe both cases in which participants attribute a state to all of the members of the group without attributing that state to the group itself and, most critically, cases in which participants attribute a state to the group itself without attributing that state to any of the individual members. See for complete dataset. For the Members-Only vignettes, a one-way ANOVA revealed a significant effect of question condition, *F*(2, 114)  =  41.2, *p* \<.001, η² = .42, such that participants were willing to attribute states to some or all of the members of a group without attributing those states to the group itself. Tukey's posthoc tests showed that participants agreed less with ascriptions in the ‘group’ question condition than in either the ‘any member’ question condition, *p* \<.001, or the ‘each member’ question condition, *p* \<.001, suggesting that attributions to the group did not simply reduce to attributions to the group's members. Critically, for the Group-Only vignettes, a one-way ANOVA again revealed a significant effect of question condition on participants' responses, *F*(2, 114)  =  91.6, *p* \<.001, η² = .62, such that participants were willing to attribute states to the group itself that they did not attribute to any of the members of the group. Tukey's posthoc tests showed that participants agreed more with ascriptions in the ‘group’ question condition than in either the ‘any member’ question condition, *p* \<.001, or the ‘each member’ question condition, *p* \<.001. Moreover, participants' responses in the group question condition were significantly above the neutral midpoint of the scale, *p* \<.001, indicating that participants were genuinely endorsing sentences ascribing mental states to group agents. These results suggest that attributions to the group agent were made over and above the attributions made to individual members. This study explored the relationship between ascribing states to group agents and their members. We observed cases in which participants attributed a state to all of the members but did not attribute that state to the group itself and also cases in which participants attributed a state to the group itself but did not attribute the state to any of the members. Together, these results demonstrate that mental state ascriptions to a group agent can diverge from those made to the group's individual members, suggesting that perceivers can attribute a property of some sort to the group agent itself. # Experiment 2: Neural processes supporting mental state ascriptions to group agents Experiment 1 suggests that that when people use expressions of the form ‘United Food Corp. wants.’, they appear to be ascribing something to the group itself, rather than to the members of the group. However, a further question concerns the processes supporting these ascriptions. That is, although such statements clearly involve the same linguistic expressions that people use when applying theory-of-mind to individual human beings, to what extent do they also involve the same cognitive processes? To investigate the processes supporting attributions of purported mental states to group agents, we scanned participants using fMRI as they considered the mental states of individuals and groups. In one task, participants read sentences that referred explicitly to the mental states of groups and individuals (along with matched, non-mental control sentences). In a second task, participants carried out a procedure that relied on mental state ascription incidentally, without the use of mental state words: making predictions about what an individual or group would do in a variety of situations. To the extent that perceivers rely on processes associated with understanding individuals when they understand and predict the behavior of groups, brain regions associated with theory-of-mind should be active both when thinking about individuals and when thinking about group agents, and they should be active to a similar degree. On the other hand, to the extent that perceivers rely on different processes to understand group agents, we should observe reduced activation in brain regions associated with theory-of-mind—RTPJ, MPFC, and precuneus—during consideration of groups versus individuals. In the design of this study, steps were taken to (a) minimize, as much as possible, the likelihood that participants would simply consider the minds of individual group members when considering group agents and (b) test sensitively the degree to which brain regions associated with theory of mind are engaged during consideration of group agents. Unlike past studies, no individuals were mentioned or shown in the group condition, and both directed and spontaneous theory of mind tasks were included. Moreover, the results of Experiment 1 show that perceivers do interpret sentences about group mental states as ascribing mental states to the group agent itself. Although MPFC, TPJ, and precuneus have all been associated consistently with theory-of-mind, finer-grained differences in the response profiles of these regions facilitate predictions about their involvement during consideration of group agents. Recent neuroimaging research has increasingly revealed that, even when mental state attributions to individuals are concerned, MPFC, TPJ, and precuneus do not all respond in the same ways under the same circumstances. In particular, there are at least two ways in which the processes associated with purported mental state reasoning about group agents may differ from those associated with individual people. One is that certain properties of the *type* of mental state content being attributed may differ. The other is that certain properties of the *target* to whom that content is being attributed may differ. The RTPJ consistently demonstrates sensitivity to the *type* of mental state being ascribed. Specifically, a series of studies has demonstrated that RTPJ is selective for processing representational mental states, such as beliefs –; see for review. The RTPJ response is high when participants read stories that describe a character's true or false beliefs but low during stories containing other socially salient information, such as a character's physical appearance, cultural background, or even internal sensations such as hunger or fatigue. Similarly, activation in RTPJ is higher during inferences about an individual's beliefs than during closely matched inferences about an individual's preferences regardless of whether such inferences are more or less constrained by external information—a response profile that is not shared by other regions associated with social cognition, such as MPFC. Moreover, activation in the RTPJ consistently tracks with thinking about mental contents, not merely seeing mental state words. RTPJ becomes engaged when participants think about others' mental states even in the absence of any mental state words, such as when they view non-verbal cartoons or read descriptions of actions that imply a particular mental state. Conversely, mental state words alone do not elicit activation in the RTPJ; for review see. Thus, mental state words are neither necessary nor sufficient for eliciting RTPJ activation. Instead, RTPJ activation during social cognition appears to be associated with the ascription of representational mental state content; for discussion see. Thus, to the extent that perceivers attribute representational mental states to group agents, we should observe similar levels of RTPJ activation during consideration of group agents and individuals, both of which should exceed that associated with a non-mental control condition. In contrast, MPFC appears to be especially sensitive to the *target* of mental state ascription. In particular, thinking about oneself, a similar individual, a familiar individual, or an individual whose perspective one has taken earlier is associated with more MPFC activation than thinking about more distant others. MPFC also appears to be sensitive to the target of consideration when theory-of- mind is not explicitly called for. For example, this region exhibits less activation during consideration of “dehumanized” than “humanized” individuals and responds more during consideration of one's own versus another person's physical attributes. Although it remains open to further inquiry whether lower MPFC response in these cases genuinely indexes a difference in the degree to which mental states are attributed or rather the use of an alternative process for doing so, the sensitivity of MPFC to the target of judgment suggests that group agents may be particularly likely to be associated with lower activation than individuals in this region. ## Method ### Participants Nineteen right-handed, native English speakers (10 female; age range 19-25, mean age 21 years) with no history of neurological problems participated for payment. All participants had normal or corrected-to-normal vision. ### Ethics statement This study was approved by the Committee on the Use of Humans as Experimental Subjects (COUHES) at the Massachusetts Institute of Technology. All participants provided written informed consent. ### Stimuli and Behavioral Procedure #### Directed theory-of-mind task During fMRI scanning, participants completed an individual vs. group agent theory-of-mind task in which they read short statements about everyday events. Participants were instructed to read each statement and were told that they would be asked a series of questions about the statements later on in the experiment. Inanimate (control) statements communicated information without reference to people (e.g., “Although there wasn't much real data on agricultural production, the statistics showed that rutabaga production was consistently going down.”). Based on each control statement, an *individual* statement and a *group* statement were constructed. *Individual* statements concerned a single person's mental state (e.g., “Although there wasn't much real data on agricultural production, George Hailwood was sure that rutabaga production was going down.”). *Group* statements concerned the ‘mental state’ of a group agent (e.g., “Although there wasn't much real data on agricultural production, United Food Corp. was sure that rutabaga production was going down.”). No participant viewed more than one version of the same base statement. In each run of this task, participants read statements organized around a single theme (e.g., one run concerned George Hailwood, United Food Corp., and food production, whereas another concerned Stephanie Ann Majors, a record company, and music sales). For full texts of the stimuli, see. Participants completed ten functional runs of eighteen statements each (six per condition), totaling 180 trials. Statements were displayed in random order within each run and remained onscreen for 8 s. Trials were separated by a variable inter-stimulus interval (2–16 s) during which participants passively viewed a black screen. #### Spontaneous theory-of-mind task Following each run of the directed theory-of-mind task, participants were asked to make a series of predictions about the individual and group about which they had just read (e.g., “The asparagus might be contaminated by bacteria. Would George Hailwood \[United Food Corp.\] be more likely to (a) recall all of the asparagus or (b) cover up the whole incident?”). This task elicited mental state reasoning indirectly by asking participants to formulate predictions about behavior, such that no mental state words were presented to participants at any point. Each question remained onscreen for 12 s, and participants were obliged to respond during that time by pressing one of two buttons on a button box held in the left hand. Each run comprised eight trials (four per condition) separated by 10 s. Each participant answered each question either for the individual or the group, but not both (question assignment randomized across participants). #### Theory-of-mind localizer In order to facilitate region-of-interest (ROI) analyses focusing on brain regions associated with theory-of-mind, participants also completed a functional localizer task in which they read short narratives and made inferences about individual protagonists' beliefs (e.g., concerning the location of a hidden object) and inferences about physical representations (e.g., the contents of an outdated photograph). Each narrative was displayed for 10 s and was followed by a statement that participants judged as true or false (e.g., Belief story: “Sarah thinks her shoes are under the dress”; Physical story: “The original photograph shows the apple on the ground”) which remained onscreen for 4 s. Participants were obliged to respond during that time by pressing one of two buttons. Trials were separated by 12 s fixation. Participants completed four runs, each of which comprised eight trials (four per condition), for a total of 32 trials. ### Imaging Procedure fMRI data were collected using a 3 Tesla Siemens scanner. Functional imaging used a gradient-echo echo-planar pulse sequence (TR  =  2 s; TE  =  30 ms; flip angle  =  90°, 30 near-axial slices, 4 mm thick, in-plane resolution  =  3×3 mm, whole brain coverage). These sequences used PACE online motion correction for movement \< 8 mm. fMRI data were preprocessed and analyzed using SPM2 (Wellcome Department of Cognitive Neurology, London, United Kingdom) and custom software. Data from each subject were motion corrected and normalized into a standard anatomical space based on the ICBM 152 brain template (Montreal Neurological Institute). Normalized data were then spatially smoothed (5 mm full-width-at- half-maximum \[FWHM\]) using a Gaussian kernel. Statistical analyses were performed using the general linear model in which the event-related design was modeled using a canonical hemodynamic response function and other covariates of no interest (a session mean and a linear trend). After these analyses were performed individually for each participant, the resulting contrast images for each participant (i.e., *individual \> control, group \> control*) were entered into a second-level analysis in which participants were treated as a random effect. Data were thresholded at *p*\<.001, *k*\>10, uncorrected. For the directed theory of mind task, conjunction analysis was performed following the procedure described by Cabeza, Dolcos, Graham, & Nyberg. Whole- brain statistical maps were created from the *individual* \> *control* and *group* \> *control* contrasts separately to identify voxels activated by each condition (thresholded individually at *p* \<.01), making for a conjoint threshold of *p* \<.001. ROIs were defined for each subject individually based on a whole-brain analysis of the theory-of-mind localizer in three regions: RTPJ, precuneus, and MPFC. Regions were defined as 10 or more contiguous voxels that were significantly more active (*p* \< 0.001, uncorrected) during stories about mental states than during control stories about physical representations. The average responses relative to rest during the individual and group conditions were then estimated in these regions. Within each ROI, the mean percent signal change (PSC  = 100 × raw BOLD magnitude for \[condition − rest\]/raw BOLD magnitude for rest) was calculated for each condition at each time point (averaging across all voxels in the ROI and all trials of the same condition) and averaged across seconds 6–10 to account for hemodynamic lag. Individual subject means for each condition of each task are available as. The full fMRI dataset is available upon request. ## Results ### Directed theory-of-mind task In order to assess the extent to which common cognitive processes subserve thinking about the minds of individuals and groups, we first conducted whole- brain, random effects analyses of BOLD signal. In whole-brain analyses, activation when participants contemplated the mental states of both individuals and groups (compared to control) was observed in brain regions associated with theory-of-mind, including MPFC, RTPJ, and precuneus. The direct comparisons between the individual and group conditions (individual \<\> group) yielded no areas of differential activation in regions typically associated with social cognition. To the extent that overlapping BOLD activation reflects the engagement of overlapping cognitive processes, these initial observations suggest that thinking about individuals and groups may draw upon shared theory- of-mind processes. Next, to test more directly the extent to which overlapping regions of cortex were recruited during contemplation of the mental states of individuals and groups, we conducted a conjunction analysis on the individual \> control and group \> control contrasts. This analysis revealed conjoint activation specifically in brain regions associated with theory-of-mind–MPFC, right and left TPJ, and precuneus–suggesting further that thinking about individuals and groups draw upon shared processes. Although the foregoing analyses suggest that similar processes subserve thinking about individuals and groups as compared to a control condition, they leave open the possibility that thinking about individual and group agents may recruit theory-of-mind processes to different degrees. In order to evaluate the degree to which processes associated with theory-of-mind were recruited when thinking about individuals versus groups, we conducted independent region-of-interest (ROI) analyses within the regions of MPFC, RTPJ, and precuneus identified by the independent theory-of-mind localizer. Because the mental states in the localizer task were attributed to individual protagonists, this analysis technique provides a particularly stringent test for whether thinking about group agents genuinely recruits processes associated with thinking about individuals. Consistent with previous research, the theory-of-mind localizer (belief \> photo contrast) yielded activation in MPFC (17/19 participants), RTPJ (19/19 participants), and precuneus (19/19 participants);. First, ROI analyses of the main task confirmed that each of these regions showed greater activation in the individual condition than in the control condition (MPFC, *t*(16) = 2.28, *p* \<.04, *d*  =  0.57; Right TPJ, *t*(18) = 2.43, *p* \<.03, *d*  =  0.57; precuneus, *t*(18) = 5.99, *p* \<.0001, *d*  =  1.41). Second, ROI analyses further revealed that each of these regions showed greater activation in the group condition as compared to control (MPFC, *t*(16) = 2.22, *p* \<.04, *d*  =  0.55; Right TPJ, *t*(18) = 2.39, *p* \<.03, *d*  =  0.56; precuneus, *t*(18) = 6.32, *p* \<.0001, *d*  =  1.49). Finally, no significant differences were observed between the responses to individuals versus groups in any of these regions, (MPFC, *t*(16) = 0.69, *p*  = .5; Right TPJ, *t*(18) = 0.09, *p*  = .93; precuneus, *t*(18) =  1.51, *p*  = .15;). Together, these analyses suggest that brain regions associated with theory-of-mind are recruited to a highly similar degree during the contemplation of individuals and groups. ### Spontaneous theory-of-mind task The design of the previous task raises the possibility that activation during the individual and group conditions may have differed from the control condition due to the explicit use of mental state words (e.g., thinks, believes, wants) in the individual and group conditions. To explore whether common theory-of-mind processes subserve attributions to individuals and groups even when no mental state terms are used, we analyzed data from the portion of the study during which participants made predictions about the behavior of individuals and groups. Specifically, we compared activation during the individual and group conditions of the prediction task in the same regions of RTPJ, MPFC, and precuneus identified by the theory-of-mind localizer. Results replicated those from the directed theory-of-mind task. Consistent with the hypothesis that thinking about the minds of individuals and groups recruit similar theory-of- mind processes, activations above baseline were observed across the network in both the individual, *t*(19)  =  2.84, *p* \<.02, *d*  =  0.65, and the group condition, *t*(19)  =  2.23, *p* \<.04, *d*  =  0.51 (averaging across regions), and no differences were observed between the individual and group conditions in RTPJ (*M*_ind  =  −.004 *M*_group  =  −.019, *t*(19)  =  0.86, *p* \>.39), MPFC (*M*_ind  = .197 *M*_group  = .180, *t*(19)  =  0.36, *p* \>.72), or precuneus (*M*_ind  = .266 *M*_group  = .231, *t*(19)  =  1.64, *p* \>.12). For individual subject data, see. These results suggest that the similar patterns of activation in the individual and group conditions observed in the first task are not simply due to the common use of mental state terms in those conditions. Here, when no mental state terms were presented, making predictions about individual and group agents' behavior also recruited the theory-of-mind network to an indistinguishable degree. ## Discussion In describing corporations, government agencies and other organizations, people sometimes use sentences of the form ‘Apple thinks…’ or ‘The CIA wants…’ The aim of the present investigation was to help illuminate how people think about group agents. The results of Experiment 1 indicate that sentences like these are ascribing something to the group agent itself. Perceivers used expressions like ‘believes’ and ‘wants,’ not merely to talk about some or all of the individual members of a group, but to talk about the group agent. Thus, attributions to the group sometimes diverged from attributions to the individual members: participants were willing to attribute a state to the group itself even when they were not willing to attribute that state to any of the individual members, and they were willing to attribute a mental state to all members of a group even when they were not willing to attribute that state to the group itself. In turn, the results of Experiment 2 reveal that that such ascriptions recruit brain regions associated with thinking about the minds of individuals, i.e., brain regions associated with theory-of-mind, both when theory-of-mind use is called for explicitly and when it arises spontaneously. Past research has demonstrated consistent engagement of a particular network of regions, including MPFC, RTPJ, and precuneus, during inferences about the minds of individual people, i.e., during theory-of-mind. Across two tasks, we observed activation in this network when participants read or made predictions about group agents. In the *directed theory-of-mind task*, participants read about the states of individuals, group agents, and inanimate objects. In the *spontaneous theory-of-mind* task, participants made predictions about what individual or group agents would do in particular situations. In both cases, activation associated with groups was indistinguishable from that associated with consideration of individuals. Whole-brain analyses, conjunction analysis, and ROI analyses all support the conclusion that cognitive processes associated with thinking about the minds of individuals were also recruited when participants thought about the ‘mind’ of a group agent. However, it is worth noting the possibility that participants may have been thinking to some degree about the minds of individual group members, and that this may have accounted for the observed activation in theory-of-mind regions during consideration of group agents. This possibility is weakened, but not completely ruled out, by (a) the fact that, unlike past studies, no individuals were mentioned or shown in the group condition and (b) the observation that perceivers interpret sentences about group mental states as ascribing mental states to the group agent itself in Experiment 1, and (c) the recent observation that the more perceivers think about the ‘mind’ of the group, the less they think about the minds of its members. Past research has documented the selectivity of the RTPJ for attributing representational mental content, such as beliefs and intentions, to others, compared to other sorts of attributions, such as those concerning a person's physical appearance, preferences, or personality traits. In this research, neither the mere presence of a person nor the need to make other types of inferences about that person was associated with as much activation in this region as attributing representational mental states. Accordingly, the fact that the RTPJ activated indistinguishably during consideration of individuals and groups (but distinguished both from the inanimate control condition) is an especially compelling suggestion that participants used similar processes for understanding the representational mental states of individuals and group agents. Although the specific contributions of MPFC to social cognition remain uncertain, this region has been observed to be sensitive to the *target* of mental state ascription. In particular, greater MPFC activation has been associated with interpersonally close others, and with humanized others, compared to those who are more distant or dehumanized. Accordingly, it would not have been surprising to observe reduced MPFC response to group agents compared to individuals. However, the current study observed indistinguishable engagement during consideration of group agents and individuals in a region of MPFC involved in attributing mental states to individuals, as identified by the theory-of-mind localizer, and similar to regions of MPFC associated with mentalizing or theory-of-mind in past studies (according to Neurosynth). Moreover, the individual condition and group condition were associated with greater MPFC activation than the inanimate control condition, suggesting that MPFC's contributions to individual-oriented social cognition are also present during social cognition concerning group agents. More generally, an abundance of past research has observed greater engagement of brain regions associated with theory-of-mind when perceivers think about certain types of target entities (humans and, to some degree, other animals) than when they think about other types of target entities (computers, food, furniture); for reviews, see. Here, we find just as much activation in brain regions associated with theory-of-mind when people think about group agents as when they think about individual humans, yet a group agent is something very different from a human being or animal, or even from a collection of human beings. Accordingly, the current results are consistent with the possibility that perceivers apply theory-of-mind generally to things that conform to a certain kind of abstract structure, and that group agents turn out to be among the things that conform to that structure. This possibility draws further support from recent research observing activation in brain regions associated with theory-of-mind during consideration of other non-human agents that display human-like properties – and is broadly consistent with the observation that brain regions engaged when people construct representations of others' mental states are also engaged when people construct other types of representations that are removed from their current, first-person experience, such as representations of the past or future. In sum, people appear in certain respects to treat groups as ‘entities’. They assign moral blame to whole organizations as a whole, treat whole financial markets as though they have minds of their own, and give corporations many of the legal rights enjoyed by individual human beings. In the current studies, we observed that perceivers were willing to attribute mental states to group agents that they did not attribute to the individual members of those groups, and that attributing mental states to group agents was associated with activation in the same brain regions that support ascriptions of mental states to individual people (as confirmed by an independent localizer task). Taken together, these results suggest that in order to understand the striking ways in which people reason about corporations, governments, and other group agents, it may be important to consider the possibility that perceivers sometimes attribute mental states such as beliefs, desires, and intentions not only to the members of such groups but also to the group agent itself. # Supporting Information The authors thank Rebecca Cox for assistance with data collection. Imaging data were collected at the Athinoula A. Martinos Center for Biomedical Imaging at MIT. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: ACJ RS JK. Performed the experiments: ACJ DDF. Analyzed the data: ACJ DDF JK. Contributed to the writing of the manuscript: ACJ DDF RS JK.

# 1. Introduction Schizophrenia, a severe psychiatric disorder characterized by positive and negative symptoms and cognitive deficits, has long been hypothesized as a disorder of brain connectivity. Disrupted brain development can lead to abnormal neural connectivity or network, which may cause abnormal information processing and integration, and clinical symptoms including psychosis\[–\]. It has been well established that the insula cortex is an anatomical gateway between the visual, olfactory, auditory, somatosensory cortices and the limbic structures. Secluded deep within the lateral sulcus of the human brain, the insular cortex is part of an extended network of neuronal pathways connecting to the anterior cingulate cortex, temporal lobe, parietal lobe, hippocampus, amygdala, olfactory cortex, and thalamus. The ventromedial prefrontal cortex, posterior cingulate cortex, bilateral inferior parietal cortex and middle temporal lobe are parts of default mode network (DMN). The central executive network (CEN) consists of mainly the dorsolateral prefrontal cortex and posterior parietal cortex. The salience network (SN) includes primarily the anterior insular cortex and dorsal anterior cingulate cortex. Apparentlythe insular cortex functions as a multimodal sensory integration region. In addition, the insular cortex plays an important role in emotion processing including interoceptive awareness, anticipation, evaluation of emotional stimuli, self-awareness, episodic memory, executive function, attention, and saliency processing. Recently, the insula cortex has attracted significant attention in schizophrenia research. The aberrant functional and structural alterations in insular cortex have been frequently reported in schizophrenia including reduced gray matter volume, thickness, and surface area, decreased white matter integrity (measured by fractional anisotropy or mean diffusivity), and altered functional activity under various tasks or during resting state. Changes in the SN may be one of the most important findings among all insular cortex related networks in schizophrenia. Recently, the anterior insular cortex within the SN has been demonstrated to be crucial to modulate DMN /CEN interactions in patients with schizophrenia. Manoliu et al. also found that the dependence of CEN and DMN interactions on SN’s right anterior insular activity is altered in patients with schizophrenia during acute psychosis or psychotic remission. Furthermore, insular cortical dysfunction might be associated with core symptoms of schizophrenia. Aberrant salience network activity in the insular and cingulate cortices has been implicated in the development of positive symptoms of schizophrenia such as delusions and hallucinations due to an inappropriate assignment of salience to stimuli that would normally be considered irrelevant. Patients with schizophrenia are usually suffer from impaired insight; both awareness and mental state attribution, two core components of insight, are associated with the function of insular cortex. Previous studies have provided compelling evidence supporting the critical role of insular cortex in schizophrenia. However, most studies examining the functional connectivity between insular cortex and other brain areas were in schizophrenia patients treated with antipsychotics. The potential confounding effect of antipsychotic treatment on the brain functional connectivity has been well established. The purpose of the present study was to examine the insular cortical functional connectivity and its relationship with the severity of clinical symptoms in drug naïve, first episode schizophrenic patients. # 2. Methods ## 2.1. Subjects The study was approved by the Ethics Committee of the First Affiliated Hospital of Zhengzhou University. All subjects provided written informed consent to participate in the study. Subjects were recruited from the consecutive admissions to the inpatient unit between November 2011 and December 2012. Inclusion criteria included: 1) diagnosis of schizophrenia according to the criteria of DSM-IV; 2) 18–45 years old; 3) in the age rangenever treated with antipsychotic medications or other psychotropics. The diagnosis of schizophrenia was confirmed by a research psychiatrist (X.S.) using the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-IV). Exclusion criteria were: 1) history of alcohol or other substance use; 2) history of brain injury; 3) any ongoing significant medical conditions. Healthy control subjects were recruited from the local community through advertisements. The same research psychiatrist (X.S.) conducted a comprehensive clinical interview to rule out any psychiatric conditions in healthy controls. A complete medical history, physical examination, and routine laboratory tests were obtained from all subjects to rule out possible medical conditions. All subjects were right-handed. ## 2.2. Clinical symptom measurement Symptoms of schizophrenia were assessed for all patients using the Positive and Negative Syndrome Scale (PANSS), which includes 3 subscales: positive symptoms, negative symptoms and general psychopathology. The PANSS was administered by the same rater (J.G.) throughout the study. ## 2.3. Image acquisition On the same day after clinical assessment and before taking any antipsychotic or other psychotropic medications, all participants were scanned on a 3.0 Tesla Scanner (Signa HDxt 3T GEHCGEHC) at the Magnetic Resonance Center of the First Affiliated Hospital of Zhengzhou University. A 6 minute ‘resting-state’ functional MRI scan was obtained, comprising of 180 time points of whole-brain functional (EPI) volumes (TR = 2000 ms; TE = 30ms; flip angle = 90; 34 contiguous AC–PC aligned axial slices; matrix = 64×64; FOV = 22 cm; acquisition voxel size = 3.4mm×3.4mm×4 mm). During this scan that acquired all neuroimaging data, participants were instructed to rest without moving. The wakefulness of the subject was verified by self-report at the end of the scan. T1-weighted Spoiled Gradient Echo (SPGR) images were also collected for the purposes of anatomical localization. The acquisition protocol included the following pulse sequence and parameters: repetition time (TR) = 12ms, echo time (TE) = 4.5ms, inversion time (TI) = 1100ms, flip angle 7°, field of view (FOV) = 256×256×220 mm³, matrix size 256×256, slice thickness 1mm contiguous, and scan time 17 min. ## 2.4. Image processing Imaging data were coded and catalogued before being transferred to the Psychotic Disorders Program of the University of Massachusetts Medical School (UMMS) for blinded analysis. AFNI was used for image preprocessing (<http://www.afni.nimh.gov/afni>), performing slice timing correction for interleaved acquisition (using Fourier interpolation), motion correction (by aligning each volume to a “base” image \[middle volume\] using Fourier interpolation) and de-spiking (detection and reduction of extreme time series outliers using a hyperbolic tangent function). All other data processing was carried out using FMRIB Software Library (FSL) (<http://www.fmrib.ox.ac.uk>), including spatial smoothing (FWHM = 6 mm), mean-based intensity normalization of all volumes, temporal bandpass filtering (highpass temporal filtering: Gaussian- weighted least-squares straight line fitting, with sigma = 100.0 s; Gaussian lowpass temporal filtering HWHM 2.8 s), and pre-whitening. Each individual’s time series were spatially normalized by registration to the MNI152 (Montreal Neurological Institute) template with 2mm³ resolution, using a 12 degree of freedom affine transformation. Nine nuisance covariates (time series for global signal intensity, white matter, cerebrospinal fluid, and six motion parameters) were regressed out of the data to minimize the contributions of artifactual physiological signals (e.g., cardiac and respiratory cycles) using the general linear model implemented in FSL program FEAT. The FreeSurfer software (<http://surfer.nmr.mgh.harvard.edu/>) was used to segment T1-weighted SPGR images into cortical and subcortical gray and white matter regions, as well as total intracranial volume, for each subject. Functional connectivity was examined using a seed-based approach. The left and right insular seed regions were generated using the Harvard-Oxford atlas, co-localized to the MNI space used in this study. For each participant, we calculated the mean time series of each seed by averaging across all voxels within the seed. A voxel-wise correlation map was generated to each of these seed-based time course reference signals. This correlation map was r-to-Z transformed in order to generate a standardized representation. Each of the remaining cortical and subcortical gray matter regions of the Harvard-Oxford atlas were then used as a mask to determine the average resting state correlation (Z-score) between the insular cortex and each ipsilateral gray matter region in each subject. ## 2.5. Statistical analysis The data were analyzed using SPSS 20.0 (SPSS Inc., Chicago, IL). Demographic and clinical characteristics were reported using descriptive statistics. The Shapiro-Wilk test was used to check the normality of the data. Group comparisons were performed using the independent samples t-test for continuous variables and Chi-square or Fisher’s exact test for categorical variables. Connectivity values (Z-scores) of the insular cortex with ipsilateral cortical ROIs were extracted for group comparison, using the independent samples t-test. Bonferroni corrections considering the number of ROIs were used to define the significance level: *p*\<0.05/48 = 0.001. Insula connectivity maps (Z-score maps) were subjected to voxel-based morphometry (VBM) analysis. To further explore the regions that show group difference in insular connectivity. A false discovery rate (FDR) corrected *p*\<0.05 was used to define the overall significant level. Peasson correlation analysis was used to examine the relationship beteween insular connectivity and clinical symptoms as measured by the PANSS; the significance level was set at *p*\<0.05. # 3. Results The original study sample included 46 patients with schizophrenia (SZ) and 30 healthy control (HC) subjects. One SZ and 2 HC individuals were excluded due to the poor quality of their T1 scans; one SZ and 1 HC individuals were excluded due to the poor quality (missing time series) of the resting-state scans; Seven SZ and 2 HC individuals were excluded because the imaging values were defined as outliers using the criteria of Mean±2SD. The final study sample used for data analysis was thus composed of 37 SZ subjects and 25 HC subjects. shows demographic and clinical characteristics of the study sample. There were no significant differences between the patient group and the HC group in age, gender, education, and intracranial volume (ICV) (*p*’s \> 0.05). ## 3.1. ROI-based analysis shows the complete results of the seed-based analysis for each hemisphere. In the left hemisphere, no regions met the p \<.001 criterion (with Bonferroni corrections) for between-group differences. However, seven regions including caudate, Heschl’s gyrus, posterior cingulate, hippocampus, anterior and posterior parahippocampal gyrus and central operculum showed a trend-level difference (p’s \<.05). In the right hemisphere, three regions met the significance criterion (Heschl’s gyrus, anterior cingulate and caudate), and five regions showed a trend-level difference (central operculum, putamen, planum polare, planum temporale and thalamus). shows the group differences in average connectivity for the regions of right insular cortical connectivity that had significant group difference. The regions include the right insular cortical connectivity to the right anterior cingulate cortex (ACC), right Heschl’s gyrus, and right caudate nucleus. Each of these regions showed a significant decrease in the SZ group compared with the HC group. ## 3.2. Confirmatory voxel-based analysis shows the results of the VBM analysis for the right hemisphere insula-seeded correlation maps. A FDR—corrected significant cluster was observed in Heschl’s gyrus. An additional significant cluster was observed in the superior temporal gyrus. FDR trend-level clusters were also observed in the anterior cingulate gyrus and caudate nucleus. Thus, of the three ROI-based regions in the right hemisphere with significant betwee-group difference, only the Heschl’s gyrus difference was confirmed in the VBM analysis. ## 3.3. Correlation with clinical symptoms The relationship between the resting state average Z-scores of insula-Heschl’s gyrus connectivity and the measures of clinical symptoms were examined in both hemispheres. As shown in, there was a significant negative correlation between the left insula-Heschl’s gyrus connectivity and the PANSS negative symptoms subscale scores (r = -0.406, p = 0.019), and a significant positive correlation between the right insula—Heschl’s gyrus and the PANSS general psychopathology subscale scores (r = 0.384, p = 0.019). # 4. Discussion Minimizing the potential confounding effect of antipsychotics, the present study provides new evidence regarding changes in insula-Heschl gyrus connectivity in drug naïve, first episode schizophrenia patients. In this study, we found that there was significant reduction in the insular cortical connectivity with Heschl’s gyrus, anterior cingulate cortex (ACC), and caudate nucleus in the right hemisphere. We also found that the insula-Heschis connectivity might be associated withclinical symptoms of schizophrenia. Reduced insular cortical connectivity with Heschl’s gyrus in the right brain hemisphere was found in the seed-based analysis and was confirmed in the VBM analysis. Similar to insular cortex, Heschl's gyrus (transverse temporal gyrus) may also play an important role in the neural mechanism of schizophrenia. It has been reported that first-episode schizophrenia patients have bilateral Heschl’s gyrus gray matter volume reduction, and decreased left Heschl’s gyrus gray matter volume over time. Heschl's gyrus contains the primary auditory cortex (PAC), which is the first cortical structure to process incoming auditory information. The PAC is in close spatial proximity to the posteroinferior insular cortex. The auditory information related to prosodic processing may be conveyed to the posterior insular cortex directly through links with the primary auditory and association areas. These findings suggest that deficits in the right insula-Heschl’s functional connectivity may be engaged in the prosody dysfunction in patients with schizophrenia. Reduced right insula-caudate connectivity is another finding in the present study based on the seed-based analysis. Caudate nucleus is related to successful goal-directed action. Studies have found that schizophrenia and their relatives have a reduced caudate volume and compromised white matter integrity. Using functional magnetic resonance imaging to study brain activation during a Monetary Incentive Delay task, Mucci et al. found that avolition in patients with schizophrenia might be related to dorsal caudate hypo-activation. Another research team has reported an abnormal hemispheric specialization of caudate nucleus connectivity in schizophrenia. Reduced insula-caudate connectivity found in our study may play an important role in the development of schizophrenia. We also found significantly decreased right insula connectivity with ACC using the ROI-based analysis. The insula and ACC together constitute the SN, an intrinsic large-scale network showing strong functional connectivity. Menon and Uddin have recently proposed that the primary role of this network is to segregate the most relevant stimuli among internal and extra-personal stimuli, to assist target brain regions and generate appropriate behavioral responses to salient stimuli, and to enable switching between the default mode and task- related states of brain connectivity. Abnormal connectivity of SN has been reported to be associated with the negative symptoms, hallucinations, delusions and other psychotic symptoms of schizophrenia\[–\]. Previous studies have suggested that functional and/or structural alterations within the insular cortex might contribute to aberrant salience processing, leading to the development of schizophrenia symptoms. In the present study, our correlation analysis within the patient group showed a significant negative relationship between left insula-Heschl’s gyrus connectivity and the PANSS negative symptoms subscale scores, suggesting the impairment in this particular functional connectivity might be related to the development of negative symptoms of schizophrenia. In addition, the correlation analysis found a significant positive relationship between the right insula- Heschl gyrus connectivity and the PANSS general psychopathology subscale scores, suggesting that an abnormally increased connectivity might contribute to the manifestation of clinical symptoms such as depression and anxiety observed in patients with schizophrenia. Symptoms of schizophrenia have been attributed to a failure of functional integration or aberrant connectivity among regions or systems of the brain. Manoliu et al. found that the right anterior insular cortical dysfunction was associated with positive symptoms of schizophrenia during the acute phase of psychosis; they further suggested that the specific SN/DMN/CEN reorganization with distinct insular cortical pathways might be related to different symptom domains of schizophrenia. Another study reported that structural alterations of the insular cortex might be related to negative symptoms during psychotic remission, which is consist with our report about the negative correlation between the insular cortex-Heschl’s gyrus connectivity in the left brain hemisphere and the PANSS negative symptoms. Lee et al. reported that progressive gray matter volume reduction in both insular cortex and temporal pole in first episode psychosis was inversely associated with changes in the overall Brief Psychiatric Rating Scale symptom scores. Some studies suggested that disturbances in emotional prosody is due to the impairment in early auditory sensory processing, which may contribute to later impairment in attention-dependent processes in schizophrenia. It is also reported that the cortical surface area and local white matter volume of posterior insula might play an important role in insight impairment in schizophrenia. Both attention deficit and insight impairment might be related to the general psychopathology symptoms (depression, anxiety etc.) observed in patients with schizophrenia. The present study has several limitations: 1) Some meaningful information may have been lost by removing global signal in preprocessing of the imaging data. 2) The patient subjects were recruited from an inpatient unit; the overall symptom severity of the study sample might be higher than patients in the outpatient setting. Therefore the findings from this study, especially the correlation between functional connectivity and clinical symptoms, may not be generalizable to all patients. 3) The correlation analysis between functional connectivity and clinical symptoms was not corrected for multiple correlations; 4) Several software tools were used for pre-processing and processing of imaging data; 5) The PANSS was the only rating scale used to assess clinical symptoms. In summary, our study demonstrates the abnormal functional connectivity of insula, insula-Heschl gyrus connectivity in particular, in drug naïve, first episode schizophrenia patients. These abnormalities might contribute to the development of schizophrenia. Future studies with a larger sample size and in combination with other techniques such as diffusion tensor imaging (DTI) are needed to further assess the regions of insular connectivity in relation to the development and treatment of schizophrenia. # Supporting Information Funding for this study was provided by the National Natural Science Foundation of China (No.81571318 to X-QS; No.81371472 to L-XL; No.81401110 to XL), Science and Technology Planning Project of Health and Family Planning Commission (No.201501015 to X-QS), International Science and Technology Cooperation Program of Henan (No. 162102410061 to X-QS), Science and Technology Planning Project of Guangdong Province (No. 2013B021800085 to Q-LW), the Fundamental Research Funds for the Central University (No. 14ykpy28 to Q-LW) and the Youth Fund of the First Affiliated Hospital of Zhengzhou University (to XL, L-JP). [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** XF XS. **Data curation:** DK MQ XL. **Formal analysis:** XF DK. **Funding acquisition:** LL XS. **Investigation:** LP QW JG YY. **Methodology:** DK MQ HL. **Project administration:** LL XS. **Resources:** XL XS. **Supervision:** LL XS. **Validation:** DK XF. **Visualization:** LP DK QW. **Writing – original draft:** LP MQ. **Writing – review & editing:** LL XL XF XS. [^3]: ‡ These authors share co-first authorship on this work.

# Introduction Metabolic profiling is essential to ensure the quality, consistency, safety and efficacy of herbal medicine products; especially for the injection dosage form. Water extraction or decoction is the most favored method of preparation of herbal medicine injections (HMIs). Saccharides, amino acids, organic acids, and other primary metabolites are unavoidably extracted along with targeted secondary metabolites during the process of HMIs, such as Qingkailing injection, Danshen injection, Guanxinning injection, and Shuxuetong injection. In general, amino acids are considered to provide tonic activities and act as key regulators of nutrient metabolism, and polysaccharides are believed to be one of the most important constitutes in some herbal materials for pharmacological activities. Some monosaccharides have the suppressive effect on cell-mediated immune reactions. However, these primary metabolites in HMIs are often ignored to detect and set corresponding quality criteria in China Pharmacopeia and national standard. Due to the strong polarity and hydrophilicity, conventional reversed- phase performance liquid chromatography (HPLC) fingerprint, and other analytical methodologies, including anion exchange chromatography, gas chromatography, and capillary electrophoresis, are limited to separate and detect primary metabolites unless with complicated pretreatment, derivatization reagents and laborious preparation procedures. To delineate various class metabolites in HMIs, different types of fingerprints are necessary for a holistic quality evaluation, which are difficult to realize during the practical industry processing. Therefore, a simple and fast approach is required to be capable of detecting saccharides, amino acids, and organic acids, together with mainly bioactive secondary metabolites in one fingerprinting spectrum simultaneously. Proton nuclear magnetic resonance (¹HNMR) spectroscopy provides access to detect all proton-bearing compounds, almost irrespective of the chemical compound class. Because of the signal intensity directly proportional to the number of nucleus contributing to a specific resonance, ¹H NMR method achieved the identification and quantification of metabolites in a one- step acquisition. With simple sample preparation, ¹H NMR facilitates high-throughput analysis for metabolic studies and quality control of various food – and herbal materials. As a universal technique with the simplicity and rapidity of implementation, ¹H NMR has expansive prospect of application in profiling of polar metabolites of HMIs. Since the complexity of chemical composition, HMIs cannot be completely represented by a limited number of certain bioactive compounds. To extract the feature variables, principal component analysis (PCA) and independent component analysis (ICA) are classical tools to reduce the dimension of multivariate. The components (PCs) in PCA method are mutually orthogonal, while ICA method contains the components to be statistically independent. ICA has been found to be a successful alternative to PCA in eliminating the overlapping information between the components. However, ICA faces some limitations due to some instability, the choice of number of components to extract and high dimensionality. As a consequence, independent principal component analysis (IPCA) was proposed by Yao et al. in 2012 to use PCA as a pre-processing step to reduce the dimension of the data, and then use ICA as a denoising process of PCA to separate relevant information. On simulation studies and real data sets, IPCA offered a better visualization of the data than ICA and with a smaller number of components than PCA. Owing to the benefit to generate denoised the loading vectors, we attempted to employ IPCA method to construct χ² and Hotelling T² control charts for multivariate statistical analysis. Danhong injection (DHI) is a patent injection made from the extracts of *Radix Salviae Miltiorrhizae* and *Flos Carthami*. It has been widely used for the prevention and treatment of cardiovascular and cerebrovascular diseases in clinic. In our previous work, ultra-performance liquid chromatography (UPLC) coupled with UV detection was adopted to identify 11polyphenolic acids in DHI. However, the total weight of identified constituents accounted for only a low proportion (about 10%) of the solid content in DHI. In this study, we describe a strategy to detect more hydrophilic primary metabolites in DHI based on quantitative ¹H NMR spectroscopy. The absolute concentration of identified metabolites was calculated by using internal standard method. Linearity, precision, repeatability, stability and accuracy were carried out to validate the method. The contents of polar metabolites were further evaluated by the multivariate analysis tool IPCA to establish χ² and Hotelling T² control charts. # Materials and Methods ## Materials and Chemicals Thirty-six batches of DHI manufactured in 2011, 2012 and 2013 were provided by Heze Buchang Pharmaceutical Co. Ltd (Heze, China). The standards of valine, threonine, alanine, pyroglutamate, procatechuic aldehyde and asparagine were purchased from the National Institute for Food and Drug Control (Beijing, China). Salvianic acid and procatechuic acid were obtained from Zhongxin Innova Laboratories (Tianjin, RP China). Succinate and malonate were obtained from Dr. Ehrenstorfer GmbH (Augsburg, Germany). Fructose and glucose were purchased from Sigma (Aldrich, America). Rutinose was purchased from Hazard Communication (Tokyo, Japan).The purities of the compounds were all above 98%, using NMR analysis. Deuterium oxide (D₂O, 99.9%) and sodium 3-trimethylsilyl \[2,2,3,3-*d*₄\] propionate (TSP) were purchased from Cambridge Isotope Laboratories (Miami, FL, USA). D₂O was used as internal lock; TSP was as internal standard for chemical shift calibration and quantification. ## Ethics No specific permission was required for the described field studies. The field locations are neither privately owned nor protected, and neither endangered nor protected species were involved. ## Sample preparation All the injection samples were subjected to freeze-drying. The dried powders (18 mg) were accurately weighed and dissolved with 600 µL of D₂O containing 0.58 mM TSP. Exactly 500 µL of sample solution was transferred into a standard 5 mm NMR tube (Vineland, NJ, USA). No buffer was used due to the stable pH of injection. ## NMR measurements ¹H NMR spectra were acquired at 298 K on a Bruker AV III 600 MHz NMR spectrometer (600.23 MHz for proton frequency) with a 5 mm broadband BBFO probehead. All pulse sequences were from Bruker pulse program library. A standard one dimensional composite pulse sequence (zgcppr) was employed to suppress the residual water signal. The 90° pulse width was adjusted to about 13 µm for each sample. Sixty-four scans were collected into 32k data points using a spectral width of 12335 Hz, a relaxation delay of 1.0 s and an acquisition time of 2.66 s. A 0.3 Hz line-broadening function was applied to all spectra for Fourier transformation (FT) followed by phasing and baseline correction. For proton signals assignment purposes, a set of two dimensional (2D) spectra, including ¹H-¹H correlation spectroscopy (COSY), ¹H-¹H total correlation spectroscopy (TOCSY), ¹H *J*-resolved (*J*-res), and ¹H-¹³C heteronuclear single quantum coherence (HSQC), were acquired for selected samples and processed with similar parameters as described previously. Spin-lattice relaxation time (*T*₁) values of the quantified protons of individual constituent and TSP were measured using a classical inversion recovery pulse sequence with 10 relaxation delays (*τ*) ranging from 0.01 to 20 s. ## Quantification of the metabolites Because the intensity of a given ¹H NMR signal is directly proportional to its contributing number of protons, the amount of metabolites in DHI can be measured by the signal areas of given metabolites and an internal reference with known concentration. Thirteen metabolites in DHI were selected for quantification. The important parameters for data acquisition and processing of ¹H NMR spectra must be set appropriately to obtain accurate and precise measurements. Most of all, the relaxation delay *τ* should be long enough to ensure complete relaxation for all the signals of interest.¹H NMR measurements are done in a longer acquisition time by choosing *τ*≥5× longest *T*₁. For shortening the acquisition time, ¹H NMR spectra can be acquired in an incompletely relaxed condition, and the absolute concentrations should be calculated taking the *T*₁ values in consideration. With the effective magnetization reading pulse of 90°, the quantification of chemical constituents in this study can be performed by using the following equation: where *P_X* and *P_TSP* are the mass concentrations of metabolite and TSP, *A_X* and *A_TSP* are the integral areas for targeted signal of metabolite and for methyl groups of TSP, *N_X* and *N_TSP* are the proton numbers of metabolite and of methyl groups of TSP, *M_X* and *M_TSP* are the molar masses of metabolite and TSP, and are the spin- lattice relaxation times for proton X and methyl protons of TSP, respectively; *t* is total relaxation time (relaxation delay plus acquisition time). The quantitative ¹H NMR method was checked for linearity, precision, repeatability, stability, and accuracy. Precision, repeatability and stability were calculated as the relative standard deviation (RSD). Recovery test was employed to determine the accuracy, and four typical compounds, alanine, glucose, salvianic acid and procatechuic aldehyde, were chosen to evaluate the average recovery. ## Assay for multivariate quality control 36 batches of DHI samples were split into two phases, Phase I and Phase II. Phase I as training set included 25 batches of qualified products manufactured continuously in the years of 2012 and 2013. Phase II as testing set included 6 batches of qualified products in 2012 and 2013, and 5 batches of expired products manufactured in 2011. The quantitative resulting data of 13 metabolites was imported to R 3.0.2 software loaded with packages of MVA, MSQC and mixOmics ([www.r-project.org](http://www.r-project.org)) for multivariate statistical analysis. For simplifying the multivariate problem, principal component analysis (PCA) and independence principal component analysis (IPCA) were performed to reduce the dimensionality of data. The scores of principal components characterized the whole data were then imported to χ² and hotelling T² control charts to calculate the upper control limits (with 99% confidence ellipsoids). Moreover, as one of the requisites in control chart is the independence of the data, the independence of selected components in PCA and IPCA models were validated by autocorrelation function (ACF). The out-of-control samples in Phase II were examined by the upper control limits achieved from Phase I. # Result and Discussion ## Proton signal assignments and chemical identification A representative ¹H NMR spectrum of DHI was shown in. The resonance signals were assigned to 30 metabolites based on the elucidation with extensive 2D NMR experiments (¹H-¹H COSY, ¹H-¹H TOCSY, ¹H *J*-resolved, HSQC),the literature data in our former work , and in-house database. In the range of *δ* 3.2–5.8, the spectrum is dominated by 5monosaccharides and 2 disaccharides, including glucose, galactose, arabinose, fructose, rhamnose, rutinose, and rutinulose. In the high-field region (*δ* 0.5–3.2), 8 amino acids (isoleucine, leucine, valine, threonine, alanine, proline, pyroglutamate, asparagine) and 3 organic acids (acetate, succinate, malonate) were observed. In the low-field region (*δ* 5.8–10.0), 7 polyphenolic acids, including salvianic acid, salvianolic acid B, salvianolic acid A, rosmarinic acid, lithospermic acid, procatechuic acid and procatechuic aldehyde, together withuridine and 5-(hydroxymethyl)-2-furaldehyde (5-HMF), were identified. Moreover, 3 organic acids (4-hydroxybenzoic acid, 4-hydroxycinnamic acid, and formate) were observed as well in the low-field region. The chemical shifts of the identified 30 metabolites by¹H NMR were listed in.To our knowledge, 7 saccharides and 6 organic acids were reported for the first time in DHI. Without the need of any sample pretreatment or pre-column derivatization, the established ¹H NMR method provided an approach to determine 7 saccharides, 6 organic acids, 8 amino acids, 1nucleoside, 1 carbohydrate derivatives (5-HMF) and 7 polyphenolic acids in DHI simultaneously. ## Quantitative ¹H NMR analysis and method validation Due to the narrow chemical shift range of ¹H NMR and frequent signal overlap, it is a challenge to quantify all the constituents in a mixture. In our study, 13 metabolites with fully separated signals were selected for quantification. In order to improve the efficiency of ¹H NMR method, the spectra were acquired in an incompletely relaxed state. As a consequence, the spin-lattice relaxation time (*T*₁) value must be accurately measured and taken into account for quantitative analysis. The *T*₁ values were determined by the inversion-recovery experiments. Accordingly, the absolute concentrations of the 13 metabolites were calculated from three parallel samples of each batch (Table S1). ### Linearity ¹H NMR as method itself is linear and no calibration is necessary for the determination of molar ratios of mixtures. Thus, the 13 metabolites in five different molar ratios confirmed the linearity of NMR spectroscopy. Good linearity was achievable, as indicated by the equations and satisfactory correlation coefficients (*r*²). ### Precision The intraday and interday precision was determined by analyzing six replicates on the same day and on three consecutive days respectively. The intraday precision for the contents of 13 metabolites ranged from 0.20% to 0.89%, and the interday precision ranged from 0.29% to 1.49%. The RSD values were adequate and indicated the suitability of the method. ### Repeatability Six samples prepared from the same batch showed RSD values ranging from 1.46% to 2.75%, indicating a high repeatability. ### Stability One sample was analyzed to determine stability on three consecutive days. The RSD values of the analytes were in the range of 0.37% to 2.21%. ### Accuracy Considering the limited volume of NMR tube and the cost of using deuterium reagents to dilute sample continuously, four metabolites of different types were employed, including alanine, glucose, salvianic acid and procatechuic aldehyde. The recovery was calculated as the ratio of the response of the selected four compounds in the spiked DHI samples against that of the standards at the same levels. The average recoveries were found to be 106.6% (±1.2), 106.6% (±2.5), 98.3% (±2.1) and 91.3% (±4.0) for alanine, glucose, salvianic acid and procatechuic aldehyde, respectively, indicating acceptable recovery. According to the results, the concentration of glucose was extremely high in DHI as shown in, and the amount of saccharides, amino acids and organic acids represented about 60% of the total solid content of DHI (Table S1). ## Control Charts based on PCA and IPCA With the enhancement in quality control of DHI, the analysis should be performed through a multivariate approach, that is, the above 13 metabolites must be analyzed together, not independently. The concentrations of 13 metabolites were mean centered and unit-variance scaled before being analyzed by PCA and IPCA. To avoid the loss of significant information, the percent specified of the principal components (PCs) cumulative proportion of explain variance is normally fixed on 80% in PCA model. Thus, the first three PCs in Phase I were selected to construct χ² and Hotelling T² control charts. Since the independence of the data is one of the requisites in control chart, we assessed the marginal independence of each necessary PC to indicate the model validation. Correlograms showed that PC1 fell outside of the confidence bands, which indicated that there was an evidence of autocorrelation or dependence of PC1, and PCA model using Phase I data was not achieved. In order to remove the autocorrelation effects of PCA, we employed IPCA to generate denoised and independent loading vectors. The kurtosis measure of loading vectors was used to decide the number of independent principal components (IPCs). The kurtosis of all extracted IPCs was plotted in, whereas the kurtosis of IPC7 was close to zero. By using the first 7 components of IPCA, the exactly choosing number of IPCs was obtained. Since the kurtosis of IPC3 was close to zero, the first 2 components were sufficient with IPCA. The presence of autocorrelation was assessed as shown in (IPC1 and IPC2), which indicated that there was no evidence of relation between the adjacent observations. Therefore the original 13 dimension of our data had been reduced to a two-dimensional problem. Then the first two IPCs were taken to establish the in-control state (Phase I). According to χ² and Hotelling T² control charts with the 99% confidence region, the upper control limits (UCL) were determined as 9.21 and 11.56, respectively. The first two IPCs were consequently controlled through 2D ellipsoids.The χ² control ellipse (UCL = 9.21) could be used as process region, and the T² control ellipse with less restrictive (UCL = 11.56) was used as tolerance region. In both cases, all the points in Phase I fell inside the confidence ellipsoids. The samples in Phase II (Table S2) were monitored by employing the UCLs of both χ² and T² charts obtained from Phase I, and the points of Phase II were added into 2D ellipsoids. The seventh sample fell outside the 99th confidence ellipsoids of both the process and tolerance regions, indicating the presence of out-of- control sample (batch 110402). The decomposition of T² value showed that the out-of-control variability was associated to the IPC1 since *p*-value was equal to 0.0032. The same result was also obtained through IPCA loading plots as shown in. Although the loading plot could not exactly determine which metabolites were responsible for the variation, it still showed the contents of succinate, malonate, glucose, fructose, salvianic acid and protocatechuic aldehyde made more contributions to the independent loading vectors of IPC1. # Conclusion Based on quantitative ¹H NMR analysis, a reliable approach for simultaneous determination of amino acids, organic acids, saccharides, and botanic secondary metabolites of HMIs in one fingerprinting spectrum has been developed and validated by using Danhong injection as a model. The method had taken*T*₁ values into account when calculated the contents of feature metabolites, which allowed the assay with good linearity, precision, repeatability, stability and accuracy. Unlike HPLC fingerprinting methods, the ¹H NMR approach has the significant advantages of less analysis time (about 5 min) without chromatographic separation, and no requirement of standard materials used to quantitative analysis. In combination with IPCA, two kinds of multivariate control charts (χ² and Hotelling T²) were also successfully carried out for detecting off-test HMI samples by employing the independence principle components. The decomposition of T² value and IPCA loading vectors can reflect the significance variations of overall metabolite profiling, although it cannot decide the mutative metabolites exactly. The established multivariate models have the prospects for extracting sufficient characterization to monitor more feature metabolites in HMIs. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MJ YW BZ LJ YZ XG. Performed the experiments: MJ YJ LX. Analyzed the data: MJ YJ YW. Contributed reagents/materials/analysis tools: BZ LJ YZ XG HP. Wrote the paper: MJ YJ YW YZ XG MW.

# Introduction The spatial distribution of a species is one of the fundamental aspects of its ecology and is a component of many conservation and management plans. For marine fishes, understanding a species’ spatial distribution is necessary for delineating Essential Fish Habitat (EFH), which is important for an ecosystem- based approach to fisheries management, and is part of the Magnussen-Stevens Act (as amended in 1996). In the United States, the National Marine Fisheries Service (NMFS) and regional fisheries councils are required to identify EFH \[, \]. However, broadly defined as “those waters and substrate necessary to fish for spawning, breeding, feeding and/or growth to maturity,” EFH quickly comes to include most waters within the exclusive economic zone (<https://www.fisheries.noaa.gov/resource/map/essential-fish-habitat-mapper>) when EFH is summed across species, making the prioritization of conservation and management actions difficult. While a species may occupy many habitats or areas, some areas are likely to be more ecologically important than others to the long-term persistence and productivity of the species. For example, protecting core adult habitat for adult rockfishes *Sebastes* spp. has been shown to lead to increases in reproductive output for these populations. Alternatively, restoration of juvenile habitats for red drum *Sciaenops ocellatus* has been shown produce larger increases in population growth rate and adult abundance than would protection and restoration of adult habitat. These ‘essential’ EFHs are termed Habitat Areas of Particular Concern (HAPC) due to their increased importance for the species’ population dynamics. Locating areas that support high densities of juvenile fishes is a first step towards understanding EFH and HAPC and being able to apply spatial management plans. Nursery areas are juvenile habitats that contribute disproportionately more individuals to the adult population than average (usually defined in terms of production per unit area) due to higher juvenile: (1) density, (2) growth rate, (3) survival, and (4) movement to adult habitat. Here, we focus on the first criteria: density of juvenile fishes. Strong recruitment events can determine age structure and stock size in marine fishes, suggesting that the location, quality, and availability of juvenile habitat are essential to the health of these populations. For example, recruitment to five populations of sole *Solea solea* in the northeast Atlantic is related to the area of suitable juvenile habitat, and the abundance of age-0 cod *Gadus morhua* is correlated with the availability of eelgrass habitat. While there has been a substantial focus on nearshore and estuarine habitats, less work has gone into quantifying the spatial distribution of juvenile habitat for deeper marine species across large spatial scales. Starting with an areal delineation of a species juvenile habitat is useful because bottom type alone does not necessarily define high quality areas. Other factors, such as larval supply, food availability, and connectivity with adult habitats can be important. For example, there are substantial seagrass beds that appear to be suitable habitat for juvenile queen conch (*Stombas gigas*) in the Bahamas. However, settlement occurs to specific areas where tidal circulation concentrates larvae, and conch growth is highest in areas where macroalgal production exceeds surrounding areas. In other cases, nursery habitat can be ephemeral and temporally variable when species recruit to biogenic habitats or respond to changes in oceanographic or climate drivers like temperature. In fact, shifts in juvenile distributions related to climate or other factors may hamper management efforts. Quantifying the spatial distribution and abundance of juvenile fishes allows us to identify juvenile habitats and potential nursery areas as well as to begin developing hypotheses concerning the factors controlling the spatial distribution of juvenile fishes (e.g., current regimes that may entrain larvae). Here, we used data from the Northwest Fisheries Science Center’s US West Coast Groundfish Bottom Trawl Survey \[WCGBTS, \] to quantify the spatial distribution of juvenile fishes for 13 species (five flatfishes and eight roundfishes) in waters off the west coast of the United States. We used vector-autoregressive spatio-temporal modeling \[VAST, \] to estimate species distribution models and quantify spatial and temporal patterns in juvenile density and identify juvenile habitat (in terms of spatial distributions) for each species. We ask: (1) are there juvenile hotspots along the coast for each of the species or are juveniles distributed homogeneously? (2) Is there spatial fidelity over time in juvenile hotspots for each of the species? And (3) are spatial patterns of juvenile biomass stable through time? Finally, (4) we also estimate an annual, coast-wide index of juvenile biomass for each species, the species effective area occupied, and its center of gravity (CoG). # Materials and methods ## Data source We used data for 2003–2018 from the Northwest Fisheries Science Center’s (NWFSC) U.S. West Coast Bottom Trawl Survey of Groundfish Resources off Washington, Oregon, and California \[WCGBTS, \]. The survey is takes depth-stratified, random samples that span approximately 32–48.58°N and 55–1280 m and is conducted in two passes from May to October with over 600 trawls per year \[see, for a detailed description of the sampling design\]. Survey vessels start at Newport, Oregon and initially head north to Cape Flattery before heading south to San Diego. The survey uses a standard Aberdeen-net with 25.9-m headrope, 31.7-m footrope, and an additional 3.8-cm liner extending from the middle of the net through the codend, to retain smaller fish and invertebrates. The net was towed at \~2.2 knots for a nominal 15 minutes \[an average of 20 minutes on bottom time including lift-off lag, \] and swept area ranged from 0.8 to 4.5 ha (median: 1.7). We included only those hauls deemed acceptable for stock assessment. ## Data selection and processing Data selection and processing involved two issues. First, ideally we would have analyzed only age-0 or the youngest observed age class fish to provide estimates of recruitment distributions and abundance. However, the survey trawls often do not catch small juvenile fish in high numbers, resulting in a large number of tows with zero catch for the smallest age classes. For some species, we had to include multiple age classes in the analysis (sum biomass across age classes) in order to achieve a positive-definite Hessian and assure that the VAST model converged (see *Statistical analyses* below). Thus, for some species (lingcod, sablefish, arrowtooth flounder, and English sole), the analysis examined a single juvenile age class (either age-0 or age-1) and the resulting pattern and index can be considered an index of annual recruitment. However, for the remainder of the species, the analysis quantifies juvenile density because we had to combine several juvenile age classes. Hereafter, we refer to both cases as juvenile density. Second, while the survey data include subsamples of individual fish lengths, weights, and age structures–with ages determined from otoliths (except for Lingcod for which dorsal fin rays are used)–there are always more length measurements than ages due to limited processing capacity. Since the size distributions for a multiple ages can overlap, we used length-at-age relationships from the aged individuals to set a maximum length for inclusion in a given age class for those individuals that were not aged. We then pooled all individuals of this maximum length or below for the statistical analyses, which do not distinguish age classes within the statistical model. To set the maximum length for each species, we examined the length distributions of aged individuals by 1-cm bins. The longest length bin where younger individuals outnumbered older individuals was set as the maximum length for the younger age class. For example, the length distribution for age-0 and age-1 sablefish overlapped between 26 and 31cm (Table A). However, 29 cm is longest length bin where there were more age-0 fish (ten) than age-1 fish (nine). At 30 cm there were more age-1 than age-0 fish. Therefore we set a maximum length of 29 cm for inclusion of unaged sablefish in the age-0 group. We then extrapolated the full tow weight for the lengths selected based on their weighted proportion in that species’ subsample that was measured for length. If the extrapolated tow-level species weights from the initial multispecies sample were already in the database, they were used as given. The length-weight relationship for each species was calculated from the available data and used when a weight was missing for an individual fish. We also used the length-weight relationships to check for weights that were outliers, for example, in cases where a basket holding the specimen was not correctly tared. In such cases, the calculated weight was used instead. The implied assumption that the length is the correct measurement and not the weight. This assumption is justified given the ease of taking length measurements using the digital length boards employed by the survey. The trawl survey data did not contain age-structure information for Pacific grenadier and the two thornyhead species. Instead, we used age-at- length information from Andrews, Cailliet \[; Pacific grenadier\], Taylor and Stephens \[, shortspine thornyhead\], and Stephens and Taylor \[, longspine thornyhead\] as a guide. For each species, lengths (in 1 cm intervals) were added to the analysis until we obtained acceptable model fits, and based on the published information, we estimated the maximum age of individuals in the analysis. ## Statistical analyses We used vector-autoregressive spatio-temporal modeling \[R package VAST, version 4.1, \] to quantify the spatial and temporal trends of juvenile biomass density (hereafter, density for brevity) for each of the 13 species. VAST is a species distribution modeling approach that effectively reproduce species distributions and provide good estimates of abundance and associated error. VAST performs well with linear covariates, but has limited ability to include non-linear relationships beyond general quadratic ones compared to generalized additive models or branched regression trees. We use VAST here because we are primarily interested in estimating species distributions and an annual index of abundance. Catch per unit effort (CPUE) was the dependent variable, calculated as the extrapolated abundance estimate of juveniles divided by the swept area of the net. The year of capture, pass, vessel, tow location (latitude and longitude), and depth were the predictor variables. The trawl survey is done in two passes between May and October. Pass therefore accounts for intra-annual variation the time of sampling, and the inclusion of pass is consistent with VAST models used to produce biomass indices for west coast stock assessments. Since the goal was to model distributions and not test for explanatory variables, we did not include additional environmental predictors in the models. We applied one common intercept across years, which allowed spatial variation to be explained by spatial and spatio-temporal variation terms, both of which were included in the model. This parameterization prevents the model from forcing biomass to increase or decrease coast-wide in a given year (thereby potentially overestimating recruitment in some areas) as would be the case for yearly intercept. We used gamma-distributed errors for the positive catch rates with a Poisson-link function. This approach approximates a Tweedie distribution, which has positive mass at zero but is otherwise continuous, but is more computationally efficient. Previous work has shown that this approach generally provides a better fit to the WCGBTS data than does a conventional delta model. The extrapolation grid or stratum area was defined based on the data extent for each species separately (Region = ‘Other’ in VAST, Table B). Trawls with zero catch for the target species and within the stratum area were included in the analysis. We used 600 knots \[locations at which spatial processes are estimated by VAST, used for computational efficiency; \] for all model runs. See Thorson for more detail on VAST and Table B in for more detail on the VAST parameterization used here. ## Hotspots We arbitrarily defined juvenile hotspots as areas where estimated juvenile density (biomass) derived from VAST was in the top 20% of its maximum density (80%+) over the study period (red areas the following distribution figures). We define secondary hotspots as areas where juvenile density was in the top third (66% +) of its maximum density over the study period (orange areas in the following distribution figures). Additionally, we examined hotspots at two temporal scales: within years and across years (‘All Years’ on the following distribution figures). All Years was the average juvenile density from 2003–2018. Therefore, areas that consistently had moderate juvenile density may appear as a hotspots across all years, though they may not necessarily qualify as a hotspot at an annual scale. # Results We observed three general, qualitative patterns of juvenile density (biomass) along the West Coast: (1) species with distinct, spatially-limited hotspots that were consistent through time; (2) species with distinct, spatially-limited hotspots that spatially varied through time; and (3) species with large regional hotspots that spanned a broad latitudinal range but were depth limited. Additionally, we estimated the time series of juvenile density for each species, which illustrated that the prevalence and intensity of juvenile hotspots was often related to density. Finally, we also report the effective area occupied and CoG for each species. Dover sole, English sole, Pacific grenadier, shortspine thornyhead, and splitnose rockfish had distinct and spatially limited hotspots of juvenile density (shortspine thornyhead, ; others Figs A-E in ; see for diagnostic plots including QQ-plots and plots of the spatial residuals for the presence/absence portion of the model). For example, the density of juvenile shortspine thornyheads was fairly consistent through time but peaked in 2009–2010, after which it declined. There was a single, distinct across-years hotspot at approximately 45° N in waters between 160–625 m, just north of an area with an expansive area of shallower shelf habitat that includes Heceta, Perpetua, and Stonewall Banks (Fig F in). The hotspot north of the banks had primary or secondary hotspots in most years until approximately 2014, after which juvenile density was very low, but it was most obvious in 2010 when juvenile density was highest. There was a small, secondary, across-years hotspot adjacent to the Columbia River mouth, with a primary hotspot at this location in 2010 when the density of shortspine juveniles peaked. Pacific hake and darkblotched rockfish (Fig G) also showed distinct hotspots of juvenile density, but there was more temporal variation in the location of these hotspots. For example, Pacific hake showed peaks in juvenile density in 2006 and 2009. Juveniles (age-0 and age-1) were caught between 50–700 m. Within a given year, they showed localized areas of high density, but the location of these hotspots varied from year to year. While the location of the hotspots did not appear related to the density of juveniles, hotspots were most apparent in 2006 and 2009 when the density of juvenile hake showed peaks. For example, in 2006 hotspots occurred north and south of Cape Blanco and around Point Conception. However, in 2009 there was a hotspot to the north on either side of the Columbia River outflow. Juvenile density was consistently moderate south of Cape Mendocino resulting in across-years hotspots north of San Francisco Bay and around Point Conception (‘All Years’). For juvenile arrowtooth flounder, sablefish, lingcod, longspine thornyhead, petrale sole, and Pacific sanddab, the areas of high juvenile density were more expansive in area (sablefish, ; lingcod, ; others Fig H-K). These spatial patterns were generally consistent through time. For example, the density of juvenile sablefish was variable with peaks in density in 2004, 2008, 2010, 2013 and 2016. Spatially, across years, age-0 sablefish had large, primary hotspots north of Cape Blanco and on either side of San Francisco Bay with a smaller hotspot around Point Conception. These hotspots to the north of Cape Blanco were most obvious in years with high coast-wide density of age-0 fishes (2004, 2008, 2013, and 2016, and to some extent 2017), suggesting that successful coast-wide recruitment relies to a large extent on successful recruitment in northern waters. The exception was 2010 when juvenile density was high in the absence of the northern hotspot, but density was high in the southern region. Age-0 sablefish were found on shelf and upper slope waters (50–475 m). Additionally, age-0 lingcod had a large, primary hotspot in shelf waters (50–240 m) located from Point Arena to the south of Point Conception. However, there were some small secondary hotspots just north of Cape Mendocino and Cape Blanco. Within years, the prevalence of the large southern hotspot varied and was related to overall density of age-0 lingcod. Again, hotspots were most apparent in years with high juvenile density. For lingcod, the density trend indicated longer-term patterns with low density from 2003–2007, followed by higher juvenile density from 2008–2012, followed again by lower density from 2013 to 2018. Effective area occupied varied for all species, but there were no Temporal trends except for Dover sole, which showed a decline in effective area occupied through time ( and Fig L). Area occupied was negatively correlated with abundance for most species (Fig M). Two species, Pacific hake and sablefish, showed larger variations in their CoG of approximately 1000 km ( and Fig N). The remaining species tended to show fluctuations in the range of 200–400 km, but there were no strong directional trends over time except for Dover sole; its CoG shifted south over time (Fig N in. Relationships between CoG and abundance varied among species (Fig O), as did the relationship between effective area occupied and the CoG (Fig P). For example, the effective area occupied by Dover sole juveniles decreased as the CoG shifted south over time. However, for longspine thornyhead, higher area occupied was associated with more northerly CoG. # Discussion We quantified the temporal variation in the spatial distribution of the juvenile abundance for 13 groundfish species. The observed spatial patterns provide important temporal and spatial information for fisheries managers that will improve our understanding of each species, which was formerly only based on adult distributions. Additionally, we estimated annual coast-wide indices of juvenile density. While evaluating the causes of the observed spatial and temporal patterns was beyond the scope of this paper, we suggest potential ecological and physical habitat mechanisms that likely play a role in the observed temporal and spatial patterns, and we describe how our results have direct application for fisheries management. Mesoscale geographic features like headlands and submarine canyons can alter local circulation, in particular the development of eddies, which may lead to increased retention of larvae and increased productivity, resulting in higher settlement and better food resources. Such processes may explain the location of distinct juvenile hotspots for some particular species and the presence of locations that are hotspots or areas of high juvenile density for many species. For example, multiple species (Dover sole, English sole, hake, lingcod, Pacific sanddab, sablefish and splitnose rockfish) showed high juvenile density near Point Conception, an area of significant eddy formation. Similarly sub-mesoscale eddies form north of San Francisco Bay between Pt. Reyes and Pt Arena where several species also had high juvenile density. Water circulation on wider, shallower sloping areas of the shelf off of the Washington coast is slower leading to higher residence times, and potentially higher retention, which may also help to explain areas of high juvenile density in areas with extensive shelf habitat. Transport mechanisms will obviously interact with an individual species’ life history, in particular the location of any spawning habitats and those advantages to larval biology. Horizontal transport and water temperature during the pelagic larval stage prior to recruitment to demersal habitat also likely drive the inter-annual variation in juvenile densities seen here. For example, inter-annual recruitment variation around the stock-recruitment relationship for sablefish and petrale sole is associated with horizontal transport and water temperature during specific phases in the pelagic life-history of these species. The current results for sablefish are consistent with stronger age-0 recruitment when conditions favored the onshore transport of eggs and northerly transport of yolk-sac larvae. Recruitment variation in bocaccio rockfishes, *Sebastes paucispinus*, is related to basin scale processes and density dependence. For stocks with weak stock- recruitment relationships and persistent low spawning, inter-annual climate variability is probably the most important driver of recruitment variation. However, for species like lingcod that have experienced recovery in spawning density during the period studied, increases in juvenile density over time were also likely due to increased spawning biomass. Depth is clearly an important factor (as a proxy for other variables such as light, pressure, temperature or oxygen saturation) in determining assemblage structure in marine fishes, and this pattern was supported by our analyses. Some species, such as English sole, Pacific sanddab, petrale sole, and lingcod had fairly restricted depth distributions of 100–200 m for juveniles on the shelf. Others like arrowtooth flounder, Dover sole, sablefish and shortspine thornyheads had broader juvenile depth ranges across the shelf and upper slope. Pacific grenadier and longspine thornyhead had broad depth ranges that encompassed deeper slope waters. While not examined here, bottom type, biogenic habitat, temperature, frontal boundaries, and other physical characteristics are known to be important in influencing species distributions, community structure, and the location of juvenile habitats. For demersal fishes early, post-settlement mortality is often high, density-dependent and mediated via competition for shelter. Bottom depth and bottom temperature are important predictors of recently settled juvenile Kamchatka flounder *Atheresthes evermanni*, and the size and location of the cold pool largely determines the distribution of arrowtooth flounder in the eastern Bering Sea. In waters off the U.S. west coast, groundfish assemblages on the slope shift to deeper waters moving north to south likely tracking changes in bottom temperature, and geographic features like Point Conception, Cape Mendocino, and Cape Blanco are known transition zones for biogeographic regions in the California Current. For semi-pelagic or mid-water species like Pacific hake, characteristics of the water column are habitat, and factors like temperature, currents, and distance from the shelf break may be important in determining both spawning location and the distribution of pelagic juveniles. Immature Pacific hake have higher biomass in cooler than average temperatures, while the response of adults varies with latitude. Additionally, hake tend to be found in areas with poleward drift, which may aid in their annual migration to northern waters. Temporal variability in these factors may explain the variability in the location of hotspots for juvenile hake seen here. The distribution of adult habitat may also explain some of the observed juvenile hotspots in that juvenile fishes need access to adult habitats as they age. Many groundfishes recruit to shallow habitats, potentially to avoid cannibalism or predation by other species, but move to deeper areas as they grow and age. For example, sablefish show ontogenetic increases in depth distributions. The location of juvenile hotspots for sablefish were inshore of adult areas, especially the shelf and upper slope waters between Cape Mendocino and the Columbia River outflow. Successful coast-wide recruitment, in terms of total abundance, appears reliant largely on recruitment to this northern hotspot, which is consistent with onshore transport during the egg stage and northerly long-shore transport during the yolk-sack larval stage in the region being correlated with stronger recruitment. Additionally, the sablefish assessment, which incorporates age structure data across multiple age classes to estimate recruitment history, also captured the high age-0 density years seen in the WCGBTS data, suggesting that these northern recruitment pulses were important to stock size. Current habitat maps of EFH for juvenile sablefish show the majority of juvenile EFH in the same area, but in slightly deeper waters farther off shore. Our results suggest that the EFH definition for juvenile sablefish could be revised to include to shallower waters in this northern area. Several species showed large-scale spatial discrepancies between the distribution of juveniles and adults, raising the question of what ecological processes (or potentially methodological mechanisms) establish adult distributions. For example, shortspine thornyheads had a single, high-density northern juvenile hotspot that was consistent across the study period. However, adults are relatively evenly distributed along the coast at mid- to deeper- depths much further south than this northerly area with a high density of juveniles. Two potential mechanisms could lead to the distribution of adult thornyheads. First, dispersal of young fish from this northern hotspot could lead to a much wider distribution of adult fish farther south. Second, juvenile density was not zero to the south, and consistently lower density in the south could lead to lower density dependence resulting in higher survival in more southerly waters. Other factors such as lower predator densities or differing trawling intensity may also influence the distribution pattern. Likewise, while the density of juvenile Dover sole was highly restricted to several hotspots in the south, adults were widely distributed along the coast. Similarly, splitnose rockfish juveniles occupied several hotspots around and to the south of Point Conception, while adults were abundant around and to the north of Point Conception as well as around and to the north of San Francisco Bay, where juvenile density was particularly low. Contrasting patterns in juvenile and adult biomass were also apparent for lingcod, which are currently assessed as two separate stocks off the US west coast. The most recent stock assessments estimated higher stock biomass off the coasts of Washington and Oregon (21,976 mt) compared to the stock off the coast of California (6,509 mt). However, the distribution of age-0 lingcod in the WCGBT survey was consistently higher south of Cape Mendocino. Longshore movement of age-1+ lingcod does not appear to explain the difference in the distributions of age-0 and adult lingcod as genetic analyses identified genetic differentiation between lingcod north and south of Cape Mendocino. Lingcod initially settle to shallow, vegetated habitats, which the trawl survey (limited to trawlable, non-rocky habitats and depths \>55 m) does not sample. So the survey may not adequately describe the distribution age-0 fish if the distribution diverges in these shallower habitats. Alternatively, these genetically dissimilar sub-stocks may have different ecology in the north and south with different levels of recruitment and survival of juvenile fish. Since lingcod move to deeper habitats as they grow and age, the observed distributions are to some extent the result of processes like post-settlement migration, predator avoidance, prey availability and habitat connectivity. Long-term changes in the latitudinal distribution of fish biomass (CoG) on in the California Current have varied among species. For example, from 1977–2013, the CoG for lingcod, rex sole *Glyptocephalus zachirus*, spotted ratfish *Hydrolagus colliei*, darkblotched rockfish, greenstriped rockfish *S*. *elongatus*, and slender sole *Lyopsetta exilis*, all shifted north—a response that might be expected due to long-term warming—while the distribution of Pacific hake and sablefish biomass moved south. Here, while the juveniles of some species like Pacific hake and sablefish showed large fluctuations in their CoG form 2004–2018, only Dover sole had a significant shift in biomass of juvenile fishes to the south. Instead, the variability in the latitudinal distribution appeared related to the fluctuation in the intensity of juvenile hotspots. For example, when sablefish had high age-0 abundance in the north, the CoG shifted north. The species distribution maps presented here should be interpreted in relation to methodological factors like gear type, survey design, and catchability. The trawl survey samples between approximately 55–1280 m, and any conclusions are limited to trawlable (non-rocky) habitat in this range. For some species, like lingcod discussed above, this limited depth range may explain mismatches between juvenile abundance patters and adult biomass. Additionally, because the trawl survey does not sample complex, rocky habitats, abundance estimates may be biased for some species, especially if there is density dependence. For many species, the effective area occupied was negatively correlated with abundance, albeit the relationship was not significant in all cases. The relationship does not mean that the latitudinal or depth range of the species contracted at higher juvenile biomass. Effective area as calculated by the VAST package is the area needed to contain the population at average biomass-density. Therefore, this relationship would be expected as densities increase in hotspots more quickly relative to other areas, thus concentrating biomass–a pattern we might expect if juveniles follow ideal free distributions. Rockfish, for example, are generally more abundant on complex, rocky habitat. Our species distribution maps likely, therefore, miss or underestimate biomass in juvenile hotspots associated with untrawlable habitat like rocky reefs or rocky pinnacles. Temporally, if juvenile abundance follows an ideal free, trawl- derived abundance indices may underestimate abundance at low densities because densities will increase in the primary and unquantified habitat (untrawlabe) before increasing in secondary, sampled habitat (trawlable). Conversely, at higher levels, juvenile density will continue to increase in trawlable areas after density in untrablable areas has stabilized, potentially overestimating abundance. Nevertheless, for species like sablefish, the observed temporal patterns largely match model derived trends for age-0 abundance in the stock assessments. Gear selectivity in relation to fish body size may also influence the results to some extent. For many species, we had to combine age classes to obtain enough data to produce acceptable model fits. Growth of recently settled individuals over the sampling period may influence the distributions as larger fish become more selected. However, the survey is conducted in two passes between mid-May and October, and we account for this effect in part by including pass in the VAST model. Including pass also accounts to some extent for variability in the within-year timing of settlement. These effects are most likely more important for species where we analyzed a single age class. For each pass, vessels begin in Newport, Oregon and initially work north to Cape Flattery; vessels then head south to San Diego. The timing of sampling within year could, therefore, miss late-season settlement in northern waters if settlement occurs after the vessels have passed. Natural mortality may lead to lower estimates of abundance within years for southern waters, as fishes in the south well have been subjected to more predation or starvation risk when they are caught comparted to more northerly fish. Catchability, the product of the efficiency of the sampling gear and the availability of taxa to that gear, likely varies among depths, bottom types, and the size of individuals among other factors. Since we have not adjusted for catchability, the estimates of abundance here should be considered indices and not absolute measures of biomass. While the effect of different sediment sizes within the range of trawlable habitats may not be large, the spatial patterns may be influenced by these factors and the distribution maps should be interpreted with this issue effect in mind. Species distribution modes have been increasingly used to define essential fish habitat in regions managed by the U.S. National Marine Fisheries Service, as mandated by the Magnuson-Stevens Act. For example, Laman et al. estimated habitat-based models for Alaskan species. Our analyses increase the information available to fisheries management practices of the 13 stocks we analyzed. The spatial and temporal results of this study can be used as leading indicators of potential future changes in stock size for both stock and integrated ecological assessments or for investigating environmental drivers of recruitment variation, particularly for those results that capture age-0 and age-1 fish. Delineation of juvenile habitats can also help to parameterize dispersal models and understand dispersal patterns and connectivity. If fact, one of the motivators for the present study was the need to identify juvenile habitats for petrale sole for use in an individual-based dispersal model. Fishery managers can use the spatially explicit information on juveniles from our analyses to further refine and improve essential fish habitat and nursery habitat definitions and delineations. Identifying juvenile habitats allows for more targeted spatial management and conservation, allows for better prediction of responses to environmental events, and forms the foundation for a more complete understanding of a species ecology. For example, fisheries closures implemented to protect adult rockfish and adult habitats led to increased spawning output near closed areas. In the North Sea, managers implemented a closed area to protect the nursery habitat of North Sea plaice *Pleuronectes platessa* from trawling. However, the closure of plaice nursery habitat to trawling was unsuccessful because the distribution of juvenile plaice shifted. The locations of juvenile hotspots for the species we analyzed were generally persistent through time, suggesting that closures or gear restrictions to protect juveniles and juvenile habitat might be effective mechanisms to protect juveniles for these species. Since many species show ontogenetic shifts to deeper waters, such spatially explicit and targeted closures might not adversely affect adult fishing effort, although adverse impacts will be dependent on gear type and the species’ life history, since one species’ adult habitat may be another’s juvenile habitat. Likewise, the multi-species nature of the fishery may also complicate spatial management, but good information on the HAPC for target species is essential for resolving such issues. Our characterization of the spatial distribution of juvenile habitats improves our understanding of how climate or physical conditions affect juvenile survival. For example, there is a large oxygen minimum zone (OMZ) in mid-depth (500–720 m) slope waters off the NW Pacific Coast of the U.S.. This OMZ affects both assemblage structure and diversity, and there is evidence that dissolved oxygen (DO) concentrations are decreasing in the OMZ. Shoaling of these low DO waters can lead to fish kills and alter the behavior and occurrence of fishes. Knowledge of the location of juvenile habitats will aid in assessing the vulnerability of these species to low DO events. # Conclusions Our analysis represents a refinement of information available for the delineation of essential fish habitat for west coast groundfishes by identifying, more specifically, the geographic locations and temporal stability of juvenile habitats (and potential nursery areas). As we address only the density of juvenile fishes, the areas we identify are best termed juvenile habitat rather than nursery habitat. Nevertheless, these areas represent potential, if not likely, nursery habitat for some species. Further work such as tagging studies or growth analyses will help us to better understand whether these areas function as nurseries and their importance to the population dynamics of each species. # Supporting information We thank two anonymous reviewers, B. Feist, K. Marshall, C. Harvey, and J. Samhouri for comments on the manuscript. We also thank A. Shelton, L. Barnett, and J. Thorson for discussions regarding VAST. Special thanks to the trawl survey crew for collecting the data and to M. C. Longino and J. Boulogne for inspiration regarding wearing gloves and masks, and maintaining distance. [^1]: The authors have declared that no competing interests exist.

# Introduction Astrocytomas are the most common primary brain tumors. The World Health Organization (WHO) classifies astrocytomas into four malignant grades: grade I, or pilocytic astrocytoma; grade II, or low-grade astrocytoma (AGII), grade III, or anaplastic astrocytoma (AGIII); and grade IV astrocytoma or glioblastoma (AGIV or GBM). Diffusely infiltrative astrocytomas (AGII-GBM) have the ability to invade the surrounding normal brain tissue, hampering tumor resection. GBM, the most malignant and frequent brain tumor in adults, can be divided into two subgroups: primary GBM, which arises *de novo*, and secondary GBM, which results from the progression of a lower grade astrocytoma. Interestingly, mutations in the gene that encodes isocitrate dehydrogenase 1 (IDH1) have been reported in diffuse gliomas, including WHO grades II and III astroglial and oligodendroglial lineages. *IDH1* mutations are strong predictors of a more favorable prognosis and serve as a highly selective molecular marker of secondary GBM that complements clinical criteria for distinguishing secondary GBM from primary GBM. Lysyl oxidase (LOX), a copper-dependent amine oxidase, catalyzes the enzymatic stage of collagen and elastin cross-linking by oxidizing primary amines into reactive aldehydes. These reactions are essential for stabilization of collagen fibrils and for the integrity and elasticity of mature elastin to ensure normal functionality of connective tissue, embryonic development and adult tissue remodeling. Importantly, biologically active compounds, hydrogen peroxide and ammonia are generated as by-products during these catalytic reactions. LOX also has intracellular functions and is involved in the regulation of cell differentiation, motility/migration and gene transcription. Aberrant expression of the *LOX* gene has been reported in multiple tumors. LOX is synthesized by several cell types as a 48 kDa protein. After signal peptide cleavage and N-glycosylation, the resulting 50 kDa proenzyme is secreted and converted into a mature, active 30 kDa form as a result of proteolytic processing by procollagen C proteinase/bone morphogenic protein-1 (BMP1). The catalytic activity of *LOX* can be specifically and irreversibly inhibited by beta-aminopropionitrile (BAPN). LOX has been identified as an important regulator of the hypoxia-induced tumor progression pathway through a HIF-1α-dependent mechanism in numerous cancer types, such as breast, head and neck, prostate and renal cell carcinomas. LOX is involved in the hypoxic upregulation of *HIF1A*, and LOX and HIF-1α potentiate each other to foster tumor progression in the colon through the PI3K-Akt signaling pathway. Secreted LOX is responsible for the invasive properties of hypoxic cancer cells, including astrocytomas, through the activation of focal adhesion kinase (FAK)/paxillin. LOX has been implicated in tumor angiogenesis *in vitro* and *in vivo* by increasing vascular endothelial growth factor (VEGF) expression and secretion as well as blood vessel formation. Recently, it was demonstrated that HIF-1α-responsive genes essential for cell growth, including *LOX*, were underexpressed in gliomas with *IDH1* mutation. Therefore, we aimed to investigate *LOX*, *BMP1* and *HIF1A* mRNA expression levels in a large series of astrocytomas of different malignant grades and compare these results between cases with wild type *IDH1* and cases with mutated *IDH1*. *LOX* knockdown by siRNA was performed for functional studies *in vitro*, and LOX protein was also analyzed in tumor samples. These data suggest that *LOX* expression increases according to malignancy grade in astrocytomas and represents a potential therapeutic target, especially for cases without *IDH1* mutation. # Methods ## Tissue Samples The samples used in this study consisted of 153 astrocytomas (grades I to IV). Tumors were graded according to the WHO classification into AGI (n = 23; mean age at diagnosis, 19.4±9.7 years; 14 males and 9 females), AGII (n = 26; mean age at diagnosis, 34.0±8.1 years; 15 males and 11 females), AGIII (n = 18; mean age at diagnosis, 35.0±12.3 years; 11 males and 7 females) and GBM (n = 86; mean age at diagnosis, 54.0±13.9 years; 58 males and 28 females). The non-neoplastic control group consisted of samples from individuals undergoing temporal lobe resection during epilepsy surgery (n = 22; mean age at diagnosis, 38.0±7.6 years; 10 males and 12 females). All samples were collected during surgical procedures by the Neurosurgery Group of the Department of Neurology at the Hospital das Clinicas of School of Medicine, University of Sao Paulo, Brazil. Fresh surgical samples were immediately snap-frozen in liquid nitrogen upon surgical removal. Before RNA extraction, a 4-μm-thick cryosection of each sample was stained with hematoxylin, and necrotic and non-neoplastic areas were removed from the frozen block of tumoral tissue by microdissection. Grey matter was avoided in the control non-neoplastic samples. Written informed consent was obtained from all patients according to the ethical guidelines approved by the Ethics Committee of the School of Medicine, University of São Paulo (0600/10). The Ethical Commission for Research Projects Analysis (CAPPesq) from the Clinical Board of Hospital das Clinicas and School of Medicine, University of São Paulo, in council session taken place at 2012, August 8th, approved the research protocol entitled: “Expression and role of lysyl oxidase family genes in astrocytomas”, presented by the Department of Neurology. ## Total RNA extraction and cDNA synthesis Total RNA was extracted from frozen tissues (tumor and non-neoplastic) using an RNeasy Mini kit (Qiagen, Hilden, Germany). The RNA concentration and purity were evaluated by measuring the absorbance at 260 and 280 nm. A 260/280 ratio ranging from 1.8 to 2.0 was considered satisfactory purity. Denaturing agarose gel electrophoresis was used to assess the quality of the samples. A conventional reverse transcription reaction was performed to yield single-stranded cDNA. First-strand cDNA was synthesized from 1 μg of total RNA that was previously treated with 1 unit of DNase I (FPLC-pure, GE Healthcare, Uppsala, Sweden) using random and oligo(dT) primers, RNase inhibitor, and SuperScript III reverse transcriptase according to the manufacturer’s recommendations (Life Technologies, Carlsbad, CA). The resulting cDNA was subsequently treated with 1 unit of RNase H (GE Healthcare, Uppsala, Sweden), diluted with TE buffer, and stored at −20°C until later use. ## Quantitative real time PCR (RT-qPCR) The relative expression levels of *LOX*, *HIF1A* and *BMP1* were analyzed by RT- qPCR using the SYBR Green approach. Quantitative data were normalized to the geometric mean of three reference genes suitable for the analysis: hypoxanthine phosphoribosyltransferase (*HPRT*), beta glucuronidase (*GUSB*) and TATA box binding protein (*TBP*). The primer sequences were as follows (from 5’ to 3’): LOX F: CCTACTACATCCAGGCGTCCA; LOX F R: CATAATCTCTGACATCTGCCCCTGT; HIF1A F: CATCCAAGAAGCCCTAACGTGT; HIF1A R: CATTTTTCGCTTTCTCTGAGCAT; BMP1 F: CTCGTAAGTCCTCCATCAAAGCT; BMP1 R: CTCTCCATCTCCCACAGGCTC; HPRT F: TGAGGATTTGGAAAGGGTGT; HPRT R: GAGCACACAGAGGGCTACA; GUSB F: GAAAATACGTGGTTGGAGAGCTCATT, GUSB R: CCGAGTGAAGATCCCCTTTTTA; TBP F: AGGATAAGAGAGCCACGAACCA, TBP R: CTTGCTGCCAGTCTGGACTGT. The primers were synthesized by IDT (Integrated DNA Technologies, Coralville, IA, USA). The minimum primer concentrations necessary were determined to give the lowest threshold cycle (Ct) and maximum amplification efficiency while minimizing non- specific amplification. The primer concentrations used were 200 nM for *LOX*, *HIF1A*, *BMP1*, *HPRT* and *TBP* and 400 nM for *GUSB*. Standard curves were established to ensure amplification efficiency, and an analysis of melting curves demonstrated a single peak for all PCR products. Additionally, agarose gel electrophoresis was employed to check the size of the PCR products amplified. SYBR Green I amplification mixtures (12 μl) contained 3 μl of cDNA, 6 μl of 2x Power SYBR Green I Master Mix (Life Technologies) and primers. The PCRs were run on an ABI Prism 7500 sequence detector (Life Technologies) as follows: 2 min at 50°C, 10 min of polymerase activation at 95°C, and 40 cycles of 15 s at 95°C and 1 min at 60°C. Quantitative data were normalized relative to the internal housekeeping control genes. The equation 2^−ΔCt was applied to calculate the expression of *LOX*, where ΔCt = Ct of the target gene—geometric mean of the Ct of the reference genes. The RT-qPCR reactions were performed in duplicate for each sample and repeated when the Ct values were not similar. The results are presented on a log10 scale for better visualization. *LOX* expression was scored according to the median expression values of each astrocytoma grade. For statistical analysis, scores equal to or higher than the median values were defined as *LOX* overexpression. For functional analysis of *LOX* knockdown after siRNA transfection, the same procedures were followed, except that only *HPRT* was used as a reference gene. The expression values were calculated relative to the scrambled non-target control (NTC). ## DNA extraction and IDH1 mutational analyses DNA was extracted from frozen tumor samples using a QiaAmp DNA Micro kit (Qiagen). Polymerase chain reaction (PCR) followed by DNA sequencing was applied to detect mutations in IDH1, as previously described. The sequences of primers (5’-3’) synthesized by IDT for PCR amplification of exon 4 were as follows: CCATCACTGCAGTTGTAGGTT and CATACAAGTTGGAAATTTCTGG. PCR products were generated in a 25 μL reaction mixture including 100 ng of DNA, 50 mM KCl, 50 mM of each dNTP, 10 mM Tris-HCl (pH 9.0), 1.5 mM MgCl₂, 10 pmol of each primer and 1 unit of Taq DNA polymerase (GE Healthcare). The PCR was performed with an initial denaturating step at 94˚C for 5 min, followed by 35 cycles consisting of 94˚C for 30 s, 54˚C for 30 s and at 72˚C for 30 s. After the final cycle, an extension period of 10 min at 72˚C was performed. The PCR products (436 bp) were purified with a GFX column (GE Healthcare) and sequenced on an ABI Prism 3130 DNA automated sequencer using the Big Dye Terminator Cycle Sequencing Ready Reaction Kit version 3.1 (Life Technologies). The primers used for sequencing were the same as those used for PCR. ## Cell culture conditions and transient transfection The human malignant astrocytoma cell lines U87MG and A172 were routinely cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Life Technologies) supplemented with 10% fetal bovine serum (FBS) (Life Technologies), 100 IU/ml penicillin and 100 μg/ml streptomycin in an atmosphere consisting of 5% CO₂ in air at 37°C in a humidified incubator. Dicer substrate small interfering RNA (siRNA) duplexes for *LOX* knockdown (5′-GUAAUUACAGAAUUGAAACACUGUGUU-3′) were diluted in buffer according to the manufacturer’s recommendations (IDT). U87MG and A172 cells (1 × 10⁵ cells/well) were seeded in a six-well plate; after 24 h, they were transfected with Lipofectamine RNAiMax (Life Technologies). Control cells were transfected with a scrambled non-target control (NTC) siRNA from IDT. siRNAs for both LOX and NCT were used at a concentration of 1 nM, and *LOX* knockdown as well as the effect of LOX silencing were evaluated at 2, 4, and 7 days after transfection by RT-qPCR and Western blotting. ## Cell migration and invasion assay The migration and invasion abilities of the cells were assessed by determining the ability of the cell lines to cross a membrane with a pore size of 8 m in non-coated (for migration) and matrigel-coated (for invasion) Transwells (BD Biosciences, Becton, Dickinson and Company, Franklin Lakes, NJ), according to the manufacturer's recommendations. After 48 hours of transfection, the cells were maintained in DMEM supplemented with 1% FBS for 2 hours. To determine the effect of LOX inhibitor treatment on migration, the cells were treated with 100 mM β-aminopropionitrile (BAPN Sigma-Aldrich, Saint Louis, MO) for 24 h prior to the migration assays and for an additional 4 h during the assay. This concentration of BAPN has no cytotoxic effects, as demonstrated previously. After trypsinization, a total of 2.5 x 10⁴ cells were suspended in 0.5 ml DMEM with 1% FBS and added to the upper compartment of the Transwell. The bottom chamber contained 10% FBS, and the cells were incubated for 18 hours at 37°C in an atmosphere containing 5% CO₂. Non-migrating and non- invading cells were wiped away from the upper surface of the membrane with a cotton swab. The cells were fixed with 4% paraformaldehyde, stained with 0.2% crystal violet (Sigma-Aldrich) in 20% methanol, and analyzed by inverted microscopy (40x magnification). The results of the migration and invasion assays were quantified by counting 18 random fields from each of the experimental inserts in duplicate. Data were generated from two independent experiments. Migration values were expressed as the average number of migrated cells per microscope field. ## Cell proliferation and anchorage-independent cell growth assay Proliferation of U87MG and A172 cells was evaluated after 2, 4, and 7 days of siRNA *LOX* transfection in duplicate experiments. The cells were stained with Trypan blue and counted using an automatic counter (Countess, Life Technologies) for determination of live cells. Anchorage-independent cell growth was analyzed via a soft agar colony formation assay. Two days after transfection, 1 x 10³ cells were resuspended in 0.3% agar on a layer of 0.6% agar in DMEM (1x) in a six-well plate and incubated in a humidified atmosphere in the presence of 5% CO₂ at 37°C. After 10 days, the colonies were fixed with formaldehyde and stained with 0.005% crystal violet in phosphate-buffered saline (PBS) for 1 h. The number of colonies was recorded for each well. Two independent experiments were performed in duplicate. ## Western blot analysis After transfection with control siRNA (NTC) and siRNA specific for *LOX*, total protein lysates were prepared from cell cultures with RIPA lysis buffer and protease inhibitor cocktail (Sigma-Aldrich) on ice. The protein concentration was determined using the Bradford reagent (Bio-Rad Laboratories, Richmond, CA). The absorbance was measured at 595 nm using a microplate spectrophotometer (Multiskan Spectrum; Thermo Labsystems, Helsinki, Finland). All assays were conducted in duplicate, and calculations were carried out with a standard curve constructed using different concentrations of bovine serum albumin (2 mg/ml to 0.125 mg/ml). Total protein lysates (30 mg) were separated by 12% SDS polyacrylamide gel electrophoresis (TGX Mini Protean, Bio-Rad) with Tris-glycine running buffer. The proteins were transferred to a nitrocellulose membrane using the iBlot dry blotting system (Life Technologies). The membrane was blocked with 5% skim milk and incubated with rabbit polyclonal primary anti-LOX diluted 1:1,000 (Sigma-Aldrich). The membrane was also incubated with mouse monoclonal anti-β-actin (1:5,000, clone AC-74, Sigma-Aldrich) as a protein loading control. The secondary antibodies used were anti-rabbit (1:1,000) and anti-mouse IgG (1:5,000) conjugated to peroxidase (Sigma-Aldrich). The immune complexes were visualized using enhanced chemiluminescence reagent (Western Lightning Chemiluminescence Reagent Plus, Perkin Elmer, Waltham, MA) and detected with ImageQuant LAS4000 (GE Healthcare). ## Immunohistochemistry For immunohistochemical detection, 4-μm formalin fixed paraffin embedded tissue sections were routinely processed and subjected to antigen retrieval. Briefly, slides were immersed in 10 mM citrate buffer (pH 6.0) and incubated at 122°C for 3 min using an electric pressure cooker (BioCare Medical, Walnut Creek, CA). Specimens were then incubated for endogenous peroxidase blocking (Novolink system, Novocastra, Newcastle-upon-Tyne, UK) and further incubated with a polyclonal antibody raised in rabbits against human LOX (ab31238, 1:100 dilution; Abcam, Cambridge, UK) at 16–20°C for 16 hours. The reaction was developed with a commercial kit (Novolink; Novocastra, Newcastle-upon-Tyne, UK) at room temperature using diaminobenzidine and Harris hematoxylin for nuclear staining. Optimization using a positive control suggested by the manufacturer (breast carcinoma) was performed to obtain the optimal dilution. The staining intensity of tissue sections was evaluated independently by two observers (SKNM and RS). A semi-quantitative scoring system considering both the intensity of staining and percentage of cells was applied as follows: for intensity of staining, 0 = negative, 1 = weak, 2 = moderate and 3 = strong; for cell percentage, 0 = no cells stained, 1 = 10–25%, 2 = 26–50%, 3 = 51–75% and 4 = 76–100%. The LOX immunohistochemistry labeling score (ILS) was obtained by determining the product of the staining intensity and the percentage of cells stained. Digital photomicrographs of representative fields were captured and processed using PICASA 3 (Google, USA). ## Statistical analysis Kolmogorov-Smirnov normality test was used to analyze the distribution of gene expression data and also the effect of *IDH1* mutations on gene expression. The Kruskal- Wallis test was used for analysis of the differences in gene expression between non-neoplastic tissues and astrocytomas of different grades of malignancy. Coexpression of genes was analyzed using the Spearman-rho test. A correlation coefficient (r) of ≥0.7 was interpreted as a strong correlation, 0.3≤r\<0.7 was interpreted as a moderate correlation, and r \<0.3 was interpreted as a slight correlation. The Mann-Whitney test and *t* test were used to compare *IDH1* mutational status and gene expression for non-parametric and parametric distributions, respectively. The Mann-Whitney test was also used for functional assays. When the Kruskal-Wallis test was used for gene expression as well as ILS analysis among the different groups and the results were significant, Dunn multiple comparison post hoc test was applied. The Kaplan- Meier survival curves were analyzed using the log rank test (Mantel Cox), excluding 9 cases of tumor relapse in GBM patients. To perform the analysis, the patients were divided into two groups characterized by high (above the median) and low (below the median) gene expression. The statistical significance was set at *p*\<0.05. All analyses were performed using SPSS version 15.0 (Chicago, IL), and scatter plots were constructed using the program GraphPad Prism version 5.0 (GraphPad Software, Inc., San Diego, CA). # Results ## *LOX*, *BMP1* and *HIF1A* Expression Levels in Astrocytomas of Different Malignant Grades *LOX*, *BMP1* and *HIF1A* expression analysis by qRT-PCR showed great variability in astrocytomas of all malignant grades when compared to non- neoplastic samples. The GBM cases had higher *LOX* expression levels relative to non-neoplastic cases, with a statistically significant difference. No difference was found in *LOX* expression level between AGI, AGII and AGIII when compared to non-neoplastic samples. On the other hand, *BMP1* expression levels were significantly higher in AGI and GBM groups when compared to non-neoplastic samples. *HIF1A* expression levels increased with the malignant grade of astrocytomas, with statistically significant values for all malignant grades of astrocytomas when compared to control samples. Coexpression of the three genes was compared for astrocytomas of all malignant grades. Interestingly, Spearman analysis demonstrated that *LOX* mRNA levels were positively correlated with *BMP1* and *HIF1A* in GBM (*r* = 0.217 and *p* = 0.045, ; *r* = 0.367 and *p* = 0.001, respectively). *BMP1* expression also presented a positive correlation with *HIF1A* in GBM (*r* = 0.389; *p*\<0.001) and in AGI (*r* = 0.477; *p* = 0.033). *LOX*, *BMP1* and *HIF1A* expression levels were correlated with the clinical outcome of the GBM cases. The analysis of overall survival of GBM cases with high and low expression did not reveal a significant correlation between the gene expression data and the prognosis of GBM patients. ## Association of *LOX*, *BMP1*, and *HIF1A* mRNA Expression Levels and *IDH1* Mutation Status The frequency of IDH1 mutation was 80.8% in AGII (21 out of 26), 61.1% in AGIII (11 out of 18) and 12.8% in GBM (11 out of 86) (, respectively), which was also described in our previous study of the frequency of *IDH1* mutations in a series of GBM patients. Our GBM cases were composed mainly by primary GBMs, which explains the low frequency of *IDH1* mutation. Of all the mutations, R132H was the most common in glioma. summarizes the comparison of the median expression levels of cases with wild-type and mutated *IDH1*. *IDH1*-mutated AGII and GBM cases showed lower *LOX* expression levels when compared to wild-type *IDH1* AGII (*p* = 0.049) and GBM (*p* = 0.008) cases. On the other hand, AGIII cases with wild-type *IDH1* presented lower *HIF1A* expression levels when compared to *IDH1*-mutated cases. (*p* = 0.038). No difference was found for *LOX* expression in AGIII groups with mutated and wild type *IDH1*. Similarly, *BMP1* expression between wild-type and mutated *IDH1* cases was not different regardless of astrocytoma group, and *HIF1A* expression was not different for AGII and GBM. ## Effect of *LOX* Knockdown in GBM Cells *LOX* knockdown was evaluated after 2 days of transfection of U87MG and A172 GBM cell lines with siRNA targeted against *LOX* or non-targeted control (NTC) siRNA. The efficiency of transfection was analyzed by RT-qPCR and western blotting for each of the duplicate experiments. Both cell lines presented an approximate 80% decrease in LOX mRNA expression when compared to NTC. Protein expression was also confirmed to be diminished after *LOX* knockdown with siRNA. Transfected cells were maintained under cell culture conditions for 2, 4 and 7 days after transfection to evaluate the involvement of *LOX* in the proliferation of A172 and U87MG cells. There was no difference in the proliferation rate of cells after *LOX* knockdown when compared to NTC in both cell lines analyzed (data not shown). To test the hypothesis that *LOX* expression was correlated with the migratory ability of tumor cells, U87MG and A172 cell lines were evaluated after knocking down *LOX* expression by siRNA and after inhibiting the active form of LOX with a specific drug (BAPN). There was a reduction in the migration ability of U87MG and A172 cells after either transfection with siRNA for LOX or treatment with BAPN, as shown in. The differences between the *LOX* siRNA and NTC groups were statistically significant in both U87MG (*p*\<0.001) and A172 (*p*\<0.0001) cells. Inhibition of LOX by BAPN significantly inhibited the migration of both U87MG and A172 cells when compared to non-treated cells (*p*\<0.0001 for both cell lines). LOX involvement in the invasion of GBM cell lines was also observed for U87MG and A172 cells (*p*\<0.0001 for both). Anchorage-independent growth assays were subsequently used to examine the role of *LOX* in the colony formation ability of U87MG and A172 cells. No reduction in colony numbers was observed for U87MG cells after knockdown of *LOX* when compared to NTC. On the other hand, there was a significant decrease in the number of colonies of A172 cells transfected with *LOX* RNAi compared to NTC (*p*\<0.0001). ## LOX Immunohistochemistry Analyses Expression of LOX at the protein level was investigated by immunohistochemistry of non-neoplastic brain tissues and astrocytomas of different malignant grades, as shown in. LOX expression was observed in the cytoplasm and nucleus of all tissues with variable intensity. Endothelial cells were stained specifically in neoformed vessels of GBM cases. The immunohistochemistry results were analyzed semi-quantitatively by the ILS, as described in Methods. The cytoplasmic ILS did not vary significantly among the different astrocytoma groups and control samples. Nevertheless, a high nuclear ILS was found in non-neoplastic samples, with staining of glial cells and neurons, while the nuclear ILS in astrocytomas increased with increasing degree of malignancy. AGI group presented statistical differences to non-neoplastic, AGIII and GBM samples. Additionally, endothelial cells were predominantly stained for LOX in GBM cases, with significant differences of this tumor group and non-neoplastic and AGI, AGII and AGIII groups. Interestingly, GBM cases with mutated *IDH1* presented lower LOX nuclear and endothelial staining, confirming the results of *LOX* transcript expression and *IDH1* mutation status analyses. # Discussion In the present study, we investigated the expression of *LOX*, *BMP1* and *HIF1A* in astrocytomas of different malignant grades. Additionally, the effect of *IDH1* mutation on gene expression was also evaluated as well as the role of *LOX* in the behavior of GBM cell lines. LOX is a secreted amine oxidase that plays a key role in modifying the primary tumor microenvironment by crosslinking collagens and elastin in the ECM thereby causing stiffening of the matrix and enhancing the invasive and metastatic properties of the tumor. The stiffness of the ECM is particularly enhanced by the active form of LOX (, and BMP1 is responsible for processing LOX into the active form. We demonstrated significantly higher *LOX* mRNA expression levels in GBM cases, when compared to non-neoplastic brain tissue samples, as reported by others in different types of tumors. Furthermore, higher *BMP1* expression that was positively correlated with *LOX* expression was also detected in GBM cases. Such findings are in agreement with the fact that GBMs are the most malignant and invasive astrocytomas. In fact, we used functional analysis to demonstrate that knocking down *LOX* with siRNA or inhibiting LOX with BAPN led to a reduction in the migration and invasion of U87MG and A172 GBM cell lines. LOX knockdown and BAPN treatment have similar effect on migration in U87MG cell line. On the other hand, the same was not observed for A172 cells even though reduction of LOX expression by siRNA treatment was confirmed. Those results might reflect the GBM heterogeneity. The mutational profile of U87MG and A172 is different according to data of COSMIC cell lines project (<http://cancer.sanger.ac.uk/cancergenome/projects/cell_lines/>) shows that U87MG has as main consensus genes mutated *PTEN*, *NF1* and *ATRX*, while A172 has *EGFR*, *RB1* and *PTCH1*, among others. Similar results were obtained by Lackzo et al. using other GBM cells (U251 and U373) treated with BAPN and catalase. These authors associated the role of active LOX in the migration/invasiveness of GBM cell lines with FAK/paxillin activation through hydrogen peroxide generated by LOX catalytic reactions. Intracellular hydrogen peroxide in excess can facilitate the phosphorylation and activation of Src, which then phosphorylates FAK, with consequent induction of various signaling pathways involved in the regulation of cell adhesion and migration. FAK phosphorylation may also explain our finding of a decrease in the number of colonies of A172 cells in which *LOX* expression was knocked down, as suggested by others~~)~~. The absence of a similar effect in the colony formation assay with U87MG cells corroborates the heterogeneity observed in this type of tumor. The irreversible LOX inhibitor BAPN was initially used to treat disorders such as hypertension, decreased wound healing, and peripheral blood mononuclear cell chemotaxis by preventing the formation of a highly cross-linked form of vascular collagen. Later, BAPN was proposed to also be useful in the treatment of cancers with LOX hyperexpression, such as melanoma, head and neck carcinoma and breast carcinoma. Recently, BAPN was used in the treatment of *in vivo* tumor models of pancreatic ductal adenocarcinoma in mice, and it was able to stabilize senescence, delay tumorigenesis, and increase survival. Moreover, another inhibitor of LOX, magnolol, presented a similar effect in a breast cancer model *in vitro*. In addition to BAPN, d-penicillamine, which depletes intracerebral copper, also exhibited antiangiogenic effects on GBM tumor growth in mice. LOX is essential for stimulating endothelial cells and angiogenesis by increasing VEGF expression, and the increase in matrix stiffness also upregulates VEGF expression. Interestingly, LOX is expressed in GBM endothelial cells, as demonstrated by immunohistochemistry. This result corroborates a recent report of LOX expression in tumor endothelial cells. The expression of LOX by tumoral and endothelial cells in GBM may be controlled by a positive feedback loop mechanism, although this will require further investigation. Taken together, these results reinforce the notion that LOX may be a target in the treatment of tumors. The high proliferative capacity of GBM leads to the development of hypoxic areas and ultimately necrosis. The reduction in oxygen availability activates hypoxia- inducible factor-1, which in turn activates the transcription of target genes, including *LOX*. Indeed, our results demonstrated that *HIF1A* is highly expressed in GBM cases, and its expression is associated with *LOX* and *BMP1* expression. Interestingly, the median expression levels of both *LOX* and *BMP1* were higher in AGI samples than in AGII samples. In contrast, the median expression of *HIF1A* in AGI cases was lower than that in AGII. AGI is non-invasive tumor, while AGII is infiltrative. Additionally, AGI presents vascular proliferation and shows greater contrast on neuroimaging compared to AGII. Taking these phenotypic characteristics together, we speculate that *HIF1A* upregulation in AGI activates a hypoxia-driven angiogenic pathway. Additionally, BMP1 can activate various other substrates that are important regulators of extracellular matrix production and quality as well as of antiangiogenic responses by producing a factor from the basal membrane compound. In contrast, stepwise upregulation of *HIF1A* observed in our cases of diffusely infiltrative astrocytomas associated with high expression of *LOX* and *BMP1* suggest progressive activation of both angiogenic and invasive pathways. Another interesting finding of the present study was the association of *IDH1* mutation with low expression of *LOX*. A list of underexpressed genes responsive to HIF-1, including *LOX*, has been recently reported in *IDH1* mutant gliomas and brain tumor stem cells. In fact, in the present series, *LOX* expression was significantly lower among AGII (p = 0.049) and GBM (p = 0.008) cases presenting *IDH1* mutations when compared to cases with wild-type *IDH1*. For GBM, a decreased level of LOX immunostaining in the nucleus was observed with the presence of *IDH1* mutations. Mutated *IDH1* is a common feature in lower grade gliomas and secondary GBMs. Our series of GBM presented only 12% of IDH1 mutation, consisting mainly of primary cases, as previously reported. The presence of mutated *IDH1* is strongly correlated with a CpG island methylator phenotype in gliomas. Therefore, promoter methylation might explain the decreased LOX expression in the presence of such a mutation. Indeed, *LOX* inactivation by methylation has already been demonstrated in gastric cancers as well as colon, lung and ovarian cancer cell lines. Using an antibody against the C-terminus of LOX that detects only the 50 kDa pro-LOX or the processed active 32 kDa LOX, we have unequivocally detected a high level of LOX nuclear staining, particularly in the most malignant astrocytomas (grades III and IV). Previous studies have described only intense perinuclear and cytoplasmic staining of astrocytomas of different malignant grades; nuclear staining has not been mentioned. Intriguingly, nuclear staining was also detected in our normal controls, including glial and neuronal cells, as described by others. These findings were in agreement with previous reports of LOX nuclear localization in smooth muscle and proliferating cells. LOX can oxidize nuclear proteins such as histone H1, leading to epigenetic effects on DNA-histone and histone-histone interactions, with consequent effects on DNA transcription analogous to the effects resulting from histone acetylation. Further studies are needed to clarify differential LOX function in normal and tumor tissues as well as in distinct intracellular compartments. In summary, our results confirmed that LOX plays an important role in migration and angiogenesis in diffusively infiltrative astrocytomas, especially GBMs. Moreover, *LOX* expression is influenced by *IDH1* mutational status, which provides new insights for the design of targeted therapies to control these tumors. # Supporting Information We are very thankful for all the neurosurgeons of the Division of Neurosurgery of the Department of Neurology at Hospital das Clinicas of School of Medicine, University of São Paulo for the therapeutic and follow-up procedures of all patients included in this study. We particularly acknowledge Thais F. Galatro for the valuable immunohistochemistry reactions. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SMOS. Performed the experiments: RDS. Analyzed the data: MU. Contributed reagents/materials/analysis tools: SKNM. Wrote the paper: RDS SMOS SKNM.

# Introduction The oral detection of sugars is presumed to occur by activation of a class 1 taste receptor heteromer, TAS1R2-TAS1R3, sometimes referred to as a ‘sweetener receptor’. Notions of how human sweet taste operates are largely based on mouse sweetener perceptual genetics and physiology. By genetic homology, binding to and activation of this receptor is believed to stimulate perceived sweetness in humans as well. In support of this idea, mutations in regulatory regions of the human T1R receptor genes are associated with slightly altered sensitivity to sucrose. However, more recent genetic studies of human sweet taste perception and intake have not replicated these observations. Based on the assumption that the human preference for sugary foods and beverages is driven by stimulation of sweet taste orally, the design and creation of artificially sweetened beverages occurred over 100 years ago by Hyman Kirsch in 1904 who made No-Cal ginger ale with calcium cyclamate to treat diabetics. Despite the century-long refinement of no-calorie or low-caloric sweeteners in beverages, diet sodas have never captured a major share of the beverage market. Reasons for this are presently unknown, but have been attributed to: 1) non-sucrose-like sweetness, 2) non- sucrose-like bitterness and other side tastes, 3) non-sucrose-like temporal profile of sweetness (lingering taste), 4) customer fears of artificial ingredients in foods (naturalism bias), and 5) concerns for increased risk of medical pathologies from use, such as cancer. We hypothesize an additional explanation that sugars may engage a second oral signaling pathway for calories that noncaloric sweeteners fail to engage. Recently, mouse taste bud cells were found to contain many of the same molecular components as do pancreatic beta islet of Langerhans cells, which use a multi- step metabolic signaling pathway to detect glucose in the blood. Beta islet cells indicate increases in blood glucose via: A) the transport of glucose into the cells via molecular carriers such as the sodium-glucose co-transporters (SGLTs) along with other transporters, such as GLUTs, B) the oxidation of glucose to produce several ATP molecules via i) glycolysis involving glucokinase and the production of ATP and pyruvate and ii) oxidative metabolism of the pyruvate via Krebs cycle in mitochondria to yield additional ATP, and lastly, C) the closing of potassium channels that are gated by ATP (K_ATP channel). The presence of glucose transporters, glucokinase, and the ATP gated potassium channels K_ATP (Kir6.1 and SUR) have all been identified in mouse taste bud cells. Thus, it appears that taste bud cells in the mouth of mice are capable of identifying when a “sweetener” is metabolizable. This system appears to have functionality in mice, as oral stimulation with sugars in the absence of T1R2-T1R3 taste receptors continues to elicit anticipatory insulin responses to sugar. Whether a similar glucose metabolic signaling pathway exists and is functional within human oral taste cells and whether it impacts sugar perception has not been determined. Were such a signaling system to exist and be functional in the human mouth, it would have implications for oral signaling of the metabolizable sugar glucose and could help explain preference for sugared beverages over non- caloric sweetener beverages (cf.). In the present study, we conducted several psychopharmacological experiments to identify whether a second signaling pathway exists in human mouths for sugars such as glucose, but not for non-caloric sweeteners, such as sucralose, and whether this pathway may involve the sodium- glucose-linked co-transporters (SGLTs) as an initial step in this additional signaling pathway. Support for the role of the specific SGLT transporters in glucose taste come from our use of added sodium, which is co-transported with glucose, use of the SGLT inhibitor phlorizin, use of the sugar fructose, which is not transported by SGLT, and use of the glucose analog α-methyl-D- glucopyranoside (MDG) which is co-transported with sodium by SGLT, but is not metabolized to produce ATP. # Materials and methods ## Experiment 1 ### Hypothesis We hypothesize that glucose engages a dual oral signal comprised of i) T1R receptor-based signaling and ii) a form of oral metabolic-pathway signaling that involves transport of glucose by an SGLT, whereas sucralose engages only the T1R receptor signaling. Therefore, we predicted that inhibiting the T1R2-T1R3 taste signal with the T1R inhibitor Na-lactisole would have a greater impact (raising the threshold concentration) on sucralose than it would on glucose detection thresholds. ### Ethics statement All research was conducted according to the principles expressed in the Declaration of Helsinki, and approved by an Institutional Review Board at the University of Pennsylvania (IRB \#701334). Subjects provided written, informed consent on forms approved by the Institutional Review Board prior to participation. ### Subjects Twelve healthy adults (7 female, 5 male; mean age  =  39 years, S.D.  =  13) participated. Subjects were recruited from the surrounding community, and paid for their time. ### Stimulus materials Stimuli included filtered water (Milli-Q Water Purification System); serial dilutions of D-(+)-glucose (\> 99.5%, Sigma-Aldrich, USA) in filtered water in 1/8 log steps ranging from 0.73–73 mM; serial dilutions of sucralose (\> 98.0%, Sigma-Aldrich, China) in filtered water in 1/8 log steps ranging from 2.0–200 μM; serial dilutions of glucose in filtered water in 1/8 log steps ranging from 0.73–412 mM with the addition of 2 mM Na-lactisole (Sodium 2-\[4-methoxyphenoxy propionate\], Endeavor Specialty Chemicals, UK); serial dilutions of sucralose in filtered water in 1/8 log steps ranging from 2.0–1124.7 μM and 2 mM Na- lactisole; serial dilutions of glucose in filtered water in 1/8 log steps ranging from 0.73–73 mM with the addition of 2mM NaCl (\>99.0% sodium chloride, Fisher Scientific, USA); serial dilutions of sucralose in filtered water in 1/8 log steps ranging from 2.0–200 μM and 2 mM NaCl; 0.8 mM Na-lactisole used as a rinse. Stimuli were presented at room temperature. ### Detection threshold method Subjects completed six conditions each with a replicate. The conditions measured absolute detection thresholds for i) glucose and ii) sucralose alone, iii & iv) each with 2 mM Na-lactisole added, and v & vi) each with 2 mM NaCl to control for the sodium associated with lactisole. In these conditions, subjects rinsed with water before and between samples. Each subject was instructed to refrain from smoking, eating, chewing gum, and drinking anything, except water, for one hour before participation. At the start of each session subjects rinsed their mouths with water 4 times for 30 seconds each and then expectorated for a total rinse time of 2 minutes. They were presented with two 10 ml samples. In the glucose condition the samples were water and glucose. While wearing nose clips, they put the whole sample in their mouth and after 1–2 seconds expectorated and then rinsed with water. Next, they repeated the tasting with the second sample. Their task was to select the sample that is different from water in a two-alternative forced-choice (2-AFC) trial. If unsure, they were instructed to guess. For the glucose + Na-lactisole condition, the samples were water + 2 mM Na-lactisole and glucose + 2 mM Na- lactisole. For the glucose + NaCl condition the samples were water + 2 mM NaCl and glucose + 2 mM NaCl. For these conditions their task was to identify which sample was stronger. Again, if unsure they were instructed to guess. The same scenario was repeated using sucralose in place of glucose comprising three glucose conditions and three sucralose conditions. Detection thresholds were measured using a modified staircase method. Starting with the average threshold for each test sample (7.33 mM for glucose, 20 μM for sucralose) subjects made their selection (either different from water or the stronger sample depending on the test condition). If their response was correct, they were presented with the same concentration until they gave 4 correct responses. If their response was incorrect, the next higher concentration was presented in ascending order until 4 correct responses were given in a row at the same concentration. After 4 correct responses, a descending order of concentrations was presented until an incorrect response was given. This pattern was followed until subjects completed 5 reversals in concentration direction (ascending to descending or descending to ascending). If the 5 reversals spread over more than 3 concentration-steps, however, testing continued until the 5 reversals remained within a three concentration-step range in order to clamp variability and avoid random walks on the staircase. The concentrations of the last 4 reversals were averaged to calculate each subject’s absolute detection threshold. ### Sweet water taste control Rinsing with Na-lactisole may result in what is known as “sweet water taste”. That is, after Na-lactisole treatment, plain water rinses are sometimes perceived as sweet. To determine if our procedure gave rise to this phenomenon, we measured subject’s detection thresholds for both glucose and sucralose under three conditions: 1) when presented against water, 2) with the addition of 2 mM Na-lactisole, and 3) with 0.8 mM Na-lactisole rinses between samples to inhibit sweet water taste. The same twelve subjects (7 female, 5 male; mean age  =  39 years, S.D.  =  13) that participated in Experiment 1 were tested in these three conditions in duplicate. The concentration of 0.8 mM Na-lactisole for the rinse was selected in preliminary studies to prevent sweet water taste from 2 mM Na- lactisole treatment. See. A repeated measures analysis of variance for the glucose detection thresholds revealed a significant effect of condition F (2, 66) = 31.50, p \<.00001. Post- hoc Tukey HSD analyses revealed the detection thresholds for all three conditions (water, Na-lactisole, and Na-lactisole rinse) were significantly different from one another, p \<.001, with the detection threshold for glucose without treatment as the lowest and the detection threshold with the 0.8 mM Na- lactisole rinses between stimuli as the highest. Therefore, there was no evidence for an effect of sweet water taste following Na-lactisole treatment in this procedure. There was no effect of replication, nor an interaction between condition and replication. The repeated measures analysis of variance for the sucralose detection thresholds also revealed a significant effect of condition F (2, 66) = 21.42, p \<.00001. Post-hoc Tukey HSD analyses revealed the detection thresholds for Na-lactisole treatment and the Na-lactisole treatment with rinse were significantly higher than the detection threshold for sucralose alone, p \<.001. Again, there was no evidence for a sweet water taste following Na- lactisole treatment in this procedure. There was no effect of replication, nor an interaction between condition and replication. ## Experiment 2A & B ### Hypothesis Based upon the outcomes of Experiment 1, we hypothesized that if the additional signaling pathway for glucose involves transport into cells via the sodium glucose co-transporter (SGLT), then adding NaCl to both glucose and MDG should enhance detection (lower detection threshold), but not enhance sucralose detection. MDG is transported by SGLT but is not metabolizable into ATP. Thus, MDG will determine whether movement of sodium with MDG by an SGLT is sufficient to enhance detection. ### Subjects The same subjects who participated in Experiment 1, participated in the detection threshold experiment comparing glucose to sucralose (Experiment 2A). A different group of subjects participated in the experiment comparing detection thresholds of glucose and MDG (Experiment 2B). Eleven healthy adults (7 female, 4 male; mean age  =  44 years, S.D.  =  12) participated. Subjects were recruited from among the surrounding community, and paid for their time. ### Stimulus materials Stimuli included filtered water (Milli-Q Water Purification System); serial dilutions of D-(+)-glucose (\> 99.5%, Sigma-Aldrich, USA) in filtered water in 1/8 log steps ranging from 0.73–73 mM; serial dilutions of D-(+)-glucose in filtered water in 1/8 log steps ranging from 0.73–73 mM with 20 mM NaCl (\>99.0% sodium chloride, Fisher Scientific, USA) added; serial dilutions of sucralose (\> 98.0%, Sigma-Aldrich, China) in filtered water in 1/8 log steps ranging from 2.0–200 μM; serial dilutions of sucralose in filtered water in 1/8 log steps ranging from 2.0–200 μM with 20 mM NaCl added; serial dilutions of α-methyl-D- glucopyranoside (MDG) (\> 99%, Sigma-Aldrich, China) in filtered water in 1/8 log steps ranging from 0.73–73 mM; serial dilutions of MDG in filtered water in 1/8 log steps ranging from 0.73–73 mM with 20 mM NaCl added. Stimuli were presented at room temperature. ### Detection threshold method For Experiment 2A, subjects completed 4 conditions each with a replicate. The conditions measured detection thresholds for glucose and sucralose each with and without the addition of 20 mM NaCl. The same threshold measurement protocol described in Experiment 1 was followed for Experiment 2A. Detection thresholds were measured using a modified staircase method with 5 reversals. In the glucose condition, the sample pairs were deionized filtered water versus glucose or deionized filtered water with 20 mM NaCl added versus glucose with 20 mM NaCl added. In the sucralose condition, the sample pairs were deionized filtered water versus sucralose or deionized filtered water with 20 mM NaCl added versus sucralose with 20 mM NaCl added. The subject’s task was to identify which sample in the pair was stronger. If unsure, they were instructed to pick one. For Experiment 2B, subjects also completed 4 conditions each with a replicate. The conditions measured detection thresholds for glucose and MDG each with and without the addition of 20 mM NaCl. The same threshold measurement protocol described in Experiment 1 was followed for Experiment 2B. Detection thresholds were measured using a modified staircase method with 5 reversals. In the glucose condition, the sample pairs were deionized filtered water versus glucose or deionized filter water with 20 mM NaCl added versus glucose with 20 mM NaCl added. In the MDG condition the samples were deionized filtered water versus MDG or deionized filtered water with 20 mM NaCl added versus MDG with 20 mM NaCl added. The subject’s task was to identify which sample in the pair was stronger. If unsure, they were instructed to pick one. ## Experiment 3 ### Hypothesis Based upon the outcome of Experiments 2A and 2B, we hypothesized if oral glucose signaling involves an SGLT, then the pharmacological SGLT inhibitor, phlorizin, should increase absolute detection thresholds for glucose and MDG, but not for sucralose, and have no effect on fructose that is transported by GLUT5 instead of SGLT. ### Subjects Eleven of the same subjects that participated in Experiment 1participated in the glucose and sucralose segments of Experiment 3 (6 female, 5 male; mean age  =  39 years, S.D.  =  13). For the MDG segment subjects were the same individuals that participated in Experiment 2B (7 females and 4 males with a mean age = 44 years with S.D. = 12). Finally, for the fructose segment subjects included 7 females and 5 males with a mean age = 36 years with S.D. = 12. ### Stimulus materials Stimuli included filtered water (Milli-Q Water Purification System); serial dilutions of D-(+)-glucose (\> 99.5%, Sigma-Aldrich, USA) in filtered water in 1/8 log steps ranging from 0.73–73 mM; serial dilutions of D-(+)-glucose in filtered water in 1/8 log steps ranging from 0.73–73 mM with 20 mM NaCl (\>99.0% sodium chloride, Fisher Scientific, USA) added; serial dilutions of D-(+)-glucose in filtered water in 1/8 log steps ranging from 0.73–73 mM with 20 mM NaCl and 0.2 mM phlorizin (\>98.0%, Cayman Chemical Company, USA) added; serial dilutions of sucralose (\> 98.0%, Sigma-Aldrich, China) in filtered water in 1/8 log steps ranging from 2.0–200 μM; serial dilutions of sucralose in filtered water in 1/8 log steps ranging from 2.0–200 μM with 20 mM NaCl added; serial dilutions of sucralose in filtered water in 1/8 log steps ranging from 2.0–200 μM with 20 mM NaCl and 0.2 mM phlorizin added; serial dilutions of α-methyl-D-glucopyranoside (MDG) (\> 99%, Sigma-Aldrich, China) in filtered water in 1/8 log steps ranging from 0.73–73 mM; serial dilutions of α-methyl-D- glucopyranoside (MDG) in filtered water in 1/8 log steps ranging from 0.73–73 mM with 20 mM NaCl added; serial dilutions of α-methyl-D-glucopyranoside (MDG) in filtered water in 1/8 log steps ranging from 0.73–73 mM with 20 mM NaCl and 0.2 mM phlorizin added; serial dilutions of D--fructose (\> 99%, Sigma-Aldrich, USA) in filtered water in 1/8 log steps ranging from 0.56–56 mM; serial dilutions of D--fructose in filtered water in 1/8 log steps ranging from 0.56–56 mM with 20 mM NaCl added; serial dilutions of D--fructose in filtered water in 1/8 log steps ranging from 0.56–56 mM with 20 mM NaCl and 0.2 mM phlorizin added. Stimuli were presented at room temperature. ### Detection threshold method There were four segments in Experiment 3. One for each sweetener: glucose, fructose, MDG, and sucralose. In each segment, subjects completed 3 conditions each with a replicate. The conditions measured detection thresholds for the sweetener alone, with the addition of 20 mM NaCl, and with the addition of 20 mM NaCl and 0.2 mM phlorizin. The same protocol described in Experiments 1 and 2 was followed for Experiment 3. Detection thresholds were measured using a modified staircase method with 5 reversals. For each sweetener type (glucose, fructose, MDG, and sucralose the sample pairs were filtered water versus sweetener alone, filtered water with 20 mM NaCl versus sweetener with 20 mM NaCl, filtered water with 20 mM NaCl + 0.2 mM phlorizin added versus sweetener with 20 mM NaCl + 0.2 mM phlorizin added. The subject’s task was to identify which sample was stronger. If unsure, they were instructed to pick one. # Results All analyses were conducted using Statistica software (Version 13.5.0.17, Tibco), using an alpha value of \<0.05 for allowing Type I errors. ## Experiment 1 Factorial repeated-measures analysis of variance (ANOVA) were conducted to compare the main effects of condition (water, Na-lactisole, NaCl), and replication (1, 2) and the interaction effect on detection thresholds for both glucose and sucralose. There was a main effect of condition for each sweetener such that the Na-lactisole treatment significantly raised threshold levels. For glucose F (2, 66) = 24.86, p \<.0001, post-hoc Tukey HSD analyses showed the detection threshold with Na-lactisole (mean = 41.8 mM, S.D. = 25.9) was significantly higher than both glucose alone (mean = 13.3 mM, S.D. = 5.5) and glucose with NaCl (mean = 13.8 mM, S.D. = 4.6), p \<.001. For sucralose F (2, 66) = 33.00, p \<.0001, post-hoc Tukey HSD analyses showed the detection threshold with Na-lactisole (mean = 100.0 μM, S.D. = 73.7) was significantly higher than both sucralose alone (mean = 11.9 μM, S.D. = 5.4) and sucralose with NaCl (mean = 11.9 μM, S.D. = 5.0), p \<.001. The water and NaCl conditions did not differ from each other for either sweetener. The 2 mM NaCl was added to control for the effects of 2 mM sodium from the Na-lactisole. There was no effect of replication nor an interaction between condition and replication for either sweetener. As expected, the Na-lactisole treatment decreased sensitivity to the sweeteners glucose and sucralose (See). Also as expected the Na-lactisole treatment had a significantly larger impact on sucralose than on glucose. After Na-lactisole treatment and with water rinsing between samples, the ‘fold’ increase in threshold concentration for glucose was x 3.1 and for sucralose x 10.0, t (11) = -2.99, p = 0.012 (See). Overall, our hypothesis is supported by these data; Na-lactisole interfered with detecting sucralose approximately three times more than it interfered with glucose detection. ## Experiment 2A Factorial repeated-measures ANOVA were conducted to compare the main effects of condition (with and without NaCl), and replication (1, 2) and the interaction effect on detection thresholds for both glucose and sucralose. For each sweetener there was a significant effect of condition. For glucose the addition of NaCl significantly lowered the detection threshold (glucose mean = 13.3 mM, S.D. = 5.5 while glucose + NaCl mean = 7.2 mM, S.D. = 1.6), F (1, 44) = 22.71, p \<.0001. Conversely, for sucralose the addition of NaCl significantly raised the detection threshold (sucralose mean = 11.9 μM, S.D. = 5.4 while sucralose + NaCl mean = 19.1 μM, S.D. = 7.7), F (1, 44) = 13.88, p \<.001. Thus, the addition of 20 mM NaCl enhanced sensitivity to glucose (lowered threshold concentration) by 46% on average, whereas the addition of 20 mM NaCl diminished sensitivity to sucralose (raised threshold concentration) by 161% on average. There was no effect of replication nor an interaction between condition and replication for either sweetener (See). The hypothesis that NaCl would enhance sensitivity to glucose but not sucralose is supported by these data. These data are consistent with SGLT participating in glucose signaling by moving glucose and sodium together into the cells. ## Experiment 2B A factorial repeated-measures ANOVA was conducted to compare the main effects of sweetener (glucose or MDG), condition (with and without NaCl), and replication (1, 2) and the interaction effects on the detection thresholds. The ANOVA revealed there was not a significant difference between the sweeteners, thresholds are similar, but there was a main effect of condition F (3, 80) = 6.28, p \<.001. Post-hoc Tukey HSD analyses revealed the addition of NaCl lowered the detection thresholds for each sweetener, glucose mean = 12.6 mM, S.D. = 4.2 while glucose + NaCl mean = 9.3 mM, S.D. = 2.8, p \<.01; MDG mean = 11.7 mM, S.D. = 2.9 while MDG + NaCl mean = 9.0 mM, S.D. = 1.8, p \<.05. There was no effect of replication nor any significant interaction effects (See). The lowering of the detection thresholds for both glucose and MDG by the addition of NaCl is further evidence that SGLT is involved in moving glucose and its analog into the cell, suggesting a second signaling pathway. Furthermore, it suggests that movement of MDG with sodium by SGLT is sufficient to alter thresholds and ATP generation is not required. ## Experiment 3 Factorial repeated-measures ANOVA were conducted to compare the main effects of condition (water, NaCl, and phlorizin), and replication (1, 2) and the interaction effects on detection thresholds for glucose, MDG, fructose, and sucralose. The results for glucose revealed a significant main effect for condition F (2, 60) = 58.27, p \<.00001. Post-hoc Tukey HSD analyses showed the detection threshold for glucose with NaCl and phlorizin added (mean = 39.9 mM, S.D. = 16.8) was significantly higher than the detection threshold for glucose alone (mean = 13.5 mM, S.D. = 5.7) and for glucose with NaCl (mean = 7.2 mM, S.D. = 1.6 mM), p \<.001. Similar to glucose, there was a main effect for condition for MDG, F (2, 60) = 62.05, p \<.00001. Post-hoc Tukey HSD analyses showed the detection threshold for MDG with NaCl and phlorizin added (mean = 37.6 mM, S.D. = 15.4) was significantly higher than the detection threshold for MDG alone (mean = 11.7 mM, S.D. = 2.9) and for MDG with NaCl (mean = 9.0 mM, S.D. = 1.8), p \<.001. For fructose, there were no significant differences in condition. Finally, for sucralose there was a main effect for condition F (2, 60) = 3.45, p \<.05. Post-hoc Tukey HSD analyses showed the detection threshold for sucralose with NaCl (mean = 18.3 mM, S.D. = 7.5 mM) was significantly higher than sucralose alone, (mean = 12.2 mM, S.D. = 5.6 mM) p \<.05, but no significant differences for the phlorizin condition (mean = 16.2 mM, S.D. = 9.6 mM). There was no effect of replication nor an interaction between condition and replication for any of the sweeteners (See). Phlorizin increased the absolute detection thresholds for both glucose and MDG in 100% of the subjects (11 of 11) by approximately a 3-fold increase on average. Whereas, phlorizin had no effect on absolute detection thresholds for fructose or sucralose. The pharmacological SGLT inhibitor phlorizin had a profound impact on impeding every subject’s ability to detect glucose and MDG but did not impede their detection of fructose nor sucralose, further supporting the idea that oral glucose signaling involves a second pathway involving SGLT. # Discussion These rinse-and-expectorate experiments point to the ability of humans to sense glucose orally via a signaling pathway that includes the sodium-glucose cotransporters (SGLTs). The SGLT is an initial step in what is described as the glucose metabolic signaling pathway, and our data suggest this exists in parallel to the traditional T1R sweet taste pathway for oral glucose sensing in humans. In the first experiment, we used the T1R (class 1 taste receptor) inhibitor Na-lactisole to block sweet taste. Na-lactisole had a much larger impact on sucralose detection thresholds than it did on glucose thresholds. illustrates that detection thresholds for sucralose increased in concentration (indicating decreased sensitivity) by more than 8-fold in the presence of Na- lactisole, whereas glucose thresholds only increased by 3-fold. We interpret this as supporting the idea that there are two sugar sensing coding channels in the mouth: one for signaling the sweet taste of sugars and a second that we hypothesized is based on the metabolic signaling pathway for saccharides. Sucralose only engages the T1R-sweet taste receptor, so inhibition of this receptor has a larger impact on its detection. Glucose, however, engages both signaling channels and in the presence of Na-lactisole continues to be signaled by a second pathway. In a second experiment, we tested whether the SGLT glucose transporters are involved with the initial step of moving the saccharide into the cell. Since the SGLTs must move sodium together with glucose, we hypothesized that adding sodium (NaCl) at levels comparable to the detection thresholds for glucose (\~20 mM at the high end) would lower the threshold for glucose but not for the non-caloric sweetener sucralose by enhancing glucose transport into the cells. The SGLT1 transporter moves two sodium ions for each glucose molecule and the SGLT2 transporter moves one sodium ion for each glucose molecule; therefore, this concentration of sodium would improve movement for both transporters. We found that glucose thresholds decreased (indicating increased sensitivity) in the presence of 20 mM NaCl by approximately 50%, but sucralose thresholds did not decrease with added NaCl (See). In fact, the addition of NaCl interfered with sucralose sensitivity and raised detection threshold concentrations modestly. This is most likely due to the weak salty taste of 20 mM NaCl cognitively suppressing detection of sucralose by acting as a sweet-taste masking agent. This influence of the salty taste would also have occurred for glucose, but the impact of added Na on the SGLT mechanism overcame this masking effect. In the third experiment, we used phlorizin as an SGLT inhibitor to interfere with glucose signaling in the mouth. Whereas the addition of sodium was expected to enhance glucose signaling and decrease detection thresholds of glucose, the addition of phlorizin was predicted to interfere with glucose signaling and increase glucose detection thresholds. We found the addition of 2 mM phlorizin increased glucose detection thresholds compared to the glucose + NaCl condition by approximately 5-fold, whereas sucralose detection thresholds were unaffected by the addition of phlorizin (See). Furthermore, there was no effect on fructose thresholds of the addition of phlorizin. This is consistent with glucose taste signaling utilizing an SGLT transporter, but fructose does not use SGLT. Phlorizin is a natural phenolic glucoside chalcone common in apple trees, which when dimerized and oxidized is responsible for the yellow-brown color of apple juice and cider. It is a non-selective inhibitor of the SGLTs, so we cannot conclude from the present data whether one particular SGLT (SGLT1 or SGLT2) is more important for glucose signaling. Note that there are currently six known forms of SGLT transporters, although SGLT3 in humans is thought to be a glucose receptor, not a transporter. There is recent evidence that phlorizin may also inhibit the GLUTs in addition to SGLTs. We do not believe this is reflected in our results, however, as phlorizin had no impact on fructose, supporting the involvement of SGLTs but not GLUT2, GLUT5, or GLUT8. We also saw NaCl enhance glucose thresholds (See), further indication of a role for SGLT and not GLUTs in this effect. Regardless, we have established that we can manipulate human oral sensitivity to glucose positively with the addition of sodium and negatively with the addition of phlorizin, all without similar impact on the sweetener sucralose. This bidirectional psychopharmacological approach has strongly implicated the SGLTs as participating in the first step of sugar transport into the cell. Logically, it is likely that other sugars besides glucose, such as galactose, can be transported by SGLT in taste tissue, and sugars that do not engage SGLT, such as fructose, may be transported via GLUT2 and GLUT8 in taste tissue and oxidized in taste tissue to produce ATP and close K_ATP channels. The glucose metabolic signaling pathway combined with the sweet taste pathway creates a striking parallel between the glucose signaling abilities of the pancreatic beta islet cells of Langerhans and the signaling abilities of the oral cavity (most likely taste tissue). It has previously been reported that mice possess within oral taste buds: 1) sugar transporters, 2) kinases required to convert sugars into ATP, and 3) an ATP sensor in the form of an ion channel. Whether any of these exist in humans has been previously unknown. In the present work, we also included MDG as a taste stimulus. This glucose analog can be transported by SGLT but does not get oxidized to produce ATP, as it cannot be metabolized. This allows us to distinguish whether the first step in the metabolic signaling pathway can produce oral signals independently of the remainder of the traditional metabolic signaling pathway, such as ATP generation and the closure of K_ATP channels. We found that MDG oral detection thresholds were enhanced by the addition of NaCl and impaired by the addition of phlorizin (See). This indicates that the transport of sodium by SGLT with glucose and MDG can activate cells irrespective of whether ATP acts on K_ATP channels. Logically, the closure of K_ATP channels in this signaling pathway may also further activate cells. Collectively, these data allow us to screen enhancers of glucose signaling via stimulation of the SGLT pathway for their impact on oral sugar signaling. These may involve any form of pharmacological SGLT enhancement including the addition of sodium to glucose, SGLT pharmacological modulation (allosteric or otherwise) to better transport glucose across the membrane, and any glucose mimetic that can be transported into cells via SGLT. We speculate that it may be possible to reduce sugar levels in beverages and foods by enhancing glucose via the metabolic signaling pathway, especially at the level of SGLT transport. We note that SGLT signals may contribute to sweet taste, but may also contribute to an independent non-sweet signal that conveys the presence of glucose. This idea is suggested by work in T13-Knock-Out mice which showed anticipatory insulin release to a glucose load, but did not show a behavioral preference for glucose in water. It is possible that oral SGLT contributes to glucose reward in humans either directly or indirectly via anticipatory metabolic regulatory reflexes, but note that oral SGLT in mice does not appear to contribute to preference In summary, the experiments presented here utilizing the T1R inhibitor Na- lactisole, the co-transport agent for SGLT added NaCl, and the SGLT inhibitor phlorizin support the idea that SGLTs are involved in the oral perception of glucose, but not the perception of sucralose or fructose. These studies show that a non-T1R oral signaling pathway profoundly affects absolute detection of glucose but not sucralose or fructose. Future studies will determine the utility of this SGLT-linked oral glucose signal in human psychology and physiology. # Supporting information 10.1371/journal.pone.0256989.r001 Author response to previous submission 17 May 2021 10.1371/journal.pone.0256989.r002 Decision Letter 0 Glendinning John I. Academic Editor 2021 John I. Glendinning This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 8 Jun 2021 PONE-D-21-16275 Evidence that Human Oral Glucose Detection Involves A Sweet Taste Pathway and A Glucose Transporter Pathway PLOS ONE Dear Dr. Breslin, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Dear Dr. Breslin and coauthors, You have done a superb job addressing most of the concerns raised by the reviewers. In particular, I commend you for conducting the additional experiments. As you note, they strengthen your study and provide robust support for the central claim of your study--that SGLT1 is involved in oral glucose detection in humans. This is clearly a novel and important finding.  Overall, I believe that you have addressed the majority of the reviewers' original concerns. You will see, however, that Reviewer 1 has some remaining concerns that need to be addressed in your revision. I also have a few editorial suggestions.  1\. Your current title is: "Evidence that Human Oral Glucose Detection Involves A Sweet Taste Pathway and A Glucose Transporter Pathway." This title implies that you have identified two pathways for detecting glucose in the oral cavity. This is not the case. What you have established is that one can systematically alter glucose detection thresholds by modulating the activity of SGLT1. While I do not mean to diminish the importance of this finding, you currently do not know whether modulating the activity of SGLT1 alters downstream voltage- sensitive events in the T1r2-r3 pathway or whether it activates a T1r2-r3-independent pathway. Thus, I recommend that you scale back the scope of your title. 2\. End of first paragraph of Introduction: You state: "Clearly sugars engage a second signaling pathway for calories that noncaloric sweeteners fail to engage post-prandially." I do not think that this statement follows logically from the five reasons you provide for why "...diet sodas have never captured a major share of the beverage market." Reviewer 1 has raised additional concerns about this statement. 3\. Line 1, second paragraph of Introduction: Change "mice" to "mouse" 4\. Line 4, second paragraph of Introduction: The following phrase seems awkward: "...the transportation of glucose". I think that it is more conventional to say "...the transport of glucose" 5\. Line 5, second paragraph of Introduction: Why don't you refer directly to the GLUTs? That would seem to be clearer than a vague reference to "...other transporters".  6\. Lines 5-6, second paragraph of Introduction: I think that the following wording is accurate: "...the conversion of glucose into several ATP molecules..." Glucose is not actually converted into ATP molecules. The energy in the glucose molecule is used by glycolysis, krebs cycle and oxidative phosphorylation to add high-energy phosphate bonds to AMP and ADP. 7\. Line 10, second paragraph of Introduction: I recommend changing "mouse tissue" to "mouse taste bud cells." 8\. Lines 10-13, second paragraph of Introduction: I do not think that the prior literature has established that "taste bud cells in the mouth of mice are capable of identifying when a “sweetener” is metabolizable." Instead, I think this literature has established that murine taste bud cells may be able to detect elevations in salivary glucose via a T1r2+r3-independent mechanism. 9\. In the last paragraph of the Introduction, you state: "Were such a signaling system to exist and be functional in the human mouth, it would have implications for oral signaling of metabolizable sugars and could help explain why humans overwhelmingly prefer sugared beverages to non-caloric beverages." I recommend that you dial-back this statement a bit for two reasons. First, because your paper addresses threshold (not suprathreshold) effects of SGLT1, you are creating false expectations in your readers. Second, the use of "metabolizable sugar" is a bit confusing. As noted by Reviewer 1, fructose may not be metabolized in taste cells, but it is certainly metabolized in the liver. Thus, fructose is a metabolizable sugar that does not activate SGLT1. 10\. I was surprised that you did not incorporate the findings from a recent mouse paper \[Yasumatsu K, Ohkuri T, Yoshida R, Iwata S, Margolskee RF, Ninomiya Y (2020) Sodium-glucose cotransporter 1 as a sugar taste sensor in mouse tongue. Acta Physiologica 230: e13560. doi: 10.1111/apha.13529\], as it is highly relevant and supportive of your general hypothesis. Please submit your revised manuscript by Jul 23 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, John I. Glendinning, PhD Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <u><https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf></u> and [https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf](about:blank) 2\. Thank you for stating the following in the Acknowledgments Section of your manuscript: "This research was funded in part by a grant from the Suntory Global Innovation Center Limited to PASB. The funder consulted on the general conception of the study and provided support in the form of salaries for the authors (PASB, AT, LJF) and a scientist from Suntory (AI) helped collect data under the supervision of the Monell Center. The specific roles of the authors are articulated in the “Author Contributions” section. The funder did not play a role in the specific study design, data analyses, decision to publish, or preparation of the manuscript." We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form. Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows: "The funder did not play a role in the specific study design, data analyses, decision to publish, or preparation of the manuscript.  " Additionally, because some of your funding information pertains to commercial funding, we ask you to provide an updated Competing Interests statement, declaring all sources of commercial funding. In your Competing Interests statement, please confirm that your commercial funding does not alter your adherence to PLOS ONE Editorial policies and criteria by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” as detailed online in our guide for authors  [http://journals.plos.org/plosone/s/competing- interests](about:blank).  If this statement is not true and your adherence to PLOS policies on sharing data and materials is altered, please explain how. Please include the updated Competing Interests Statement and Funding Statement in your cover letter. We will change the online submission form on your behalf. Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: [http://journals.plos.org/plosone/s/competing-interests](about:blank) \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This revised manuscript is substantially enhanced by the addition of MDG and fructose experiments which support the involvement of SGLT1 sensing in oral glucose detection in humans. However, the characterization of this oral SGLT1-glucose sensing pathway as a "caloric" or " metabolic signaling pathway" is misleading. The SGLT1 pathway does not respond to fructose which is a "caloric" and "metabolizable" sugar. It is also unclear how much sucrose activates the SGLT1 taste pathway given that sucrose must be first hydrolyzed to glucose in the mouth to have this action. The authors should revise their description of the SGLT1 taste pathway. Also, for completeness, the authors should mention that galactose is a ligand for SGLT1 although, given the relatively low amount of free galactose in foods, this sugar may have minimal effects on the oral SGLT1 pathway. The authors describe MDG as a ligand for SGLT1 but don't mention that it is also a ligand for the T1R2/T1R3 receptor and thus has a sweet taste to humans as it apparently has to mice as judged by their avidity for this glucose analog. The authors speculate that the SGLT1 taste pathway may contribute to the sweetness (or palatability) of sugars in humans and contribute to their preference for sugars over non-nutritive sweeteners. However, this does not appear to be the case in mice. SGLT1 KO and WT mice showed identical preferences for MDG over water in 3-min, 2-bottle tests (Sclafani, Koepsell and Ackroff, 2016). Furthermore, SGLT1 KO and WT mice showed similar preferences for a glucose + saccharin solution over a saccharin in a 1-h choice test. The KO mice consumed less glucose and MDG than WT mice which was attributed to their failure to normally absorb these sugars in the intestine. The authors should acknowledge that the oral glucose-sensing SGLT1 pathway does not appear to mediate oral sugar preference in mice although, as discussed next, it is essential for post- oral glucose preference conditioning. This does not preclude them from speculating about a role of SGLT1 taste pathway in sugar preference in humans. The following statement in the Introduction should be supported with citations: "Clearly sugars engage a second signaling pathway for calories that noncaloric sweeteners fail to engage post-prandially." Presumably, the authors are referring to rodent and human studies showing that the post-oral actions of glucose and glucose-containing carbohydrates (sucrose, maltodextrin) condition preferences. However, describing this post-prandial pathway as "for calories" is questionable given the many rodent studies showing that glucose is much more effective than fructose in activating the post-prandial pathway. Of particular relevance here are studies showing that mice learn to prefer glucose but not isocaloric fructose over non-nutritive sweeteners (sucralose, Ace K), an outcome thought to be mediated by intestinal SGLT1 signaling (e.g., Sclafani & Ackroff, 2017; Tan et al., 2020). The effectiveness of post-prandial fructose in conditioning sugar preferences in humans is uncertain because the few published human conditioning studies all used maltodextrins. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0256989.r003 Author response to Decision Letter 0 17 Aug 2021 1\. Your current title is: "Evidence that Human Oral Glucose Detection Involves A Sweet Taste Pathway and A Glucose Transporter Pathway." This title implies that you have identified two pathways for detecting glucose in the oral cavity. This is not the case. What you have established is that one can systematically alter glucose detection thresholds by modulating the activity of SGLT1. While I do not mean to diminish the importance of this finding, you currently do not know whether modulating the activity of SGLT1 alters downstream voltage- sensitive events in the T1r2-r3 pathway or whether it activates a T1r2-r3-independent pathway. Thus, I recommend that you scale back the scope of your title. We respectfully disagree. We can take out T1R2/3 with lactisole and still have glucose transduction occur. We can also take out SGLT with phlorizin and still have sucralose be sweet. Therefore, the T1R2/3 and SGLT participate forin glucose transduction and sweet taste transduction with inhibition of the other and are, therefore, independent. They may converge at some point, but at least initially they are able to function are two independently of each other as taste systems. 2\. End of first paragraph of Introduction: You state: "Clearly sugars engage a second signaling pathway for calories that noncaloric sweeteners fail to engage post-prandially." I do not think that this statement follows logically from the five reasons you provide for why "...diet sodas have never captured a major share of the beverage market." Reviewer 1 has raised additional concerns about this statement. We changed to: We hypothesize an additional explanation that sugars may engage a second oral signaling pathway for calories that noncaloric sweeteners fail to engage.. 3\. Line 1, second paragraph of Introduction: Change "mice" to "mouse" OK 4\. Line 4, second paragraph of Introduction: The following phrase seems awkward: "...the transportation of glucose". I think that it is more conventional to say "...the transport of glucose" OK 5\. Line 5, second paragraph of Introduction: Why don't you refer directly to the GLUTs? That would seem to be clearer than a vague reference to "...other transporters". We changed to…..Other transporters, such as GLUTs. 6\. Lines 5-6, second paragraph of Introduction: I think that the following wording is accurate: "...the conversion of glucose into several ATP molecules..." Glucose is not actually converted into ATP molecules. The energy in the glucose molecule is used by glycolysis, krebs cycle and oxidative phosphorylation to add high-energy phosphate bonds to AMP and ADP. Changed to "...the oxidation of glucose to produce several ATP molecules..." 7\. Line 10, second paragraph of Introduction: I recommend changing "mouse tissue" to "mouse taste bud cells." OK 8\. Lines 10-13, second paragraph of Introduction: I do not think that the prior literature has established that "taste bud cells in the mouth of mice are capable of identifying when a “sweetener” is metabolizable." Instead, I think this literature has established that murine taste bud cells may be able to detect elevations in salivary glucose via a T1r2+r3-independent mechanism. We are asked to switch the verb capable to the verb able. Given our wording that the "system appears capable" we believe our wording is still appropriate. 9\. In the last paragraph of the Introduction, you state: "Were such a signaling system to exist and be functional in the human mouth, it would have implications for oral signaling of metabolizable sugars and could help explain why humans overwhelmingly prefer sugared beverages to non-caloric beverages." I recommend that you dial-back this statement a bit for two reasons. First, because your paper addresses threshold (not suprathreshold) effects of SGLT1, you are creating false expectations in your readers. Second, the use of "metabolizable sugar" is a bit confusing. As noted by Reviewer 1, fructose may not be metabolized in taste cells, but it is certainly metabolized in the liver. Thus, fructose is a metabolizable sugar that does not activate SGLT1. We changed sugars to glucose and emphasize preference here as an implication of our work rather than suprathreshold sweetness. Changed to: Were such a signaling system to exist and be functional in the human mouth, it would have implications for oral signaling of the metabolizable sugar glucose and could help explain preference for sugared beverages over non-caloric sweetener beverages. 10\. I was surprised that you did not incorporate the findings from a recent mouse paper \[Yasumatsu K, Ohkuri T, Yoshida R, Iwata S, Margolskee RF, Ninomiya Y (2020) Sodium-glucose cotransporter 1 as a sugar taste sensor in mouse tongue. Acta Physiologica 230: e13560. doi: 10.1111/apha.13529\], as it is highly relevant and supportive of your general hypothesis. We cite this paper now. 2\. Thank you for stating the following in the Acknowledgments Section of your manuscript: "This research was funded in part by a grant from the Suntory Global Innovation Center Limited to PASB. The funder consulted on the general conception of the study and provided support in the form of salaries for the authors (PASB, AT, LJF) and a scientist from Suntory (AI) helped collect data under the supervision of the Monell Center. The specific roles of the authors are articulated in the “Author Contributions” section. The funder did not play a role in the specific study design, data analyses, decision to publish, or preparation of the manuscript." We note that you have provided funding information that is not currently declared in your Funding Statement. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form. Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows: "The funder did not play a role in the specific study design, data analyses, decision to publish, or preparation of the manuscript. " The revised funding statement should read: This research was funded in part by a grant from the Suntory Global Innovation Center Limited to PASB. The funder consulted on the general conception of the study and provided support in the form of salaries for the authors (PASB, AT, LJF), and a scientist from Suntory (AI) helped collect data under the supervision of the Monell Center. The funder did not play a role in the specific study design, data analyses, decision to publish, or preparation of the manuscript. Add to online-submission form Additionally, because some of your funding information pertains to commercial funding, we ask you to provide an updated Competing Interests statement, declaring all sources of commercial funding. In your Competing Interests statement, please confirm that your commercial funding does not alter your adherence to PLOS ONE Editorial policies and criteria by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” as detailed online in our guide for authors <http://journals.plos.org/plosone/s/competing-interests>. If this statement is not true and your adherence to PLOS policies on sharing data and materials is altered, please explain how. Please include the updated Competing Interests Statement and Funding Statement in your cover letter. We will change the online submission form on your behalf. \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This revised manuscript is substantially enhanced by the addition of MDG and fructose experiments which support the involvement of SGLT1 sensing in oral glucose detection in humans. However, the characterization of this oral SGLT1-glucose sensing pathway as a "caloric" or " metabolic signaling pathway" is misleading. The SGLT1 pathway does not respond to fructose which is a "caloric" and "metabolizable" sugar. It is also unclear how much sucrose activates the SGLT1 taste pathway given that sucrose must be first hydrolyzed to glucose in the mouth to have this action. The authors should revise their description of the SGLT1 taste pathway. Also, for completeness, the authors should mention that galactose is a ligand for SGLT1 although, given the relatively low amount of free galactose in foods, this sugar may have minimal effects on the oral SGLT1 pathway. We now state near the end of the discussion: Logically, it is likely that other sugars besides glucose, such as galactose, can be transported by SGLT in taste tissue (Harada et al., 2012), and sugars that do not engage SGLT, such as fructose, may be transported via GLUT2 and GLUT8 in taste tissue (Sukumaran et al, 2016) and oxidized in taste tissue to produce ATP and close KATP channels (Berger et al, 2020). The authors describe MDG as a ligand for SGLT1 but don't mention that it is also a ligand for the T1R2/T1R3 receptor and thus has a sweet taste to humans as it apparently has to mice as judged by their avidity for this glucose analog. The authors speculate that the SGLT1 taste pathway may contribute to the sweetness (or palatability) of sugars in humans and contribute to their preference for sugars over non-nutritive sweeteners. However, this does not appear to be the case in mice. SGLT1 KO and WT mice showed identical preferences for MDG over water in 3-min, 2-bottle tests (Sclafani, Koepsell and Ackroff, 2016). Furthermore, SGLT1 KO and WT mice showed similar preferences for a glucose + saccharin solution over a saccharin in a 1-h choice test. The KO mice consumed less glucose and MDG than WT mice which was attributed to their failure to normally absorb these sugars in the intestine. The authors should acknowledge that the oral glucose-sensing SGLT1 pathway does not appear to mediate oral sugar preference in mice although, as discussed next, it is essential for post- oral glucose preference conditioning. This does not preclude them from speculating about a role of SGLT1 taste pathway in sugar preference in humans. We now acknowledge that SGLT1KO mice have sugar preference unaffected but still speculate about human oral influences on preference for sugars. We write: It is possible that oral SGLT contributes to glucose reward in humans either directly or indirectly via anticipatory metabolic regulatory reflexes, but not that oral SGLT in mice does not appear to contribute to preference (Sclafani et al., 2016). The following statement in the Introduction should be supported with citations: "Clearly sugars engage a second signaling pathway for calories that noncaloric sweeteners fail to engage post-prandially." We now cite the paper showing that T1R KO mice have no responses to non-caloric sweeteners but continue to show responses to glucose and sucrose. Margolskee et al. Presumably, the authors are referring to rodent and human studies showing that the post-oral actions of glucose and glucose-containing carbohydrates (sucrose, maltodextrin) condition preferences. However, describing this post-prandial pathway as "for calories" is questionable given the many rodent studies showing that glucose is much more effective than fructose in activating the post- prandial pathway. Of particular relevance here are studies showing that mice learn to prefer glucose but not isocaloric fructose over non-nutritive sweeteners (sucralose, Ace K), an outcome thought to be mediated by intestinal SGLT1 signaling (e.g., Sclafani & Ackroff, 2017; Tan et al., 2020). The effectiveness of post-prandial fructose in conditioning sugar preferences in humans is uncertain because the few published human conditioning studies all used maltodextrins. In this context, we avoid discussing post-prandial feedback mechanisms in humans. \_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_\_ 10.1371/journal.pone.0256989.r004 Decision Letter 1 Glendinning John I. Academic Editor 2021 John I. Glendinning This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Aug 2021 Evidence that Human Oral Glucose Detection Involves A Sweet Taste Pathway and A Glucose Transporter Pathway PONE-D-21-16275R1 Dear Dr. Breslin, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, John I. Glendinning, PhD Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0256989.r005 Acceptance letter Glendinning John I. Academic Editor 2021 John I. Glendinning This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 10 Sep 2021 PONE-D-21-16275R1 Evidence that Human Oral Glucose Detection Involves A Sweet Taste Pathway and A Glucose Transporter Pathway Dear Dr. Breslin: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. John I. Glendinning Academic Editor PLOS ONE [^1]: The authors declare that we have no competing interests in the work described here. The participation of the Suntory Global Innovation Center Limited does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Cells recognize and respond to external mechanical signals in their surrounding microenvironment such as an extracellular matrix (ECM). There have been various reports into the effects of mechanical forces *in vitro*, including applied substrate strain and substrate stiffness, on cell behaviors. Various cell types such as endothelial cells, fibroblasts, and smooth muscle cells exhibit morphological changes and migration that are dependent on substrate stiffness although the preferential stiffness range is dependent on cell type. Recent studies have revealed that neural cells including several types of neurons and glial cells, which are the main constituents of the central nervous system (CNS), respond to mechanical cues. There have been various reports on how the mechanical properties of ECM affect neuronal development, growth, disorders and regeneration. One example is traumatic injury to the CNS, which changes the mechanical properties of the surrounding environment. Injury to the CNS induces the formation of glial scars, which are mainly composed of glial cells that are responsible for the local immune response and wound healing processes. It has been reported that a glial scar prevents neurons from regenerating and elongating their neurites due to many obstacles including an injury environment filled with deleterious factors. Although the role of the glial scar is under discussion, a change in its mechanical properties seems to act as a mechanical barrier to axon regeneration. As another example, devices implanted into the body become encapsulated due to a foreign body reaction. In the CNS, this can lead to loss of functionality in electrodes since a mechanical mismatch between nervous tissue and the devices induces the formation of glial scars. From the above, mechanical matching between nervous tissue and a lesion or implant should be considered as regards developing neuronal regeneration and bio-interfaces. Although there have been reports on neurite elongation and the dependence of its molecular pathways on the mechanical properties of the surrounding environment, less is known about how the mechanical properties affect the first morphological step of neurite formation. Neurite extension from a cell body is the first step in forming a functional nervous system and realizing neuronal regeneration. When neurons generate and grow, they initially attach themselves to the surrounding ECM and sprout neurites from a spherical cell body. Neurite extension, which is differentiated in the axon and dendrites, forms the basis of proper neuronal connectivity and brain function. Morphological changes in hippocampal neurons from neurite initiation to neurite extension have been well studied *in vitro*. The neuronal development stages can be classified as follows. First, immature neurons with a localized bud from a cell membrane are classified as stage 1. The transformation of a bud into neurites of approximately equal lengths is classified as stage 2. Neurons that possess one neurite considerably longer than the rest are classified as stage 3. Neuritogenesis occurs during stage a 1 to 2 transition. First, the rearrangement of F-actin assembly in the bud induces the leading edge of the cell body to protrude. Next, microtubules and other components must be engaged with the F-actin assembly; specifically, F-actin bundles can facilitate microtubule elongation. Finally, the proper consolidation of the proteins leads to neuritogenesis. Actin and microtubules are the main components of neurites, and the proper organization of F-actin assembly occurs prior to microtubule organization in the process of neurite formation. Although a variety of neuronal shapes appear during development, the initial sprouting of neurites seems to follow the morphological criteria described above. When neurons are cultivated on a substrate *in vitro*, they initially attach themselves to the substrate and form lamellipodia and filopodia, which are a sheet-like extension of a crosslinked F-actin meshwork at the leading edge of cells and a thin protrusion of F-actin-bundles at the periphery of cells and growth cones, respectively. F-actin organization plays an important role in initiating neurites from the cell body. Here, we investigated how the mechanical properties of the microenvironment affected neuronal morphologies and the F-actin structures of neurons during neurite initiation by preparing polyacrylamide gel substrates with the various stiffness ranging from similar to greater than brain stiffness. We hypothesized that the substrate stiffness can affect neurite initiation if injury-induced mechanical changes in the surrounding environment are related to the suppression of axonal regrowth or regeneration in the CNS. Since the F-actin cytoskeleton is an important structure for neurite initiation as described above, we focused on the organization of the F-actin cytoskeleton of neurons on gel substrates. A detailed investigation of the structures of the F-actin cytoskeleton reveals that the F-actin organization depends on the substrate stiffness, and the stiffness-dependent F-actin structures regulate neuritogenesis. # Materials and methods ## Materials The sources of the materials, chemicals and antibodies used in this work were as follows. Coverslips (No. 1, 18 mm x 18 mm) were purchased from Matsunami Glass, Japan. Acrylamide (AAm), N,N’-methylenebisacrylamide (Bis), and 3-(trimethoxysilylpropyl) methacrylate were purchased from Tokyo Chemical Industries, Japan. A photoinitiator, lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP) was synthesized in accordance with a previous report. Poly-*D*-lysine hydrobromide (PDL) and triton X-100 were obtained from Sigma-Aldrich. Blebbistatin (BS) and cytochalasin D (CD) were purchased from WAKO Pure Chemicals, Japan. ProLong Diamond as a mounting reagent, Sulfosuccinimidyl 6-(4'-azido-2'-nitrophenylamino)hexanoate (sulfo- SANPAH), *L*-glutamine, trypsin, glutamate, gentamycin, neurobasal medium, and B27 supplement were obtained from Thermo Fisher Scientific, USA. Mouse anti-β- III tubulin and phalloidin labeled with green fluorescence dye were purchased from R&D Systems, USA and Cytoskeleton, USA, respectively. Wistar rats were obtained from Charles River Laboratories, Japan. ## Preparation of hydrogel substrate Hydrogel substrates were prepared using the following procedures. All cover glasses were used after overnight treatment with 0.1 M NaOH followed by O₂-plasma treatment. The cleaned cover glasses were immersed in silane coupling solution containing 0.5% 3-(trimethoxysilylpropyl) methacrylate for 2 hours at room temperature. The silanized cover glasses were annealed at 110 ^oC for 30 min. We prepared two types of hydrogel precursor solutions that contained 5% w/v AAm with 0.05, 0.1, 0.15, 0.2, 0.25, and 0.5% w/v Bis for soft gels and 0.5% w/v Bis for stiff gels. For photo-initiated polymerization, LAP was added as a photoinitiator to the precursor solutions at a final concentration of 1 mM. The precursor solution was dropped on the silanized coverslip and sealed with a cleaned coverslip with 11-μm-thick spacers. The substrate was irradiated with 360 nm wavelength UV light at 10 mW/cm² for 10 min. After the light irradiation, the substrates were immersed in PBS and gently agitated at 50 rpm and 25 ^oC over 3 days. For comparison with previous reports about the neuritogenesis of neurons on glass substrates, we used poly-D-lysine as adhesive molecules. To display the PDL on the surface of the hydrogel, sulfo-SANPAH was used as a crosslinker. After a 10-min UV irradiation of the sulfo-SANPAH solution on the hydrogel, 1 mg/mL of PDL was applied to the hydrogel surface. ## Cell culture All animal experiments were approved by Biological Safety and Ethics Committee of NTT Basic Research Laboratories (approval ID 2014–04). Wistar rats (embryo day 18) were used to obtain hippocampal cells. They were used immediately after receipt, and anesthetized with a gas mixture of 1–3% isoflurane and air during preparation. Every effort was made to minimize suffering. The hippocampus was extracted from rat brain, and then treated with 2.5 mg/ml trypsin for 10 min at 37 ^oC. The cells were then centrifuged at 1000 rpm for 5 min and triturated with a pipette. The culture was carried out with a neurobasal medium that consisted of 74 μg/ml *L*-glutamine, 25 μM glutamate, 50 μg/ml gentamycin and 2% B27 supplement. The cell suspension was applied to gel substrates with an initial cell number of 15000 cells/substrate, and cultured at 37 ^oC and in 5% CO₂, with saturated humidity. ## Measurement of elastic modulus of hydrogel substrates A customized atomic force microscope (AFM) equipped on an upright optical microscope (Eclipse FN1, Nikon, Japan) was used to measure the Young’s modulus, *E*, of the gel substrates. A rectangular cantilever (BioLever mini, BL- AC40TS-C2, Olympus, Japan) with a nominal spring constant of less than 0.1 N/m was used. The loading force was determined using Hook’s law by multiplying the cantilever deflection by the spring constant calibrated using a thermal fluctuation method. The force curve measurements were examined in a 300 μm × 300 μm scan region at a maximum loading force of around 1.5 nN. *E* was estimated from the observed force-distance curves with the Sneddon’s modulation of a Hertzian contact model as a conical indenter, which is expressed as: $$F = \frac{2\tan\alpha}{\pi}\frac{E}{(1 - \nu^{2})}\delta^{2},$$ where *F* is the loading force, *δ* is the indentation depth, α is the half angle of the conical probe of 17.5°, and *ν* is the Poisson’s ratio of the gel substrate, assumed here to be 0.5. ## Fluorescence staining To count neurons without neurites on a polyacrylamide gel substrate, we used calcein-AM and propidium iodide (PI) to stain living and dead neurons, respectively. These fluorescent dyes were directly added to samples in a culture medium. After 30 minutes’ incubation at 37 ^oC in a 5% CO₂ atmosphere, the cells were observed with a fluorescent microscope. The cells were immunofluorescently stained as follows. All the samples were fixed with 4% paraformaldehyde in PBS, permeabilized with 0.5% Triton X-100 in PBS, and then blocked with a mixture of 5% NBS and 1% BSA in PBS at room temperature. After the blocking treatment, primary antibodies were applied to the cells followed by washing and incubation with appropriate secondary antibodies bound to fluorescent dyes. For F-actin staining, the cells were treated with 0.5% TritonX-100 in PBS, and then stained with phalloidin bound to fluorescent dyes. Fluorescently labeled samples were mounted with a mounting reagent. To distinguish neurons from the other hippocampal cells, neuron-specific marker β-III tubulin with anti-β-III tubulin antibody and 500 times dilution. It was a marker for neuritogenesis because it is the main component of a microtubule in neurites. The antibodies were labeled with species-specific secondary antibodies conjugated with Alexa 568 with 400 times dilution. ## Fluorescent imaging Fluorescent images were obtained using a fluorescent microscope (Eclipse TE3000, Nikon, Japan) with a CMOS camera (ORCA flash 4.0, Hamamatsu Photonics, Japan), and a laser scanning confocal microscope (LSM510, Carl Zeiss, Germany or IX81, Olympus, Japan). The microscope setup for the spinning disc superresolution microscope (SDSRM) is based on a disk-scanning confocal microscope system, which includes an IX81 and a disc-scanning unit (IX2-DSU, Olympus, Japan). To investigate neuritogenesis and F-actin bundle formation, acquired images were analyzed with ImageJ (NIH, <http://rsb.info.nih.gov/ImageJ>). ## Statistical analysis All experiments were performed using at least three independent donors and three replicate gel substrates. To obtain a statistical analysis of neuritogenesis, we classified neurons that had protrusions with a diameter less than that of the cell body as stage 1. Protrusions whose diameter exceeded that of the cell body were classified as neurites. Neurons with neurites of approximately equal length were defined as stage 2. Neurons that possessed one neurite at least twice as long as the rest were classified as stage 3. Statistical comparisons were performed using an independent t-test when filopodia density and length were compared, and one-way analysis of variance (ANOVA) with Bonferroni’s *post hoc* testing was used to make pairwise comparisons between multiple groups. The statistical significance was set at *p* \< 0.05. # Results ## Characterization of gel substrates A hydrogel substrate with a homogeneous elastic modulus and defined thickness is required if we are to investigate how substrate stiffness affects neuritogenesis. To characterize the gel substrate, the Young’s modulus, *E*, was measured with AFM with different crosslinker concentrations. The results were shown in. E for 0.05% BIS concentration was 1.7 × 10² (± 0.3 × 10²) Pa, which was comparable to *in vivo* brain stiffness. E for 0.1% BIS concentration was 2.2 × 10³ (± 0.3 × 10²) Pa, and E for more than 0.15% BIS concentration was 3.2 × 10³ (± 0.5 × 10³) Pa, which was one order of magnitude higher than the brain’s stiffness. ## Neuritogenesis on polyacrylamide gel substrates As described in the Introduction, neuronal developmental stages *in vitro* can be divided from neuronal morphologies. To obtain a statistical analysis of neuritogenesis on the gel substrates, we defined neurons with the protrusions that were less than the diameter of the cell body as stage 1 and elongation of the protrusions equal to the diameter of the cell body as neurite initiation or neuritogenesis. We counted the number of neurons in stage 1 from fluorescent images of neurons stained with calcein-AM and PI, which allowed us to distinguish living neurons from dead ones in stage 1. Fluorescence observations showed that the cellular viability on each gel substrate was not statistically significant. The neurons in stage 1 as a percentage of the total number of cells on the gel substrates are shown in. The graph shows that the percentage of neurons in stage 1 on stiff substrates at 26 hours after plating was 17%±1.8% for 0.05% BIS, 19%±3.2% for 0.1% BIS, 55%±2.8% for 0.15% BIS, 66%±2.5% for 0.2% BIS, 65%±2.8% for 0.25% BIS, and 66%±1.0% for 0.5%BIS. The results indicate that the neuritogenesis on the gel substrates with elastic moduli more than 3.0 kPa is significantly suppressed. To investigate the stiffness dependent suppression of neuritogenesis, we hereafter used gel substrates prepared from 0.05% BIS as the standard substrate (soft substrate) and 0.5% BIS as the stiffer gel substrate (stiff substrate). We statistically analyzed the further development of neurons on soft and stiff substrates at 20 hours after plating. For the analysis, protrusions that exceeded the diameter of the cell body were classified as neurites. Neurons with neurites of approximately equal length were defined as stage 2. Neurons that possessed one neurite at least twice as long as the rest were classified as stage 3. The percentages of neurons in stage 2 were 48% ± 10% for the soft substrate and 23% ± 4% for the stiff substrate, whereas the percentages in stage 3 were 26% ± 14% for the soft substrate and 2.6% ± 3% for the stiff substrate. The differences are statistically significant. The results suggested that substrate stiffness affected neurite elongation and development as well as neuritogenesis. To investigate neuronal morphologies, the neurons were immunofluorescently labeled with phalloidin and anti-βIII-tubulin for F-actin and microtubule, respectively. Fluorescent images of neurons on soft and stiff substrates at 20 hours after plating are shown in, and. respectively. On soft substrates, fluorescent images revealed the co-existence of neurons with and without red fluorescence from anti-βIII-tubulin, indicating the formation of neurites. The neurons had thin and short protrusions with green fluorescence from phalloidin, indicating F-actin fibers (black arrow head). Compared with the soft substrates, fewer neurons had protrusions with red fluorescence, and F-actin meshworks with green fluorescence were formed at the cell periphery on stiff substrates (white arrow head). These fluorescent images indicated that the stiffening of the surrounding area induces neuronal morphological changes, especially F-actin organization. ## F-actin cytoskeleton structure To assess whether F-actin organization induced by substrate stiffness is involved in neurite initiation, the F-actin structures were observed in detail with a spinning disk superresolution microscope (SDSRM). show representative fluorescent images of neurons on polyacrylamide gel substrates. At developmental stage 1, neurons on stiff substrates had larger areas of circumferential F-actin meshworks as shown in compared with the soft substrates shown in. Some neurons on the stiff substrates had a F-actin structure with two distinct parts. One part indicated by a black arrowhead consisted of F-actin meshworks and the other indicated by a white arrowhead consisted of condensed F-actin meshworks, known as F-actin arcs. F-actin arcs were not observed on the soft substrates with our experimental setup. At developmental stage 2, neurons on the stiff substrates had larger growth cones at the tips of neurites and larger areas of F-actin meshworks than soft substrates. On the other hand, fewer F-actin bundles sprouted from the leading edge of a cell body and neurites on the stiff substrates. The fluorescent images showed that F-actin bundle formation was suppressed on stiff substrates. To analyze the density and length distribution of F-actin bundles statistically, we counted the number and measured the length of F-actin bundles protruding from the perimeter of neurons on a soft or a stiff substrate. The graph in, shows the number of F-actin bundles per 10 μm at the perimeter of a neuron. Neurons on stiff substrates have F-actin bundles with significantly lower densities than those on soft substrates. In addition, the average length of the F-actin bundles on stiff substrates (2.5±1.6 μm) are significantly shorter than those on soft substrates (4.4±2.5 μm). It has been reported that the prevention of filopodia formation suppressed neuritogenesis because filopodia can guide microtubule elongation during neuritogenesis. Our results imply that the suppression of F-actin bundle formation on stiff substrates is involved in neuritogenesis. Moreover, stiffness of the gel substrates induced a change in the cytoskeletal F-actin organization including the formation of a larger area of F-actin meshworks and F-actin arcs. Therefore, we hypothesized that F-actin organization induced by substrate stiffness can play a role in regulating neuritogenesis. ## Blebbistatin treatment We investigated the way in which the organization of the F-actin cytoskeleton, and in particular the circumferential F-actin arcs and the suppression of the F-actin bundles, are associated with neuritogenesis. We used blebbistatin (BS), which is known to be a myosin II inhibitor and inducer of filopodia formation. Since myosin II activity contributes to the compaction of F-actin meshworks and the consequent formation of F-actin arcs, BS treatment can suppress F-actin arc formation. BS was exposed at a concentration of 100 μM, which was sufficient to attenuate the interaction between F-actin and myosin II, immediately after cell plating. The neuritogenesis of neurons on gel substrates after BS exposure was analyzed statistically by counting the number of neurons in stage 1 at 20 hours after plating. As control experiments, neurons on gel substrates were exposed to dimethyl sulfoxide (DMSO), which did not affect neuritogenesis or the F-actin structures. The graph shows that the percentage of neurons in stage 1 on both the soft and stiff substrates decreased significantly after the BS (from 36% to 17% for soft substrates, and from 55% to 29% for stiff substrates). Although the BS treatment improved neuritogenesis, the percentage of neurons in stage 1 on stiff substrates was higher than that on soft substrates, indicating the suppression of neuritogenesis. To investigate BS-induced neurite protrusions, an SDSRM was used for a detailed observation of the F-actin structures after BS treatment on the stiff substrate. We could not find significant structural changes of F-actin organization on the soft substrates after BS treatment when compared with neurons without BS exposure. On the other hand, on stiff substrates, although the circumferential actin arcs disappeared, F-actin meshworks were observed at the edge of a cell body. The F-actin meshworks were segmented, and the widely spread growth cones observed before BS treatment were barely seen. Parallel arranged F-actin organization was observed in the segmented F-actin area. These F-actin structures were not observed on the soft substrate. These results indicate that F-actin structures induced by the attenuation of myosin II activity are associated with neuritogenesis, although the myosin II activity is strongly involved in neurite initiation. ## Cytochalasin D treatment Although the exposure of neurons on gel substrates to BS revealed that F-actin arcs at the leading edge of a cell body suppressed neuritogenesis, it remained unknown whether the other aspects of F-actin organization, which was observed after the BS treatment, were related to neuritogenesis. Therefore, we attempted to completely suppress F-actin organization by treating neurons with cytochalasin D (CD), which induces the disruption of actin filaments and prevents actin polymerization by binding to the barbed end. CD was exposed at a concentration of 100 nM, which was sufficient to disrupt the F-actin structures at the leading edge of the cell body, immediately after cell plating. To assess whether CD treatment affected neuritogenesis, we counted the neurons without neurites on the hydrogel substrates. The control results were the same as those in. The exposure of CD to neurons on soft substrates significantly impaired neuritogenesis whereas it significantly improved neuritogenesis on stiff substrates (36% to 41% for soft substrates and 55% to 40% for stiff substrates). In addition, the percentages of the soft and stiff substrates after CD treatment had no statistical significance. To investigate the way in which CD treatment affected the F-actin organization of the neurons on the soft and stiff substrates, the structure of the F-actin cytoskeleton was observed in detail with an SDSRM. Neurons on both soft and stiff substrates had similar morphologies after CD treatment. There was no F-actin bundle formation and no meshwork on the gel substrates, indicating that the CD concentration in was sufficient to disrupt the F-actin structures at the leading edge of the cell body. The suppression of F-actin organization induced the suppression of neuritogenesis on the soft substrates but the acceleration of neurite initiation on the stiff substrates. As F-actin bundles can accelerate neuritogenesis, the suppression of F-actin bundle formation on the soft substrates resulted in the suppression of neuritogenesis. On the other hand, whether or not CD treatment improved neuritogenesis on the stiff substrates remains controversial. CD- induced changes in the F-actin organization on the stiff substrates consisted of the disruption of F-actin meshworks and arcs at the leading edge as well as the suppression of F-actin bundles. This suggests that F-actin structures formed on the stiff substrates help prevent the suppression neuritogenesis. # Discussion It has been reported that the mechanical properties of ECM in various tissues varied during several events such as external stimuli, aging, and diseases. With respect to the CNS, understanding the relationship between ECM mechanical properties and neural functions is an important issue in the field of tissue engineering and neuroscience. This study demonstrates the effects of substrate stiffness on the neuritogenesis of hippocampal neurons because neuritogenesis is the first step towards development and regeneration. Also, we investigated structure of the F-actin cytoskeleton as proper arrangements of the F-actin assembly are required prior to organization of the other proteins related to neurites. To accomplish this, we employed a polyacrylamide gel as a substrate with a mechanical property ranging from similar to a hippocampus to a stiffer one. As an adhesion molecule, we used PDL which can act as a non-selective focal adhesion activator. Previous study reported that adhesion molecules derived from ECM proteins such as laminin, fibronectin and collagen, initiate integrin- mediated cell binding via the formation of focal adhesions, which is a key structure as regards regulating mechanotransduction and activating signaling pathway of neuritogenesis. For example, laminin can facilitate neurite initiation and axonal outgrowth compared with fibronectin. Moreover, laminin can rescue the neuritogenesis of neurons whose neurite initiation is genetically inhibited whereas collagen and fibronectin have no effect. These findings indicate that distinct ECM proteins activate distinct signaling pathways related to neuritogenesis. On the other hand, PDL modulates cell adhesion via an electrostatic interaction between a negatively charged cell membrane and a positively charged PDL. The PDL-mediated non-selective activation of focal adhesion allows us to eliminate the possibility that specific activation of a certain signaling pathway induced by ECM proteins affects suppression or acceleration of neritogenesis. We found that a stiff substrate with an elastic modulus exceeding 3.0 kPa changed the neuronal morphologies and organization of the F-actin cytoskeleton, and suppressed neuritogenesis. The further development of neurons from stage 2 to 3 was suppressed on the stiff substrate. Noteworthy features of the morphologies of neurons on the stiff substrates were that they had fewer F-actin bundles and the formation of more circumferential F-actin meshworks and arcs at the leading edge. According to a previous report stating that sprouts of F-actin bundles are required for neuritogenesis, the suppression of F-actin bundle formation on stiff substrates prevents microtubules from invading and protruding from the leading edge. Together, F-actin meshworks and arcs on stiff substrates appeared to suppress neurite protrusions. Regarding the further development from stage 2 to 3, the neurite elongation at the leading edge of the growth cones can be suppressed by the stiffer substrate-induced F-actin structures. To assess our assumption that stiffer substrate-induced F-actin organization plays a critical role in neuritogenesis, we treated neurons on polyacrylamide gel substrates with BS to inhibit the formation of actin arcs. The disruption of F-actin arcs is observed on stiff substrates after the BS treatment. It is known that neurons, which genetically lack the ability to form filopodia, fail to initiate neurites, and they form F-actin arcs at the leading edge. In the report, BS treatment disrupts the F-actin arcs and rescues neuritogenesis. These results are similar to ours. Although we must analyze protein expression if we are to discuss the relationship between previous reports and our results, substrate stiffness may regulate protein expression relating to cytoskeletal organization. Also, BS treatment accelerates F-actin bundle formation and improves neuritogenesis regardless of the substrate stiffness. Since an F-actin bundle assists microtubule protrusion at the leading edge of the cell body as described above, the acceleration of F-actin bundle formation can help to improve neuritogenesis. After BS treatment, neuritogenesis on stiff substrates was suppressed compared with that on soft substrates even though the formation of actin arcs was prevented and F-actin bundle formation was accelerated. This suggests that the F-actin structure at the leading edge of neurons on the stiff substrates after BS treatment prevents microtubule protrusion. To prevent the F-actin organization at the leading edge, neurons on the gel substrates were treated with cytochalasin D. After the CD treatment, the percentages of neurons in stage 1 on the soft and stiff substrates are similar, indicating the suppression of neuritogenesis on soft substrates and the improvement on stiff substrates. Fluorescent images revealed that the structures of F-actin cytoskeletons on stiff substrates were similar to those on soft substrates after the CD treatment. The suppression of neuritogenesis on the soft substrates strongly supports the view that F-actin bundles play a beneficial role in sprouting neurites. On the other hand, neuritogenesis on the stiff substrates was improved even though the formation of the F-actin bundles was suppressed. This supports the idea that the F-actin meshworks and arcs prevent microtubules from sprouting from the leading edge. It is known that the elongation process in the growth cones takes place preferentially where the actin meshwork is unstable, whereas the stable actin meshwork tends to impair microtubule protrusion. We propose a neuritogenesis mechanism that is dependent on substrate stiffness as shown in. F-actin cytoskeletons such as F-actin meshworks and arcs at the leading edge of the cell body were preferentially organized on the stiff substrates, whereas F-actin bundle formation was suppressed. The F-actin structures on the stiff substrates are obstacles to the protrusion of microtubules from the leading edge of the cell body. Therefore, on a stiff substrate, the disruption of F-actin organization by CD had an effect on neuritogenesis even though F-actin bundle formation was suppressed. # Conclusions We investigated how substrate stiffness affected neuritogenesis. The high- resolution microscopy observations showed that stiff substrates significantly suppressed neuritogenesis due to their stronger myosin II-based contractility and stabilized the F-actin cytoskeleton. To our knowledge, this is the first observations regarding the organization of neuronal F-actin cytoskeleton regulated by the mechanical properties of the surrounding environment. The stiffness-induced F-actin structures played a key role in regulating neuritogenesis. Our findings suggest that the suppression of neuronal regeneration at a glial scar could be influenced in part by a stiffening microenvironment, which suppresses neuritogenesis and development to the later stage. Further investigations are required to understand the neuronal development because various proteins are properly associated and dissociated during neurite development. Of the several processes, the interaction of multiple proteins with F-actin assembly can be important as regards protrusions maturing into neurites and developing neurons from stage 2 to stage 3, corresponding to axon formation. In this study, we qualitatively discuss the circumferential F-actin meshworks and arcs because we did not observe the structures on the soft substrates with our experimental setup. It is known that the low contractile force at the interface between a cell and a substrate suppresses the formation of stable adhesion and cytoskeleton assembly at the interface. This is why no F-actin meshworks were observed on the soft substrates. We must investigate the dynamics of F-actin structures and other proteins such as microtubules at the leading edge during neuritogenesis by time-lapse imaging to verify our proposed model. The 3D cultivation of neurons on artificial platforms such as extracted ECM, collagen, and biocompatible polymers has been reported. However, less is known about how the mechanical properties of ECM affect neuronal development in 3D cultivation. Although our experimental methods cannot be applied to 3D experiments due to the toxicity of gel monomers, our findings suggest that ECM stiffness can affect the F-actin structure of neurons in 3D cultivation. This study provides an insight not only for developing a scaffold for neuronal regeneration, but also for designing a compliant interface between tissue and a device such as a brain-machine interface. # Supporting information We thank Ms. Yuriko Furukawa for preparing the samples, Dr. Misaki Takahashi for helpful artwork, and members of the Molecular and Bio Science Research Group at NTT Basic Research Laboratories. We also thank Mr. Kohei Hirono for experimental assistance, and members of the Cellular and Tissue Engineering Laboratory at Hokkaido University for fruitful comments and discussions. [^1]: NTT does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction Chronic hypertension is a relatively common disorder occurring in approximately 1–5% of pregnant women; rates depending on the population studied and the criteria used for the diagnosis. Because of increasing maternal age, obesity, and type 2 diabetes worldwide, it is expected that the prevalence of chronic hypertension in pregnancy will continue to increase. The study ENNS (National Nutrition Health Survey) cross-sectional survey in France between 2006 and 2007 reveals a prevalence of chronic hypertension of 4.1% in women between 18 and 34 years and of 8.3% between 35 and 44 years. Hypertension was known to the patient in 22.3% of cases between 18 and 34 years and in 55.5% of cases between 35 and 44 years. Pregnancies complicated by chronic hypertension are at increased risk of superimposed preeclampsia, abruptio placenta, fetal growth restriction, preterm delivery, and perinatal death,. In women with chronic hypertension, the risk of superimposed preeclampsia is increased in black ethnic origin, raised body mass index (BMI), smoking, booking systolic blood pressure of 130 to 139 mm Hg, and diastolic blood pressure of 80 to 89 mm Hg, and in women with chronic hypertension ≥4 years. Conflicting results have been published on the relationship between a history of preeclampsia and the rate of superimposed preeclampsia in subsequent pregnancies. These differences might be related to inclusion of heterogeneous population of women with hypertension; in some studies women had only essential hypertension whereas in others women had all forms of hypertension. In addition, most of the studies included women who were diagnosed with chronic hypertension on the basis of either having hypertension prior to pregnancy or during the first 20 weeks gestation. Moreover, none of the studies reported to date have included only women who received antihypertensive medication prior to conception. The objective of this study was to identify risk factors for superimposed preeclampsia at first prenatal visit in women with essential chronic hypertension receiving antihypertensive medication prior to conception. # Methods This retrospective study included women with chronic hypertension delivered between 1 January 2004 and 31 December 2007 who were identified from the hospital computer databases of two university hospital centers (CHI Creteil and AP-HP Cochin Port-Royal Paris). Every medical chart was reviewed to collect the data. The criteria used to select women with chronic hypertension was a diagnosis of hypertension that needed a treatment before the onset of the pregnancy. Exclusion criteria were: women with multiple pregnancies, women with secondary hypertension, women with proteinuria at less than 20 weeks’ gestation, women considered as having a chronic hypertension but without any treatment at first prenatal visit, women transferred from other maternities, pregnancies complicated by fetal malformations. The data collected from medical records included: age, pre-pregnancy BMI, parity, ethnic origin, tobacco use during pregnancy, duration of hypertension, past obstetric history, antihypertensive treatment, treatment with low dose aspirin, maternal systolic and diastolic blood pressure at booking, presence or absence of proteinuria at first prenatal visit, maternal and neonatal outcomes. The blood pressure was obtained with automated device, patient in sitting up position. The mean arterial pressure was calculated from brachial systolic and diastolic blood pressure (BP), according to the following formula: Mean arterial pressure = \[systolic BP+(2\*diastolic BP)\]/3. Superimposed preeclampsia was defined as a new onset proteinuria (0.3 g of protein or more in a 24-hour specimen) after 20 weeks’ gestation and without proteinuria early in pregnancy (less than 20 weeks’ gestation). Fetal growth restriction (FGR) was defined as a birth weight \<5th percentile. Abruptio placenta was diagnosed according to clinical findings and/or placental examination. HELLP syndrome (hemolysis, elevated liver enzymes, low platelet count) was defined according to Sibai’s criteria. We analyzed the risk factors that may influence the rates of superimposed preeclampsia. Categorical variables are presented as percentage, and continuous variables as mean and SD. Categorical variables were compared with *X²* square or Fisher’s exact test and continuous variables with a two-tailed student *t* test. Variables with a p\<.1 were included in a logistic regression analysis. Data are expressed as odds ratio (OR) with 95% confidence interval (CI). P\<.05 was considered as significant. Positive likelihood ratio with 95% CI was calculated for the prediction analysis. The statistical software Statview 5.0 (SAS Institute) was used for statistical analysis, and REALbasic (2002) for Receiver Operative Characteristics (ROC) analysis. The evaluation of chronic hypertensive women did not need approval of our ethical committee because it was a retrospective analysis of data in women who received standard management at both hospitals. Therefore, our ethical committee has waved requirement for approval. # Results The study population consisted of 362 women with chronic hypertension. Women transferred from other centers (n = 61), women with secondary hypertension or with chronic hypertension and pre-existing proteinuria before 20 weeks’ gestation (n = 20), women with fetal anomalies (n = 4), and women not receiving antihypertensive medication prior to pregnancy (n = 66) were excluded. The analysis concerned exclusively the 211 women with essential chronic hypertension who were receiving antihypertensive therapy prior to conception. describes maternal characteristics at booking (first visit) and summarizes overall maternal and perinatal outcomes. No maternal death occurred among the women. The rate of superimposed preeclampsia was 23.2%. All women with HELLP syndrome had preeclampsia. shows perinatal outcomes according to the occurrence of superimposed preeclampsia. The following variables obtained at first prenatal visit: black ethnicity, nulliparity, previous preeclampsia, previous FGR, systolic and diastolic blood pressure at booking, mean arterial blood pressure ≥95 mmHg at booking, and duration of hypertension ≥4 years were significantly associated with the occurrence of superimposed preeclampsia in univariate analysis. In contrast, the rate of superimposed preeclampsia was not affected by maternal age, BMI, tobacco use during pregnancy, antihypertensive medication (one drug only versus several drugs), and aspirin treatment. We have chosen to include mean arterial blood pressure ≥95 mmHg at multivariate analysis and to analyze its predictive status, rather than systolic or diastolic blood pressure, because mean arterial blood pressure had the best prediction analysis for superimposed preeclampsia \[area under curve at ROC analysis of.72 (p\<.0001) for mean arterial blood pressure.69 (p\<.0001) for diastolic blood pressure, and.68 (p = .0001) for systolic blood pressure\] and 95 mmHg was the best cut-off for sensitivity and specificity. Multivariate analysis showed that previous preeclampsia, and mean arterial blood pressure ≥95 mmHg were independently associated with the occurrence of superimposed preeclampsia. We analyzed the prediction for the occurrence of superimposed preeclampsia using the variables that were selected in logistic regression for superimposed preeclampsia at first prenatal visit: previous preeclampsia and mean arterial blood pressure. The ROC analysis of mean arterial blood pressure and previous preeclampsia showed an area under curve at of.77 (p\<.0001). The prediction analysis of previous preeclampsia and MAP≥95 mmHg is shown in. When both variables were present, sensitivity, specificity, positive predictive value, negative predictive value, and likelihood ratio for superimposed preeclampsia were 43%, 94%, 70%, 85%, and 7.71 (95% CI: 3.20–18.57), respectively. Of the 49 patients with superimposed preeclampsia, 19 occurred at less than 34 weeks’ gestation. Of all variables analyzed, only SBP at booking was associated with increased risk of early onset compared to late onset of preeclampsia (SBP≥140 mmHg: 74% vs. 27%, SBP 130–140 mmHg: 21% vs. 40%, and SBP\<130∶5% vs. 33%, respectively, p = .004). # Discussion The main goal of this study was to analyze factors obtained in the first half of the pregnancy that may be associated with superimposed preeclampsia in women with essential chronic hypertension that needed a treatment before the onset of pregnancy. In contrast to other studies, we have focused on essential chronic hypertension women receiving antihypertensive therapy prior to pregnancy for two reasons: 1- the diagnosis of chronic hypertension was already made, 2- the results are not biased by the primary disease as it may be observed in secondary chronic hypertension related to nephropathy or auto-immune disease. In addition, only two studies have focused on women with chronic essential hypertension, one of these studies did not specify whether or not patients were treated by antihypertensives, and in the second one almost half of the women had no antihypertensives before the onset of pregnancy. The main findings of the study were: 1- superimposed preeclampsia occurred in 23.2% of women, 2- previous preeclampsia, and mean arterial blood pressure ≥95 mmHg at booking was associated with increased risk of superimposed preeclampsia. The rate of superimposed preeclampsia of 23.2% observed in our study is in agreement with that reported in other studies, ranging from 17% to 34.9%. In several of these studies, conflicting results concerning the relationship between a history of previous preeclampsia and the recurrence of preeclampsia in a subsequent pregnancy have been reported. Sibai et al. showed that the risk of superimposed preeclampsia was greater in chronic hypertensive women who had preeclampsia during a previous pregnancy (32% vs. 23%; OR: 1.6; 95% CI: 1.1–2.3), as well as Chappell et al. (OR = 1.95, 95% CI: 1.25–3.04). A recent study, however, reported that in women with chronic hypertension a history of preeclampsia was not associated with increased risk of superimposed preeclampsia (OR = 1.28, 95% CI: 0.78 to 2.1). This latter study, however, had only 61% power to detect differences in rate of superimposed preeclampsia between groups. In our study, among the women who developed preeclampsia 53.3% had a history of preeclampsia in a previous pregnancy, and previous preeclampsia was clearly associated with superimposed preeclampsia in subsequent pregnancy in the logistic regression analysis. Along with others, blood pressure at first prenatal visit (systolic, diastolic, and mean arterial) was associated with increased risk of superimposed preeclampsia at univariate analysis. The prediction of a history of preeclampsia, and mean arterial blood pressure ≥95 mmHg for superimposed preeclampsia in our population were encouraging, particularly, when both variables were present at first prenatal visit. Although sensitivity was only of 43%, positive predictive value of 70% was extremely high, and the likelihood ratio to have a superimposed preeclampsia was 7.71-fold increase. These two variables simply obtained early during the pregnancy may help to select women at very high risk of superimposed preeclampsia that might be eligible for low dose aspirin treatment. Although the usefulness of low dose aspirin in the prevention of superimposed preeclampsia is not established in women with chronic hypertension, this may be related to the late start (after 20 weeks’ gestation) of the treatment; and the use of low dose aspirin in this subgroup could be beneficial, particularly if started before 16 weeks’ gestation. Maternal age and parity had no significant effect on the risk of superimposed preeclampsia in our study. A similar lack of effect of maternal age in women with chronic hypertension was found by Sibai and colleagues. This is in contrast to recent reports where nulliparity, and maternal age were associated with increased risk of preeclampsia in unselected women. Therefore, it is not clear whether multiparity or lower maternal age are protective in pregnant women with essential chronic hypertension, highlighting if necessary that surveillance should be heightened for these women. Duration of hypertension prior to pregnancy of more than 4 years has been associated with increased risk of superimposed preeclampsia. Our study, however, did not confirm this finding, and is in agreement with a recent one. The prevalence of black ethnicity (53.6%) in chronic hypertensive women is quite high and it does not reflect our general population. It is, however, important to note that ethnicity was not independently associated with superimposed preeclampsia in our study. In several studies, pregnant women with essential chronic hypertension had high rates of fetal growth restriction (\<5th birth weight percentile) without superimposed preeclampsia. In our study where 44 babies were \<5th birth weight percentile, only 12 (20.7%) were observed in women having superimposed preeclampsia whereas the latter 32 (17.2%) were found in women without superimposed preeclampsia (non significant difference). These findings show that women with essential chronic hypertension are at risk for severe fetal growth restriction irrespective of the occurrence of superimposed preeclampsia. The explanation may be an overcorrection of blood pressure suggested by some authors, and in our study, treatment was stopped for 16 patients because of a SBP under 120 mmHg and DBP under 70 mmHg. A recent meta-analysis, however, did not find such relationship. Our study has some limitations. The number of patients is relatively small. It is to note, however, that our population concerned exclusively women with essential chronic hypertension who had a treatment at first prenatal visit, and this is not the case of the majority of published studies. Another limitation is related to the retrospective structure of the study, for instance, we did not make any confirmation on past obstetric history that were mentioned in the medical charts, and this may induce some errors in the analysis. In addition, an important number of women transferred from other centers (n = 61), and not followed in both maternities since the beginning of their pregnancy were excluded from the study because the management of antihypertensive treatment in these women could have been different. Moreover, almost all these patients had an adverse pregnancy outcome, and therefore the results would have been biased. Finally, gestational age at first prenatal visit is relatively late (18.1 wks). This result, however, reflects our clinical practice and it still remains possible to introduce some preventive treatment such as aspirin and to plan a close monitoring of the pregnancy. ## Conclusion In essential chronic hypertensive women treated before the pregnancy, previous preeclampsia, and mean arterial blood pressure ≥95 mmHg are associated with increased risk of superimposed preeclampsia. Together, these two variables may select women at extreme high risk of superimposed preeclampsia. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: EL BH. Performed the experiments: EL BH. Analyzed the data: EL BH VT FG DC BS. Contributed reagents/materials/analysis tools: VT FG DC BS. Wrote the paper: EL BH.

# Introduction The receptor tyrosine kinase EGFR (ErbB1) plays a crucial role in both cancer initiation and progression and is discussed as a promising target for cancer therapy. New insights into EGFR biology have established that EGFR signals through two distinct pathways: (i) canonical membrane-associated signaling and (ii) non-canonical nuclear signaling to regulate gene expression, DNA replication and DNA damage repair. The most intriguing step associated with nuclear EGFR signaling represents the physical translocation of the EGFR protein from the cell membrane to the perinuclear/nuclear location in response to cellular stress or stimulation with EGF. All four members of the ErbB family have been reported to undergo nuclear translocation. Several distinct functions have been described for nuclear EGFR. Nuclear EGFR (nEGFR) is reported to act as a co-transcriptional activator for cyclin D1. Furthermore, nEGFR controls proliferating cell nuclear antigen (PCNA) activity during DNA replication. In the context of DNA repair, EGFR regulates DNA-PK activity. In addition, recent data suggests a role for nEGFR in regulating mRNA stability and protein translation. Mechanistically, nEGFR regulates the ribonuclease activity of polynucleotide phosphorylase through DNA- PK-mediated phosphorylation, and consequently, the expression of the *c-myc* mRNA is increased. Moreover, the same study reported an inhibitory effect of EGFR-mediated argonaute 2 (AGO2) phosphorylation on miRNA maturation in response to hypoxia. Importantly, miRNA-mediated RNA silencing represents an effective pathway to prevent the active translation of mRNAs. This process is performed by the RISC (RNA-induced silencing complex) in the perinuclear cytoplasm. Major elements of the RISC are AGO2 and GW182 (TNRC6), which are organized in protein complexes within the nucleus, where these RNAi factors are presumably involved in regulating gene transcription and mRNA splicing. The GW182 protein acts as a platform that recruits and activates the deadenylase complex CCR4-NOT to the miRNA-directed target mRNA and facilitates the removal of the poly(A) tail. This process promotes mRNA degradation by AGO2 and inhibition of translation. Deadenylation is the major step triggering mRNA decay in eukaryotic cells. The poly(A) tail and associated poly(A)-binding protein (PABP) interact with the 5’ m⁷G-cap/cap-binding complex to form a closed loop that enhances translation initiation and protects mRNA ends from nuclease attack. Consequently, deadenylation represents an important control point for both mRNA degradation and translational silencing. As shown in the present study, nEGFR is part of the miRNA-directed cNOT1 deadenylase complex and regulates the stability and translation of mRNAs through its kinase activity. In summary, we propose a mechanism by which nEGFR triggers a fast and efficient switch in protein translation in a miRNA-directed manner. # Material and methods ## Cell culture and irradiation Experiments were performed using the A549 human bronchial carcinoma cell line (ATCC CCL-185, Manassas, Virginia, USA), the FaDu head and neck tumor cell line (ATCC, HTB43), and HSF7 normal fibroblasts. Irradiations were performed using the X-ray cabinet RS 225 (X-Strahl, Surrey, United Kingdom). The voltage was set to 200 kV, with a current of 15 mA (dose rate = 1 Gy/min at a 49 cm distance from the X-ray tube). The X-ray beam was hardened with a 0.5 mm removable copper filter. Dosimetry was performed with a farmer chamber (PTW, Freiburg, Germany). Irradiation was conducted at 37°C. Erlotinib was purchased from Selleck (Houston, Texas, USA) and EGF was purchased from Sigma-Aldrich (St. Louis, Missouri, USA). ## Subcellular protein fractionation Cytoplasmic and nuclear extracts were prepared using the NE-PER^® kit (Thermo Scientific, Waltham, Massachusetts, USA), according to the manufacturer’s instructions. ## Quantification of mRNAs enriched in complexes containing nuclear EGFR EGFR-IgG or IgG was covalently bound to an agarose support (direct IP-kit, Thermo Fisher Scientific/Pierce, \#26148). Nuclear proteins were isolated from cells treated with the indicated compounds and EGFR complexes were precipitated by direct IP, in which the EGFR antibody (clone 13, BD Biosciences) was directly linked to agarose beads. We proved a lack of nonspecific complex formation by binding a nonspecific mouse IgG (isotype IgG1, Sigma) to agarose. Enriched mRNA templates were primed with oligo(dT)_12-18 primer (Thermo Fisher Scientific) and transcribed into cDNAs using ImProm-II reverse transcriptase (Promega) in the presence of ³³P-dATP. Free ³³P-dATP was separated from labelled cDNAs using an Illustra NICK G50 Column (GE Healthcare). The incorporation of ³³P-dATP in eluted cDNAs was quantified using a scintillation counter. Incorporated radioactivity was normalized to the nuclear EGFR protein concentration. ## Western blot analysis and immunoprecipitation After irradiation, cells were lysed and proteins were resolved by SDS-PAGE. Western blotting was performed using standard procedures. All primary antibodies were diluted 1:1000 and purchased from the following sources: anti-EGFR (BD Transduction Laboratories, clone 1F4); anti-EGFR-pY992 (Abcam, ab81440), anti- lamin B1 (Biozol, clone ZL-5), anti-NFATC4 (Abcam, ab3447), anti-cNOT1 (Sigma, HPA046577), anti-GW182 (Biorbyt, orb183979), anti-AGO2 (Cell Signaling Technology, clone C34C6), anti-AGO2-pY393 (ECM Bioscience, AP5311), and anti- Actin (Sigma, A2066). Quantification was performed with the LI-COR detection system (LI-COR, Odyssey Fc). Immunoprecipitations were performed using the Pierce direct IP kit (#26148), according to the manufacturer’s instructions. Cell lysates were pre-cleared with nonspecific IgG bound to agarose beads. ## Expression of recEGFR-GST 3000–4000 containing the epitope for binding to the EGFR antibody clone 1F4 Total mRNAs were isolated from A549 cells and transcribed into cDNAs by RT-PCR. The GST coding sequence was amplified using the forward primer (Sall) `` 5`-ACGCGTCGACATAGTCGCCCAAAGTTCCGTGA-3` `` and the reverse primer (Not1) `` 5`- ATAAGAATGCGGCCGCATGCTACCAGCAAGCTTCTTCC-3´ ``. After digestion at the restriction sites, the product was cloned in frame into the pGEX-4T vector (GE Healthcare) and transformed into competent *E*. *coli* cells. Positive colonies were selected and the proper orientation and correct reading frame were confirmed by sequencing. The expression of GST fusion protein and affinity purification were performed using standard procedures. ## DNA microarray analysis A549 cells were irradiated with 4 Gy of radiation or sham irradiated, and nuclei were isolated from both samples 24 h after irradiation. Nuclei were lysed and EGFR or IgG immunoprecipitation was performed with a direct IP kit (Thermo- Fisher/Pierce, \#26148). RNA was eluted from biological triplicates, transcribed into cDNAs, and hybridized to the Human Gene 2.1 ST Array (Affymetrix). The DNA microchip analysis was performed in cooperation with the Microarray Facility Tübingen (MFT, now c.ATG <http://www.c-atg.de>), and the statistical analysis was conducted in cooperation with the Quantitative Biology Center Tübingen (QBIC, <https://portal.qbic.uni-tuebingen.de/portal/>). Bioinformatics analyses of the microarray data were performed in the statistical language R (R version 3.1.1). For QC and data normalization, the R package ‘oligo’ (version 1.28.3) was used to quantitatively normalize probe intensities across all samples using the robust multi-array average (RMA) procedure. The ‘genefilter’ package (version 1.46.1) was used to remove probes with very low variability between samples. Differential expression was analyzed using the package ‘Limma’ (version 3.20.9,). For the Limma analysis, a linear model was fitted to the log2 expression data for each probe using the following formula: expr \~ construct \* treatment. This model examines the effect of the factors construct (immunoprecipitation with EGFR versus immunoprecipitation with IgG) and treatment (non-irradiated versus irradiated) and their interaction on gene expression (expr). This model allowed us to extract coefficients/ratios for immunoprecipitation with nEGFR versus immunoprecipitation with IgG in cells that were not treated (contrast 1), the treatment effect for IgG (contrast 2) and the difference between contrast 1 and contrast 2 (the interaction term, contrast 3). Contrast 3, the interaction term in the model, equals the treatment effect for nEGFR. A post hoc interaction analysis was then performed with Limma and p-values were adjusted for multiple tests using. For each of the three above- mentioned coefficients, gene expression was considered significantly different at an adjusted p value \< 0.05. A log fold change cut-off was applied afterwards to only look at genes with a fold change \> +1 or \< -1. Raw data and metadata from the project were deposited into Gene Expression Omnibus (GEO) with the identifier GSE92428. Pathway analyses were performed using the DAVID functional annotation software. ## RT-PCR A549 cells were pretreated with erlotinib (2 micromolar) for 2 h and subsequently irradiated with 4 Gy of radiation or sham irradiated. Twenty-four hours after irradiation, RNA was isolated from cells using an RNeasy Mini Kit (Qiagen). The cDNAs were generated with an RT2 First Strand Kit (Qiagen). Quantitative PCR was performed with the RT² Profiler^™ PCR Array for Human VEGF Signaling (Qiagen, \#330231), according to the manufacturer’s instructions. ## Quantification of deadenylase activity Nuclear fractions were isolated and the CCR4-NOT deadenylase complex was enriched by immunoprecipitation with a cNOT1 antibody, as described above. The assay was performed using a previously described method, with the following modifications. High-performance liquid chromatography-purified oligonucleotides were purchased from Eurofins. We used a 16-mer RNA substrate oligonucleotide (`5’-CCU UUC CAA AAA AAA A-3’`) containing a 5’-CY5 label, the 15-mer (`Cy5-CCU UUC CAA AAA AAA-3’`) and the 14-mer (`Cy5-CCU UUC CAA AAA AA-3’` as a marker of deadenylation. In a standard reaction, cNOT1 precipitates were dissolved in 5 microliter of reaction buffer (20 mM Tris-HCl, pH 7.9, 50 mM NaCl, 2 mM MgCl₂, 10% glycerol, and 1 mM beta-mercaptoethanol) containing 1.0 microliter RNA substrate in nuclease-free water. For gel-based detection and quantification of deadenylase activity, reactions were incubated at 30°C for 60 min, stopped by the addition of 12 microliter of RNA loading buffer (95% formamide, 0.025% bromophenol blue, 0.025% sodium dodecylsulfate and 5 mM ethylenediaminetetraacetic acid (EDTA)), and heated for 3 min at 85°C. A 3 microliter sample of the RNA mixture was analyzed by denaturing PAGE using a 20% acrylamide:bisacrylamide (19:1) gel containing 50% (w/v) urea. Intact mRNA labelled with the fluorescent dye Cy5 and deadenylation products were visualized and quantified using the LI-COR detection system. ## miRNA-mediated inhibition using the MISSION synthetic microRNA inhibitor Inhibition of hsa-miR-1180-5p (NFATC4) was performed by transfecting A549 cells with the MISSION synthetic microRNA inhibitor (Sigma). Cells were seeded in 24-well plates at densities of 1 × 10⁵ cells in DMEM containing 10% FBS on the day before transfection. Cells were then transfected in triplicate with Lipofectamine 2000 and SYN MIRNA INHIB HUM hsa-mir-1180-5p (`UUUCCGGCUCGCGUGGGUGUGU`) or SYN MIRNA INHIB NEG. CONTROL 1 (`GGUUCGUACGUACACUGUUCA`) (25 nM). After 24 h, transfected cells were irradiated with 4 Gy of radiation, and after an additional 24 h, cell lysates were prepared for western blot analysis with an antibody against NFATC4. ## siRNA treatment For EGFR or NFATC4 silencing, cells were treated with a specific siRNA for 72 hours before irradiation (ON-TARGETplus SMARTpool L-003114-00-0005 human EGFR, or ON-TARGETplus SMARTpool LQ-009584-00-0002, human NFATC4 siRNA, Dharmacon) and with on-TARGETplus Non-Targeting Pool siRNA (Dharmacon) as a control. Transfections were performed with Lipofectamine 2000 transfection reagent according to the manufacturer’s protocol (Invitrogen). ## Statistics All data represent the means ± s.d. of three independent experiments. All statistical analyses, with the exception of the analysis of the microarray data, were performed using two-sample t-tests and Bonferroni’s correction for multiple testing. All raw data are given in supplementary file. # Results The observation that nEGFR is detected in a complex with mRNA binding proteins suggests a potential role for nEGFR in mRNA processing. We synthesized cDNAs from mRNAs enriched in nEGFR complexes that were immunoprecipitated from nuclear extracts to elucidate the role of nEGFR in this process. Immunoprecipitation with an anti-EGFR antibody enriched the EGFR protein in nuclear fractions. The addition of a recEGFR-GST-fragment (68 kda) containing the epitope that binds the EGFR antibody reduced the precipitation of endogenous EGFR (180 kda), indicating the specificity of the applied antibody. As described in a previous study, several cellular stress treatments increased the expression of the nuclear EGFR protein. Immunoprecipitates were used to transcribe mRNAs complexed with nuclear EGFR into cDNAs in a reaction containing poly (TT) primers and radioactive dATP. Free radioactivity was separated from labelled cDNAs by gel filtration and the incorporation of radioactively labelled dATP into cDNAs was quantified. After normalization to the concentration of the nuclear EGFR protein, the expression of mRNAs enriched in complexes with immunoprecipitated nEGFR was increased compared to the NRT-IP-EGFR-control. However, irradiation and cisplatin treatments decreased the relative amount of enriched mRNAs complexed with nEGFR compared to untreated cells. Notably, an immunoprecipitation with nonspecific IgG was also positive for mRNA enrichment–as presented in —, although to a much smaller extent. Enrichment of mRNAs in complex with nEGFR was not only observed in A549 cells but also in the FaDu head & neck tumor cell line and in HSF7 normal skin fibroblasts. We performed an immunoprecipitation of nEGFR and the IgG control at time 0 without irradiation (0 Gy, non-irradiated) and also at 24 h with irradiation (4 Gy, irradiated) to elucidate the functional relevance of complexes between nEGFR and mRNAs. The mRNAs were extracted after immunoprecipitation from the complexes and characterized with the help of Affymetrix microarrays. After quality control of the microarray data, quantile normalization of the probe intensities was performed. Subsequently, normalized intensities were used to assess differential expression (DE) of genes and determine whether genes respond differently to immunoprecipitation with nEGFR versus immunoprecipitation with IgG in the absence of irradiation (contrast 1) or whether genes respond differently to radiation in IgG-precipitated samples (contrast 2). An interaction term was also added to the linear model (contrast 3, see the section) to determine whether a difference in gene expression was observed in nEGFR samples in response to radiation that differs from IgG samples. Following the DE analysis, we observed a strong effect on contrast 1, as 15068 probes of 53617 probes were DE (p \< 0.05), when considering only oligos referring to genes. 8920 out of these 15068 probes have in addition a log fold change either \> +1 or \< -1 (S1_Tab.xls). No genes were DE with a multiple adjusted p value \< 0.05 for contrast 2 and the interaction term (contrast 3). This finding indicates a strong influence of immunoprecipitation with either nEGFR or IgG on gene expression (contrast 1). However, a treatment/radiation effect that differed between IgG and EGFR samples was not observed (contrast 2 and 3). Heatmap-based visualization of the log2 normalized expression values for the 8920 DE probes in contrast 1 revealed the differentially expressed genes when either immunoprecipitated with nEGFR (two left columns) or IgG (two right columns). The heatmap also indicated that most DE genes in contrast 1 were expressed at rather low levels (indicated by the green color). We performed a functional annotation of all 8920 identified mRNAs in complex with nEGFR that were significant different in non-irradiated cells compared to IgG complexes (p\<0.05) using the DAVID Gene Ontology Analysis to further evaluate the binding of these mRNAs to nEGFR. DAVID matched 5515 of those mRNAs and the analysis resulted in the following top scoring KEGG pathway groups: Protein processing in the endoplasmic reticulum, Endocytosis, Proteasome, Lysosome, Ribosome, Cell cycle and HIF-1A signaling pathway According to our previous study, nEGFR is involved in regulating HIF-1A and VEGF expression; therefore, we focused in a first approach on mRNAs involved in HIF-1A /VEGF signaling. We isolated mRNAs from irradiated or non-irradiated A549 cells 24 h after irradiation and performed a quantitative RT-PCR to determine the expression of a panel of 84 mRNAs associated with HIF-1A /VEGF signaling in response to irradiation. We correlated mRNA expression quantified by RT-PCR with constitutive enrichment of distinct mRNAs at nEGFR measured using an Affymetrix chip analysis. Eighty-three of 84 mRNAs transcribed from the genes assigned to the RT² Profiler^™ PCR Array for Human VEGF/ HIF-1A -signaling (Qiagen) were detected in a constitutive complex with nEGFR in the absence of irradiation. We isolated mRNA from cells 24 h after irradiation and performed a quantitative RT-PCR for this panel of RNAs to elucidate the relevance of the formation of this complex. Fifty-three of the 84 mRNAs showed increased expression in response to irradiation (mean expression 1.68 ± 0.86) (Tab 1, column 2). Furthermore, treatment with the EGFR kinase inhibitor Erlotinib reduced the radiation-induced increase in the expression of 45 of 53 mRNAs (mean expression 1.1 ± 0.3, , column 3). Hence, we propose that nEGFR and its kinase activity are involved in regulating the increased expression/stability and translation of mRNAs associated with VEGF signaling in response to irradiation. We also determined the expression of representative proteins encoded by 10 of the 84 genes assigned to the RT² Profiler^™ PCR Array for Human VEGF/HIF-1A signaling to further elucidate the relevance of the formation of the complex between nEGFR and mRNAs. The increased mRNA expression observed in response to irradiation correlated with increased protein expression. Moreover, we detected over 1000 different miRNA species in complex with nEGFR (GEO identifier GSE92428). Notably, the amount of mRNA-specific miRNAs within the complex with nEGFR decreased in response to irradiation, whereas the amount of the corresponding mRNA and protein increased. Based on this observation, the net amount of mRNA-specific miRNAs in the complex with nEGFR negatively regulates the corresponding mRNA levels and protein translation. In summary, we hypothesize that nEGFR must be part of the P-bodies, which regulate mRNA translation and degradation in a miRNA-regulated manner. Consequentially, we immunoprecipitated nEGFR and assessed the presence of the GW182 protein, a marker of RISC and P-bodies. In addition to a radiation-induced increase in nEGFR expression, we observed the formation of a complex with the AGO2, PABPC1 and cNOT1 proteins, which all belong to the RNA-induced silencing complex (RISC). The relative amounts of these proteins in the EGFR-complexes were approximately equivalent, suggesting that the radiation-induced increase in nEGFR expression is associated with increased formation of the complex with RISC proteins. None of these proteins were detected in the precipitate with the nonspecific IgGs. Neither the RNAse nor DNAse treatment resolved the protein complex, which argues for a direct protein interaction. These observations strongly suggest the functional relevance of nEGFR in the context of regulating mRNA stability. As further proof of the hypothesis that EGFR kinase is involved in AGO2 phosphorylation in response to irradiation, as has previously been reported for hypoxia, we detected increased amounts of AGO2 within the nuclear fraction, which was phosphorylated at residue Y393. Y393 phosphorylation is a well characterized phosphorylation event performed by EGFR kinase. Importantly, blockade of EGFR kinase activity by pretreatment with erlotinib, as visualized by reduced auto-phosphorylation of nEGFR at Y992, reduced AGO2 phosphorylation at Y393 in response to irradiation. The deadenylase cNOT1 is critically important in the regulation of mRNA stability. The cNOT1 protein shortens the poly-A tails of mRNAs and initiates mRNA degradation. We applied an in vitro deadenylase assay to investigate the role of nEGFR in cNOT1 activity. We immunoprecipitated cNot1 from nuclear protein extracts and incubated it with a recombinant mRNA molecule terminally labelled with the Cy5 dye. Deadenylated products were separated by urea-PAGE gel electrophoresis and quantified using LICOR. Deadenylation by cNOT1 was detected in the nuclear protein preparation, which included proteins from the perinuclear region. Irradiation increased deadenylase activity with a maximum at 6 hours and decreased 24 hours after irradiation. Deadenylation was visualized by detecting smaller, deadenylated versions of the Cy3-labelled recombinant mRNA, and the quantified activity was inversely correlated with the amount of residual intact mRNA substrate. Bands were allocated to deadenylation products using specific markers, and cNOT1 antibody specificity was confirmed by incubating the substrate with a precipitate with nonspecific IgG. The specificity of the deadenylation reaction was confirmed by knocking down cNot1 expression with a specific siRNA. Incubation of recRNA-polyA(8) with cNot1 immunoprecipitates for 1 h resulted in the appearance of a band corresponding to the first deadenylation product. This deadenylation reaction was blocked by the addition of poly A-oligonucleotides. Incubation of the intact recRNA-polyA(8) template with the control IP for 1 h did not produce a deadenylated product. Pretreatment with erlotinib, which blocks EGFR kinase activity, strongly increased radiation-induced cNot1 activity, whereas EGFR knockdown with a specific siRNA inhibited deadenylase activity. Moreover, pretreatment of cells with EGF for 16 hours also induced cNot1 deadenylase activity. Likewise, irradiation induced cNot1 deadenylase activity in several tumor cell lines, e.g., FaDu, SAS, and PC3 cells, and in normal fibroblasts. We transfected A549 cells with an inhibitor of a miRNA, hsa-miR1180-5p, which is enriched in the nEGFR complex and was validated to regulate stability and translation of the NFATC4 mRNA, to examine the hypothesis that nEGFR regulates mRNA stability and translation in concert with the cNot1 deadenylase and miRNAs. We selected this miRNA as a target, since hsa-miR1180-5p was the only miRNA within the nEGFR complex that targets the NFATC4 mRNA. NFATC4 regulates HIF-1A/VEGF signaling and acts as a transcription factor that regulates cell survival, differentiation, angiogenesis, invasive migration, and the tumor microenvironment. Inhibition of hsa-mir1180-5p increased the expression of NFATC4 protein to a maximal level and prevented a further radiation-induced increase. Since NFATC4 expression responds to irradiation, we tested for a possible role for NFATC4 in the radiation response. Clonogenic survival after radiation treatment was determined to elucidate the effect of increased levels of the NFATC4 protein induced by hsa-mir1180-5p inhibition. Increased NFATC4 protein expression reduced the plating efficiency and increased radiosensitivity of A549 cells. Comparable results for NFATC4 protein expression were achieved after EGFR expression was knocked down with a specific siRNA. EGFR knockdown resulted in increased expression of the NFATC4 protein, which was accompanied by a lack of a radiation-induced increase in NFATC4 expression, similar to the control siRNA. Since hsa-miR1180-5p does not exclusively target the NFATC4 mRNA, we directly knocked down NFATC4 expression with a specific siRNA. The knockdown of NFATC4 in A549 cells was efficient and protein expression was reduced. This knockdown resulted in a radioprotective effect on both A549 and FaDu cells. # Discussion Here, we described a mechanism by which nEGFR regulates the stability of mRNAs associated with VEGF signaling in a miRNA-directed manner. We immunoprecipitated EGFR from nuclear protein preparations and identified bound proteins using MALDI-TOF to elucidate the molecular function of nEGFR. Under control conditions, nEGFR had already complexed with heterogeneous nuclear ribonucleoproteins (HNRNP) and PML bodies, which are involved in regulating gene transcription and mRNA translation during the cellular stress response,. Indeed, based on published results from our group and other researchers, nuclear translocation of EGFR is induced by cellular stress, e.g., radiation, hypoxia, cisplatin, as well as its natural ligand, EGF. Interestingly, the binding of mRNAs to EGFR was reduced significantly in response to radiation and cisplatin treatment. As shown in previous studies and the present study, nuclear EGFR translocation was accompanied by the activation of the tyrosine kinase. Importantly, EGFR tyrosine kinase activity seemed to negatively correlate with mRNA enrichment. In addition, mRNA binding to nEGFR was observed not only in tumor cell lines but also in normal skin fibroblasts, suggesting a general role for nEGFR in regulating mRNA stability. Using a microarray analysis, we were not able to identify a radiation-induced enrichment of mRNAs bound to nEGFR at 24 h that differed from IgG (control) in the present study. However, these results do not exclude the possibility that irradiation might alter mRNA binding to nEGFR compared to IgG at different time points. Although a treatment effect was not observed during the microarray analysis, we detected a significant enrichment of mRNAs in the EGFR complex compared to control IgG samples in the absence of stress treatment. Differential expression (DE) of 15068 of a total of 53617 mRNAs tested was detected. 8920 of them code for genes and have in addition a log fold change either \> +1 or \< -1 and were therefore estimated as potential biological relevant. We hypothesize that this large biological variation might only partially be related to the differential binding of mRNAs to nEGFR. Among other factors, nonspecific binding to nEGFR, genetic polymorphisms, and changes in mRNA levels due to age and genotype-environment interactions most likely also contribute to the observed high rate of differential expression. The observation that mRNAs are also expressed in IgG-precipitated samples is consistent with previously published data and also indicates that many tumor cells, including A549 cells, produce endogenous IgGs, which are involved in regulating tumor growth. The hypothesis that the list of DE genes contains specific and nonspecific signals is supported by gene ontology analysis, which pointed towards the regulation of terms such as cell cycle and ribosome. These categories are enriched in many experiments, indicating that these GO terms are not specific and do not need further interpretation. In contrast, the GO term HIF-1A/VEGF signaling pathway is potentially attributed to binding to nEGFR, as a link to nEGFR was observed in previously published data. An RT-PCR analysis of 83 DE genes related to the HIF-1A/VEGF signaling pathway and the expression of proteins encoded by 10 selected genes from this group confirmed microarray data. Preliminary data also revealed a potential link between nuclear EGFR and mRNA stability, as the expression of genes involved in the HIF-1A/VEGF signaling pathway was slightly increased in response to irradiation, which was abrogated again by the EGFR kinase inhibitor erlotinib. However, based on the results from the microarray analysis, the radiation-induced increase in the expression of the candidate genes is not due to altered binding to nEGFR in response to irradiation. Therefore, we propose that additional factors play roles. In this context, a potentially important observation is that the radiation-induced increase in the expression of mRNAs associated with VEGF signaling and protein translation was associated with a simultaneous loss of miRNAs from the nEGFR complex. This loss was not significant for a single miRNA species, as determined by the array analysis. However, when we summed up the radiation-induced losses of different miRNAs validated to target specific mRNAs (up to 339 species per mRNA) from the nEGFR complex, the loss was apparent. Nevertheless, the interplay between nEGFR, mRNAs and miRNAs has not yet been completely resolved yet. Our data indicate that nEGFR may participate in a central process that negatively regulates mRNA stability and translation in a miRNA dependent manner. Based on these observations and the perinuclear location of EGFR in complex with mRNAs and miRNAs, we hypothesized that nEGFR is part of the P-bodies. P-bodies are involved in mRNA storage and RISC-regulated mRNA degradation. Therefore, we assessed the localization of EGFR within P-bodies. Indeed, we identified the RISC proteins cNot1, AGO2, GW182 and PABPC1 in complex with EGFR. All proteins are located in P-bodies or GW-bodies in cells from higher eukaryotes and govern miRNA-mediated silencing to inhibit the active translation of mRNAs. Translational repression of mRNAs is initiated by mRNA sorting to P-bodies for decapping-dependent decay or sequestration. Alternatively, active translation of mRNAs is inhibited by miRNA-mediated RNA silencing in the presence of GW182, the endonuclease AGO, and the CCR4-Not deadenylation complex. However, although AGO 1–4 isoenzymes are capable of loading miRNA and exhibit endonuclease activity, RNAi-dependent gene silencing is exclusively executed by AGO2 (RISC). An interaction between AGO2 and nEGFR in response to hypoxia has been described. Mechanistically, hypoxia activates EGFR kinase activity, which phosphorylates AGO2 at residue Y393, inhibits AGO2 activity and subsequently blocks miRNA maturation and loading to the miRNA-induced silencing complex (miRISC). Our data also suggest in addition to an effect on miRNA maturation, a role for the AGO2/EGFR interaction in the regulation of mRNA stability by the miRISC. This complex is a miRNA-recruited protein complex that regulates mRNA stability and subsequent protein translation. The core of this complex consists of miRNA-loaded Argonaute and GW182 proteins. In mammalian cells, RISC contains both poly(A)-binding proteins (PABP) and a deadenylase complex, which initiates the mRNA degradation process by inducing deadenylation. The mechanism by which protein translation is repressed is still not defined. The CCR4-Not protein complex recruited by the GW182 protein was recently shown to release PABP from the poly(A) tail, thereby disrupting mRNA circularization and facilitating translational repression and deadenylation. As shown here, nEGFR complexes with GW182 and cNot1 proteins. Inhibition of EGFR kinase activity with erlotinib promoted cNot1 deadenylase activity. Thus, nEGFR kinase activity acts as a negative regulator of deadenylase activity. This idea agrees with the observation that nEGFR is a negative regulator of AGO2 in response to hypoxia. In this context, nEGFR kinase activity is responsible for the phosphorylation of AGO2 at residue Tyr 393 in response to hypoxic stress, which suppresses dicer binding that is essential for microRNA maturation and RISC function. The present data suggest an inhibitory function of AGO2 phosphorylation in the AGO2/GW182 complex, which regulates deadenylase activity. On the other hand, downregulation of EGFR by a specific siRNA was associated with the absence of the EGFR protein in the nuclear fraction and the subsequent inhibition of cNot1 deadenylase activity. Thus, nEGFR is required for deadenylase activity. Moreover, the regulatory effect of nEGFR kinase activity on deadenylase activity was also observed after activation of EGFR kinase by EGF. We detected deadenylase activity in several tumor cell lines and in normal fibroblasts. Consequently, general roles for nEGFR kinase activity in regulating mRNA stability and protein translation are postulated. After irradiation, the expression of mRNAs associated with VEGF signaling increased. Interestingly, pretreatment with erlotinib blocked the radiation- induced increase in mRNA expression. This finding supports the hypothesis that inhibition of EGFR kinase activity prevents AGO2 phosphorylation/inactivation and active cNot1-deadenylase degrades mRNAs located in the EGFR complex. The hypothesis that nEGFR is involved in regulating the expression of proteins involved in HIF1A/VEGF signaling is supported by previous studies showing that radiation-induced expression of the HIF-1A protein is reduced in response to erlotinib-mediated inhibition of nEGFR kinase activity. The mRNA expression profile obtained from RT-PCR analysis of irradiated cells favors the increased expression of genes involved in angiogenesis and anaerobe glycolysis, which are markers of the metastatic tumor cell phenotype in response to irradiation. In this context, EGFR is reported to be involved in regulating VEGFA expression in transformed cells and human NSCLCs. Moreover, as shown in our previous study, radiation induces HIF-1A expression to stimulate a metabolic shift toward lactate production, which is associated with treatment resistance. We hypothesize that the mRNA/nEGFR complex may be relevant to radiation-induced tumor cell resistance. According to the data presented in, the expression of mRNAs released from the complex with nEGFR was increased in response to irradiation. This finding may have resulted from mRNA stabilization and is consistent with the hypothesis that radiation-induced EGFR translocation and activation of kinase activity blocks AGO2 activity and deadenylase activity, subsequently resulting in mRNA stabilization. Importantly, we were able to identify validated miRNAs in complex with mRNAs and nEGFR. In response to irradiation these miRNAs were removed from nEGFR complex as AGO2 was phosphorylated and inactivated. These observations suggest the presence of a miRNA-driven deadenylase activity in complex with nEGFR. A miRNA-driven model for the regulation of cNot1 activity has been reported previously. The present study postulates a novel role for nEGFR, which accumulates in perinuclear and nuclear regions. The nEGFR protein complexes with mRNA and proteins involved in the RISC and negatively regulates the deadenylase cNOT1 in a miRNA-directed manner through its kinase activity. Notably, EGFR kinase activity and nuclear translocation are induced by cellular stress. Moreover, this novel function of EGFR kinase is also observed in several EGF- stimulated tumor cell lines and normal skin fibroblasts. The effect of hsa-mir1180-5p knockdown on NFATC4 protein expression links the presence of miRNAs in the EGFR/cNot1 complex with the translation of the corresponding protein. Furthermore, miRNA-regulated NFATC4 protein expression acts as a negative regulator of the post-radiation cell response and radiosensitivity. In general, NFATC4 acts as a negative regulator of growth and its expression is EGFR-dependent. Consequently, as shown here for A549 and FaDu cells, knockdown of NFATC4 expression results in a radioprotective effect. These data suggest that NFATC4 controls cell survival in tumor cells after exposure to stress. # Conclusions In summary, the present study is the first time to show that nEGFR kinase- regulated cNOT1 deadenylase activity enables cells to immediately respond to cellular stress by interfering with mRNA stability in a miRNA-directed manner. We postulate a three-step regulatory mechanism. In the first step, mRNA species are loaded into a complex that contains nEGFR, GW182 and AGO2. The selectivity of this process has not yet been defined, but we assume that the regulation of the nuclear translocation of EGFR and constitutive formation of the complex with HNRN-proteins play roles in selectivity. In the second step, the deadenylase cNOT1 is loaded onto EGFR-bound mRNAs in a miRNA-guided manner. Third, deadenylase activity is regulated by EGFR kinase activity and determines mRNA stability and the translation frequency. Thus, a membrane bound tyrosine kinase receptor, such as EGFR, can directly increase the stability of mRNAs involved in regulating survival in a miRNA-dependent manner. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction The Wallacean shortfall, which suggests our understanding of geographical distribution patterns of the species at large scales is generally poor, can be minimized by sampling ecosystem gradients at smaller scales and expanding our knowledge outward. Despite some progress in achieving global strategic goals of the United Nations and the Aichi Targets, biodiversity has declined, particularly in developing countries. The lack of information regarding biodiversity status precludes recognition of impacts of anthropogenic activities on biodiversity and implementation of conservation strategies and targets. Therefore, to gain a large-scale understanding of any taxonomic group and to determine threats and conservation potential within their spatial distribution in accordance with environmental variability, there must first be fine-scale (local/regional) baseline data. Amphibian occurrence and abundance are greatly influenced by localized variation in geomorphic, geologic, and environmental characteristics. Species Distribution Modelling (SDM) approaches based on Geographical Information System (GIS) have been widely used to predict distribution of species of amphibians and other vertebrate groups such as rodents and passerine birds for conservation purposes. However, with presence-only records often being the only available data, and potential for imperfect detection within a study location, one must account for distribution and absence given contributing environmental covariates. Amphibians in Pakistan have long been ignored in research, conservation, management, and policy and legislation. Pakistan’s National Climate Change Policy (2012), National Biodiversity Strategy and Action Plan (2015) and Pakistan Wetland Action Plan (2000) proposed guidelines for the conservation of natural resources, including fauna and flora, and mitigation of threats. Currently, 21 species of amphibian (all anurans) have been documented in Pakistan of which nine are believed to be endemic. However, no progress on integrating anurans in wildlife conservation, and policy development or legislation has been made. There is no national assessment of conservation status of anurans of Pakistan which cautions the use of global conservation status. Only a few published studies report the richness and abundance of various Pakistani anuran species. For example, the common Skittering Frog (*Euphlyctis cyanophlyctis*) was reported as abundant in the rice fields of Gujranwala, Punjab Province, and nine anuran species were recorded from Margalla Hills National Park, Islamabad. Six anuran species were reported from Rawalpindi and Islamabad areas including high abundance of Indus Valley Bull Frog (*Hoplobatrachus tigerinus*) and Skittering Frog from Rawal Lake, Islamabad. In the current study, we aimed to estimate niche suitability, niche overlap and model distribution of anuran species within the Rawalpindi District and Islamabad Capital Territory, which encompasses a unique mosaic of deserts and xeric shrublands, montane grasslands, temperate broadleaf and mixed forests, temperate conifer forests, and tropical/subtropical coniferous forests. Our findings were expected to generate new information on factors affecting niche suitability, spatial distribution size, and important ecoregions of the studied species and for anuran diversity to inform future conservation plans in Pakistan. # Materials and methods ## Study area The study was carried out in seven administrative units (Gujar Khan, Kahuta, Kallar Sayedan, Kotli Sattian, Murree, Rawalpindi, and Taxila) of the Rawalpindi District and Islamabad Capital Territory, Pakistan. The Rawalpindi District (33.4620° N, 73.3709° E) is located in the north-west of Punjab Province and covers an area of 5312 km². Islamabad Capital Territory (ICT) (33.7205° N, 73.0405° E) is in the north-east of the country and covers an area of 906.50 km². The elevation ranges from 457–2286 m and 457–610 m in Rawalpindi District and Islamabad Capital Territory, respectively. The climate of the study area is humid subtropical (Koppen climate classification). The summers produce more rain than the winter due to the monsoon season (July- August). The average rainfall in Islamabad Capital Territory is about 940 mm, in areas of Rawalpindi, Gujar Khan and Taxila ranges from 970–990 mm and in Murree, Kotli Sattian, and Kahuta is 1249 mm. The study area is dominated by tropical and subtropical coniferous forests in the north; broad-leaf, mixed forest and montane grasslands and shrublands in the mid and midwest; and arid shrublands in the east and south. The tropical and subtropical coniferous forests are dominated by *Pinus wallichiana* and *Pinus roxburgii* and have relatively fewer human settlements. The proximal, central and southern regions feature urban and semi-urban areas with vegetation species such as *Acacia modesta*, along with *Olea cuspidata*, and *Dodonea viscosa*. Of the 9 anuran species that are endemic to Pakistan, and therefore are of increased conservation interest, our study area encompasses most of the range of two of these endemics within Pakistan. Hazara Torrent Frog (*Allopaa hazarensis*) is endemic to the springs and streams of Northern Pakistan. Murree Hills Frog (*Nanorana vicina*) is endemic to Pakistan and India. ## Data collection We surveyed 87 randomly selected sites in the seven administrative units of Rawalpindi District and Islamabad Capital Territory and gathered anuran presence data from 2016 to 2018 (March–September) on a monthly basis; we visited each site at least twice from during the study period. The duration and number of observers varied during the surveys; however, on average each visit consisted of two days of non-standard field observations from early morning till late night carried out on a weekly basis, with 4–6 observers. Each visit had at least one observer familiar with the identification of anurans of the area. Anuran species richness was recorded using time-constrained visual encounter survey (VES) technique. Observers actively and thoroughly searched the sampling locations for a predefined time (\~60–120 min.) in order to record the species. Adult specimens and tadpoles were collected with dip nets or simply picked up by hand, were examined, and identified following descriptive identification by Khan and released back. ## Data analysis and species distribution modelling All analyses, models, maps and plots were generated using R statistical software version 3.6.3. A preliminary check on observed environmental values for each species was conducted by using the extract function in the ‘raster’ package. Environmental ranges of species were checked before running models in order to identify potential flaws in the output (e.g. high spatial distribution probability in montane ranges when the species observations have initial low- elevation values). Bioclimatic variables (BIO 1–19) were retrieved from the WorldClim repository via the ‘raster’ package at 0.5 arcmin resolution, and were tested for collinearity using Variation Inflated Factors (VIF). The following bioclimatic variables were retained: BIO7: temperature (C°) annual range; BIO8: mean temperature (C°) of wettest quarter (3 consecutive months); BIO9: mean temperature (C°) of driest quarter; BIO15: precipitation seasonality; BIO18: precipitation (mm) of the average warmest quarter; and BIO19: precipitation (mm) of the average coldest quarter. To map vegetation, we used MODIS (https://develo pers.google.com/earthengine/datasets/catalog/MODIS_006_MOD13Q1#bands; https://code.earthengine.google.com/e3a10b1ec6086c3ee7c598cfaca7dd98; resolution: 250m; scale 0.0001). MODIS has a spatial resolution of 250 m and provides a Vegetation Index (VI) value at a per pixel basis. We used two primary vegetation layers: Normalized Difference Vegetation Index (NDVI), which is the continuity index to the existing National Oceanic and Atmospheric Administration-Advanced Very High-Resolution Radiometer (NOAA-AVHRR) derived NDVI; and Enhanced Vegetation Index (EVI), which minimizes canopy background variations and maintains sensitivity over dense vegetation conditions. In Google Earth Engine (GEE), both MODIS layers The MODIS NDVI and EVI products were computed from atmospherically corrected bi-directional surface reflectance that had been masked for water, clouds, heavy aerosols, and cloud shadows. A 30 m Digital Elevation Model (DEM) was retrieved from the USGS earth explorer database (<https://earthexplorer.usgs.gov/>). The DEM was used to generate slope, aspect, and terrain roughness via the *terrain* function in the ‘raster’ package. Distance to rivers was calculated using a local shape file of all waterways in Pakistan and the *fasterVectToRastDistance* function of the ‘fasterRaster’ package. River distance values (in meters) were log normalized to prevent outlier values from overwhelming the model contribution estimates. All environmental variables were resampled to 0.5 arcmin resolutions thereafter using the bilinear method, and masked to a shapefile of Pakistan regional boundaries. The explanatory variables were then stacked and tested for multicollinearity using the *vifstep* and *vifcor* functions of the ‘usdm’ package with a threshold VIF of 10 and a threshold correlation coefficient with an absolute value of 0.7; whereby we excluded predictors that were strongly correlated with ⸺ but considered to be less important for anurans ⸺ than other predictors. The Maximum Entropy or “Maxent” modelling method has been shown to perform as well as, or better than ensemble models when modelling species distributions, with additional benefits of lower computational power requirements and increased simplicity of use. Species Distribution Models for this study were created following guidelines on Maxent parameterization. Species occurrence data was thinned to one point per raster cell to omit spatial biases. Maxent models were created using the Maxent software wrapper through the ‘dismo’ package. A kernel density bias mask was created by querying all available anuran occurrences from the Global Biodiversity Information Facility (GBIF) database for Pakistan as a measure of restrictive effort using the ‘MASS’ package using the reference bandwidth smoothing factor. We used the *occ_search* function in the ‘rgbif’ package in R to access the GBIF database, using the taxon “Anura”, the country Pakistan (“PK”), and “coordinates = TRUE” as filters to create a text file containing 345 dataset keys and number of observations per dataset). We generated 10,000 geographically randomized background points within the bias mask estimate. Models were tuned using the *ENMeval* function in the ‘ENMeval’ package to identify feature selection variables and regularization multiplier (beta-multiplier) values selected from the lowest delta Akaike Information Criterion (AIC_c) using five random *k*-folds. Maxent datasets were partitioned into training (3/4) and testing (1/4) data using a 4-way partitioned *k*-fold. Features selected in the final models were linear and quadratic (lq) with a beta-multiplier set to between 1–2 for sensitivity testing, and 1 for the final model. We selected these features for the final Maxent models for each species based on the statistics we used to validate our model: area-under-the- curve (AUC) values within receiver-operating characteristic (ROC) curves and the true skill statistic (TSS \[= TP + TN—1, where TP = proportion of true positive predictions and TN = proportion of true negative predictions\]). Models were replicated with replacement using the bootstrap method. Final models were visualized using the raw model output using the ‘ggplot2’ package and each model prediction was condensed using the highest true skill statistics (TSS) threshold value as a filter (hereafter called “threshold model”). The true skill statistics is defined based on the components of the standard confusion matrix representing matches and mismatches between observations and predictions. Model outputs were defined as *niche suitability* on a 0–1 probability scale (1 = highest niche suitability; 0 = lowest niche suitability), and threshold models were considered to be the niche distribution of a given species in terms of overlap with others. Each threshold model was converted to points and extracted from the explanatory variables to derive summary statistics of environmental values for each species within their niche distribution. All species threshold predictions were then combined to create niche overlap raster to determine which eco-regions were most suitable for anuran species in Pakistan. High-density niche overlap areas were identified by filtering 50% (4.5 species per raster cell) and 100% (9 species per raster cell). We then converted the two final niche spaces to a spatial polygon to identify percentage overlap between each ecoregion and high-density niche distributions. # Results ## Niche suitability and overlap Final predictions for niche suitability of all nine anuran species using Maxent models indicated good fit in terms of AUC. Environmental variables varied among species; however, distance to rivers and precipitation in the warmest and coolest quarters were shown to have significant contribution (% contribution \>10; higher permutation importance and variable importance in jacknife-testing) for *Duttaphrynus bengalensis* and *M*. *nilphamariensis;* distance to rivers and precipitation of the warmest quarter for *Duttaphrynus stomaticus;* distance to rivers, precipitation of the warmest quarter, and NDVI for *A*. *hazarensis* and *N*. *vicina*; and precipitation of the warmest quarter and NDVI were most influential for *H*. *tigerinus*, *Euphlyctis* spp., *Minervarya* spp., and *Sphaerotheca pashchima*. Habitat suitability for most species was higher in locations with greater precipitation in the warmest quarter and less precipitation in the coolest quarter. Habitat suitability for *A*. *hazarensis* and *N*. *vicina* was higher in locations with greater ecosystem productivity (higher NDVI) closer to rivers. ## Spatial distribution Nine anuran species showed a divergent pattern of spatial distribution. We found that *Euphlyctis* spp., *Minervarya* spp., and *H*. *tigerinus* showed preference for the lowlands in proximal, central and southern parts of the study with high urban settlements, little vegetation and higher average temperatures. The toads (*D*. *bengalensis* and *D*. *stomaticus*) had scattered distributions throughout the study area with no clear preference for elevation. *S*. *pashchima* showed a patchy distribution in the midwestern extent of the study area as well as the foothills to the north. *Microhyla nilphamariensis* also showed a wide distribution throughout the study area with a preference for both lowlands and montane terrain. Endemic species such as *N*. *vicina* and *A*. *hazarensis* were observed in locations with higher elevations, proximal to streams and lower average temperatures as compared to the other seven species sampled. Niche overlap at 50% (4.5 species per area) within the studied ecoregions showed 0.2% overlap in montane grasslands and shrublands, 4.8% in temperate broadleaf and mixed forests, 23.6% in temperate coniferous forests and 33.9% in tropical and subtropical coniferous forests. Whereas, the niche overlap (9 species per area) revealed 0% overlap in montane grasslands and shrublands, 3.2% overlap in deserts and xeric shrublands, 9.4% overlap in temperate broadleaf and mixed forests, 17% overlap in temperate coniferous forests, and 70.2% overlap in tropical and subtropical coniferous forests, providing evidence that tropical and subtropical coniferous forests support the highest diversity of anuran species in the region. # Discussion Development of analytical habitat distribution models has rapidly increased in ecology due to the invention of new GIS tools and statistical techniques. Such models statistically relate the geographical distribution of species or communities to their present environment. Neither any scientific study on the distribution patterns nor any species occurrence database of anurans of Pakistan exists in the country. In forested mountainous regions outside of but similar to those of Pakistan, amphibians, particularly *Ascaphus truei* and *Dicamptodon tenebrosus*, were more abundant in stream habitats within older coniferous forests. Precipitation and soil temperature influences probability of occurrence for Polish amphibian species while vegetation contributed significantly in the prediction for salamander species of central Portugal. Our results suggest similar trends, with precipitation (of the warmest and coldest quarter), distance to rivers and vegetation being highly deterministic factors of suitability for anurans in Pakistan. While Maxent models can accurately predict grid-based habitat suitability and presence of species when observation data of those species is limited, predictions from such models could also be used to identify the likeliest locations of species for further monitoring and obtain actual presence/absence or abundance data. Actual presence/absence or abundance data at precise locations enables biologists to use finer-scaled environmental variables that influence habitat suitability and abundance of species. For anurans and other aquatic wildlife, the type of wetland habitat or changes in wetland variables over time at repeatedly monitored sites can increasingly be quantified or classified at fine spatial scales over large regions away from ground-truthed locations. Increased availability of wetland data is due to: 1) the accumulation of decades of remotely sensed environmental data by satellites; 2) the development of free, open-source, online platforms (e.g., Google Earth Engine) for processing remotely sensed data and extracting this data to survey locations; and 3) development and sharing of open-source machine learning techniques for predicting and classifying wetlands. To improve the chances of detecting anurans or other vocalising species when they are present at sites, monitoring could involve obtaining multiple visits per site by using passive acoustic monitoring with pre-programmed acoustic recorders. Acoustic data could then be transcribed to obtain detections per visit and hierarchical models can be used to account for varying detection probability of species among sites when estimating effects of environmental variables on occupancy or abundance. The toads (Family Bufonidae) of the study area: *D*. *bengalensis* and *D*. *stomaticus*, had somewhat scattered distribution throughout the region with no clear preference for elevations; however, *D*. *stomaticus* showed a clear preference to the northwestern extent of the study area, mostly concentrating within and proximal to the lowlands. *D*. *bengalensis* and *D*. *stomaticus* have been recorded as widespread species found up to 1800 m and 4500 m elevations, respectively. *D*. *bengalensis* is adapted to various types of habitats, even degraded ones, and around human habitations. The geographic range of *D*. *bengalensis* has now been extended due to its introduction and *D*. *bengalensis* has attained a status of invasive species in various parts of the world. The two toads are found in the plains, lowlands, sub mountain areas as well as hilly areas in Pakistan which experience monsoon season (July-August) during which the species breed. We recorded *D*. *bengalensis* at elevations higher than previously reported. One reason for their widespread distribution in our study area is their adaptation to a wide range of habitats. *M*. *nilphamariensis*, a diminutive frog species (Family Microhylidae), showed a widespread distribution throughout the study area with a preference for both lowlands and montane terrain. The species has been recorded from areas up to 2000 m elevation as well as from lowlands, sub mountain areas and foothills in Pakistan. The dicroglossid frogs of the study area showed varied distribution patterns. *S*. *pashchima*, a burrowing frog species, showed a patchy distribution in the midwestern extent of the study area as well as the foothills to the north. *S*. *pashchima* has been recorded as a widespread species from lowlands and forested areas up to 1500 m. This species remains under soft soil for most parts of the year and emerges during summers to breed during the monsoon and avoids high altitude areas possibly due to their burrowing habit, since hard substrate makes it difficult for them to dig. Further, mountains in the north are under less influence of monsoon, which likely provides less suitable breeding conditions. Other dicroglossid frogs such as *Euphlyctis* spp. showed scattered distribution probability through the lowlands. *Minervarya* spp. was shown to prefer proximal, central, southern, and western lowland areas. *H*. *tigerinus* also showed variability in preference between lowland and elevated areas, with a concentrated distribution toward the middle to northern extents of the study area. *H*. *tigerinus* has previously been recorded from areas up to 2000 m. These areas are amongst most built up parts of the region in addition to encompassing other human modified habitats such as croplands. *Euphlyctis* spp. has been recorded from areas up to 2500 m while *Fejervarya* spp. from areas up to 2000 m (Dijk 2004), but these dicroglossid frogs are also widespread in lowlands and forested areas. *Euphlyctis cyanophlyctis* and *Zakerana syhadrensis* occur along stream banks and water pools between forest edges, agricultural areas, and residential gardens. Our findings are consistent with the available information. However, we have thus forth provided empirical data on the response of these species to the studied environmental factors for the first time. The endemic frogs, *A*. *hazarensis* and *N*. *vicina*, showed restricted occurrence within the northern and north-eastern mountain ranges in areas of high elevation compared to the proximal lowlands to the southeast. Most of these areas feature subtropical pine forest (900–1500 m) dominated by *Pinus roxburghii* trees, while the northernmost areas possess Himalayan moist temperate forest (1500–3000 m) dominated by *Pinus wallichiana* and *Pinus roxburghii*. The wetlands throughout this range exist in the form of freshwater streams. *A*. *hazarensis* is endemic to Pakistan while *N*. *vicina* is known from Pakistan and India. *A*. *hazarensis* and *N*. *vicina* are known from streams and pools in forested mountainous areas as high as 1500 m and 3000 m, respectively. During our study, we found that *A*. *hazarensis* could occur at a higher elevation (\>1500) than the previously reported range. # Conservation implications and suggestions Urban developments have been shown to have negative impacts on amphibians. Several types of culverts, tunnels and corridors have been developed and their effectiveness assessed. Many amphibian species in North America and Europe have used these structures, resulting in reduced mortality from vehicular collisions. Since anurans in Pakistan enjoy no legal protection, no such consideration is given during urban planning. With recent urban expansion in Rawalpindi District and the Islamabad Capital Territory, there has been a noticeable reduction in forests, open spaces and watersheds. As shown by our results, some of the anuran study species are tolerant to habitat degradation while others are not. The creation of human modified habitats may further facilitate the spread of native species such as *Euphlyctis* spp., while also accommodating invasive species such as *D*. *bengalensis* and *D*. *stomaticus* which may pose the threat of resource competition against native species. We recommend studying the effectiveness of such existing amphibian tunnels and corridors or designing new ones tailored to the needs of our species. These could then be incorporated in future urban development programs. Conservationists put more emphasis on conserving threatened species. Studies, however, have shown that even common species are subjected to population decline and local extinction especially if these species are associated with a particular type of habitat or set of environmental conditions. A 29% decrease in the population of the Moor Frog (*Rana arvalis*), which is found in heathlands and moorlands in parts of Europe, was reported between 1950–2006. This decline was attributed to cultivation of heathlands and moorlands, lowering of ground water levels and intensification of agricultural practices. In a forest landscape study, Brown Frogs (*Rana arvalis*, *R*. *temporaria*) bred more in various wetland habitats (e. g., naturally flooded areas, beaver ponds, mitigation pools with shallow littoral zones, cleaned ditches) than in ditches overgrown with forest vegetation. Climate change during the past two decades has affected several species of plants and animals in Pakistan. The tadpoles of *A*. *hazarensis* and *N*. *vicina* responded (under laboratory conditions) to higher temperature (*\>*26°C) through faster metamorphosis, reduction in the body size, more frequent developmental complications or deformities such as edema and tail kinks, lower fitness and higher mortality. Being associated with a particular set of environmental conditions in the north and northeast of the study area, it is feared that these two species endemic frogs, which are currently evaluated as least concern in the IUCN Red List of threatened species, may experience local extinction in the future. Land use simulators can be used to project changes over space and time in environmental variables used both in Maxent models (e.g., climate) and in models based on surveys whose locations were informed by Maxent model predictions. Thus, Maxent models can be used to project changes in distribution of species over time under different climate scenarios, and raster layers based on these distributions can be used to identify potential refugia for anurans either under current or future climate conditions. These raster layers may also be used as inputs in raster overlay-based conservation planning tools (e.g., Marxan, Zonation) to prioritise locations for protection or management of threatened species. # Supporting information We would like to especially thank Mr. Wajehuddin and Mr. A. Bhutta, Punjab Forest Department, Government of Punjab, Pakistan for providing assistance during the field work. We are thankful to Snow Leopard Foundation, Pakistan, for providing logistics. We are also thankful to Dr. Hussain Ali, Snow Leopard Foundation, for his help. [^1]: The authors have declared that no competing interests exist.

# Introduction The membrane skeleton is the basis of erythrocyte morphology and deformability. It is the hexagonal lattice structure formed by 6 spectrin tetramers connecting to the short actin filaments at the junctional complex. The membrane skeleton is anchored to the lipid bilayer via ankyrin and band 4.1. The short actin filaments or protofilaments in the junctional complexes have the constant length of \~35–37 nm, which plays important roles in keeping the hexagonal structure and the mechanical property of the membrane skeleton. The short actin filament consists of 6 pairs of actin monomers and is coated by tropomyosins (TMs), with erythrocyte tropomodulin (E-Tmod) capping its slow growing end (pointed end). As a TM-binding protein and the only capping protein at the pointed end in erythrocytes, E-Tmod plays a critical role in restricting the length of the short actin filament (F-actin). E-Tmod is a 41 kDa protein (E-Tmod41), which consists of N-terminal actin binding domain (E-Tmod_1–92) and C-terminal actin binding domain with six leucine repeats. E-Tmod41 binds to TM5/5b (35 nm) at 39–138 residues and the complex functions as a “molecular T ruler” metering off long actin filaments to short filaments of 37 nm. Recently, a short E-Tmod isoform of 29 kDa (E-Tmod29) was discovered. It lacks the N-terminal actin-binding domain but retains the C-terminal actin-binding domain. It can bind to TM5 or G-actin and is localized in the cytosol of erythrocytes. The expressions of E-Tmod41 and E-Tmod29 are driven by the alternative promoters of *E-Tmod* gene. E-Tmod41 null mice display a mild sphero-elliptocytic anemia with osmotically fragile erythrocytes, due to the misregulation of F-actin lengths and a disrupted spectrin-actin lattice of membrane skeleton. In addition to E-Tmod, there are three members in Tmod family, neuronal Tmod (N-Tmod), ubiquitous Tmod (U-Tmod), and skeletal muscle Tmod (sk-Tmod). E-Tmod is the only Tmod isoform present in human and mouse mature erythrocytes. But U-Tmod is found in erythroid progenitors and exists in the erythrocytes of TOT (Tropomodulin overexpressing transgenic) / E-Tmod^-/- mice. It may be due to the wide range of expression and the weak capping activity of U-Tmod. The topology of membrane skeleton is formed during the development and maturation of erythrocytes. The synthesis and expression of major cytoskeletal proteins in erythrocyte membrane occur in an asynchronous manner and the remodeling of the membrane skeleton begins at a very early stage during erythrocyte development. In the erythroid differentiation induced by erythropoietin, interleukin-3 (IL-3) and stem cell factor, etc., the gene expressions for many membrane skeleton proteins are significantly upregulated. In addition to chemical factors, the biomechanical force, fluid shear stress, has been found to contribute to the hematopoiesis in the embryos. In adult, the reticulocytes have to circulate in the blood vessels before they become mature. Thus, these indicate that physical environment like fluid shear stress may also be involved in regulating of gene expressions of membrane skeleton proteins. microRNAs (miRNAs) are the small non-coding RNA molecules containing about 22 nucleotides and play key roles in the regulation of gene expression. Many miRNAs, such as miR-451, miR-221/222, and let-7d, etc., are shown to be involved in the erythroid differentiation. There are mechano-sensitive miRNAs, such as miR-126, miR-23b, miR-10a, etc., which could be regulated by fluid shear stress in endothelial cells, vascular smooth muscle cells and macrophages. Therefore, it is possible that miRNAs may play roles in regulating the gene expressions of membrane skeleton proteins in the erythroid differentiation induced by fluid shear stress. In the present study, we examined the effects of fluid shear stress on the erythroid differentiation, E-Tmod41 expression, and F-actin cytoskeleton remodeling. Our data showed that fluid shear stress could induce the erythroid differentiation and E-Tmod41 expression, thus contributing to the F-actin cytoskeleton remodeling. Furthermore, shear stress could upregulate E-Tmod41 expression by suppressing E-Tmod41-targeting miR-23b-3p and activating the alternative promoter upstream of exon 0. # Materials and Methods ## Cell culture and animals Mouse erythroleukemia (MEL) cell line derived from Friend virus-infected mice were maintained in RPMI 1640 medium containing 10% fetal bovine serum (FBS). Terminal erythroid differentiation was induced in MEL cells by adding 2% dimethyl sulphoxide (DMSO) in the culture medium for 24 hours. C57BL mice of 8 weeks old were paired. The vaginal plug was checked on the second day and counted as embryonic day 0.5 (E0.5). On E10.5, the pregnant mice were sacrificed by cervical dislocation. Embryonic erythroblasts were isolated from the yolk sacs embryos and maintained in IMDM medium containing 10% FBS, 10 units ml^-1 penicillin, 10 μg ml^-1 streptomycin, 0.2 mg ml^-1 Fe-saturated transferrin, and 0.5 mg ml^-1 ascorbic acid. The animal protocol was approved by the Ethical Committee of Peking University Health Science Center. ## Cell exposure to shear stress MEL cells or erythroblasts were plated in a 6-cm Corning petri dish and the dish was placed in a cone-plate shearing system (Fig A). When the cone spins, the cells will be subjected to shear stress. The shear stress, τ, can be calculated based on the equation, *τ* = *μω*/ *α*, where *μ* is the viscosity of the cell suspension (2.21 mPa∙s), is the angular velocity, and *α* is the angle of the cone (0.5°). The shear stress was set to 5 dyn/cm², which is the shear stress yielded in the dorsal aorta of E10.5 embryos. ## Flow cytometry Cells were fixed in 4% paraformaldehyde, washed with phosphate buffered saline (PBS), and incubated with 0.1% Triton X-100 for 5 min. After blocked in 1% bovine serum albumin (BSA) for 30 min at room temperature, the cells were incubated with rhodamine phalloidin (Cytoskeleton Inc., USA) for 20 min in the dark at 37°C. Then the cells were resuspended in 300 μl PBS and analyzed in a BD FACS Calibur (USA). The mean fluorescent intensity was measured to represent the F-actin content in the cells. ## Laser scanning confocal microscopy The cells were stained with rhodamine phalloidin following the procedures mentioned above. The nuclei were counterstained with 4',6-diamidino-2-phenylindole (DAPI, Beyotime Biotechnology, China) for 10 min and then washed with PBS. The samples were resuspended in the mixture of PBS and glycerol (v/v, 1:1) and placed into a coverglass bottom dish. The cells were observed with a laser scanning confocal microscope (Leica TCS SP8 MP FLIM, Germany). The mean fluorescent intensity of the images were calculated by using ImageJ 1.46r software (National Institutes of Health, USA). ## RNA isolation and Quantitative RT-PCR Total RNA was extracted from cells using TRIzol reagent (Invitrogen, USA) according to the manufacturer’s instructions. Two microgram of total RNA was reverse transcribed to cDNA using a ReverAid First Strand cDNA Synthesis Kit (Thermo, USA). Real-time PCR was performed on a Mx3000 Multiplex Quantitative PCR system (Stratagene, USA) using Brilliant II SYBR Green QPCR Master Mix (Agilent Technologies, USA). The sequences of the primers used for E-Tmod41 were as follows: forward: 5’-GAC ACA GCC TCA CAC AAT GT-3’; reverse: 5’-CTT GGT GGT CTG ATC CTT CT-3’. The sequences of the primers for the markers of erythroid differentiation, β-major globin (*Hbb-b1*), glycophorin A (*GYPA*), and *GATA1*, and miR-23b-3p host gene, *2010111I01Rik*, were listed in Table A in. GAPDH (forward: 5’-ACC ACA GTC CAT GCC ATC AC-3’; reverse: 5’- TCC CCA CCC TGT TGC TGT A-3’) was used as an internal control. A relative fold change in the gene expression was calculated using the method of 2^-△△CT. ## Western blot analysis Cells were lysed in RIPA buffer and centrifuged at 12000×*g* for 5 min at 4°C. The supernatant was collected and the protein concentration was quantified by BCA assay (Applygen Technologies Inc., Beijing, China). Proteins (20 μg) were separated by SDS-PAGE, and then transferred onto nitrocellulose membranes. The membranes were incubated with anti-E-Tmod41 antibody (prepared by AbMax Biotechnology Co., Ltd, Beijing, China), anti-glycophorin A antibody (Beijing Biosynthesis Biotechnology Co. Ltd, Beijing, China), anti-GAPDH antibody, or anti-β-tubulin (Santa cruz Biotech., USA), followed by HRP-conjugated goat anti- rabbit or mouse IgG. The signals were detected by using an Enhanced Chemiluminescence Detection (ECL) kit (Evergreen, Beijing, China). ## Adenovirus infection MEL cells (1×10⁶ cells) were infected with adenovirus, Ad-E-Tmod41 and Ad-Null (SinoGenoMax, Beijing, China) at 0, 25, 50, 75, and 100 MOI for 48 hours. The cells were either lyzed for protein extraction or fixed in 4% paraformaldehyde for F-actin content analysis. ## siRNA, mimic, and inhibitor transfection Small interfering RNA (siRNA) specific to E-Tmod41 was designed using Invitrogen Block-iT RNAi Designer. The sequence is 5’-GGA AUU UAA GGA CCG AGA A-3’. The mimic and inhibitor of miR-23b-3p were purchased from RiboBio (Guangzhou, China). MEL cells were transfected with siRNA, mimic, or inhibitor (50 and 100 nM) by using riboFECT CP transfection kit (RiboBio, Guangzhou, China). Transfected cells were harvested at 24 hours for mRNA extraction or at 48 hours for protein isolation. ## MicroRNA array analysis MEL cells sheared for 12 hours and the unsheared controls (3 pairs) were harvested and lysed in TRIzol reagent (Life technologies, Carlsbad, CA). After RNA extraction and quality analysis, miRNAs were labeled and hybridized on the miRCURY LNA Array (v.18.0, Exiqon, Danmark) at KangChen Bio-tech Inc. (Shanghai, China). The images were analyzed and data were extracted using GenePix Pro 6.0 software (Axon). The microarray data was validated by using TaqMan Assay (Life technologies). ## Plasmid construction, site-directed mutagenesis, and dual luciferase assay P_E0 (\~700 bp) and P_E1 (\~1000 bp) promoter sequences were amplified from mouse genomic DNA and inserted to pGL3 basic vector (Promega, USA) at the multiple cloning sites, respectively. The E-Tmod 3’-untranslated region (3’UTR) was amplified from mouse cDNA and subcloned to pGL3 control vector at XbaI site. The mutagenesis of potential miR-23b-3p targeting site was introduced by using site-directed mutagenesis kit (Tiangen Biotech., Beijing, China). For transfection, 1 μg of plasmid and 100 ng of renilla luciferase vector were co-transfected into MEL cells by FuGENE 6 (Promega, USA). Cells were collected after 48 hours for the detection of luciferase activity by the dual- luciferase reporter assay system (Promega, USA). For miRNA study, pGL3 control vector containing wild type or mutated E-Tmod 3’UTR were co-transfected into MEL cells with miR-23b-3p mimics (50 nM) or control mimics. Then the cells were collected after 24 hours for the dual luciferase assay. ## Statistical analysis All experiments were performed in duplicate with data averaged from at least three independent experiments. The data are presented as the mean ± SEM. Direct comparisons were made using paired or unpaired Student’s *t*-test, and multiple group comparisons were made using one-way analyses of variance (ANOVA). Statistical significance was defined as *p*\<0.05, 0.01, or 0.001 (indicated as \*, \*\*, or \*\*\*, respectively). Prism 5.0 software (GraphPad, Inc., USA) was used for data analyses. # Results ## Fluid shear stress induces erythroid differentiation and F-actin cytoskeleton remodeling in MEL cells We employed MEL cell, a cell model commonly used for the study of erythroid differentiation, to examine the effects of fluid shear stress on the erythroid differentiation and cytoskeleton remodeling. A cone-plate shearing device was developed to generate the laminar shear stress of 5 dyn/cm² (Fig A), mimicking the blood flow in the aortas of embryos at embryonic day 10.5. MEL cells treated with 2% DMSO served as the positive control for erythroid differentiation. Wright-Giemsa staining showed that the dark blue color became reddish in the cytosol of MEL cells treated with shear stress or DMSO and that the nuclei became condensed. Further analysis showed that their cellular area significantly decreased as compared to control. Quantitative RT-PCR data showed that markers for erythroid differentiation, *Hbb-b1*, *GYPA*, and *GATA1*, were all significantly upregulated by both shear stress and DMSO. These data indicate that, although it was less competent than DMSO, fluid shear stress indeed induced the erythroid differentiation in MEL cells. Since the morphological change is accompanied with F-actin cytoskeleton remodeling during erythroid differentiation, we examined the F-actin content in MEL cells treated with DMSO and shear stress. Flow cytometry analysis showed that the F-actin content increased \~1.5 fold in both DMSO- and shear stress- treated MEL cells. Confocal images also showed the similar results. ## Fluid shear stress upregulates E-Tmod41 in MEL cells As the capping protein at the pointed end of F-actin, E-Tmod41 is critical for F-actin polymerization and stability. It may play a role in the F-actin cytoskeleton remodeling induced by fluid shear stress. Therefore, we examined E-Tmod41 expression in shear stress-treated MEL cells. Quantitative RT-PCR showed that E-Tmod41 mRNA expression was upregulated by 3 h of shearing. As the shearing time extended, E-Tmod41 mRNA level increased and reached the highest level at 12 h. As detected by Western Blot, E-Tmod41 protein level did not change at 12 h (data not shown) but markedly increased 2 fold at 24 h. The data indicate that fluid shear stress could upregulate E-Tmod41 expression. ## E-Tmod41 regulates F-actin content in MEL cells The upregulation of E-Tmod41 expression by shear stress may contribute to the F-actin cytoskeleton remodeling. We next performed the gain or loss of function experiments to study the effect of E-Tmod41 in F-actin remodeling. E-Tmod41 was over-expressed by adenovirus infection. Both flow cytometry and confocal microscopy showed that the F-actin content was significantly increased in MEL cells infected by 50 MOI Ad-E-Tmod41. On the contrary, when E-Tmod41 was knocked down by E-Tmod41-specific siRNA, the F-actin content was greatly reduced. These findings demonstrate that E-Tmod41 affected the F-actin content in MEL cells, which may mediate the F-actin cytoskeleton remodeling induced by shear stress. To explore whether the expression of E-Tmod41 would affect the erythroid differentiation, we detected the expressions of *Hbb-b1*, *GYPA* and *GATA1* in MEL cells with E-Tmod41 overexpression or suppression. But quantitative RT-PCR data showed that none of these markers changed (Figs A and B). ## Fluid shear stress suppresses the expression of an E-Tmod-targeting miRNA, miR-23b-3p miRNAs play important roles in hematopoiesis and there are many miRNAs that could be regulated by fluid shear stress. So we collected MEL cells sheared for 12 h and performed miRNA microarray analysis. Data showed that 24 miRNAs were upregulated and 58 miRNAs were down-regulated \>1.5 fold (with *p* value \<0.05) by shear stress (Table B). Among the down-regulated miRNAs, miR-23b-3p was predicted to target E-Tmod mRNA 3'UTR by TargetScan and Pictar. Quantitative RT- PCR data showed that, in addition to miR-23b-3p, fluid shear stress could also suppress the expression of miR-23b-3p’s host gene, *2010111I01Rik*. Importantly, *2010111I01Rik* is not a target of miR-23b-3p (Fig D). Sequence analysis showed that the targeting site of miR-23b-3p on E-Tmod 3’UTR is highly conserved in mammals. To test whether miR-23b-3p targets E-Tmod, two reporter vectors were constructed with a luciferase-coding sequence followed by wild type or mutant E-Tmod 3’UTR , which were named as pGL3-E-Tmod 3’UTR and pGL3-E-Tmod 3’UTR mutant, respectively. Dual luciferase assay demonstrated that the luciferase activity of pGL3-E-Tmod 3’UTR was reduced by miR-23b-3p mimic, while that of pGL3-E-Tmod 3’UTR mutant did not change. The data suggest that fluid shear stress downregulated the expression of miR-23b-3p and that miR-23b-3p was an E-Tmod targeting miRNA. ## miR-23b-3p regulates E-Tmod41 expression and F-actin content in MEL cells Since miR-23b-3p was shown to be an E-Tmod targeting miRNA, we next examined whether miR-23b-3p could regulate E-Tmod41 expression and thus affect F-actin content in MEL cells. miR-23b-3p was overexpressed or knocked down with mimic or inhibitor (Figs). Quantitative RT-PCR showed that mRNA level of E-Tmod41 was not changed by miR-23b-3p mimic or inhibitor (Figs). But the E-Tmod41 protein level was decreased or increased by mimic or inhibitor. F-actin was stained in MEL cells transfected with miR-23b-3p mimic or inhibitor. Confocal microscope images were taken (Figs) and the mean fluorescent intensities (representing F-actin content) were measured. Data showed that miR-23b-3p mimic reduced F-actin content in MEL cells, while miR-23b-3p inhibitor had the opposite effect. ## Fluid shear stress regulates *E-Tmod* alternative promoters The expression of *E-Tmod* gene was driven by two alternative promoters. One is located up-stream of non-coding exon 0 (E0), named as promoter E0 (P_E0), the other is located up-stream of exon 1 (E1), named as promoter E1 (P_E1). Our previous study showed that P_E0 preferably regulated the transcription of E-Tmod41, while P_E1 preferably regulated the transcription of E-Tmod29. Therefore, we constructed two reporter vectors, with the luciferase coding sequence driven by P_E0 or P_E1. MEL cells were transfected with two vectors, respectively, and then subjected to shear stress for 12 hours. Dual luciferase assay showed that P_E0 activity was augmented over 10 fold by shear stress , while P_E1 activity was reduced to 1/10 of the control level. ## Fluid shear stress induces erythroid differentiation and E-Tmod41 expression in mouse embryonic erythroblasts To test the results obtained in MEL cells, mouse erythroblasts were isolated from mouse E10.5 embryos and subjected to shear stress for 24 hours. Wright- Giemsa staining showed that un-sheared erythroblasts had the characteristics of polychromatic erythroblasts. After shearing, the dark blue in the cytosol became much lighter and the nuclei became condensed. Western blot results demonstrated that the protein expressions of E-Tmod41 and glycophorin A were greatly upregulated in sheared erythroblasts. These data suggests that shear stress could induce erythroid differentiation and E-Tmod41 expression in primary erythroblasts, which is consistent with the data obtained in MEL cells. # Discussion Previous efforts have proved that hemogenic endothelial cells could be induced to differentiate into hematopoietic cells both *in vivo* and *in vitro* by fluid shear stress in zebra fish and mouse. Furthermore, hematopoietic stem cells (HSCs) or multipotent progenitor cells circulate and differentiate in the embryonic blood vessels. Therefore, erythroid differentiation should also be regulated by shear stress. Here, by using both MEL cells and embryonic erythroblasts, we showed that fluid shear stress yielded in aortas of E10.5 embryos, could indeed induce the erythroid differentiation (Figs), which is consistent with Adamo’s observation. The effect of fluid shear stress on erythroid differentiation may come from its regulation on the transcriptional factor Krüppel-like factor 2 (Klf2), which has been proved to play a role in erythropoiesis. Our preliminary result showed that Klf2 mRNA level was significantly upregulated in MEL cells sheared for 6 hours (data not shown), which is consistent with previous studies. Findings of flow cytometry and confocal microscopy analyses showed that MEL cells treated with both fluid shear stress and DMSO had increased F-actin cytoskeleton content as compared to the control cells. This means fluid shear stress contributes to F-actin cytoskeleton remodeling during erythroid differentiation. Our previous work showed that DMSO treatment could upregulate the expressions of E-Tmod41 and its transcription factor GATA1 in MEL cells. In this study, we found fluid shear stress could upregulate the expression of E-Tmod41. Further studies with overexpression or knockdown of E-Tmod41 indicate that E-Tmod41 affects F-actin content (Figs), which may explain the mechanism for the effect of fluid shear stress on F-actin remodeling. E-Tmod29, an isoform of E-Tmod that can bind to TM and G-actin in the cytosol, was found downregulated in MEL cells after shearing (data not shown), suggesting that it may play a role in F-actin cytoskeleton remodeling. The downregulation of E-Tmod29 may free more TMs and G-actins, which are available for F-actin assembly. As the capping protein at the pointed end of F-actin, E-Tmod and its family members play important roles in regulating F-actin structures. Recent studies are mainly focused on E-Tmod and U-Tmod due to their wide expressions. The abilities of U-Tmod that capping the pointed end of F-actin and binding to TMs (including α/βTM, TM5b, TM5NM1) are weaker than E-Tmod. Therefore, U-Tmod is involved in the dynamic change of F-actin structure, and regulates the cell motility in endothelia and insulin-stimulated GLUT4 exocytosis in adipocytes; while E-Tmod is widely expressed in the terminally differentiated cells, such as erythrocytes, cardiomyocytes, and skeletal muscle cells, etc. U-Tmod was found in the hematopoietic stem and progenitor cells and plays an important role in terminal differentiation of fetal liver erythroid cells. During the erythroid differentiation, U-Tmod was downregulated and E-Tmod41 was upregulated, finally, leaving E-Tmod41 to be the sole protein that caps the pointed end of F-actin. This information suggests that U-Tmod may be downregulated during the erythroid differentiation induced by fluid shear stress, which would provide more pointed ends for E-Tmod41 to cap and accelerate F-actin remodeling. In addition, our data suggested that E-Tmod41 does not affect the erythroid differentiation (Figs A and B). This is consistent with previous findings that E-Tmod null mice only have a mild spherocytic elliptocytosis and seem to have no defect in erythroid differentiation. miRNAs are involved in both erythroid differentiation and the mechanical responses in endothelial cells, vascular smooth muscle cells, and macrophages. By using microRNA array, we found that many miRNAs are differentially regulated by fluid shear stress in MEL cells. Among them, mechanosensitive miR-23b-3p, which was proved to be an E-Tmod targeting miRNA, was found suppressed by shear stress. The suppression of miR-23b-3p resulted in the upregulation of E-Tmod41 protein expression and contributed to F-actin remodeling in MEL cells (Figs). We noticed that miR-23b was found upregulated by pulsatile shear stress in human umbilical endothelial cells and participated in the regulation of cell proliferation. The contradictory results may come from the different responses of one miRNA to different flow patterns in various cell types. It was reported that miR-23b could enhance the connections between breast cancer cells and that the inhibition of miR-23b enhanced their migration and deformation abilities. MiR-23b was found to directly target p21-activated kinase 2 (PAK2) and increase the phosphorylation of myosin II. In addition, miR-23b was found to target many genes that participate in cell cytoskeleton remodeling. These studies suggest that miR-23b can regulate cytoskeleton and deformation of cells. In our case, it contributes to F-actin cytoskeleton remodeling in sheared MEL cells by targeting E-Tmod41. Our data showed that fluid shear stress could activate P_E0 promoter to promote E-Tmod41 expression and suppress P_E1 promoter to inhibit E-Tmod29 expression. This is consistent with our previous observations that P_E0 activity is increased and becomes dominant in maturating reticulocytes, while P_E1 activity is high in undifferentiated erythroblasts. The differential regulations of E-Tmod41 and E-Tmod29 by fluid shear stress would be beneficial to F-actin cytoskeleton remodeling and morphological change in erythroid cells. Besides, it should be pointed out that, in our study of miRNA, the protein level of E-Tmod29 was not changed by miR-23b-3p (data not shown), which means E-Tmod41 but not E-Tmod29 is the target of miR-23b-3p. Therefore, the down-regulation of E-Tmod29 is mainly through the regulation of alternative promoters by fluid shear stress but not through miR-23b-3p. In our study, we used the cone-plate shearing device to shear the cells. Since MEL cells and embryonic erythroblasts are suspending cells, the shear stress acting on them in the shearing device may be much more complicated as compared to adherent cells, e.g., hemogenic endothelial cells. In addition, we used the shear stress yielded in the aortas of E10.5 embryos, 5 dyn/cm². But for the erythroblasts in bone marrows and reticulocytes circulating in the blood, the mechanical forces they sense (including shear stress and hydrostatic pressure) and their physical environments may not be the same as in the embryonic aortas. Therefore, it would be useful if we design devices mimicking the environments of bone marrow or blood vessels by microfluidics and micro-nano techniques. In conclusion, by using our experiment setup, our findings suggest that fluid shear stress induced the differentiation and F-actin cytoskeleton remodeling in nucleated erythroblasts. Fluid shear stress upregulated E-Tmod41 expression in MEL cells and embryonic erythroblasts and the change of E-Tmod41 contributed to F-actin cytoskeleton remodeling. The upregulation of E-Tmod41 by shear stress could be mediated by two mechanisms: the inhibition of its targeting miRNA, miR-23b-3p, and the activation of its promoter, P_E0. Our work would help us to better understand the factors that regulates erythroid differentiation, the molecular mechanisms for cytoskeleton remodeling in erythroid cells, and the mechanisms for the mechanical regulation of E-Tmod. # Supporting Information We thank Drs. Jing Zhou and Wei Kong in our department for their useful suggestions. We thank Dr. Wei Kong for using her microscope to take bright field images. We thank Ms. Fang Yu for her assists in flow cytometry analysis and Dr. Qihua He for her assists in confocal microscopy analysis. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: WY XW LAS. Performed the experiments: WM. Analyzed the data: WM WY. Contributed reagents/materials/analysis tools: XZ SZ DS WK. Wrote the paper: MW XW LAS WY.

# Introduction Cloud computing is popular in industry due to its ability to deliver on-demand resources according to a pay-as-you-go model. Usually, three basic service models are included in cloud computing: Infrastructure as a Service (IaaS), Platform as a Service (PaaS) and Software as a Service (SaaS). Namely, SaaS provides access to complete applications as a service. PaaS provides a platform for developing other applications on top of it, such as the Google App Engine (GAE) and Azure. IaaS provides an environment to deploy the managed virtual machines. Technically, when the users submit the requests, the providers would provide the resources depending on the users’ demand. As a key technique in cloud computing, the elasticity has the ability to acquire and release the resources according to the users’ demand. Generally, the providers implement an automatic provisioning approach via the virtualization technique. Virtualization makes it possible to rapidly scale the resources up or down. The aforementioned approaches present a reactive method, which is triggered by a certain threshold, such as CPU utilization or memory utilization. Actually, two or more thresholds should be used as a performance metric. In addition, it is important to provision the correct amount of the resources efficiently using a suitable threshold. In fact, the fluctuating workload would lead to an overprovisioning state or an underprovisioning state. To avoid these problems, researchers usually use a predictive technique, such as the proactive method. These feasible predictive approaches, such as machine learning, Moving Average, and Auto-Regression, would track the dynamic resource requirement and effectively minimize the energy consumption. This predictive policy would quantify the requirement in advance in order to flexibly scale the resources up or down. However, it is a challenging issue to improve the accuracy of the predictive technique. Additionally, an estimation error would lead to an overprovisioning or underprovisioning state. When in the sudden workload, this predictive method is especially inaccurate. Thus, combining this with an automatic method and a proactive method would be more agile for provisioning the resources. For example, the Elastic VM architecture provisions the resources dynamically to reduce the SLA violation. However, the elasticity is necessary to meet the users’ demand from different perspectives. Some researchers would take the performance metrics into consideration, such as the SLA and the profit of the providers. However, more metrics are used in the elasticity to evaluate the performance. For example, from the purpose of the providers, they might consider more related elements. That is, they would seek to minimize the renting cost, the energy consumption and the Service Level Agreement (SLA) violation. In summary, the elasticity would be implemented for one or two purposes, such as saving energy or reducing the cost. However, it is difficult to make an elasticity solution by considering multiple objectives. To solve the mentioned issues, we propose the ERP approach to provision the resources by the performance threshold, including the CPU and the memory. According to the threshold, we would flexibly scale the resources up or down by considering multiple perspectives. From the perspective of the provider, the goal is aimed at minimizing the amount of the resources to reduce the energy consumption. From the perspective of the users, the goal is aimed at rapidly scaling the resources up or down. In brief, the ERP approach is aimed at maximizing the utilization and minimizing the SLA violation. Then, the main contributions would be summarized in the following. First, this approach solves the suitable threshold to determine the users’ demand. We present the performance threshold by using the GRA method, which considers such multiobjectives as the CPU utilization and memory utilization. Meanwhile, it is instructed on the cloud layer model using the MAPE loop. Usually the MAPE loop includes four phases, such as Monitoring (M), Analysis (A), Planning (P) and Execution (E). Second, this approach solves the issue of scaling the resources flexibly. According to the proposed threshold, we could efficiently scale the resources up or down. That is, we propose a fine-grained algorithm, which means to scale up the resources from the PM-level or the VM-level in order to flexibly meet the users’ demand. Third, this approach solves the issue of reducing the overheads. When it is overprovisioned, we would shut down the extra machines to reduce the energy consumption via a simple predictive technique, such as the weighted moving average (WMA). The remainder of the paper is described as listed below. Section 2 analyses the related literature on the elastic techniques in cloud computing. Section 3 presents the ERP framework based on the layer model. Section 4 provides the performance threshold via the cloud layer model. Section 5 presents the effective ERP algorithm, which would scale the resources up or down from different granularities. Section 6 proves the results by comparing them with the aforementioned approaches. Finally, section 7 draws conclusions and describes future development. # Related work Usually the elastic solution is implemented by scaling the resources in or out. By analyzing some related works, we would divide the elastic resource provisioning approaches into two major aspects, including automatic scaling methods and elastic mechanisms on the predictive technique. ## Automatic scaling methods In the automatic policy, the resources would be provisioned and released automatically according to the demand. Generally, the action is triggered by the fixed thresholds, such as the utilization. The common techniques are provided by Amazon and Scalr. However, they provision the resources only based on the utilization, when in fact more elements have taken effect. Additionally, its advantage is a kind of coarse-grained provisioning strategy to scale the virtual machines. When considering the fine-grained provisioning strategy, some researchers focus on the reactive methods by resizing the resources dynamically and minimizing the response time and executing cost in cloud computing. However, they focus more on the fine-grained scaling strategy, and less on multiple perspectives. Kingfisher proposed an elastic mechanism to reduce the transition of time and cost. This approach exploits the available resources on the virtual machines to scale in or out, and uses an integer linear program formulation to optimize the cost. Leitner et al. proposed the SLA-aware scheduling algorithm, which would reduce the request execution time. It presents a cost-efficient method to scale up from the perspective of the providers. In contrast, our approach considers more factors to formulate the threshold by the cloud layer model, such as CPU utilization, memory utilization, etc. Additionally, we aim to scale the resources by minimizing the renting cost and response time. This would shut down the spare machines from the perspective of saving the consumption. By analyzing the mentioned works, we determined that most recent elastic strategies focus on the horizontal elasticity. Therefore, it is important to scale the resources from different granularities, including horizontal elasticity and vertical elasticity. By considering the fine-grained elasticity, we present the ERP algorithm to scale up the resources in the PM-level or VM-level by the performance threshold. Moreover, when it is in overprovisioning, it would scale down the resources in the VM-level. ## Elastic mechanisms based on the prediction In fact, elasticity is essential to meet a fluctuating workload, and it is necessary to determine the suitable amount of the resources in order to scale the resources. Actually, the proactive approaches are used to determine the next demand, such as the Autoregressive moving average model (ARMA) and Holt winter. These predictive techniques have the advantage of giving an accurate prediction value in the stable workload. However, these predictive techniques focus more on the accuracy, but ignore the complexity. Moreover, when a sudden workload appears it might be in estimation error. To reduce the complexity of the prediction algorithm, some techniques are used to determine the repetitive patterns and predict the next values. PRESS is a predictive elasticity system that analyzes and extracts the workload patterns and provisions the resources automatically. The advantage of this policy is that improves the prediction accuracy, and it reduces the resource waste efficiently. However, it only makes emphasis on the overhead. CloudScale is a system that automates the fine-grained resources in cloud computing infrastructures, determining the adaptive resources by the prediction. In addition, it integrates the dynamic CPU voltage scaling to saving the consumption by migration. This technique puts more emphasis on the proactive method based on the prediction, which would minimize the energy consumption and avoid the Service Level Object (SLO) violation. In fact, more elements should be taken into consideration. Hence, in our approach, we consider more elements, such as reducing the renting cost, energy consumption and SLA violation. Additionally, we increase or decrease the resources automatically from different granularities to meet the demand, including fine-grained scaling and coarse-grained scaling. Namely, when it is underprovisioning, our approach scales up the resources from different granularities by the performance threshold, such as in the PM-level or VM-level. In contrast, we scale down the VMs by the WMA predictive technique efficiently. # Proposed approach In this section, we present our proposed approach for the detailed description. Our approach is designed on the cloud layer model. That is, this policy is implemented to determine the performance threshold to flexibly scale the resources up or down. Additionally, the formulation of the performance threshold is presented in detail in the next section. Then the ERP framework is explained in detail in the following. ## Cloud layer model In this section, our approach describes a cloud layer model to scale the resources rapidly. The cloud layer model focuses more on the quantitative analysis, whereas the Delphi method depends more on the subjective assessment. The ERP approach is implemented on the cloud layer model. The layer model is composed of three parts: SaaS, PaaS and IaaS. The SaaS determines a series of requests offered by the users. In the PaaS the broker is responsible for provisioning the infrastructure resources according to the users’ demandwhich is presented by the MAPE loop. In IaaS, the datacenter is composed of some PMs and VMs. The provider would provision the resources according to the requests. As depicted in, the key components of the MAPE are described in detail as follows. ### Monitor (M) The monitoring component collects some metrics, such as the CPU utilization, memory utilization and some available resources. It monitors the information every five seconds. The key information is collected, aggregated and calculated by the performance model, which is described in detail in the next section. ### Analyze (A) The analyzing phase is responsible for analyzing the collected information. The obtained data is aggregated and calculated by the performance model, and we achieve the performance value to decide whether the scaling action is triggered. Moreover, we use the WMA predictive technique to determine the correct number of the servers and shut down the spare machines. ### Plan (P) This component is the core of the cloud layer model. According to the users’ demand, it implements the scaling strategy by minimizing the renting cost and reducing the energy consumption. Additionally, it would increase or decrease the resources by the performance threshold. ### Execute (E) In the executing phase, the Nginx load balancing server balances the web requests by provisioning the servers in the infrastructure. Since the VMs are hosted in the PMs, the provider would provision the resources according to the demand by using the proposed plan. ## Proposed framework In our approach we propose a novel framework to flexibly increase or decrease the resources aiming at minimizing the renting cost, energy consumption and response time, as illustrated in. The ERP algorithm is mainly composed of two phases. In the first phase, the performance model constructs a baseline threshold, which is aggregated and calculated by the gathered data. From this the resources would be rapidly scaled up or down. In the second phase, the ERP algorithm is used to scale the resources by the performance threshold for the purpose of minimizing the renting cost and saving power consumption. Then, we explain these two phases in detail. In the first step, the monitoring component monitors the CPU utilization, memory utilization, CPU clock speed and some available resources. We aggregate the gathered data to make a performance evaluation by the proposed cloud layer model. In the second step, we make a further description on the ERP approach. In the analyzing component, we scale the resources by the performance threshold. Actually, the planning phase may lead into two states, including an underprovisioning state or overprovisioning state. When it is in an underprovisioning state, we execute the action on increasing the resources at the PM-level. If it continues, we go on scaling up the resources at the VM-level. The PM-level scaling depends on the available resources in the same host. The VM-level scaling is based on the VMs hosted on the PMs. Additionally, the VM could come from the same PM or another PM. Otherwise, when it is in overprovisioning we scale down the resources by the prediction. Then the extra spared machines would be shut down by saving the energy consumption. Moreover, our approach implements the elastic scaling from different granularities with the consideration of minimizing the cost and the SLA violation. # Performance threshold In this section, we present a performance threshold on multiple elements. From this we would rapidly scale the resources up or down in cloud computing. ## TOPSIS and GRA policy This policy presents a multicriteria threshold that takes five related criterion into account, as shown in. The criteria on the TOPSIS and GRA policy would include the cost type and benefit type. After the matrix is normalized, the TOPSIS method evaluates them by the positive ideal solution and negative ideal solution. Then, the GRA method makes the decision from less information and explores the system behavior by analyzing the related degree. Usually the information on the PMs is gathered every 5 seconds to form the decision matrix, as shown in. The gathered data is described as depicted in. Then we construct and implement the performance threshold in detail as follows. <img src="info:doi/10.1371/journal.pone.0216067.e001" id="pone.0216067.e001g" /> R = \[ r c y c l e 1 r r e s t M e m 1 r c p u % 1 r m e m % 1 r a v a i l % 1 r c y c l e 2 r r e s t M e m 2 r c p u % 2 r m e m % 2 r a v a i l % 2 r c y c l e 3 r r e s t M e m 3 r c p u % 3 r m e m % 3 r a v a i l % 3 \] ### Normalization of the decision matrix In the first step we normalize the decision matrix. Namely, the decision matrix is normalized by achieving the average value of every column as listed in. <img src="info:doi/10.1371/journal.pone.0216067.e002" id="pone.0216067.e002g" /> R = \[ r c y c l e 1 S U M c y c l e r r e s t M e m 1 S U M r e s t M e m r c p u % 1 S U M c p u % r m e m % 1 S U M m e m % r a v a i l % 1 S U M a v a i l % r c y c l e 2 S U M c y c l e r r e s t M e m 2 S U M r e s t M e m r c p u % 2 S U M c p u % r m e m % 2 S U M m e m % r a v a i l % 1 2 S U M a v a i l % r c y c l e 3 S U M c y c l e r r e s t M e m 2 S U M r e s t M e m r c o u % 3 S U M c p u % r m e m % 3 S U M m e m % r a v a i l % 3 S U M a v a i l % \] ### Improved TOPSIS This is the abbreviation of the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). The traditional TOPSIS method depends more on subjective weights, while the improved TOPSIS solutions depend more on key factors. In the second step, the ideal solution would be determined by and. That is, for the cost type the ideal solutions are the smaller ones, and the negative solutions are the larger ones. It is the opposite situation for the benefit type. Then, we achieve the positive ideal solution and the negative ideal solution, respectively. <img src="info:doi/10.1371/journal.pone.0216067.e003" id="pone.0216067.e003g" /> P j \+ = { ( max ( r ˜ i j ) \| i ∈ I ) , ( min ( r ˜ i j ) \| i ∈ J ) } <img src="info:doi/10.1371/journal.pone.0216067.e004" id="pone.0216067.e004g" /> P j − = { ( min ( r ˜ i j ) \| i ∈ I ) , ( max ( r ˜ i j ) \| i ∈ J ) } ### Grey relational analysis Grey theory is an effective way to solve multiobjective decision problems in the engineering areas. In the following step, we determine the difference between the comparative series *r*_*jk* and the standard series $P_{k}^{+}$ or $P_{k}^{-}$. Additionally, the distinguish coefficient *ρ* is usually 0.5, and is generally between \[0, 1\]. Then, the Grey relational coefficients *ς*⁺and *ς*⁻ are constructed by and, respectively. <img src="info:doi/10.1371/journal.pone.0216067.e007" id="pone.0216067.e007g" /> ς \+ ( k ) = min j min k \| r j k − P k \+ \| \+ ρ max j max k \| r j k − P k \+ \| \| r j k − P k \+ \| \+ ρ max j max k \| r j k − P k \+ \| <img src="info:doi/10.1371/journal.pone.0216067.e008" id="pone.0216067.e008g" /> ς − ( k ) = min j min k \| r j k − P k − \| \+ ρ max j max k \| r j k − P k − \| \| r j k − P k − \| \+ ρ max j max k \| r j k − P k − \| Actually, the weight coefficients are determined by the analytic hierarchy process (AHP) method. Then we determine the degree of relation *r* on the weight coefficients *ω* by multiplying them by Grey relational coefficient *ς*(*k*). Additionally, the degree of relations *r*⁺ and *r*⁻ are formulated by Eqs and, respectively. <img src="info:doi/10.1371/journal.pone.0216067.e009" id="pone.0216067.e009g" /> r \+ = ∑ k = 1 m ω k ς \+ ( k ) <img src="info:doi/10.1371/journal.pone.0216067.e010" id="pone.0216067.e010g" /> r − = ∑ k = 1 m ω k ς − ( k ) Then we formulate the relative closeness coefficient *u*⁺ to the ideal solution by, which is implemented on the ideal relational coefficient *r*⁺ divided by the sum of the positive relational coefficient *r*⁺ and negative relational coefficient *r*⁻. <img src="info:doi/10.1371/journal.pone.0216067.e011" id="pone.0216067.e011g" /> u \+ = r \+ r \+ \+ r − ## Performance model The performance threshold is constructed by the entropy method, which is an effective method to calculate the deviation degree. The smaller the entropy value is, the better the performance is. Similarly, the larger entropy value is, the worse the performance is. Therefore, we determine the performance threshold by the entropy method, which is listed as in. <img src="info:doi/10.1371/journal.pone.0216067.e012" id="pone.0216067.e012g" /> Δ P = − log 2 ( P 2 P 1 ) = − log 2 ( u 2 \+ u max / u 1 \+ u max ) = − log 2 ( u 2 \+ u 1 \+ ) Where Δ*P* is the performance threshold by the entropy method, *P*₁ is the probability before the demand varies, and *P*₂ is the probability after the demand varies. Additionally, the probability is constructed on the ideal relational coefficient *u*⁺ divided by the max relational coefficient *u*_max. In the scheduling, a current performance value below 0.1 denotes a better performance environment. When it is above 3, it denotes a poor performance environment. In fact, a normal value is between 0.1 and 3, which is described in. In our experiments, when the performance value is lower than 0.1, we would scale down the servers. Then, we set 0.1 as the lower threshold *P*_*d*. When the value is greater than 0.2, we would scale up the servers for the purpose of reducing the response time by reserving slightly more resources. Then, we set 0.2 as the upper threshold *P*_*u*. # The ERP algorithm In this section we describe the ERP algorithm to scale the resources from different granularities according to the users’ demand. ## ERP algorithm To provision the resources flexibly, we first discuss some related definitions on the elasticity, such as the resilience and scalability. Next, we define them and clarify the difference between them. **Scalability** means to the ability of the system to deal with an increasing amount of the servers in a capable manner. However, it focuses more on the increasing ability, and less on the response time. **Resilience** means to provision the resources rapidly in a flexible way. Elastic scheduling refers to two core conditions, including the time and speed. In this paper, we define an elastic scheme *S*, which is represented as *S* = (* clock*,*U*_*cpu*%,*U*_*mem*%,*P*_*u*,*P*<sub>*d*</ sub>), where *clock* is the CPU cycle, *U*_*cpu*% and *U*_*mem*% are the CPU utilization and the memory utilization, respectively, which are gathered by monitoring the system, and *P*_*u* and *P*_*d* are the upper and lower thresholds, respectively. In brief, the main algorithm (refer to Algorithm 1) provisions the resources rapidly via the MAPE loop. In the monitoring and analyzing components, some key elements are collected to determine the performance threshold. Then in the planning and executing components, the elastic scheme would scale the resources by the performance threshold. To make the ERP algorithm understood for the further step, lists the main parameters of the ERP algorithm as below. Next, the ERP algorithm is described in detail. It implements an elastic resource provisioning approach in the datacenter. This algorithm takes the performance threshold as the baseline to scale the resources up or down. At first, the monitoring component would collect and gather the information as listed in (lines 1–2) every few minutes. In fact, the ERP algorithm would increase or decrease the resources to meet the users’ demand. When the performance value *P* is larger than the upper threshold *P*_*u*, the algorithm would be triggered to scale up the servers (SUS) (lines 4–5).In contrast, once the current performance value is below the threshold *P*_*d*, the scaling down the servers (SDS) algorithm is triggered (lines 6–7). **Algorithm 1**. **ERP (Elastic resource provisioning)** 1: Initialization: Server, P 2: while (the allocation is deploying) 3: monitor the performance value P 4: if P \> *P*_*u* 5:     Scaling up the servers (SUS) 6: else if P \< *P*_*d*       Scaling down the servers (SDS) 7: End The proposed ERP algorithm has included two aspects. First, the scaling up the servers (SUS) algorithm proposes a scaling method that is based on different granularities. That is, we scale up the VMs in the same available PMs or from some different PMs. Second, the scaling down the servers (SDS) algorithm presents the approach to shut down the extra machines. ## The SUS algorithm The SUS algorithm is intended to scale up the resources in a flexible way, including from the PM-level or VM-level. The SUS algorithm is described by Algorithm 2. The monitoring component collects some metrics related to the resources (lines 1–4). If the performance evaluation reaches the upper threshold *P*_*u*, it scales up more available resources on the PM (lines 5–7). When the updated performance value continues past on the upper threshold *P*_*u*, we would provision slightly more resources (lines 8–10). Additionally, the VMs might come from different PMs. **Algorithm 2**: **SUS (Scaling up the servers)** 1: Begin 2: Initialization: Server, P 3: while (the allocation is deploying) 4: monitor the performance value P 5: if P \> *P*_*u* 6:     Scaling up the PMs 7:     update the performance value P 8: while (P \> *P*_*u*) 9:     Scaling up the VMs 10:     update the performance value P 11: End ## The PLI algorithm The purpose of the PM-Level increasing (PLI) algorithm is to increase the VMs on the available PMs (refer to Algorithm 3). Then we explore the PLI algorithm in detail. The monitoring component aggregates the information and calculates the performance value (lines 1–3). Once the triggered action appears we scale up the residual resources on the available PMs. Then we would choose the PMs aimed at minimizing the renting cost (lines 4–7). Additionally, the cost function is described by Eqs and. Finally, it updates the performance value (line 8). **Algorithm 3**: **PLI (PM-level increasing)** 1: Begin 2: Initialization: Server, P 3: Calculating the performance value 4: while (P \> *P*_*u*) 5:     if PM is available 6:     select the min cost PM to increase 7:     update the performance value 8: End In this phase, is aimed at minimizing the renting cost, where *u*_*cpu*% presents the CPU utilization of the VM. The binary variable *v*_*j* indicates whether or not the VM is selected, and the binary variable *p*_*i* indicates whether or not the PM is selected. The parameter *m* is responsible for the amount of VMs hosted on the current host, and *c*(*p*_*i*) is the expending cost of the current host. <img src="info:doi/10.1371/journal.pone.0216067.e013" id="pone.0216067.e013g" /> M i n i m i z e p i • ∑ j = 1 m v j • ( 1 − u c p u % ) • c ( p i ) / ∑ j = 1 m v j <img src="info:doi/10.1371/journal.pone.0216067.e014" id="pone.0216067.e014g" /> s u b j e c t t o : ∑ j = 1 m v j ≤ m , v j ∈ { 0 , 1 } , P i ∈ { 0 , 1 } ## The VLI algorithm In this section, we propose the VM-level increasing (VLI) algorithm (refer to Algorithm 4) to continue increasing the resources to meet the fluctuating demand. It consists of three parts: monitoring the component, increasing the resources and updating the state. Then we describe the algorithm 4 in detail. First, the monitoring component gathers the information to calculate the performance value (lines 1–3). Second, we choose the suitable VMs to increase (lines 4–5), which would determine minimizing the expending cost by. That is, we calculate the remaining utilization by the CPU and the memory. We implement the cost in by multiplying the remaining utilization by the single VM renting cost. The purpose of the function is to achieve the VM with a minimum cost, where the binary variable *v*_*j* indicates whether or not the VM is selected in. Then we make a global search to find a suitable VM to increase. Finally, we update the state and calculate the performance value (lines 6–7). **Algorithm 4**: **VLI (VM-level increasing)** 1: Begin 2: Initialization: Server, P 3: Calculating the performance value 4: while (P \> *P*_*u*) 5:     select the min cost VM to increase 6:     update the performance value 7: End $$C_{vm} = \arg\min(v_{j}(1 - u_{cpu\%}) \bullet (1 - u_{mem\%}) \bullet c_{}(v))$$ $$j \in \left\{ 1,2,\ldots,m \right\},v_{j} \in \left\{ 0,1 \right\}$$ ## The SDS algorithm The aforementioned algorithms (refer to Algorithms 2–4) implement increasing the resources from a different granularity according to the users’ demand. In this section, the Scaling-down servers (SDS) algorithm is described for the detailed steps. In the first step we monitor the component and gather some information to achieve the performance threshold (lines 1–3). Once the SDS algorithm is triggered we would scale down the resources. Then we select the extra machines to shut down for the purpose of minimizing the cost (lines 4–5). Hence, we shut down the machines that occupy the maximum expending cost. is as listed below. Finally, we update the state and determine the current performance threshold (lines 6–7). <img src="info:doi/10.1371/journal.pone.0216067.e017" id="pone.0216067.e017g" /> C v m ' = arg max ( v j ( 1 − u c p u % ) • ( 1 − u m e m % ) • c ( v ) ) **Algorithm 5**: **SDS (Scaling-down servers)** 1\. Begin 2\. Initialization: Server, P 3\. Calculating the performance value 4\. while (P \< *P*_*d*) 5\.     select the max cost VM to decrease 6\.     update the performance value 7\. End # Experiments In this section, we implement the elastic resource allocation strategy based on the performance criterion. Meanwhile, the proposed approach proves that it is appropriate for meeting the demand in different kinds of workloads. In addition, this approach considers both reducing the renting cost and improving the utilization. ## Environment setup We use the CloudStack platform and simulated real-world workloads to evaluate the ERP approach. We deploy a cluster composed of ten PMs. One installs the CloudStack platform. The other nine PMs use Xenserver as the management nodes (2.20GHz Intel(R) Xeon(R) 8 CPU, 8 G memory, running CenOs 6.9). We create 27 VMs (1 VCPU, 1 G memory, running CenOs 6.9) in the cluster. Then, the database is run off MySQL. When the workload is fluctuating, the Nginx has the function off balancing the servers. All the configuration information is listed in. To evaluate the proposed approach, we design two kinds of workloads: synthetic workloads and real-world workloads. We use the Jmeter to generate the requests based on the TPC benchmark. First, the synthetic workload would vary from the users’ demand. The fluctuating process of the workload is described as below. The load generator would implement 600, 900, 600, 1200, 600, and 1800 users, which is shown in, which lasts for over 30 minutes. Second, the simulated real- world workload is extracted from the EPA and NASA traces. The two kinds of real- world workload traces are generated as shown in. Additionally, the monitoring service is implemented by the Jmeter plugins, such as monitoring the response time, CPU utilization or memory utilization. The experiment would last for over 40 minutes. ## Evaluation metric In the experiments, we consider some performance indicators as the metrics, such as the renting cost, energy consumption, resource utilization and SLA violation. ### The cost This metric might be measured by the reserved and on-demand VMs. For example, the basic unit of the CPU is set at 1 GB in Aliyu. It is charged 0.059 ¥/hour in the reserved plan and 0.28 ¥/hour for the on-demand plan. The renting cost is defined in, where *C*^*r* and *C*^*o* are responsible for the renting cost in the reserved or on-demand plan, respectively. Then in the scheduling the average overhead is described in, where it is calculated by the sum of the cost divided by the time interval *T*. <img src="info:doi/10.1371/journal.pone.0216067.e018" id="pone.0216067.e018g" /> C = C t r x t r \+ C t o x t o <img src="info:doi/10.1371/journal.pone.0216067.e019" id="pone.0216067.e019g" /> C a v g = ∑ i = 1 t C T ### Energy consumption This metric might be measured by the average energy consumption, which is defined as the energy consumption ratio as listed in, where *N* is the total number of the intervals. Additionally, the energy consumption is expressed in. <img src="info:doi/10.1371/journal.pone.0216067.e020" id="pone.0216067.e020g" /> P o w e r = k × P max \+ ( 1 − k ) × P max × u Where the idle power consumption coefficient *k* is equal to 0.7, and the parameter *P*_max represents the peak power. Additionally, *u* is based on the CPU utilization. <img src="info:doi/10.1371/journal.pone.0216067.e021" id="pone.0216067.e021g" /> P o w e r a v g = ∑ i = 1 n P o w e r N ### The utilization The utilization is one of key indicators to evaluate the performance in the scheduling. The average utilization is defined as the ratio between the total CPU utilization and the total number of the intervals, as shown in. <img src="info:doi/10.1371/journal.pone.0216067.e022" id="pone.0216067.e022g" /> U t a v g = ∑ i = 1 n U t N ### SLA violation The SLA violation can be calculated by the percentage of the difference between the actual requests and allocated requests divided by the total requests, as described in. Generally, the SLA violation might be measured by the CPU utilization, just in. Then the average SLA violation is defined as the ratio between the total SLA violation and the total number of the intervals, expressed by. In fact, the SLAV is expressed by the average SLA multiplied by the average response time, as shown in. <img src="info:doi/10.1371/journal.pone.0216067.e023" id="pone.0216067.e023g" /> S L A = Re q t o t a l − Re q a l l o c a t e Re q a l l o c a t e <img src="info:doi/10.1371/journal.pone.0216067.e024" id="pone.0216067.e024g" /> s l a = 1 1 \+ e ( U c p u − 0.8 ) <img src="info:doi/10.1371/journal.pone.0216067.e025" id="pone.0216067.e025g" /> s l a a v g = ∑ i = 1 n s l a i N <img src="info:doi/10.1371/journal.pone.0216067.e026" id="pone.0216067.e026g" /> S L A V = s l a a v g × t i m e ## Algorithms in comparison To validate the ERP algorithm, we compare it with other algorithms, such as lightweight resource scaling (LS) algorithm, the proactive method, and the reactive method. ### Reactive method The traditional algorithm is scaled by the CPU utilization, obeying the simple principle by a rule-condition-action. In the experiments, the threshold is usually fixed at 0.8 or 0.2. Namely, when the utilization is higher than 0.8, the VMs would be increased. In contrast, when the utilization is lower than 0.2, the resources would be decreased. ### Proactive method The proactive method means that it would scale the servers up or down by the prediction technique, such as ARMA. That is, it could scale the resources up or down by the ARMA. ### LS The LS algorithm focuses more on the response time. When it is higher than the upper threshold the number of the VMs increases. In contrast, the number of the VMs would be scaled down. Additionally, the algorithm would shut down the spare machines by a simple predictive technique. ## Experiment results Actually, our proposed algorithm is constructed on the performance value, which is calculated by the GRA and TOPSIS policy. In more experiments we determine that the performance threshold range is between 0.1 and 0.2. Namely, when it is greater than 0.2, we would scale up the servers, and when it is lower than 0.1 we would scale down the servers. Moreover, the performance evaluation considers multiple angles, such as maximizing the utilization, and minimizing the power consumption and the SLA violation. The results prove the effectiveness of the ERP approach. ### The number of the servers In the synthetic load, the reactive algorithm puts a greater emphasis on the scalability of the servers and reacts quickly at first. The proactive algorithm would obtain the suitable number of the servers in the regular load test, and the LS algorithm spends less resources. Our proposed algorithm could occasionally occupy slightly more resources than the LS algorithm to meet the multidimension requirement in the simulated experiment at the beginning, as shown in. In the real-world load, including EPA and NASA, our algorithm would occupy slightly more resources at first. Next it would outperform other algorithms in the normal level, as illustrated in Figs and. We determine that the LS algorithm is unsuitable for various loads. That is, because the LS algorithm depends more on the response time. When a sudden load appears, it would increase the overhead. However, our approach has the advantage of efficiently avoiding a sudden load efficiently by reserving slightly more resources. ### The renting cost We measure the renting cost using. As shown in, in the synthetic load the LS algorithm puts a greater emphasis on the time to scale the resources. Namely, in the stable workload it gains the smallest average renting cost. We find that the ERP algorithm spends a slightly higher cost than the LS due to reserving few resources at first. The proactive algorithm would obtain a better result in the regular load test by the prediction. Our proposed algorithm obtains a lower cost than the reactive algorithm. As shown in, in the real-world load we find that our proposed algorithm obtains a lower cost than the other algorithms, and the LS algorithm obtains a higher cost depending on the response time. When it appears in the sudden load, the LS algorithm would scale up the resources more quickly, which makes the occupied resources greater than in other algorithms. ### Resource utilization We measure the average resource utilization based on. Figs and show the CPU utilization during the experiments under different workloads, including the synthetic load and real-world loads. In these experiments, we determine that the proposed approach utilizes the resources more fully, which is depicted in Figs and. In the experiments the ERP method consumes slightly more resources at first and simultaneously guarantees a lower SLA violation rate. Additionally, it releases the servers by the WMA prediction by guaranteeing the performance in the varying workloads. We see that no resource utilization is higher than 100%, which proves that our approach efficiently reduces the underprovisioning state. ### Response time The response time is another performance metric that needs to be studied. As depicted in Figs and, in the synthetic workload, when considering the maximum response time, we determine that our proposed algorithm obtains a quicker response than the other algorithms by reserving few resources at first. For the average response time, these algorithms are in the acceptable level at the stable workload. As depicted in Figs and, in the real-world loads we find that our algorithm presents a lower maximum or average response time than others by reserving slightly more resources at first, while the LS algorithm obtains slightly higher time due to a longer monitoring time. Additionally, it is unfit for the sudden load. In the NASA load the variable workload leads to inaccurate prediction values, so the proactive algorithm obtains a longer average response time. ### SLA violation We measured the SLA violation based on. As shown in Figs and, in the workloads our algorithm presents a lower SLA violation ratio than the other algorithms. In addition, the error rate is another metric that evaluates the performance. As listed in, we see that our algorithm produces a slightly lower error ratio and efficiently avoids the sudden load. ### Average energy consumption We measure the average energy consumption based on. As shown in, in the synthetic load our algorithm achieves a lower power than the LS and proactive algorithms. Since it is in the stable workload, the reactive algorithm obtains a better result than the other algorithms only by the utilization. As shown in, in real-world loads the proposed algorithm presents a lower power than the LS and reactive algorithm. The proactive method consumes less energy consumption than others, but it cannot meet the demand due to the inaccurate prediction. This is because that it achieves a higher error rate in. # Conclusion Traditional elasticity is often used as a reactive method, which is implemented by the rule-condition-action. However, it would be a better strategy to combine this with the prediction. In this paper, we present an elastic strategy that increases or decreases the resources by the performance threshold in a flexible manner. To further elaborate, the ERP approach makes the following contributions. First, we present the performance threshold depending on the CPU and the memory. By this, we could flexibly scale the resources up or down. This solves the issue of deciding a suitable threshold on multiple elements. Second, we propose an SUS algorithm that implements the fine-grained scaling in the PM- Level or VM-Level to increase the resources flexibly. This solves the issue of an elastic scaling strategy from different granularities to reduce the SLA violation and response time. Third, combining this with the WMA prediction we propose the SDS algorithm to scale down the servers. Then we would shut down the spare machines to save energy consumption. This solves the issue of effectively saving the overheads. Finally, we evaluate the proposed ERP approach in the simulated and real-world workloads. The results show that the ERP method improves the utilization, minimizes the renting cost, saves the energy consumption and gives a quicker response time. In fact, we implement the scaling approach on the premise of regarding the servers as the available resources. However, no cloud provider offers unlimited resources, except for Google and Amazon. Thus, a further study should be made on some aspects. First, it is necessary to find an effective way to minimize the renting cost by reserving some available resources in advance. However, more servers would be wasted by reserving too many resources. Therefore, it is necessary to balance the reserved plan and the on-demand plan. Second, from the perspective of minimizing the energy consumption, a reasonable dynamical provisioning approach might efficiently consolidate the available resources by the migration technique. Then in the future it will be necessary to explore the dynamical provisioning approach in the complex workloads. Perhaps some typical types of the workflow would be an interesting extension in the future. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Peste des petits ruminants (PPR) is a contagious viral disease that primarily affects domestic and wild small ruminants. Because of its high morbidity and mortality, PPR is responsible for heavy economic losses in livestock husbandry across many developing countries. It is considered a significant threat to the global goat and sheep industry. PPR is caused by peste des petits ruminants virus (PPRV), a member of the family *Paramyxoviridae* and genus *Morbillivirus* along with Distemper, Rinderpest, and Measles viruses. To date, four PPRV lineages (I-IV) have been identified worldwide. Lineage IV is the dominant strain in Asia, including the entire Trans-Himalayan region (THR). PPRV has a tropism for epithelial and lymphoid cells. The virus can exist in different host body tissues and is discharged from the body through various secretions and excretions. These secretions and excretions, including the respiratory droplets, become the source of PPRV, allowing the transmission of the virus through close contact and aerosols. The clinical symptoms of PPR typically begin with dullness and fever (\>40 °C). Subsequently, there is the development of oral mucopurulent discharge, ocular discharge, and eventually, oral lesions, bronchopneumonia, and diarrhea. The severity of this disease is determined by the strain of the virus, local environmental features, and the immune status of the infected host. The morbidity and mortality of PPR can vary between 10%–90% and 50–90%, respectively. The typical latency period of PPR is 4–6 days, whereas the longest incubation period reported is 21 days. The major hosts of PPRV are livestock, such as sheep and goats. Wildlife is also an important target for PPRV. The main targets are antelope (such as saiga antelope—*Saiga tatarica mongolica*), ibex (such as Siberian ibex- *Capra sibirica*, Sindh ibex—*Capra aegagrus blythi*), gazelle (such as goitered gazelle—*Gazella subgutturosa*, Arabian gazelle—*Gazella arabica*). In Asia, clinical signs and mortality of PPR in wildlife have been reported essentially as the same as those in livestock, which provides a basis for interspecific transmission. This situation is different from the non-clinical infection in Africa, i.e. no viral shedding even if an antibody is produced. Especially in THR, bharal (*Pseudois nayaur*), markhor (*Capra falconeri*), blackbuck (*Antilope cervicapra*), and Himalayan goral (*Naemorhedus goral*) have been found to exhibit obvious clinical signs. Among them, strains from bharal and markhor were successfully isolated. Phylogenetic analysis showed that they were closely related to the strains isolated from livestock, suggesting a potential relationship between them. PPR was first reported in West Africa in 1942 and spread across Africa and Asia. Available research indicates that China, India, and Nepal are all PPR epidemic countries. Two PPR epidemics have been documented in China; the first occurred in 2007 in the Tibet Autonomous Region of China. The more severe outbreak occurred between the end of 2013 and the first half of 2014, which was first identified in the Xinjiang Uygur Autonomous Region and later spread to more than 20 provinces in total. This consequence in more than 30,000 sheep infections, of which 10,000 animals died. In response to the heavy losses caused by PPR, the World Organization for Animal Health (OIE) and the Food and Agriculture Organization (FAO) have set the goal of eradicating PPR globally by 2030. It has been suggested that the global eradication of PPR could return benefits of about \$74 billion over 15 years. However, the continuous epidemiological cycles of PPR worldwide constitute a great challenge to eradicating the disease. This calls for a deeper understanding of its temporospatial characteristics. The THR is an important endemic and high-risk area of PPR where monitoring and prevention measurements are difficult to implement. This can be majorly attributed to the rugged natural geographical conditions and the low effective livestock system. The poor veterinary services further complicate the situation in the region. Thus, a forecasting system would be a strong aid in defining the key points of prevention to save time, labor, and products for underdeveloped countries and regions. According to reports, the PPR risk exists across the THR, in which livestock serve as the maintenance hosts. Wildlife possibly plays the role of bridge hosts, in which virus transmission is not maintained but can persist for a while and be transmitted back (spillback) to livestock. Although scientists argue for the direct epidemiological linkage at the interface of livestock and wildlife, the interspecies transmission of PPRV during grazing and at water sources has been confirmed. Sharing the use of rangelands by livestock and wildlife can lead to disease transmission. Abubakar et al. pointed out that an outbreak of PPR in Sindh ibex was due to the spillover of the virus from a recent outbreak of PPR in nearby domestic small ruminants. Similar PPR spillovers to wild hosts are reported in Tibet and the Ngorongoro Conservation Area in northern Tanzania. Except for trade, free migration of wildlife and nomadism-driven back and forth movement of livestock enable their meeting in the same space (contaminated or not) across time. Both animals and humans prefer low-energy-consuming surfaces during movement, which has become its driving force. If the migration and grassland sharing of the multiple PPRV hosts last, reliance on the contamination of habitats within the latent period of the disease is expected, and the probability of direct contact via contaminated grassland is increased. While it would be arbitrary to conclude that the infection occurred on the cross-country paths, the potential communication of risk among the Trans-Himalayan neighboring countries is worth monitoring. We assume that interspecies transmission of PPRV occurs on small ruminants in THR, which forms the basis for the transboundary transmission of PPR. Initially, we predict the distribution of PPR on both sides using the maximum entropy model (MaxEnt) and the connectivity of landscapes among different PPR-contaminated regions using the LCP model, thereby revealing the potential transboundary communication of PPR. # Materials and methods ## Research area Our research area is defined as the THR, which mainly included the Himalayan mountains, a part of the Tibet plateau, the Ganges plains, a part of the Indus plains, a part of the Indian Peninsula, and the Arakan Mountains. Altogether, seven countries are in this region, including China, India, Nepal, Pakistan, Bhutan, Bangladesh, and Burma, covering approximately 6.89×10⁶ km². The northern, central, and southern parts of the THR differ in natural geography, ecology, and climatology. The central part, i.e., the Himalayan mountains, has the highest elevation of approximately 4000–8800 m. It is a long and narrow mountain range with a length of approximately 3000 km and a maximum width of only 400 km, covering an area of more than 1×10⁶ km². In the south-facing slope of the mountain, lower elevation regions were covered by the evergreen broad-leaved forest, and higher elevation regions were covered by coniferous forests, shrubs, and alpine meadows. The north-facing slope displayed the alpine climate, dry and cold with little precipitation. The northern and the southern parts are divided by the Himalayan mountains. The northern part is constituted by the Tibetan Plateau, with an elevation of approximately 2500–5000 m. This area is dominated by plateaus and mountains interspersed with plains and basins. The intense radiation, low temperature, large daily temperature range, and small annual temperature range verify a typical plateau climate feature. The elevation in the southern part was below 1500 m. It mainly consists of flat, fertile plains with a tropical monsoon climate and a subtropical grassland climate. There are two different livestock systems in THR. The mixed crop-livestock farming system is represented by the Ganges plains, the Indus plain, and the Indian peninsula (low-elevation). The other one consists of the grazing system, represented by the Tibetan Plateau (high-elevation). The Tibetan Plateau is a traditional pasture (the number of sheep ≈ 10 million, goats ≈ 5 million). India (sheep ≈ 65 million, goat ≈ 135 million), Pakistan (sheep ≈ 3.7 million, goat ≈ 7.4 million), and Nepal (sheep ≈ 0.8 million, goat ≈ 10.9 million) are also the core areas of animal husbandry in the world. With respect to free susceptible hosts, bharal is distributed in the Tibetan frontier, Nepal, and Bhutan, and population densities in Nepal were found to be 0.9–2.7 individual/km², increasing to a maximum of 10 in the winter, as herds congregate in the valleys. The number of mature individuals is approximately 47,000–414,000. Himalayan goral occupies the south-facing slope of the Himalayan mountains, and its population density varies from 2.6–10.5 individuals/km². Blackbuck is widely distributed in the Indian subcontinent, and the number of mature individuals is around 35,000. Markhor is mainly distributed in Pakistan. It is also found in small numbers in India (Jammu Kashmir). The number of mature individuals of this species is approximately 5,754. ## Research data There were 1135 recorded PPR outbreak locations collected from the OIE reports and published studies, including 107 records collected from the latter. The host datasets were used alongside four fundamental environmental predictor categories relevant for habitat modeling of terrestrial macro-fauna, i.e., climate, terrain, vegetation, and human impact to construct the environmental model in this study. The preprocessing and calculation of all spatial data were conducted in ArcGIS 10.6 and projected in UTM-WGS-1984 with standard settings or resampling to 30 arc-seconds. ## PPR spatial distribution model The MaxEnt model is regarded as one of the best-performing specialty distribution modeling techniques for analyzing presence-only data. It creates ecological niche models by combining presence-only data with environmental variables using a machine-learning approach known as maximum entropy. The reliability of MaxEnt has been confirmed by its good capacity to predict novel presence localities for poorly known species/diseases. It has been widely used in many diseases, including PPR and African swine fever. The MaxEnt model is applied to the spatial distribution model building to explore the risk situation of PPR in the THR. In the construction of the model, the regions with significant differences in elevation are treated separately to overcome the problem that the model is not robust enough to deal with the DEM with large differences. The low-elevation model (Model 1) and a high-elevation model (Model 2) were constructed for regions below and above 1500 m, respectively, according to the elevation standard of highland climate. The spatial autocorrelation was minimized by filtering all recorded PPR locations using the SDM Toolbox v1.1c in ArcGIS 10.6. Filtering was performed by limiting the minimum distance between each pair of points. In addition, the filtering program plays the role of systematic sampling. It can delete adjacent records to reduce spatial aggregation, which is regarded as the most effective method in correcting sampling bias. Multicollinearity was reduced for both the climate and non-climate predictors. First, major predictors were selected using principal component analysis (PCA). The variables with eigenvalues larger than 1.0 and the scree plot criterion or ‘broken stick’ stopping rule for PCA in item-level factoring were adopted. Suppression of unnecessary loading and rotation of factor pattern of variables was used to retain predictors for subsequent analysis in MaxEnt. Next, variables with low contribution rates were filtered out using the MaxEnt model. Finally, variance inflation factor (VIF) analysis was conducted to evaluate the multicollinearity among predictors after the reduction. A VIF value below 10 indicates low and acceptable multicollinearity. The filtered PPR locations and predictors were then used as input data to construct the PPR model using the MaxEnt algorithm. The present models were developed using occurrence data and 10,000 random background points, representing the distribution of environmental conditions in the study region. We divided the selected presence records into 70% training and 30% testing portions to build and validate the models based on 10 bootstrap replicates. For the remaining parameters, we kept the default settings in the pilot study. Predicted PPR risk maps obtained by models 1 and 2 were overlaid using the fuzzy overlay to construct the final PPR risk map of the THR. For visualization, the Jenks natural break optimization method was used to classify the model output to identify high-risk areas. Smoothing was followed for map visualization. The key component of the model validation procedures is the criterion that evaluates the model performance. We use threshold-dependent and threshold- independent criteria. The area under the ROC curve (AUC) is a threshold- independent criterion based on plotting the true positives against the false- positive fractions for a range of thresholds in prediction probability. Currently, the AUC is considered as the best criterion for assessing model success for presence/absence data. As a threshold-dependent validation measure, we used confusion matrix-based measures, including the Kappa test and correctly classified instances (CCI). The Kappa statistic normalizes the overall accuracy by the accuracy that might have occurred by chance alone. The percentage of CCI was defined as the rate of correctly classified cells. The thresholds of these two criteria are determined using the sensitivity-specificity sum maximization approach. ## LCP model The least-cost paths (LCPs), the shortest paths between two points with maximum efficiency for a moving individual, have been advocated as an effective, operational, and flexible approach to analyzing connectivity in heterogeneous landscapes. The LCP model allows the integration of multi-dimensional information, including geographic and behavioral information, to comprehensively predict the potential transboundary (transregional) path of the animals. LCPs are employed mainly to determine sites that are potentially used as dispersal routes for terrestrial animals and have been proven to be applicable in ruminants. To predict the potential transmission paths of the PPR in the research area, we created a cost/resistance surface for the migration of small ruminants using land cover type and elevation as cost variables according to their movement capability. Two variables were reclassified using the Jenks natural break method. Cost measurement scale of 1 (lowest cost) to 9 (highest cost) is determined according to the number of the land cover type (except for water bodies, which has been assigned as “restricted’ due to its relative barriers to ruminants). The goat still maintains the same hoof structure as the wild goat, which is designed for movement and grip in rugged environments. Similar climbing skills, food and shelter requirements make domestic and wild goats have similar movement capabilities and habitat preferences. Cost values were assigned to each classification based on small ruminant habitat preferences (see and Tables for more details). Land cover and elevation were then combined using a logical overlay operation. Recorded PPR outbreak locations were then clustered by K-nearest neighbor cluster analysis, and LCPs between the clusters were analyzed using the constructed cost surface. After the calculation, transboundary paths were highlighted by removing the internal paths. A sensitivity analysis was performed to assess the robustness of the outputs when they were affected by uncertainty. The main source of uncertainty in evaluating potential paths for host movements is the cost value used to constitute the cost surface. The one-at-a-time method was employed in the sensitivity analysis. This common approach involves changing the input criteria one at a time to observe the effect it produces on the output. This process was repeated for each variable. We changed the cost of the different land cover types/elevation classifications one at a time by adding or subtracting an amount Δ = 5% or Δ = 10% from the original cost value to build iterative models. Raster datasets presenting cost surfaces were produced through every iteration. To measure the outputs, we rely on the Spearman rank correlation coefficient to compare the ranking of countries obtained with original cost values with those obtained with different land cover/elevation cost values. The closer the Spearman rank correlation coefficient is to 1, the more similar the iteration is to the original model. That is, this classification has less impact on the model. # Results ## Results of PPR spatial distribution models Model 1 (≤1500 m): Altogether, 129 recorded PPR outbreak points at a distance of at least 10 km away from each other were obtained after filtering. After PCA and MaxEnt filtering, minimum temperatures of August (Min T Aug.), minimum temperatures of November (Min T Nov.), human population, land cover, distance to the river, and slope angle were left. The VIF values among the remaining predictors were 1.014–1.742, which was in line with the low multicollinearity standard (\<10). Moreover, AUC = 0.892, SD (standard deviation) = 0.002, Kappa = 0.869, and CCI = 0.869, indicated the robustness of the model. The response curves of the different predictors are shown in, and the relative contributions of each predictor are shown in (left). Model 2 (\>1500 m): A total of 96 recorded PPR outbreak points remained after filtering for 5 km. After PCA and MaxEnt filtering, the mean temperature of April (Mean T Apr.), human population, land cover, distance to the river, and slope angle predictors were left. The robust VIF values among the remaining predictors were 1.006–1.062. For validation of the model, AUC = 0.934, SD = 0.010, Kappa = 0.880, and CCI = 0.881, indicated the robustness of the model. The response curves of the predictors are shown in, and the relative contributions of each predictor are shown in (right). PPR high-risk areas in the THR were predicted using both models and are shown in. High-risk areas were distributed along the Himalayas, covering northern India, Nepal, and central Pakistan. In addition, PPR high-risk areas were scattered throughout Bangladesh and central India. It is worth noting that in Tibet, China, high-risk areas show an obvious trend of distribution along rivers. The results show that the risk of PPR around Pakistan, India, Nepal, and China borders is extremely high. The possibility of transboundary spread cannot be ignored, especially since it may be facilitated by wildlife. ## Results of the LCP model The LCP analysis revealed eight potential transboundary paths in the research area. The eight identified livestock transboundary paths were: A. Mandi (India)-Ali region (Tibet, China); B. Almora (India)-Ali region (Tibet, China); C. Khalanga (Nepal)-Ali region (Tibet, China); D. Chukha (Bhutan)-Lhasa (Tibet, China); E. Chandpur (Bangladesh) -India-Burma-Dali (Yunnan, China); F. Mahakali (Nepal)—Uttarakhand/Uttar Pradesh (India); G. Seti (Nepal)—Uttar Pradesh (India); H. Lumbini (Nepal)—Uttar Pradesh/ Madhya Pradesh (India). The cost value of each classification is changed by 5% (incremental percent change) within the range of −10% to +10% (range percent change). Thus, the sensitivity analysis consisted of 72 model iterations. The results are visually represented by comparing the iterative models with the original cost value model through the Spearman rank correlation coefficient, as shown in. indicate that the cost value of elevation has little effect on ranking. In contrast, the sensitivity of the cost value of land cover is only slightly higher than the former (see and Tables for more details). Sensitivity analysis underlined the significant stability of the rankings with respect to the variation in the cost value for the land cover and elevation perspectives. # Discussion ## PPR risk and variable analysis Human population density is the most important predictive variable in the low- elevation model, with a contribution of 58.2%. Indeed, the low-elevation areas in this study are mainly distributed in the Indus plains, which are heavily populated. Although animal husbandry is the major source of income in this area, considering the large local population of the herds, both the variables of sheep and goat density were excluded by the MaxEnt model due to their low contribution rate (0–0.1). However, this does not indicate that host density has nothing to do with PPR risk. It may also be due to the correlation between human and host density that leads to the deletion of collinear variables in the model, which needs further investigation. In contrast, human population densities were considered as the third highest prediction variable in the high-elevation model. Nonetheless, the response curves of the human population density for both models displayed a similar trend, with a rapid increase in PPR incidence as the population density increases, followed by a plateau. Therefore, despite the variations in the contribution of human population density in both models, its close correlation with the incidence of PPR cannot be overlooked. The temperature has also been suggested to play an important role in transmitting and spreading infectious diseases. In this study, the mean temperature in April had the highest contribution rate in the high-elevation model. Small ruminants in the Tibetan plateau are mainly raised by transhumance, and summer pastures in high-elevation areas are commonly used during the warm season (April to May). Transhumance and migration of wildlife have intensified the direct and indirect contact between wildlife and livestock, which might increase the risk of PPRV transmission. In contrast, domestic small ruminants are kept in farms in low-elevation areas. The effect of seasonal temperatures was minimized accordingly. However, the response curves remind us that the appropriate temperature in summer (Min T Aug.: 15–30°C) and winter (Min T Nov.: 0–20°C) can increase the risk of PPR. These alerted us to pay additional attention to seasonality in preventing PPR, especially the risks of seasonal pasture transfer in transhumance areas. Many landform variables related to gathering contributed to the prediction of the PPR risk in our models. According to the response curves, habitats with deciduous broad-leaved forests, urban areas, or shrublands had the highest probability of PPR. Both the deciduous broad-leaved forests and shrublands could provide food and shelter for small ruminants. In addition, they are mainly distributed in the temperate zone, which is consistent with the optimum temperature shown by the climate variables. The insignificant importance of the slope angle to our models can be explained by the good climbing skill of small ruminants, i.e., the slope hardly restricted their distribution. For this reason, slope angles were not included in our LCP model. The variations in the distances to the river contributed significantly in the high elevation model (8.9%) than in the low elevation model (1.8%). From the high elevation areas response curve, it could be understood that a farther distance from the river decreases the risk. This is different from that in the low elevation areas. The high-risk areas distributed along the rivers in Tibet can be explained by the prevalence of a cluster of wildlife around the water holes, which would increase contact and spread of PPRV. The accessibility of water resources and the lush vegetation in plain areas dispense the need for rivers. Our model shows that the mixed crop-livestock farming system has a very high PPR risk in areas close to the Himalayas, and human influence (population) is the main variable in such cases. Most ruminants in mixed crop-livestock farming systems are found in rural areas and have frequent contact with farmers due to production demand. Therefore, the risk of PPR being dominated by the human impact is expected. For grazing systems, high-risk areas are only scattered around the river valley, and the natural environment (temperature) is the dominant variable. Transhumance became the link between temperature, ruminants, and PPRV. The communication of risk between the two livestock systems and two different landscapes may play a potential role in driving PPR transboundary transmission. ## The impact of wild and domestic hosts on PPR The host populations are important for PPR maintenance, bridges, and transmission. Because of the complex migration of wild susceptible hosts, obtaining high-quality data for model construction is not easy. However, population profiles (see the second paragraph of the study area section for more detail) can still help us analyze its impact on PPR. In , we observe that bharal is distributed in the Tibetan plateau, providing sufficient bridge hosts. Himalayan goral occupies the Himalayas with a high population density. In contrast, the populations of blackbuck and markhor are relatively small. Moreover, the other seven paths, except path E, are within the territory range of wild small ruminants, which might become bridge hosts for PPR transboundary transmission. ## LCP The LCP analysis returned eight transboundary paths between India, Bhutan, Bangladesh, Nepal, and China. Next, we describe the two-way communication of PPR risk from outside China to inner China. One end of Path A connects to Mandi city in northwest India, which is known to have a prosperous livestock industry with large populations of small ruminants but poor animal health and veterinary services. The predicted risk of PPR is extremely high in this area. Path A further extends southwest to the Himalayas and passes through the middle section of the Sino-Indian border into China. While the elevation along path A is generally high with a peak of 5733 m, many wild ruminants (viz. bharal and Himalayan goral) can cross such rugged terrain. At the other end of path A is the vast alpine pasture area of the Tibetan Plateau, where nomadic domestic small ruminants are widely distributed, which provides a sufficient host for PPRV. Path B is like A, from northwestern India to Tibet, but its length is shorter, and it might be the fastest path contributing to the spread of PPR across borders. Path C extends from midwest Nepal, where PPR frequently occurs, to Tibet. In this path, the porous border and unrestricted animal movement within the country during festive seasons (August to October) may also aid in spreading the disease. Path D extends from Chukha (Bhutan) along the river valley to Lhasa (Tibet), with wild small ruminants distributed along the way. At one end of path E is Chandpur (Bangladesh) that follows the Jamuna River and the Brahmaputra rivers to Parshuram Kund (India), and then crosses the Burmese section of the Arakan Mountains to Dali (Yunnan, China). This path is also mainly distributed along the river valley, and the bushes on both sides of the valley make it easier for the animals to cross. Both paths F and G start from the edge of the Himalayas (within Nepal) and reach the Ganges plain. Path H crosses the Ganges plain and extends to the Indian plain in central India, and the upper two paths are in low elevation areas, which do not offer any obstruction to the movement of small ruminants. The constructed LCP model involved the main variables that affect the movement of ruminants. The complicated secondary variables (such as hunting and natural enemies) were not included because they could not be measured. At the same time, the merits of the LCP model do exist, especially for a large geographical scale prediction. The cost value commonly depends on the literature and the researchers. It is worth noting that identifying the animal corridors is not easy. LCP is still an effective and universal quantitative method to solve this problem. Mutual verification between the model and reference confirmed that the sensitivity of the LCP model is resistant to slight changes in the values of variables. We put forward a set of methods for countries with data limitations and regions too vast and/or too difficult to access to provide a quick risk assessment. # Supporting information We appreciated the input from Dr. Hein van Gils that helped shape our early thinking about the work described here. We would like to thank Tony Wang from the Garvan Institute of Medical Research for the English editing. The final version of this manuscript was also improved by the useful comments from Dr. Simon Clegg, Prof. Richard Kock, and two anonymous reviewers. [^1]: The authors have declared that no competing interests exist.

# Introduction Macrophages are a heterogeneous and plastic cell population that respond to environmental signals in various cholestatic liver diseases. Tissue-resident macrophages of the liver, also termed Kupffer cells, are self-renewing cells that are present in the liver at birth and promote tolerance in homeostasis. In the setting of liver injury, tissue-resident macrophages can adopt a pro- inflammatory state and additional monocyte-derived macrophages may be recruited from the peripheral circulation to the liver. This leads to a heterogeneous population of macrophages that may have distinct functions in disease. Prior studies have presented conflicting evidence for a role of macrophages in obstructive cholestasis. Recruited monocytes have been shown to have a protective role against infection in the setting of murine bile duct ligation. In contrast, C-C chemokine receptor type 2 (CCR2)-mediated recruitment of monocyte-derived macrophages in a murine model of primary sclerosing cholangitis has been implicated in the mechanism of liver injury and fibrosis. Similarly, macrophages have been associated with the pathogenesis of murine parenteral nutrition-associated cholestasis via toll-like receptor 4 (TLR4)-mediated activation and production of interleukin-1 beta (IL-1β). Furthermore, reduced farnesoid x receptor (FXR) signaling is thought to induce activation of the macrophage inflammasome in cholestasis and endotoxemia, thereby promoting IL-1β release and increasing immune susceptibility in cholestasis. However, the precise subsets of macrophages responsible for cholestatic liver injury and repair have not been fully characterized. Macrophages have also been more specifically implicated in biliary atresia (BA), an obstructive cholangiopathy of infants thought to arise from an aberrant immune response to a self-antigen. While there are two major forms of BA, isolated BA (iBA) and syndromic BA (BASM) with associated malformations, evidence supports a similar antigen-driven immune response in both subtypes. Evidence supporting a role for macrophages in this mal-adaptive immune response include the observation that increased numbers of macrophages correlate with poor prognosis in human BA. Hepatic macrophages are also increased in the rotavirus-induced murine model of BA. In addition, macrophage depletion in a murine model of BA improved bile duct obstruction. These studies demonstrate a central role for macrophages in promoting liver injury in BA but fail to identify the specific pathogenic versus pro-restorative macrophage subsets. In our current work, we define human liver macrophage heterogeneity in pediatric cholestasis by analyzing single-cell RNA-sequencing (scRNA-seq) from patients with cholestasis from BA or Alagille Syndrome (ALGS, a non-immune etiology of obstructive cholestasis) and comparing these with non-cholestatic pediatric liver and previously published normal hepatic macrophages. We identify novel hepatic macrophage subsets in obstructive cholestasis that are distinct from non-diseased macrophages by leveraging the ability of scRNA-seq to define cell sub-populations. We further demonstrate reduced expression of regulatory genes across all cholestatic macrophage subsets that may contribute to loss of immune tolerance in cholestasis. Taken together, our results lay the foundation for future mechanistic studies and development of macrophage-specific immune modulatory therapies. # Experimental methods ## Human tissue samples Formalin-fixed, paraffin-embedded liver tissue sections from non-diseased donor liver (n = 5), and BA (n = 6), and ALGS (n = 6) patients at the time of liver transplantation were obtained from the pathology archives of Ann & Robert H. Lurie Children’s Hospital of Chicago. Fresh liver tissue was obtained from the explanted liver of 3 patients with cholestatic liver disease (2 with BA and 1 with ALGS) and 1 patient with a hepatic tumor at the time of liver transplantation. Laboratory data was collected retrospectively from the hospital admission for liver transplantation. Written informed consent was obtained from each patient’s legal guardians including in the study. The study protocol conforms to the ethical guidelines of the Declaration of Helsinki as reflected in a prior approval by the Institutional Review Board of Lurie Children’s Hospital of Chicago. All methods were conducted in accordance with the Institutional Review Board’s guidelines and regulations. ## Macrophage quantification by immunohistochemistry We performed immunohistochemistry to CD68 (Dako M0876), a cell surface marker on macrophages in normal, BA, and ALGS patients to determine if the hepatic macrophage population is expanded in cholestatic liver disease. De-waxing and antigen retrieval were performed on formalin-fixed, paraffin-embedded tissue sections following the Leica Bond-Max automated protocols. Image capture was performed with a 40x objective (400x) on a Nikon 80i (Nikon, Melville, NY, USA) microscope with a DS-Ri2 color camera. Images were stitched with a Prior Proscan III (Rockland, MA, USA) 8-slide stage and digital encoder, which allows capture of the entire tissue biopsy to a single image. The surface area was determined, and cell quantification was calculated with a thresholding algorithm on RBG images using NIS-elements AR (version 5.1). The same algorithm was utilized for all specimens, and the investigator who performed the quantification was blinded to disease classification. We performed a pairwise comparison between control and cholestatic groups and determined the level of significance by unpaired t-test. ## Human liver tissue digestion We obtained fresh liver samples from explanted liver tissue for iBA, BASM, ALGS, and non-cholestatic pediatric liver (NC) at the time of transplantation. To account for variable disease throughout the liver, three 1 cm³ samples per patient were obtained from different areas of the explant by the clinical pathology team. Non-cholestatic liver tissue was taken from a patient with a hepatic neoplasm distal to the site of tumor. Matched histology slides from the explant were prepared by random sampling simultaneously from the explant by the pathology technician. Samples were stored in Tissue Storage Solution (Miltenyi Biotec) for 6 hours prior to mechanical and chemical digestion. Each of the three 1 cm³ samples per patient were split into thirds and infused with RPMI. We cut each sample into small pieces in a c-tube and added 2.5 mL digestion buffer per c-tube: 2 mg of DNase I (Sigma), 585 μL of Liberase TL (Sigma), and 9.215 mL of RPMI-1640 (Sigma R8758). Liver tissue was further digested using a both the Miltenyi Biotec gentleMACS Dissociator and incubation with shaking at 37°C for 1 hour. We strained the liver homogenate through a 40 μm filter into a 50 mL conical tube with grinding and washing to optimize yield. The samples were spun at 300 rcf for 10 min (4°C), the supernatant was aspirated, and the pellet was resuspended in Pharm Lyse for 1 minute to lyse remaining red blood cells. The reaction was stopped with HBSS, the samples were spun again at 300 rcf for 7 min (4°C), and the supernatant was aspirated. We resuspended the pellet in HBSS and strained twice over a 40 μm filter into a 15 mL tube. Cell count was performed and the cell suspension was prepared for fluorescence activated cell sorting (FACS). ## Flow cytometry and scRNA-seq library construction A total of 1.9 x 10⁷ cells were obtained from digestion of ALGS liver, 2.2 x 10⁷ from BASM, 4.92 x 10⁷ from iBA, and 1.1 x 10⁸ from CL. We stained single cell suspensions from each sample with antibodies to detect cell viability and expression of the CD45 common leukocyte antigen. 90–100,000 live CD45+ cells were collected by fluorescence activated cell sorting with a viability of 94% for ALGS, 84% for BASM, 76% for iBA, and 87% for NC. scRNA-seq libraries were prepared using the Single Cell 3’ v2 Reagent Kit for BASM and ALGS and the v3 Reagent Kit for iBA and NC. Gel Beads in Emulsion containing single cells were generated by the 10x Genomics Chromium Controller in the Northwestern Next Generation Sequencing Facility. Barcoded libraries were sequenced on the Illumina HiSeq 4000 platform. Raw sequence data was processed using the 10X Genomics Cell Ranger 3.1.0 pipeline for de- multiplexing, trimming, aligning, and mapping to genes. After filtering of the scRNA-seq data 5,027 immune cells in ALGS, 2,633 immune cells in BASM, 5,927 immune cells in iBA, and 4,691 immune cells in NC were detected. ## Single-cell RNA-seq analysis To define the hepatic immune cell heterogeneity, we analyzed each single cell library using the Seurat version 3.0 R toolkit. Filtering parameters for each sample were set to include genes expressed in \> 3 cells. Cells were included with gene counts \>200 and \< 5000, and with \< 20% mitochondrial genome. We next ran the functions LogNormalize (scale factor 10,000), ScaleData, and RunPCA on each dataset. Variability in each principal component was visualized by the ElbowPlot function. Based on this analysis we clustered the cells by the FindNeighbors function (15 dimensions for ALGS, 17 for BASM, 12 for iBA, and 10 for NC) and FindClusters (resolution of 0.5 for each cholestatic sample and 0.2 for NC). Cell clusters were visualized by Uniform Manifold Approximation and Projection (UMAP) using the function RunUMAP. Using lineage-specific marker genes, we annotated each cluster as myeloid (*CD68*, *CEBPB*, *CLEC9A*), T and natural killer (NK) cells (*CD3D*, *CD8A*, *NKG7*), and B cells (*CD79A* without *MZB1*) plasma cells (*CD79A* co-expressed with *MZB1*), and dividing cells (*TOP2A*). To confirm our cell assignments we used SingleR to compare all clusters from each patient to the reference bulk transcriptome data from Immgen. We also separately compared our disease-specific myeloid clusters to the Immgen database to further refine our myeloid subset annotations prior to integrated analysis. We next performed integrated clustering on the mononuclear phagocyte cells from each cholestatic patient and ran FindIntegrationAnchors and IntegrateData on ALGS clusters 5, 8, 10, BASM clusters 0, 6, 7, and iBA 3, 6, 9, 11, and 12. We determined the conserved genes within each integrated myeloid cluster by the function FindConservedMarkers. To compare our diseased macrophages to normal macrophages, we imported previously published single-cell data on non-diseased adult human liver from 5 donors with a median age of 41.0 years (interquartile range 23.5 to 54.5 years). We used the same cell-specific annotations and assigned this normal data-set as the reference in further SingleR analysis of our cholestatic macrophages. The degree of similarity between groups was further assessed visually by UMAP and by correlation analysis of shared genes (Morpheus, <https://software.broadinstitute.org/morpheus>). To infer pseudotime values, we used Monocle 3 for trajectory analysis of non- diseased macrophages. By grouping cells into 5 clusters based on their pseudotime values, we applied the function FindGeneModules to identify 2 gene modules upregulated at the beginning (pseudotime 0–5) and end (pseudotime of the trajectory to best represent the non-inflammatory and inflammatory macrophage profiles, respectively. Finally, to ascertain if differences in the transcriptional signatures may be secondary to patient age, we compared macrophages from the NC liver sample to the adult normal macrophages by correlation and pseudotime analyses as described above. ## Immunofluorescence and quantification of macrophage subsets We next characterized protein expression for genes that differentiated the 3 cholestatic macrophage subsets by immunofluorescence using the Vectra Multispectral Imager in the Northwestern Immunotherapy Assessment Core. Baking and dewaxing was performed on formalin-fixed, paraffin-embedded tissue sections. Using the Opal 7-color automation kit (Akoya Biosciences, Marlborough, MA, USA) slides were stained for CD68 (Abcam ab955), CD69 (Abcam ab233396), C1Q (Abcam ab268120), and S100A8/9 (Abcam ab22506). Whole slide fluorescent imaging was performed followed by multispectral imaging of three 2.01 mm x 1.5 mm areas per slide (Phenochart and Vectra software). We next used inForm software to phenotype the cells and analyzed the cell data with phenoptrReports 0.2.9 package in R. Based on gene expression data we defined LAM on histology as CD68⁺C1Q⁺S100A8/9^-CD69^-, MLM as CD68⁺C1Q^-S100A8/9⁺CD69^-, and AM as CD68⁺C1Q^+/-S100A8/9^-CD69⁺. Using these definitions, we compared abundance on histology by disease group. # Results ## Increased macrophage numbers in obstructive cholestasis as compared to healthy liver controls We performed immunohistochemistry on histology samples from donor livers, and BA and ALGS patient livers at the time of liver transplantation to determine whether the hepatic macrophage population is expanded in cholestatic liver disease. No histologic abnormality was present among donors with the exception of one individual liver which exhibited hepatocyte swelling. Mean age for donor pediatric patients rounded down to the nearest month was 68 months (SD 113, n = 5). No laboratory data was available for donor controls. All BA and ALGS liver samples had prominent fibrosis or cirrhosis at the time of tissue collection. Mean age rounded down to the nearest month for BA patients was 7 months (SD 1, n = 6) and 105 months (SD 78, n = 6) or 8 years and 9 months for ALGS cases. Difference in age between the 3 groups was not statistically significant by ANOVA (p = 0.12). Mean direct bilirubin within 24 hours prior to transplant was not significantly different between disease groups at 9.6 mg/dL (SD 8.0, n = 6) for BA and 12 mg/dL (SD 7.7, n = 6) for ALGS (p = 0.65 by paired t-test). We found increased number of CD68⁺ macrophages in BA liver as compared to control with a mean of 1332 cells/mm² in BA versus 601 cells/mm² in non-diseased pediatric liver tissue (p = 0.04). While ALGS samples also exhibited greater numbers (1040 cells/mm²) of CD68⁺ macrophages, it did not reach significance compared to control. The pronounced influx of macrophages in cholestatic liver disease suggests they may play a pathogenic role in cholestatic-induced liver injury. ## Clinical information We obtained liver tissue at the time of transplantation from three pediatric patients with cholestatic liver disease. Cholestasis with variable elevation of liver enzymes was present at the time of tissue collection for the iBA, BASM, and ALGS patient samples based on laboratory data close to the time of sample collection. The iBA case was female and presented at nearly 6 months of age with a diagnosis of BA and did not receive a Kasai Portoenterostomy. She developed cirrhosis and portal hypertension and was thereby evaluated for primary liver transplantation. The patient with BASM was also female and received a late diagnosis of BA without Kasai Portoenterostomy. She developed cirrhosis and had a pre-transplant course complicated by portal hypertension, liver synthetic dysfunction, and infection. She received a transplant at 6 months of age. The patient with ALGS was male and met criteria for liver transplantation due to refractory fat-soluble vitamin deficiency leading to severe hepatic osteodystrophy and malnutrition. He had preserved liver synthetic function and while he did not have clinically evident portal hypertension at the time of transplant at 22 months of age he had findings of mild to moderate portal fibrosis with numerous bridges on histology. ## Variable immune cell composition between BASM and ALGS We next performed scRNA-seq on CD45⁺ live cells isolated from each liver sample to better evaluate immune cell infiltration in obstructive cholestasis. We classified single-cell clusters into 5 immune cell types and a population of dividing cells in the cholestatic liver samples using lineage- specific marker genes. Different clusters of the same cell type were highly correlated within each sample and between the samples thereby supporting the lineage annotation. Further, the lineage annotations were confirmed by Single-R, which compares each cell against a reference dataset of population-level transcriptional profiles (in this case, the Immgen database). Lastly, one cluster in each patient expressed high levels of cell cycle genes, which would indicate dividing cells. T and NK cells were the most abundant immune cell population in all samples, comprising 73%, 48%, and 54% of total immune cells in ALGS, BASM, and iBA respectively. Mononuclear phagocytes were the next largest population in BASM and iBA, but not in ALGS. This discrepancy may reflect the difference in disease etiology. ## Three distinct macrophage populations in obstructive cholestasis To better understand macrophage heterogeneity in obstructive cholestasis, we focused our analysis on the clusters annotated as MNP and other myeloid cells. Our Single-R results suggested these clusters contained a mixture of macrophages, dendritic cells (DCs), and neutrophils. For further analysis, we excluded neutrophils, which were found in BASM (cluster 8) and iBA (cluster 7) and defined by distinct expression of neutrophil genes, such as *FCGR3B* and *S100P*, and lack expression of macrophage genes, such as *CD68* and *CTSB*. We then performed integrated clustering on the remaining cells from all patients to define 3 macrophage subsets and 3 dendritic cell subsets. Three macrophage clusters were identified by the lineage-specific markers *CD68*, *CEBPB*, *CD14*, and *CD69*. The dendritic cells were annotated using markers described previously to identify a *CD1C* positive subset, *CLEC9A* positive subset, and plasmacytoid DC (pDC) subset. All macrophage populations were represented in each patient. Together, these findings suggest common macrophage subsets may arise from environmental cues in the setting of cholestasis. We next sought to characterize the cross-disease transcriptional signature of each inflammatory macrophage subset and defined MΦ1 as lipid-associated macrophages (LAM), MΦ2 as monocyte-like macrophages (MLM), and MΦ3 as adaptive macrophages (AM). LAM demonstrated the highest expression of genes associated with lipid metabolism including *APOC1*, *APOE*, *LGMN*, *FABP5*. There was also high overlap with genes previously reported in LAM from human adipose tissue including *TREM2*. MLM were defined by genes previously identified in monocytes, including *S100A8*, *S100A9*, *VCAN*. Finally, AM were enriched for genes associated with lymphocyte activation including *CD2*, *CD7*, *CCL5*, *CCL4*, *CD3D*, *IL7R*. As we have previously defined these immune cells as macrophages, the increased expression of genes involved in adaptive immunity suggest these cells may have engulfed lymphocytes or play a role in regulation of lymphocyte response. To validate these three populations across cholestatic liver disease, we performed immunofluorescence on a large cohort of patients. We chose markers for each population based on their differential gene expression by scRNA-seq. Using these markers, we demonstrated the presence of all subsets across the fixed BA and ALGS samples from through overlap with CD68 expression. Since not all individual cells in a population expressed the relevant marker, we expect this approach to have lower sensitivity than specificity as supported by differences between histology and gene expression analyses for the BASM, iBA, and ALGS samples. Thus, the percent of each population is likely to be an underestimate and may explain the proportion of CD68+ cells not assigned to any population. Despite these differences, comparing the number of cells in each population between 6 BA and 6 ALGS patients shows that LAM tends to account for a greater proportion of macrophages in BA. In contrast, the AM population is larger on average in ALGS patients. Further study is required to determine whether this difference reflects disease pathogenesis. ## Reduced expression of immune-regulatory genes in obstructive cholestasis as compared to non-diseased human liver We took advantage of single-cell data that was previously published using non- diseased adult livers to determine how macrophages from cholestatic livers compared to those from healthy livers. We reproduced the 20 clusters from the original study of which 2 were labelled as “inflammatory” (IM) and “non- inflammatory” (NM) macrophages. Although this data included all cell types, not just CD45+ cells, annotation of immune cell types using lineage-specific markers led to analogous results. To overcome technical variability between data-sets limiting the utility of co-clustering, we used Single-R, Correlation analysis, and single gene and gene set comparisons to evaluate similarities and differences between macrophage subsets. All 3 populations of cholestatic macrophages were more similar to the IM than NM; AM was the least correlated overall (0.84) compared to LAM (0.89) and MLM (0.89). To determine whether the differences between datasets was due to older age of the controls, we performed scRNA-seq on a pediatric non-cholestatic (NC) pediatric liver sample. The NC case was an 11-year-old female whose explanted liver demonstrated some areas of necrosis consistent with changes after chemotherapy and chronic inflammation with margins negative for tumor. Through a comparable scRNA-seq analysis work-flow, we identified two populations of macrophages, which we label Ped1 and Ped2. Unlike the cholestatic macrophages, these populations clearly recapitulate the dichotomy of adult NM and IM. Moreover, while all cholestatic macrophages demonstrated decreased expression of immunoregulatory genes (*MARCO*, *HMOX1*, and *CD5L*), Ped2 expressed these genes at comparable levels to NM. The cholestatic populations, LAM and AM, exhibited distinct transcriptional signatures from both adult and pediatric macrophages subsets. In contrast, the genes that defined MLM were also increased in adult IM and Ped2. Interestingly, expression of *NR1H4*, which encodes FXR and is thought to play a role in macrophage inflammasome activation in cholestasis, is negligible across all macrophages. Taken together, our findings support the emergence of disease-specific macrophages in cholestasis that may mediate inflammation via different pathways than FXR signaling. Although transcriptionally distinct, macrophages in the diseased liver may be derived from their healthy counterparts. Using Monocle, we defined a pseudotime trajectory beginning in NM (pseudotime 0) and ending in IM (pseudotime 25). We then identified 2 modules associated with high expression at these endpoints: the non-inflammatory module included genes such as *CD5L*, *MARCO*, and *VCAM1* whereas the inflammatory module included *LYZ*, *S100A8*, and *VCAN*. In support of the limited effect of age on healthy macrophage heterogeneity, the former modules were highest in Ped1, while the latter was highest in Ped2. In contrast, we found that no cholestatic macrophage subset expressed high levels of the non- inflammatory module. However, MLM exhibited high expression of the inflammatory module, possibly indicating a common origin with IM. This analysis demonstrates the transcriptional variability across cholestatic macrophages beyond the dichotomy of healthy liver macrophages. # Discussion We are the first to perform scRNA-seq on pediatric cholestatic liver to define the macrophage transcriptional profile in obstructive cholestasis. Hepatic macrophages play a critical role in maintaining immune tolerance in the setting of persistent exposure to bacterial antigens from the intestine. Loss of this tolerogenic phenotype in the setting of inflammation may be of particular importance in ongoing hepatic injury in obstructive cholestasis. Here, we identify three populations of pathogenic macrophages independent of underlying etiology that may contribute to liver injury in obstructive cholestasis. No cholestatic macrophage subset was characterized by expression of immune regulatory genes as seen in normal adult NM and a subset of macrophages in non- cholestatic pediatric liver (Ped1). Our data suggest that tissue-resident macrophages, such as NM previously reported, may be absent or transcriptionally altered by the tissue microenvironment in cholestatic liver injury. Instead, all 3 cholestatic macrophage subsets were most similar to IM, which are likely to be monocyte-derived macrophages. In particular, the population of monocyte-like macrophages had the greatest upregulation of genes encoding the S100 proteins in addition to *TREM1*, known to amplify the innate immune response, suggesting this population may have recently infiltrated. We also defined a subset of lipid-associated macrophages that had the highest expression of genes involved in TLR signaling (e.g. *GPNMB*, *MT1G* and *MT1X*). Lastly, we demonstrate the presence of a novel adaptive macrophage subset with increased *RORA* gene expression, which has been shown to promote anti-inflammatory polarization of hepatic macrophages in a murine model of nonalcoholic steatohepatitis and a human monocyte cell line. The transcriptional profiling of these distinct subsets may identify macrophage-specific targets to ultimately inhibit monocyte recruitment, block TLR-mediated macrophage activation, or re-program macrophages to an anti-inflammatory phenotype. While macrophages have been implicated in immune-mediated hepatic injury from cholestasis, the exact mechanism is not well known. Current medical therapies for cholestatic liver disease include FXR agonists, which in addition to regulating the bile acid pool may also inhibit macrophage inflammasome activation based on prior studies. However, we demonstrate an absence of *NR1H4* encoding FXR in cholestatic macrophages despite evidence of TLR signaling and inflammasome activation. As macrophages play a role in cholestatic liver injury, this finding highlights the lack of current cell-specific immune-modulatory strategies and the need for a deeper understanding of the immune response to cholestasis. LAM in our samples had a gene signature that was similar to recently published data on *TREM2+* lipid-associated macrophages in murine adipose tissue. This gene signature was also similar to human hepatic macrophages during obesity and may represent a conserved response to loss of metabolic homeostasis. As hypercholesterolemia is a common sequela of cholestasis, the *TREM2*+ LAM in our samples may arise in response to similar metabolic derangements. However, LAM in our study differed in that they had an overall inflammatory gene signature despite expression of *TREM2* previously shown to promote anti-inflammatory macrophage polarization. They also were identifiable by C1Q expression, which was similarly expressed by NM from the healthy adult dataset. It is possible that LAM arise from inflammatory activation of healthy tissue-resident macrophages. Targeting the *TREM2* molecular pathway may be an important therapeutic target to re-program hepatic macrophages to an immune regulatory phenotype and reduce the consequences of hypercholesterolemia in cholestasis. The mechanism of disease pathogenesis in BA is hypothesized to be multifactorial, including an aberrant immune response to a cognate antigen whereas ALGS is a genetic disease resulting in bile duct paucity. Thus, while the aim of the current study was to identify a common cholestatic macrophage phenotype, there are likely etiology-specific differences in the immune pathways for macrophage activation that require further investigation in larger studies. However, despite this limitation, we provide important insight into hepatic macrophage heterogeneity in cholestatic liver disease compared to healthy livers. Despite age differences, it is worth noting that macrophages from a non- cholestatic pediatric patient demonstrated a similar dichotomy as adult hepatic macrophages. This finding suggests that the distinct transcriptional signature of cholestatic macrophages is not a result of age-specific differences or technical differences between datasets such as variation in sample isolation, processing and digestion protocols, or experimental design. Lastly, we acknowledge that our findings may not be limited to obstructive cholestasis and may overlap with other causes of end-stage liver disease characterized by cirrhosis and portal hypertension. A recent study on adult cirrhotic livers described a population of scar-associated macrophages in cirrhosis that appear most similar to cholestatic LAM and express higher levels of *TREM2*, *CD9*, *LGALS3*, and *SPP1*. Future studies will more clearly define the similarities and differences in subset-specific macrophage function by patient age, stage of cholestatic liver disease, and etiology of cirrhosis. In this study, we have used single-cell transcriptional analysis of pediatric cholestatic liver samples to identify macrophage subsets at greater resolution than previously described. With ongoing work, we will strengthen conclusions on the hepatic macrophage transcriptional signature in different cholestatic liver diseases and identify common therapeutic targets to reprogram macrophages and slow disease progression. More specifically, we highlight expression of the immune regulatory genes *RORA* and *TREM2* within these inflammatory subsets that may be potential therapeutic targets to ameliorate inflammatory injury in obstructive cholestasis. Future work to correlate our findings to the immune cell subsets present earlier in disease will provide important insight into cell-specific therapeutic strategies to improve prognosis shortly after disease onset. Identifying molecular targets to reprogram hepatic macrophages in cholestasis may also have therapeutic implications for other etiologies of liver diseases and reduce the medical burden of end-stage liver disease. # Supporting information The authors acknowledge the support provided by the Northwestern University Flow Cytometry, Pathology, Immunotherapy Assessment Core, and Sequencing Core Facilities. [^1]: The authors have declared that no competing interests exist.

# Introduction Achilles tendon (AT) is the largest and strongest tendon in humans. Its major function is to transmit the force produced by calf muscles to heel bone. Although AT is primarily made up of the type-1 collagen fiber which has high strength and flexibility, exposure to excessive mechanical loadings during vigorous exercises including running and jumping may easily result in tendon injuries such as rupture and tendinopathy. Ultrasound is a useful imaging tool for the assessment of musculoskeletal structures because of its high image resolution, non-invasiveness and real-time capability. Shear-wave elastography (SWE) is a new ultrasound technology which allows quantitative evaluation of soft tissue stiffness (Young’s Modulus). Its principle is to send acoustic radiation force impulses through soft tissues with a particular density (ρ) and to compute the tissue shear elasticity (μ) in kilopascal from the velocity of shear wave (cT) travelling within the tissue based on the equation: μ = ρcT². SWE has been used for the assessment of breast and liver tissues. However, there are only a few studies reported SWE assessment of AT, and they studied the characterization of normal and abnormal AT, repeatability of stiffness measurement, AT stiffness augmentation and stretched AT elastic anisotropy. The feasibility of using SWE in the assessment of AT properties was supported by these studies due to good agreement in reported results and fair repeatability in measurements as well as clear morphological delineation. Static stretching is commonly performed prior to athletic activities to avoid musculoskeletal injuries by reducing tendon stiffness and enhancing functional range of movement. It has been reported that static stretching has no significant immediate effect on cross-sectional area (CSA) and slack length of AT. However, the immediate effect of static stretching on AT stiffness is controversial in previous studies. Some studies reported a significant decrease in AT stiffness after static stretching, whilst Nakamura found a significant increase in AT stiffness, and other studies showed no alteration in AT stiffness after static stretching. The inconsistent results of previous studies may be attributed to the indirect and complex conventional methods of using grey-scale ultrasound and dynamometer for the measurement of AT stiffness in these studies. With the advantages of real-time capability and quantitative measurement of soft tissue stiffness, SWE is an ideal imaging tool for accurate assessment of AT. However, the value of SWE in assessing the immediate effect of static stretching on AT has not been reported. Leg dominance arises from predominant loading of unilateral leg in specific activities. Previous study has suggested the definition of leg dominance depending on the nature of activities–manipulative (kicking a soccer ball) or stabilization (standing on one leg). Tendons alter its structural and biochemical properties to adapt the magnitude and habit of mechanical loading during activities. The asymmetric loading profiles between two legs may cause difference in tendon properties, such as higher AT stiffness in dominant leg. Higher incidence of tendon rupture in left Achilles tendon was noted in which the left leg was considered as the dominant leg which supported body stability. The mechanical properties of AT are different in dominant and non-dominant legs. Static stretching is important for preventing and treating injuries, and is commonly used as a therapeutic tool in physical rehabilitation and sports. Understanding the variation of the mechanical properties of AT between dominant and non-dominant legs in response to static stretching can help to devise appropriate treatment protocol for dominant and non-dominant legs. However, the effect of leg dominance on immediate effect of static stretching on AT properties has not been evaluated in previous studies. With the use of SWE, the present prospective study aimed to investigate the immediate effect of static stretching on normal Achilles tendon (AT) properties; to compare the potential difference of the effect of static stretching on stiffness of AT between dominant and non-dominant legs, and to examine the inter- and intra-operator reliability of SWE measurements of AT stiffness. The primary aims of the study are to provide a new imaging perspective in determining immediate effect of static stretching on AT properties with the use of SWE, and support further research on the effectiveness of pre-exercise static stretching and designation of side-specific stretching programme to optimize athletes’ performance and prevent AT injuries. # Methodology ## Subject Recruitment and Experimental Design A total of 21 Chinese young healthy subjects were recruited on a voluntary basis via acquaintance. One subject had a previous history of lower leg injury and Achilles tendon rupture and thus the subject was excluded from the study. Finally, the remaining 20 subjects were included in the study (13 male and 7 female, mean age = 21.3 ± 1.4 years, mean height = 168.2 ± 8.0 cm, mean weight = 60.8 ± 9.6 kg, body mass index = 21.4 ± 2.5 kg/m²). All recruited subjects were informed on the study aims, examination procedures, rights of volunteers, and safety of ultrasound by an information sheet, and signed a written consent before the commencement of the study. The study was approved by the Human Subject Ethics Subcommittee of the Hong Kong Polytechnic University. Achilles tendinopathy would affect the morphological and mechanical properties of AT, and may affect the effect of static stretching on AT stiffness. Therefore, subjects who were symptomatic of Achilles tendinopathy and have dysfunction of Achilles tendon were identified. All subjects completed the Victorian Institute of Sports Assessment–Achilles questionnaire (VISA-A). The VISA-A evaluates the effect of tendinopathy on function and quantifies the symptoms and dysfunction of Achilles tendon. According Robinson et al., a recreational person with Achilles tendinopathy will have a VISA-A score of 70 or less. In addition, a patient with Achilles tendinopathy who has VISA-A score reached 70 indicated that he/she was cured. Thus, subjects with VISA-A score lower than 70 were excluded from the study. Other exclusion criteria were history of tendon rupture, trauma and surgery in lower leg, experiences of strength training or flexibility training, history of systemic, metabolic, endocrine and inflammatory diseases and on hormonal treatment, corticosteroid drugs and contraceptive pills. The leg dominance of subjects was determined by asking the subject the preferred leg to kick a ball. It was suggested that the non-dominant leg refers to the preferred leg to kick a ball while the dominant leg is defined as the supporting leg for stability. ## Reliability Tests To evaluate the inter-operator reliability of SWE measurement of AT stiffness, another 6 subjects were recruited and scanned by 6 operators. To evaluate the intra-operator reliability, each operator scanned the subjects three times in the same scanning session. The operator with the highest intra-rater reliability was selected to perform the ultrasound scanning in the main study, and performed the intra-operator reliability test of AT thickness and CSA measurements. ## Ultrasound examination of AT In the main study, the 21 subjects were reminded to avoid vigorous exercises in lower limbs for at least 2 hours prior to the examination and were asked to rest for 30 minutes before pre-stretching ultrasound examination. All ultrasound examinations were performed by the supersonic shear wave elastography ultrasound unit (Aixplorer; Supersonic Imagine, Aix-en-Provence, France) with a SuperSonic SuperLinear^™ 4–15 MHz linear-array transducer. In the ultrasound examination of AT, the subject laid in a prone position on the examination couch with the feet hanging freely over the edge of the couch to avoid tendon stress. A customized ankle fixer was used to standardize the position of the ankle for ultrasound scanning. A generous amount of ultrasound gel was applied over the AT to form a gel gap which can ensure good probe-tissue coupling and prevent tissue distortion due to transducer compression. B-mode ultrasound and SWE were performed to assess the morphological and mechanical properties of AT respectively. In the B-mode ultrasound examination, the AT was scanned transversely until transverse scans at the level of medial malleolus were obtained. To ensure the ultrasound beam is perpendicular to the tendon with minimum anisotropy, the transducer was angled cranially and caudally until a scan plane which showed the maximum echogenicity of the tendon was obtained, and the thickness and cross- sectional area of the tendon were measured with the electronic calibers. The tendon thickness was defined as the maximum anteroposterior diameter of the tendon, and the tendon CSA was measured by manual outline of the tendon boundaries. With the SWE function of the ultrasound unit activated, the stiffness of AT of the subject was evaluated and the AT was examined with longitudinal scans. To standardize the region of AT to be assessed, the inferior border of the 1.56 cm x 1.56 cm acquisition box was placed 2 cm proximal to calcaneal insertion. SWE scanning was performed until the color display in the acquisition box became steady and homogenous, and the elastogram was obtained. For each AT, two shear- wave elastograms were obtained at different longitudinal planes of the tendon. On each elastogram, the measurement of tendon stiffness was performed using Q-box measurement tool which was an in-built program for automatic calculation of mean, maximum, minimum and standard deviation (SD) of stiffness values. A total of 4 Q-boxes with a diameter of 3 mm were placed along the long axis of tendon in the image for the measurement (Figs). ## Stretching protocol Each subject had a pre-stretching ultrasound examination on both legs. The subject then performed a 5-minute static stretching on the left leg. In the 5-minute static stretching, the subject stepped on a 30° inclined platform with the body trunk leaned forward until the lower leg was aligned vertically, and the subject maintained in this posture for 5 minutes. Immediately after the stretching, a post-stretching ultrasound examination was performed on the left leg using the same scanning protocol. After the examination of the left leg, the same stretching and ultrasound protocols were performed on the right leg. ## Data Analysis All statistical data were expressed as mean ± SD and the effect of static stretching was defined as the difference in tendon properties after stretching. For each AT, the thickness, CSA and mean stiffness were obtained and analyzed. The overall mean stiffness value of each AT was the average of the mean stiffness measured by the 8 Q-boxes in the two shear-wave elastograms. Statistical Package of Social Sciences software version 20.0 (SPSS, version 20 for Windows, Chicago, IL) was used to calculate the results of all statistical tests with a 95% confidence interval. Intra-class correlation (ICC) was used to evaluate the reliability of AT properties measurements among and within the 6 operators. The normality of the data was analyzed by Shapiro-Wilk test. Data following or violating normal distribution are analyzed by paired t-test or Wilcoxon signed rank tests respectively. Paired t-test or Wilcoxon signed rank test was used to examine change in tendon properties after static stretching and to compare the difference in tendon properties between dominant and non-dominant legs. # Results In the 20 subjects included in the study, The VISA-A scores were above 70 (Range: 74–98). Seventeen subjects had the dominant leg on the left side whereas the other 3 subjects had a right dominant leg. The calculated statistical power for ultrasound stiffness measurement was 0.796 with the 20 subjects (G\*power, version 3.1.5, Franz Faul, Uni Kiel, Germany). The mean thickness and CSA of bilateral AT before and after stretching are summarized in ( and Files). There was no significant difference in the AT thickness and CSA before and after static stretching (p \> 0.05). The mean stiffness values of AT before and after stretching exercise are summarized in. The pre-stretching stiffness of dominant AT (491.3 ± 86.9 kPa) was significantly higher than that of non-dominant AT (392.6 ± 83.1 kPa) (p \< 0.05) whilst such difference became insignificant after static stretching (p = 0.830). There was a significant increase in the stiffness value in the non- dominant legs (from 392.6 ± 83.1 kPa to 495.0 ± 96.1 kPa) after performing stretching exercise (p \< 0.05). Although there was also increase in the stiffness of AT in the dominant legs but the difference was not significant (p \> 0.05). Results showed that the intra-operator reliability \[ICC (3, k) score\] of AT stiffness measurement ranged from 0.751 to 0.941, whereas the inter-operator reliability \[ICC (2, k) score\] among the 6 operators was 0.749. The intra- operator reliability of AT thickness and CSA measurements was 0.991 and 0.962 respectively. # Discussion In the present study, the CSA and thickness of AT did not show significant change immediately after static stretching. These findings were consistent with previous studies adopting the same stretching protocol with the present study. Results indicated that static stretching does not have immediate effect on the morphology of AT.A significant difference in pre-stretched AT stiffness between dominant leg (i.e. 491.3±86.9 kPa) and non-dominant leg (i.e. 392.6±83.1 kPa) was found in the present study which is consistent with Bohm’s study, and their result showed significant higher resting stiffness in dominant leg than non- dominant leg (p\<0.05). The inherent higher stiffness of dominant AT can be explained by unequal mechanical loading leading to higher shear stress caused by micro-tearing and promotion of gene expression for type-I collagen production. Different mechanical loading of dominant and non-dominant legs during daily activities may account for the different mechanical properties of AT between both legs. Therefore, there was difference in AT stiffness before stretching between dominant and non-dominant legs.Achilles tendon consists of 30% collagen, 2% elastin, and 68% extracellular matrix. Performing the function of stability and supporting, higher frequency of repeated mechanical loading on dominant AT had been suggested to contribute to more frequent intra-tendon micro-tearing, leading to subsequent increased blood supply and extracellular matrix content to the tendon for structural reconstruction, accounting for a higher hydrostatic pressure in the tendon region. Stress is defined as function of force per unit area. Since no significant difference had been found for CSA of bilateral AT, increased hydrostatic force hence contributes to a higher stiffness of dominant AT. Increased loading at dominant AT had been suggested to stimulate expression of insulin-like growth factor-I production (IGF-I) which stimulates synthesis of type I collagen and cell proliferation, resulting in an increase in cross- linking along the tendon and therefore increased the stiffness of AT. The AT stiffness in non-dominant legs was significantly increased after static stretching whereas there was no significant changes in AT stiffness in dominant legs after static stretching. The substantial stretching effect in non-dominant legs may be due to the significant difference in the pre-stretch stiffness of the tendon between dominant and non-dominant legs. It has been reported that strain magnitude applied to Achilles tendon must exceed a particular threshold in order to trigger adaptation effects on the mechanical properties of the tendon. The strains must be high enough to produce sufficient stimulus that beyond the strains triggered by the mechanical load applied during daily activities to trigger adaptation effects on the Achilles tendon. Since the pre- stretch AT stiffness in dominant legs was higher than that in non-dominant legs, the strains produced by the 5-minute static stretching on the AT in dominant legs might not be higher than the mechanical load applied during daily activities and were not high enough to trigger adaptation effects on the mechanical properties of the tendon. However, the pre-stretch AT stiffness in non-dominant legs is lower. Therefore, lower magnitude of strain can produce sufficient stimulus to trigger adaptation effects on the mechanical properties of the Achilles tendon in non-dominant legs. The present study showed increased AT stiffness after static stretching, and this finding is in consistent with the results of previous studies. Using a machine for material tests to perform mechanical tensile test of rat’s Achilles tendon, de Almeida et al. found that Achilles tendons with stretching were stiffer than those without stretching. Using dynamometer and grey scale ultrasound to measure the passive torque and the displacement of Achilles tendon after ankle dorsiflexion respectively, Nakamura et al. found that there was a significant increase in Achilles tendon stiffness after static stretching. Compared with the fair inter-operator reliability of SWE measurement of AT stiffness in Aubry’s study (inter-operator reliability is 0.46), the inter- operator reliability of the AT stiffness measurement using SWE found in the present study (inter-operator reliability is 0.749) was higher than the previous studies. Compared with the highest ICC values in intra-operator reliability test for SWE measurement of AT stiffness in Peltz’s study (highest intra-operator reliability is 0.42), the intra-operator reliability of the six operators in the present study ranged from 0.751 to 0.941, which is higher than Peltz et al.. The improvement in intra-operator and inter-operator reliability in SWE measurement of AT stiffness in the present study may be due to the standardization of scanning protocol and the special equipment used to standardize the ankle position. Peltz et al. suggested that the measurement repeatability is mainly influenced by the relative position of transducer and tendon. Angle between transducer and axis of tendon was suggested to be a factor influencing shear wave velocity in the tendon and leading to inaccurate calculation of Young’s modulus. Ankle fixer and standardized measurement site (2 cm proximal to the AT insertion) were used for the measurement of AT stiffness in the present study which standardized the position of feet and degree of flexion of ankle throughout the scanning process. Therefore, higher measurement reliability was resulted. The significant improvement in the SWE measurement reliability in present study suggests that standardization of scanning protocol and technique is crucial for SWE measurement of AT stiffness so that reliable measurements can be obtained. The present study demonstrated the asymmetry of stiffness change of dominant and non-dominant AT where non-dominant AT had significant increase in stiffness after static stretching. This finding is useful in the design of stretching protocol in which it should consider the difference in the AT stiffness between the dominant and non-dominant legs. There are limitations in the present study. The sample size of the study was small and only young subjects were recruited. Different static stretching parameters including loading, dose, magnitude, torque, and number of stretching cycle and duration of stretching had been reported to have influence on tendon stiffness. In the present study, only 5 minute continuous and 30 degree dorsiflexion stretching protocol was adopted. Further SWE investigations to assess the effect of the number of stretching cycles, stretching angles, magnitude and duration on AT stiffness are suggested. The present study investigated the immediate effect of static stretching but the delayed effect was not studied. Further studies to investigate the delayed effect of static stretching on AT are suggested. In conclusion, SWE is a useful and reliable imaging tool for the assessment of immediate stiffness change of AT upon static stretching. Dominant leg AT has a higher baseline stiffness than non-dominant leg AT. A 5-minute static stretching with 30 degree dorsiflexion can lead to significant increase in non-dominant AT stiffness which may help enhance performance of manipulative activities. Asymmetry of stiffness change of bilateral AT provides scientific ground for supporting static stretching as a means to even out the inherent bilateral AT stiffness difference and provides insight for future studies to manipulate various stretching parameters in order to optimize athletes’ performance. # Supporting Information [^1]: All authors have declared that no conflicts of interest exist for the data, results and conclusions described in this study. [^2]: Conceived and designed the experiments: TCRC HCN LWL KWL MHL HFY MY. Performed the experiments: TCRC HCN LWL KWL MHL HFY. Analyzed the data: TCRC HCN LWL KWL MHL HFY MY. Wrote the paper: TCRC, MY.

# Introduction CD8⁺ T cells are critical for control of viral infections and tumors and their efficient induction requires coordinated signaling through a number of pathways, including T cell receptor (TCR) ligation with peptide in the context of major histocompatibility complex class I (MHC I), costimulatory molecules and cytokines. One of the key effector functions acquired by CD8⁺ T cells upon activation is the ability to produce antiviral and pro-inflammatory cytokines, including IFNγ and TNF. Typically, cytokine production by antiviral CD8⁺ T cells occurs in an hierarchical fashion, with the majority producing IFNγ, and a subset of those producing TNF. Such ‘polyfunctionality’ within a T cell response is used to indicate an increased quality of response, and has been associated with heightened affinity of TCR-pMHCI recognition. Tumor necrosis factor (TNF) can substantially influence antiviral CD8⁺ T cell responses. TNF can be expressed as a membrane bound protein (mTNF) or cleaved and released as a soluble protein (sTNF). Following infection, TNF is expressed by a range of cells, including epithelial cells, natural killer (NK) cells, macrophages, dendritic cells (DCs), CD4⁺ and CD8⁺ T cells. TNF binds to two receptors, ubiquitously expressed TNFR1, and TNFR2, which is more restricted to haematopoetic tissues and is upregulated on activated CD8⁺ T cells. TNFR1 has a death domain to drive apoptosis and it also triggers NFκB driven inflammatory pathways. TNFR2 does not have a death domain and only weakly stimulates NFκB, but coordinated signaling of TNF through TNFR1 and TNFR2 has been shown to have cytotoxic effect on activated CD8⁺ T cells, suggesting that TNF:TNFR2 signaling plays an immunoregulatory role. It has been shown that global TNF/TNFR2 signaling inhibits the secondary CD8⁺ T cell response to influenza in the lungs. Studies investigating the role of TNF in anti-influenza immune responses, viral clearance and immunopathology have indicated that TNF is not required for viral clearance in the lungs, but is essential in controlling lung damage. Others reported that sTNF is responsible for limiting the extent of lung injury and this interaction was mediated via TNFR1. Moreover, the latter study demonstrated that TNF expression is required early during infection to regulate the magnitude of CD8⁺ T cell responses. However, studies with TNF knockout (*Tnf-/-*) mice are limited as this genotype causes defects in the follicular DC network as well as in B cell follicle and germinal centre formation. Consequently, *Tnf-/-* mice have a profound defect in their immune architecture and cellular composition. Therefore, studies using global *Tnf-/-* mice do not allow us to investigate the role of intrinsic TNF produced by CD8⁺ T cells and its role in the infection. Recently, Wortzman *et al*. used an adoptive transfer model of TNF- or TNFR2-deficient transgenic CD8⁺ T cells, and demonstrated that TNF produced intrinsically by CD8⁺ T cells enhanced effector functions and regulated contraction of those cells via TNFR2 signaling. In our study, we show that the rapid and robust production of TNF after *in vitro* stimulation is dependent on co-stimulation and is associated with changes in histone post-translational modification (PTM) deposition at the *Tnf* gene locus. We also demonstrate that, following intranasal infection with influenza A virus (IAV), global TNF deficiency increased the magnitude of IAV-specific CD8⁺ T cell responses, as measured in the periphery, but did not significantly affect the recruitment of IAV-specific CD8⁺ T cells to the lungs. Moreover, this TNF-mediated attenuation of the IAV-specific CD8⁺ T cell response was found to be largely dependent on extrinsic TNF production, with only a moderate contribution by intrinsic CD8⁺ T cell-derived TNF. These data clearly indicate an immunoregulatory role for TNF during IAV infection, which occurs at the global, rather than local, level and is mediated predominantly by extrinsic TNF production. # Materials and methods ## Mice Female 6–12 week old C57BL/6J (WT), ovalbumin transgenic-I (OT-1), *Tnf -/-* and *Tnf -/-/*OT-I mice were bred and housed in specific pathogen-free conditions at the Biomedical Research Facility, Department of Microbiology and Immunology, The University of Melbourne (Parkville, VIC, Aust.). *Tnf -/-* mice were obtained from the Heath Laboratory (University of Melbourne) with permission from the Centenary Institute (Sydney, Australia). Within experiments, mice were age- matched to within 1 week. Mice were killed by CO₂ asphyxiation using a slow fill rate of 20% per volume per minute. All animal experimentation was conducted following the Australian National Health and Medical Research Council Code of Practice for the Care and Use of Animals for Scientific Purposes guidelines for housing and care of laboratory animals and performed in accordance with Institutional regulations after pertinent review and approval by the University of Melbourne Animal Ethics Committee. ## In vitro stimulation and intracellular cytokine staining Sorted CD8⁺CD44^lo T cells stimulated at 2 x 10⁴ cells per well in mAb-coated 24-well plates for up to 72 hours (h) were incubated in the presence of 10 U/mL recombinant human IL-2 (Roche Diagnostics, Mannheim, Germany). The mAb were coated onto Nunc plates overnight in PBS at the following concentrations: anti-CD3ε (clone 145-2C11) at 10 μg/mL; anti-CD8α (clone 53–6.7) at 10 μg/mL; anti-CD11a (clone I21/7.7) at 5 μg/mL; anti-CD28 (clone 37.51) at 10 μg/mL. To permit intracellular cytokine staining, GolgiPlug at a 1:1000 dilution (BD Biosciences, San Diego, CA, USA) was added for the last 5 h of the incubation. After incubation, cells were surface-stained using anti- CD8α-PE (BD Pharmingen; clone 53–6.7), fixed and permeabilised using Cytofix/Cytoperm buffer and 1 × Perm/Wash buffer (BD Biosciences) according to manufacturer’s instructions and stained with anti-IFNγ-FITC (BD Pharmingen; clone XMG1.2) and anti-TNF-APC (BD Pharmingen; clone MP6- XT22). Cells were then acquired using FACSCalibur flow cytometer (BD Biosciences) and data were analyzed by using FlowJo software (Versions 9&10) (FlowJo LLC, Ashland, Oregon). ## Chromatin immunoprecipitation (ChIP) for histone modifications and RNA polymerase II Sort purified lymphocytes from LNs and spleen (\~5 x 10⁶ cells total) were fixed in 1% formaldehyde, resuspended in ChIP lysis buffer (1% v/v SDS, 10mM EDTA, 50 mM Tris-HCl) and sonicated to generate 200–1000 base pair fragments. Samples were then precleared using Protein A-agarose/salmon sperm DNA (Millipore, 16–157), split into 5 and incubated overnight with either 5 ug anti-H3K27me3, 3 ug anti-H3K4me3 or 4 ug anti-RNA polymerase II (all Invitrogen). A no antibody control and a total input positive control were also included. After washing, all samples (except the ‘total input’) were incubated with Protein A-agarose/salmon sperm DNA with rotation for 1 hour followed by a series of washes in low salt, high salt, lithium chloride, and TE buffers. DNA was eluted before crosslink reversal with 0.2M NaCl at 66°C overnight, followed by protein digestion with proteinase K (Promega). Immunoprecipitated DNA was extracted by phenol: chloroform:isoamyl (25:24:1) extraction and resuspended in HPLC water. For analysis, real-time PCR was used to measure the levels of ChIP- DNA, such that resulting cycle threshold (Ct) values were converted to copy number (#copies = 10⁵/2^Ct-17) and samples were normalised to their corresponding total inputs with background subtraction (no-antibody control). ## Influenza A virus infection and determination of viral titer Mice were anesthetized by isofluorane inhalation and infected intranasally with 1x10⁴ PFU of the HKx31 (H3N2) IAV strain in 30 μL of PBS. Weights were monitored daily and mice typically lost \<20% of original starting body weight. Viral titres were determined at the indicated timepoints using a plaque assay on monolayers of Madin-Darby Canine Kidney (MDCK) cells. Briefly, lungs were homogenised in 2 mL of incomplete RPMI, the cell suspension was centrifuged and a titration of supernatant was applied to the MDCK monolayers, overlaid, incubated for up to 3 days before plaques were counted. ## T cell adoptive transfer CD8⁺CD44^lo T cells (OT-I or *Tnf -/-/*OT-I) were sort purified and 1x10⁴ cells were transferred via intravenous injection into WT or *Tnf -/-* recipient mice. Recipient mice were infected the following day with IAV as described above and spleen and bronchoalveolar lavage (BAL) were taken for tetramer analysis at day 10 after infection. ## Tetramer staining Spleen or BAL from IAV-infected mice at day 10 were processed to single cell suspensions, red blood cells were lysed and the remaining leukocytes stained. We used PE-labeled MHCI tetramers (D^bNP₃₆₆ or K^bPB1₇₀₃) and APC-labeled MHCI tetramers (D^bPA₂₂₄ or D^bPB1-F2₆₂) (University of Melbourne Tetramer Facility), stained with Fixable Live/Dead AquaBlue viability dye (Life Technologies), blocked with anti- CD16/32 mAb (clone 2.4G2), and stained with anti-CD3ε-PerCPCy5.5 (BD Pharmingen; clone 145-2C11), anti- CD8α-PacBlue (BD Pharmingen; clone 53–6.7) and anti-CD4-AF700 (BD Pharmingen; clone RMA4-5). Cells were acquired on a FACS Canto II flow cytometer (BD Biosciences), and data were analyzed by using FlowJo software (Treestar). ## Statistical analyses All experiments contained a minimum of 4 mice (4–10) and were repeated at least twice, with similar results. Each symbol represents either data from an individual sample or the mean of multiple samples. Error bars represent standard error of the mean (SEM). Mann-Whitney test with Bonferroni correction for multiple comparisons was used to compare multiple samples and statistically significant differences between groups are indicated as follows: ns = not significant (p\>0.05), \*p ≤ 0.05, or \*\*p ≤ 0.01. # Results ## TNF expression kinetics correlate with epigenetic regulation upon TCR stimulation We first assessed the kinetics of TNF expression following TCR engagement with various levels of co-stimulation, in a population CD8⁺ T cells isolated from pooled lymph nodes of naïve WT mice. Cells were stimulated with anti-CD3 monoclonal antibody (mAb) in combination with co-stimulation from anti- CD28 mAb alone, anti-CD8/CD11a mAbs or anti-CD8/CD11a/CD28 mAbs. At 0, 5, 24, 48 and 72 h after stimulation, cells were harvested and the percentage of CD8⁺ T cells that produced TNF or IFNγ was assessed by flow cytometry. Total cell counts for a distinct but representative experiment were also determined. There was extremely rapid production of TNF by naïve CD8⁺ T cells after stimulation, with TNF detectable under all stimulation conditions within 5 h. In the presence of anti-CD8/CD11a/CD28 mAb, TNF production appeared biphasic, with an immediate peak at 5 h, followed by a drop at 24 h and then a steady sustained increase that continued out to at least 72 h. The immediate production of TNF appeared dependent on CD8/CD11a co- stimulation, since the only condition that didn’t induce robust immediate TNF production lacked anti-CD11a, while the later sustained production correlated with CD28 co-stimulation, since the condition lacking anti-CD28 costimulation showed relatively poor TNF production at the later timepoints. In contrast, IFNγ⁺ cells, although detectable at 5 h, steadily increased in proportion over the course of the assay under all conditions. The extremely rapid and robust production of TNF after CD8⁺ T cell stimulation suggested that naïve CD8⁺ T cells are poised for TNF production, as noted previously and supported by the observation that naïve CD8⁺ T cells contain substantial levels of pre-existing TNF mRNA. To determine whether differential kinetics of TNF and IFNγ expression were associated with distinct epigenetic signatures, we stimulated sort purified naïve CD8⁺ T cells as before and performed chromatin immunoprecipitation (ChIP) for histone modifications followed by quantitative PCR targeting the proximal promoter regions (PPRs) of *Ifng* and *Tnf* loci. We targeted histone 3 lysine 4 tri-methyl (H3K4me3) and H3K27me3 as these are associated with permissive or repressive chromatin structure and thereby transcriptional activity or inactivity, respectively. The patterns of histone modification largely correlated with the frequency of cytokine-producing cells, with rapid deposition of the activating H3K4me3 mark at the *Tnf* PPR at 5 h post-stimulation, corresponding to the peak frequency of TNF⁺ CD8⁺ T cells, and a subsequent removal (at 48 h) of the repressive H3K27me3 mark at this site. In contrast, the *Ifng* locus retained enrichment of the repressive H3K27me3 mark at 5 h but showed an increase in the relative enrichment of H3K4me3 at 48 h, corresponding with a substantial increase in the frequency of IFNγ⁺ cells at this time point. Binding of RNA polymerase II (pol II) is a further indication of transcriptional potential. Prior to stimulation the PPRs of *Ifng* and *Tnf* loci appeared poised for transcription, with both showing enrichment of pol II. Upon stimulation pol II binding was only retained at the *Tnf* PPR at 5 h, and was reduced but still present at 48 h. In contrast, pol II was lost from the *Ifng* PPR at 5 h, only becoming detectable again at 48 h, broadly corresponding to the relative kinetics of production of the two cytokines. Thus, these data suggest that the differential kinetics of TNF and IFNγ production by CD8⁺ T cells after stimulation is, at least in part, controlled by the accessibility of local chromatin structure determined by deposition of key histone modifications at the PPR. ## TNF attenuates virus-specific CD8⁺ T cell responses To investigate the role of global TNF on antiviral CD8⁺ T cell responses, we intranasally infected WT and TNF knockout mice (*Tnf*-/-) with the HKx31 (H3N2) strain of IAV. Global loss of TNF in influenza-infected mice led to an increased level of protein in cell-free bronchoalveolar lavage fluid (BAL) on d7 post-infection, typically used as an indication of loss of alveolar barrier integrity and accumulation of protein-rich fluid in the alveolar space but may also reflect increased levels of inflammatory mediators. Despite differences in local lung injury, there was no difference in overall disease severity, as indicated by weight loss, or in viral clearance between WT and *Tnf*-/- mice during primary influenza infection. We then analysed the endogenous IAV-specific CD8⁺ T cell response in the spleens and BAL of WT and *Tnf*-/- mice at day 10 (d 10) post-infection, which represents the peak of the acute response. To measure CD8⁺ T cell response magnitude, we used MHC class I tetramers, D^bNP₃₆₆, D^bPA₂₂₄, K^bPB1₇₀₃ and D^bPB1-F2₆₂, to identify four influenza-specific populations. This analysis revealed significantly higher proportions and a consistent trend toward increased absolute numbers of IAV- specific CD8⁺ T cells in the spleen following infection of *Tnf*-/- mice. However, we found no noticeable difference in the proportions or absolute numbers of IAV-specific CD8⁺ T cells in the BAL of infected mice at d 10. Total splenic or BAL-derived CD8⁺ T cell numbers were not different between WT and *Tnf-/-* mice. This suggests that TNF-mediated attenuation of splenic CD8⁺ T cell responses does not impact the efficiency of IAV-specific CD8⁺ T cell recruitment into the lung. ## Both intrinsic and extrinsic TNF can attenuate IAV-specific CD8⁺ T cell responses A global absence of TNF affects lymphoid secondary structure and has a profound effect on a range of cell types and their function, especially given that the predominant sources of TNF are macrophages and dendritic cells. Thus, to understand the relative impact of intrinsic versus extrinsic TNF on the antiviral CD8⁺ T cell response, we used an adoptive transfer model in which WT or *Tnf*-/- TCR transgenic CD8⁺ T cells expressing a K^b-restricted ovalbumin (OVA_257-264)-specific TCR (OT-I) were transferred into either WT or *Tnf*-/- mice, followed by IAV infection (gating strategy included as). As controls, we included intact WT and *Tnf*-/- mice that did not receive a transfer (No Tx) and we included WT OT-I→WT and *Tnf*-/- OT-I→*Tnf*-/- transfers to control for the effect of cell transfer. Both of these sets of controls were able to recapitulate our observations from, with higher proportions and numbers of OVA-specific CD8⁺ T cells in the global absence of TNF in the spleen at d10 post-infection. In WT OT-I→*Tnf*-/- transfers, only responding CD8⁺ T cells can produce TNF and the OTI CD8⁺ T cell response was significantly higher than that observed when intrinsic and extrinsic TNF were present in WT OTI→WT transfers. This suggests that CD8⁺ T cell-derived TNF plays a modest role in attenuating the IAV-specific response. In *Tnf*-/- OT-I→WT transfers, CD8⁺ T cells selectively lack the capacity to produce TNF but the CD8⁺ T cell response magnitude was comparable to that observed in the complete absence of TNF in the *Tnf*-/- OT-I→*Tnf*-/- transfers. Finally, the loss of global, extrinsic or intrinsic TNF had no impact of the proportion of CD8⁺ T cells that were OVA-specific in the BAL. Again, this finding is consistent with and further reinforces that CD8⁺ T cell- or extrinsically-derived TNF has no impact on local antigen-specific CD8⁺ T cell recruitment. Collectively, these data indicate that extrinsically-derived TNF has the dominant impact on antiviral CD8⁺ T cell responses by attenuating the magnitude in the secondary lymphoid organs, but CD8⁺ T cell-derived TNF can mediate a modest attenuation. # Discussion TNF is one of the major mediators of inflammation and its role during influenza infections has been described in several studies. While our work and previous studies in *Tnf-/-* mice demonstrate that TNF is not needed for efficient clearance of virus from the lungs as it is not required for the antiviral activity of CD8⁺ T cells, TNF is important for other aspects of antiviral immunity as it displays functional duality, being both pro- inflammatory and immunoregulatory. With regard to its proinflammatory role, TNF has been linked to increased morbidity upon influenza infection, possibly through exacerbation of lung inflammation by promoting recruitment of immune cells. The source of this TNF may be stromal cells, macrophages or dendritic cells but TNF derived from virus-specific CD8⁺ T cells has also been shown to contribute to lung damage. With regard to its immunoregulatory role, TNF may simultaneously limit lung immunopathology by attenuating CD8⁺ T cell responses. In the global absence of TNF, infected mice displayed numerically higher and prolonged IAV-specific CD8⁺ T cell responses in the lungs, and this accumulation was specifically linked to the action of sTNF. In addition, it has been shown that TNF produced by activated CD8⁺ T cells can act in an autocrine fashion via TNFR2, to regulate the peak magnitude of the CD8⁺ T cell response and promote contraction. The latter study used the adoptive transfer model of OT-I transgenic cells thus eliminating all other factors that can affect the immune response in global *Tnf -/-* mice. In this study, we showed, in accordance with previous studies, that TNF is produced by CD8⁺ T cells in two waves; namely an immediate burst of TNF that peaks at \~5 h, followed by a more rapid accumulation of TNF production that continues to increase out to 72 h. Early TNF upregulation is consistent with an observation that TNF expression is necessary during priming of T cells in order to attenuate the subsequent effector response. Moreover, early inflammatory events have been shown to control contraction of CD8⁺ T cell responses and TNF has been implicated as a critical factor, sensitizing activated T cells for subsequent apoptosis. Intriguingly, the two waves of TNF production were associated with different patterns of histone modifications, with early TNF production associated with rapid enrichment of the permissive H3K4me3 mark, and later TNF production associated with a loss of the repressive H3K27me3 mark. This is in line with a previous global analysis of epigenetic control of CD8⁺ T cell function. In that study, the gain of H3K4me3 and the loss of H3K27me3 represented two distinct epigenetic mechanisms associated with transcriptional activation, with gain of H3K4me3 correlating with more rapid gene transcription. Apart from rapid epigenetic changes, we have also previously demonstrated that naïve OTI CD8⁺ T cells contain substantial levels of TNF, but not IFNγ, mRNA transcript, which is also likely to contribute to the very early (5h) production of TNF by CD8⁺ T cells after activation. It is difficult to compare these very early epigenetic changes and cytokine production characteristics immediately after *in vitro* stimulation with the events that occur *in vivo*. This is largely because *in vivo* cell division is initiated in very rare, antigen-specific T cell population 2–3 days after infection, which means cells cannot be readily detected *in vivo* until well beyond the timepoints analysed here. However, we have previously shown that OTI cells either activated *in vivo* by infection with HKx31-OVA or *in vitro* by a 5h peptide stimulation, produce TNF protein prior to IFNγ, which supports the current findings. Characterization of H3K4me3 and H3K27me3 deposition on the *Tnf* and *Ifng* loci in naïve, effector (10 d) and memory (\> 70 d) OTI cells has also been assessed. Although these *in vivo* activated populations are not directly comparable to our analyses immediately after stimulation, naïve OTI cells exhibited elevated H3K27me3 deposition at the *Ifng* compared to *Tnf* locus, as well as detectable levels of H3K4me3 only at the *Tnf*, but not the *Ifng*, locus, consistent with our data and with rapid TNF production after stimulation. Our analysis of the impact of global loss of TNF on IAV-specific CD8⁺ T cell responses was consistent with previous studies in several respects. Firstly, we found that TNF was not required for efficient clearance of virus from the lung, but did reduce lung injury following intranasal infection. Additionally, we found that the absence of TNF resulted in an increased acute IAV-specific CD8⁺ T cell response. Collectively, these data support the idea that TNF plays an immunoregulatory role following IAV infection at a number of points: early after infection (by d 7) to mitigate lung damage, later (d10) to attenuate peak CD8⁺ T cell response magnitude, and thereafter to drive the contraction of the IAV-specific CD8⁺ T cell response. Our study also contrasted with previous studies that had demonstrated that the loss of TNF increased the severity of influenza infection, as assessed by weight loss, and increased CD8⁺ T cell response magnitude in the lung. We found that an absence of TNF had no impact on disease severity and caused an increase in the proportion and number of IAV-specific CD8⁺ T cells in the spleen, but not in the lung. This suggests that TNF modulates the overall magnitude of the CD8⁺ T cell response but does not affect the recruitment of CD8⁺ T cells to the site of infection. This second discrepancy may be explained by the fact that previous studies investigated the role of TNF in response to relatively virulent strains of IAV, associated with high levels of virus replication and pulmonary inflammation. The HKx31 strain, in comparison, is known to induce a relatively mild infection when delivered intranasally. The requirement for TNF by the CD8⁺ T cell response has previously been shown to differ depending on the inflammatory context, with T cell responses to weak tumor antigens showing greater TNF dependence than a strong anti-LCMV response. Our dissection of the relative impact of CD8⁺ T cell extrinsic versus intrinsic TNF during IAV infection demonstrates that CD8⁺ T cell- derived TNF mediates a modest attenuation of CD8⁺ T cell responses. This is supported by a previous study which concluded that intrinsic TNF potentiated the contraction of the CD8⁺ T cell response after viral clearance. Our study extends these observations by demonstrating that CD8⁺ T cell-derived TNF is not sufficient for full contraction of virus-specific CD8⁺ T cells and that extrinsically-derived TNF was largely responsible for this attenuation in the periphery. Thus, although intrinsically-derived TNF does not appear to play a critical role in this context of mild IAV infection, the fact that its role varies in a context- dependent manner means that an understanding of the kinetics and epigenetic control of TNF expression in CD8⁺ T cells is essential. In conclusion, TNF is essential for attenuating the peak CD8⁺ T cell response following acute IAV infection, with extrinsically-derived TNF having the greatest impact on response magnitude and quality. This study also found that the effects of TNF were most apparent in the global peripheral response but had little effect on recruitment of CD8⁺ T cells to the site of infection. # Supporting information This work was supported by Australian National Health and Medical Research Council (NHMRC) funding (APP1071916). and a Sylvia and Charles Viertel Senior Medical Research Fellowship awarded to N.L.L.G. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: The Walter and Eliza Hall Institute, Parkville, Victoria, Australia [^3]: Current address: Broad Institute of MIT and Harvard, Cambridge, MA, United States of America [^4]: Current address: Babraham Institute, Babraham Research Campus, Cambridge, United Kingdom

# Introduction Breast cancer is the leading cause of cancer in women with an estimated 1,383,500 new cases and 458,400 deaths worldwide. Despite improvements in treatment strategies recurrence rates are still high among breast cancer patients. This may be attributed to heterogeneous nature of breast cancers representing varied morphologic and biological features, behavior, and response to therapy. Even among breast tumors of similar histologic type and grade, prognosis varies. The clinical decisions for management of breast cancer patients rely on the availability of robust well validated clinical and pathologic prognostic factors to support treatment related decision making. Routine physical examinations along with imaging, histopathological analysis and clinical parameters (tumor size, lymph node status, stage and grade) largely impact the management of breast cancer patients. Currently, breast cancer prognosis assessment methods have limited accuracy, are expensive, and in 20–30% of cases lead to over-treatment with adverse effects. None of the currently known prognostic factors has the ability to predict accurately which breast cancer patients are at high risk of recurrence. Thus, there is an increasing need for identification and validation of prognostic markers for assessment of risk for disease recurrence in breast cancer patients. Tumors are characterized by alterations in the epithelial and stromal components, which both contribute to disease progression. Recent reports demonstrate synergy between stromal and epithelial interactions, even at the initial stages of breast carcinogenesis, appears necessary for the acquisition of malignancy and provides novel insights into where, when, and how the tumor stroma develops, allowing development of new molecular markers and therapeutic targets. It is now well recognized that stromal cells within and surrounding pathologic lesions also actively contribute to malignant phenotypes through elevated expression of cytokines and growth factors. They exert their effects through increased deposition and remodelling of the extracellular matrix (ECM). The clinical impact of changes in ECM on tumor aggressiveness and disease outcome needs in depth investigation. Transglutaminase 2 (TG2), a member of multifunctional enzyme family, modifies glutamine residues by cross-linking proteins, demonstrates protein disulphide isomerase and kinase activities, mediates transmembrane signal transduction and interacts with cell surface and extracellular matrix proteins. TG2 overexpression has been reported in cytoplasm, nucleus, membrane or ECM in tumor cells. Increased expression of cytoplasmic TG2 is associated with increased cell survival, anchorage-independent growth, loss of cell polarity, increased invasion and resistance to chemotherapy in mammary epithelial cells. TG2 promotes tumor progression by initiating a comprehensive program of de- differentiation by inducing epithelial mesenchymal transition (EMT) and cancer stem cell like phenotype,. The resulting tumors remain dependent on TG2-regulated pathways for their growth and survival. Increased TG2 induces expression of transcription repressors including Snail1, Twist, Zeb1, and Zeb2, the key regulators in development of EMT phenotype in cancers,. TG2 overexpression results in constitutive activation of NFKB, the inflammatory transcription factor known to regulate various genes involved in cancer initiation and progression. Nuclear TG2 in association with pRb, p53 and histones regulates cellular functions. Cell surface TG2 in association with β-integrins serves as a co-receptor for integrin-mediated binding to fibronectin (Fn), thereby regulating cellular adhesion, spreading, motility and survival. Extra-cellular matrix TG2 regulates cell–matrix interactions. TG2 serves as a signalling molecule transmitting signals from outside the cell through Alpha1B adrenergic receptors to a downstream cytoplasmic target, phospholipase C, through hydrolysis of GTP. These findings suggest differential localization of TG2 in cancer cells impacts tumor development, growth, survival or invasion by different cellular mechanisms. Till date, most investigations on determining clinical relevance of TG2 overexpression in epithelial malignancies including breast cancer are limited to its expression in cytoplasm of tumor cells. However, studies demonstrating an association of TG2 overexpression in ECM with disease recurrence (locoregional/metastasis) are lacking. In this study, we focussed on evaluating the prognostic significance of TG2 overexpression in ECM in breast cancer patients. Further, to evaluate the crosslinking i.e. transamidating activity of TG2 (stroma/cytoplasm), we determined the expression of N-epsilon gamma-glutamyl lysine amino residues (to detect any potential TG2-mediated protein crosslinking events) in the same cohort of the breast cancers using immunohistochemistry. In addition, we stained representative tissue sections (where TG2 is expressed in the stroma) with anti-phospho-FAK or anti-phospho-ERK antibodies to evaluate the effect of stromal TG2 on activation of integrin dependent downstream signaling in breast cancer tissues. # Materials and Methods ## Patients, Clinicopathological Data Collection and Tumor Specimens The study was approved by Mount Sinai Hospital Research Ethics Board, Toronto, Canada. Written informed consent was obtained for the acquisition and use of patient tissue samples and anonymized clinical data. The breast cancer database maintained in the Department of Pathology and Laboratory Medicine (PLM), Mount Sinai Hospital (MSH), Toronto, Canada was reviewed for the last 12 years to select breast cancer cases wherein complete clinical, pathological and follow up data were available. Tissue specimens were retrieved from the archived blocks of 253 breast cancer patients (mean age: 59 years; range: 29 to 89 years) undergoing curative cancer surgery during the period 2000–2002. Comprehensive clinicopathologic data were available in digital databases for each of these cases including demography, clinical tumor staging (American Joint Committee on Cancer staging guidelines), surgical; histological grade; recurrence including local, regional, locoregional or distant; treatment, subsequent management and disease status at last clinical review. The hematoxylin and eosin (H & E) stained slides of these cases were reviewed and tumor tissues confirmed by the pathologist (MC). These 253 breast cancer cases were classified as ductal carcinoma in situ (DCIS, n = 60), invasive ductal carcinomas (IDC, n = 168), invasive lobular carcinoma (ILC, n = 16) and invasive mucinous carcinoma (IMC, n = 9). In addition, archived blocks of normal breast tissues (n = 40) obtained from patients undergoing breast reduction surgery were retrieved from MSH tissue bank. ## Treatment and Follow-up Breast cancer patients (n = 253) were treated with a primary surgery i.e. either breast conserving surgery (BCT), or a mastectomy, as per the hospital protocol. Breast cancer patients who were ER⁺/PR⁺ were given hormonal treatment. Pre-menopausal women were given tamoxifen as their primary treatment option. Post-menopausal patients were given an option of using tamoxifen followed by aromatase inhibitors, which included anastrozole, letrozole, and exmestane. Patients were given tamoxifen for 5 years and then an aromatase inhibitor for 5 years for risk reduction. Patients who received BCT were treated with radiation therapy (RT). Radiation therapy was given from 40 Gy to 50 Gy in fractions of 1.8 to 2.0 Gy. Patients receiving adjuvant chemotherapy (CT) were defined as patients, who were ER⁻/PR⁻ with a tumor size of \<0.5 cm, patients who were node negative with a tumor size \>2 cm, and patients who had a positive nodal status. These patients were given CT regimens regardless of histology grade, and tumor size. Patients with rapidly progressive disease or visceral crisis received combination chemotherapy (CT) including AC (doxorubicin, cyclophosphamide)/CEF (cyclophosphamide, epirubicin, 5-flourouracil)/CMF (cyclophosphamide, methotrexate, 5-fluorouracil)/FAC (5-fluorouracil, doxorubicin, cyclophosphamide). Patients with metastatic disease were treated with single agents (doxorubicin, docetaxel or paclitaxel). Follow-up data were available for all 253 breast cancer patients. Survival status, loco-regional relapse or distant metastasis of the breast cancer patients was verified and updated from the records of the Tumor Registry, Mount Sinai Hospital (MSH), Toronto, Canada as of August, 2012. Breast cancer patients were monitored for a maximum period of 143 months (range: 4–143 months; mean 83.9 months and median 93 months). The patients were reassessed on a regular basis and the time to recurrence was recorded. If a patient died, the medical history, clinical examination, and radiological evaluation were used to determine whether the death had resulted from recurrent cancer (relapsing patients) or from an unrelated cause. Disease-free survivors were defined as patients free from clinical and radiological evidence of local, regional, or distant relapse at the time of the last follow-up. Disease-free survival (DFS) was evaluated in the present study for statistical analysis. Disease-free survival was expressed as the number of months from the date of surgery to loco- regional relapse or till date distant metastasis was diagnosed. ## Immunohistochemistry (IHC) Serial paraffin embedded tissue sections (4 µm thickness) were deparaffinized in xylene, hydrated through graded alcohol series, pre-treated in a microwave oven for 15 min in Tris-EDTA (0.1 M, pH = 9.0) containing Tween 20 (0.05% v/v) for antigen retrieval. Slides were washed with Tris-buffered saline (TBS, 0.1 M, pH = 7.2) containing Triton X-100 (0.1%) followed by treatment with 0.3% H₂O₂ at room temperature for 10 minutes to block the endogenous peroxidase activity. Thereafter, sections were incubated with normal horse serum (10%) prepared in 5% bovine serum albumin (BSA) to preclude any non- specific binding. The sections were incubated with either TG2 antibody (mouse mAb cat \# MS-300-PABX, 1∶4,000 dilution, Lab Vision Corporation, Fremont, CA)/N-epsilon gamma-glutamyl lysine amino residues antibody (mouse mAb cat \# ab424, Abcam, Cambridge) for 60 minutes/anti-FAK (phospho Y397) antibody (rabbit pAb cat \# ab4803, Abcam, Cambridge)/anti-ERK1+ERK2 (phospho T202+ T185+ Y187) antibody (rabbit mAb cat \# ab32538, Abcam, Cambridge). Slides were washed with Tris-buffered saline (TBS, 0.1 M, pH = 7.2) containing Triton X-100 (0.1%) followed by incubation with biotinylated secondary antibodies for 20 minutes. The sections were finally incubated with VECTASTAIN Elite ABC Reagent (Vector labs, Burlingame, CA) and diaminobenzidine was used as the chromogen. All procedures were carried out at room temperature unless otherwise specified. Slides were washed with Tris-buffered saline (TBS, 0.1 M, pH = 7.4), 3–5 times after every step. Finally, the sections were counterstained with Mayer’s hematoxylin and mounted with D.P.X mountant. In negative control tissue sections, the primary antibody was replaced by isotype-specific non-immune mouse IgG. The sections were evaluated by light microscopic examination. ## Evaluation of Immunohistochemical Staining IHC scoring was performed under supervision of the pathologist (MC). Immunopositive staining was evaluated in five pathological areas of the tissue sections as described earlier. Immunostaining for all the proteins in this study was evaluated independently in tumor cell cytoplasm, nucleus and stroma by the intensity and percentage of positive staining. Sections were scored as positive if TG2/N-epsilon gamma-glutamyl lysine amino residues/anti-FAK (phospho Y397)/anti-ERK1+ERK2 (phospho T202+ T185+ Y187) immunostaining was observed in the tumor cell cytoplasm or in the stroma when observed by two evaluators (JA & GS) who were blinded to the clinical outcome. These sections were scored as follows: 0, \<10% cells; 1, 10–30% cells; 2, 31–50% cells; 3, 51–70% cells; and 4, \>71% cells showed immunoreactivity. Sections were also scored semi- quantitatively on the basis of intensity as follows: 0, none; 1, mild; 2, moderate; and 3, intense. Finally, a total score (ranging from 0 to 7) was obtained by adding the scores of percentage positivity and intensity for each of the breast cancer tissue sections. This integrated scoring has proven to work well in our previous investigations. ## Statistical Analysis The IHC data was subjected to statistical analysis using SPSS 20.0 software (SPSS, Chicago, IL) and GraphPad Prism 5.0 software (GraphPad Software, La Jolla, CA). Scatter plots were used to determine the distribution of total score of cytoplasmic or stromal TG2 expression in all tissues examined. The p-value \<0.05 was considered significant for statistical analysis. The cut-off of IHC score ≥3.0 for cytoplasmic/stroma TG2 immunostaining was considered as overexpression for further analysis. For N-epsilon gamma-glutamyl lysine amino residues immunostaining, the cut-off of IHC score ≥2.0 for cytoplasmic/stroma was considered as overexpression for further analysis. Expression data thus generated was analyzed to determine significant correlations between TG2 overexpression, clinical parameters and prognosis of breast cancer patients. The correlation of TG2 expression with patient survival (i.e. disease free survival) was evaluated using life tables constructed from survival data with Kaplan-Meier plots as described earlier. Multivariate analysis was carried out using Cox regression models to determine the performance of TG2 overexpression as a marker in comparison to other clinical and pathological prognostic parameters including age, histological grade, tumor size, stage, grade and nodal status of breast cancer patients. # Results ## Immunohistochemical Analysis of TG2 Expression in Breast Cancer To determine the clinical significance of TG2 overexpression in cytoplasm/stroma, immunohistochemistry was performed in breast normal (n = 40) and cancer tissues (n = 253). Scatter plot analysis shown in depicts the distribution of IHC scores for TG2 immunostaining in breast normal and cancer tissues. Of the 40 breast normal tissues, 14 cases (35%) showed weak to moderate immunostaining for TG2 in cytoplasm of epithelial cells. However, no TG2 immunostaining was observed in stroma of the breast normal tissues used in this study. Immunohistochemical analysis of 253 breast cancers revealed 199 cases (78.6%) showing strong TG2 immunostaining either in cytoplasm (33.6%) or stroma (45.0%). Among DCIS, 22 of 60 (36.7%) showed cytoplasmic TG2, while majority of the cases (50/60; 83.3%) showed no detectable TG2 expression in stroma. Fifty four of 168 (32.1%) IDCs showed cytoplasmic TG2, while 97 cases (57.7%) showed TG2 expression in stroma. Of 16 invasive lobular carcinomas, 6 (37.5%) showed cytoplasmic TG2, while only 4 cases (25.0%) showed TG2 overexpression in stroma (,). Among IMCs analyzed in this study, 3 of 9 (33.3%) showed cytoplasmic TG2 while only 3 cases (33.3%) showed TG2 expression in stroma. Negative control sections, wherein primary antibody was replaced by isotype IgG, no immunostaining was observed in cytoplasm/stroma of breast cancer tissue sections (data not shown). Box plot analysis revealed significant increase in stromal TG2 with advancing stage (p = 0.020), tumor size (p\<0.001), lymph node metastasis (p\<0.001) and recurrence (loco-regional recurrence/distant metastasis) (p\<0.001) ( respectively;). ## Potential of TG2 as a Marker for Breast Cancer Recurrence and Distant Metastasis Follow up data of 253 breast cancer patients for up to 12 years was used to assess the prognostic relevance of TG2 for predicting Disease free survival (DFS) including both loco-regional recurrence and distant metastasis. Recurrence (loco-regional/distant metastasis) was observed in 57 breast cancer patients including DCIS (n = 15), IDC (n = 37) and (ILC = 5) over a time period of 4–143 months. Kaplan-Meier survival analysis showed significantly reduced DFS of breast cancer patients showing TG2 accumulation in stroma (mean DFS = 112 months, p = 0.002) in comparison with patients showing lower expression (mean DFS = 127 months). Among the clinicopathological parameters, T-stage (p\<0.001), nodal status (p = 0.001), and histology grade (p = 0.002) correlated with reduced DFS in breast cancers. Among IDC patients, increased TG2 expression in tumor stroma correlated significantly with reduced DFS (mean DFS = 110 months) in comparison with patients showing lower stromal TG2 (mean DFS = 130 months, p\<0.001). All the 5 ILCs showing reduced DFS showed increased TG2 accumulation in tumor stroma. However, Kaplan Meier analysis could not be performed for ILC due to the small sample size. No significant association of cytoplasmic TG2 overexpression with recurrence was observed in all breast cancers analyzed as well as in DCIS and IDCs. In multivariate Cox regression analysis using TG2 overexpression (cytoplasm or stroma), age, ER/PR status, tumor stage, grade and nodal status as variables in the model, TG2 accumulation in tumor stroma (p = 0.014, Hazard’s ratio, H.R. = 2.7, 95% C.I. = 1.2–5.9), T-stage (p = 0.001) and grade (p = 0.034) emerged as independent factors associated with poor prognosis of breast cancer patients. In IDC, TG2 stromal accumulation was associated with poor prognosis in a Cox multivariate analysis (p = 0.006, H.R. = 3.79, 95% C.I. = 1.4–9.8) and tumor stage (p\<0.001, HR = 5.26, 95% C.I. = 2.1–13.7). ## Evaluation of TG2 Transamidating Activity in Invasive Ductal Carcinomas (IDCs) To evaluate the transamidating activity of TG2 overexpression in stroma/cytoplasm of IDCs, we determined the expression of N-epsilon gamma- glutamyl lysine amino residues in representative tissue sections of IDCs showing either low or high scores of stromal TG2 immunostaining. Of the 40 IDCs of the breast demonstrating high immunostaining scores of stromal TG2, 34 cases (85%) showed stromal staining of N-epsilon gamma-glutamyl lysine amino residues. Interestingly, all IDCs (n = 35) showing low stromal/cytoplasmic TG2 staining also showed weak immunostaining for N-epsilon gamma-glutamyl lysine amino residues in cytoplasm and stroma of IDCs. No immunostaining was observed either in cytoplasm or stroma of breast cancer tissues used as negative controls. A correlation (R = 0.671, p = 0.016) was observed for co-localization of TG2 and N-epsilon gamma-glutamyl lysine amino residues immunostaining in stroma in tissue sections of IDCs. ## Evaluation of Phospho-FAK and Phospho-ERK1+ ERK2 in IDCs Showing Overexpression of Stromal TG2 To determine the effect of TG2 on activation of integrin dependent downstream cell signaling in IDCs showing overexpression of stromal TG2, we performed immunohistochemical analysis of phospho-FAK (Y397) and phospho-ERK1+ ERK2 (phospho T202+ T185+ Y187) in representative tissue sections. Our results of phospho-FAK (Y397) showed strong nuclear staining in all IDC sections analyzed, whereas no immunostaining in nucleus/cytoplasm was observed in IDC tissue sections used as negative controls. No significant difference in the nuclear expression for phospho-FAK (Y397) was observed among these IDC sections. Immunohistochemistry of phospho-ERK1+ERK2 (phospho T202+ T185+ Y187) showed no detectable immunostaining in IDCs showing overexpression of stromal TG2. However, thyroid cancer tissue sections used as a positive controls showed strong nuclear expression of phospho-ERK1+ERK2 (phospho T202+ T185+ Y187). Negative control tissue sections showed no immunostaining in cancer cells. # Discussion The development of metastasis poses a major clinical problem in treatment of breast cancer patients leading to poor cancer free survival. Identification of molecular markers which can predict recurrence is imperative for developing effective therapies for more effective disease management. Our study provides clinical evidence that TG2 expression is up-regulated in the primary tumor of patients likely to develop distant metastasis. The majority of IDC patients showed TG2 accumulation mainly in tumor stroma. Notably, significant association of stromal TG2 in IDC was observed with clinicopathologic features including increased tumor size, grade, stage and nodal metastasis which contribute to aggressive phenotype in these patients. Further, our results clearly demonstrated accumulation of TG2 in tumor stroma associated with loco-regional recurrence, distant metastasis, and hence decreased DFS of IDC patients. Our findings are supported by the report of Girgoriev *et al*., which showed stromal TG2 expression in 50% of the human breast carcinomas, while 15% cases stained positive for cytoplasmic TG2; however the correlation of TG2 with disease outcome was not assessed in this study. Mehta *et al*., reported TG2 expression was significantly higher in lymph nodes than in primary breast tumors, but their study was limited by a small size of 30 cases only and no follow up data were provided. Further in another independent study by same group, Mangala *et al*., compared stromal TG2 expression in 189 early stage breast cancer cases and demonstrated significant association of higher stromal TG2 expression with negative lymph nodes (p\<0.001). Although, authors reported median follow-up of 4 years for these patients with negative lymph nodes, statistical analysis including multivariate Cox regression for evaluating prognosis in breast cancer patients was not provided. Thus, our study assumes importance as the first report demonstrating association of TG2 accumulation in tumor stroma with poor disease outcome and its significant association with aggressive features such as increasing tumor size, grade, stage, lymph node and distal metastasis in a large cohort of breast cancers with special emphasis on invasive ductal carcinomas. TG2 overexpression in metastatic breast cancer promotes apoptosis-resistance phenotype, cell migration and invasion by initiating integrin-mediated cell attachment and cell survival signalling pathways,. A review of TG2 expression in other human cancers revealed overexpression of TG2 in pancreatic tumor cells associated with nodal metastasis, lymphovascular invasion and poor overall patient survival. Further, TG2 overexpression in ovarian cancer patients was associated with poor overall survival. Increased expression of TG2 in ovarian cancer cells enhanced their adhesion to fibronectin and promoted directional cell migration, whereas knockdown of TG2 showed diminished tumor dissemination on the peritoneal surface and in mesentery in an intraperitoneal ovarian xenograft mouse model. Extra-cellular TG2 along with β1 and β3 integrins serves as a co-receptor for fibronectin. Interestingly, this integrin mediated interaction of TG2 and fibronectin promotes adhesion, migration, and spreading of cells on fibronectin- coated surfaces and is independent of the TG2 enzymatic activity. TG2 in ECM associates with integrins inducing activation of anti-apoptotic protein Bcl-2, focal adhesion kinase (FAK) dependent signal transduction pathways including PI3K/Akt, and Ras/Erk, pathways which contribute to cancer aggressiveness. Moreover, TG2 overexpression in ECM leads to increased accumulation of matrix bound transforming growth factor beta 1(TGF-β1), both in vitro and in vivo. TG2 expression signals the onset of EMT in epithelial cells and contributes to their increased survival and metastatic potential. The association of stromal TG2 with lymph nodal metastasis in breast cancer patients in our study provides clinical evidence in support of its utility as a marker of metastatic potential in these patients. The mechanistic basis of aberrant stromal TG2 expression contribution to EMT and metastatic capabilities of breast cancers warrants investigation in future studies. Our results also demonstrated overexpression of N-epsilon gamma-glutamyl lysine amino residues in cytoplasm and stroma of IDCs demonstrating presence of active TG2 in these breast cancers. Notably, most of these breast cancer cases had poor prognosis indicating a plausible role of active TG2 in stroma in recurrence among breast cancer patients. However, lack of significant difference of phospho-FAK expression and absence of phospho-ERK in IDCs showing overexpression of stromal TG2 suggests that stromal TG2 may not activate integrins. Taken together our results suggest the crosslinking function of stromal TG2 might be important in IDCs of breast. # Conclusions Our study clearly demonstrates the clinical significance of stromal TG2 overexpression in breast IDCs and may serve as an independent risk factor for identifying patients with high risk of recurrence and metastasis. These patients can be followed more closely and managed appropriately by selecting other treatment modalities and thereby potentially reducing the morbidity due to recurrence. Further, it may also help avoid overtreatment of patients at low risk of disease recurrence reducing harmful side effects of therapy and reduce the economic burden on health care providers as well. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RR PGW AM. Performed the experiments: JA GS. Analyzed the data: JA GS AM. Contributed reagents/materials/analysis tools: PGW RR MC. Wrote the paper: JA AM. Histopathology reporting: MC. Chart review (clinical and followup): JA.

# 1. Introduction Canadian boreal forests represent 24% of the world’s boreal forest. In these forests, anthropogenic disturbances pose serious threats for boreal flora. This is particularly true for sensitive plant species such as bryophytes, which have been recognized as reliable indicators of environmental changes. Bryophytes are key constituents of biodiversity in Canadian boreal forests, promoting species richness and supporting important ecosystem functions. Forest management pressure is however affecting bryophyte diversity and community composition in the boreal biome, either through direct species removal or by altering habitat conditions originally suitable for bryophytes. Forestry practices are also reducing the ecological continuity of forests, jeopardizing the recolonization processes after disturbance events. Highly habitat-specific and/or dispersal-limited bryophyte species harbored by old-growth boreal forests may therefore be at risk. Despite their ecological importance and sensitivity to disturbances, bryophytes are part of the vast unseen biodiversity that is currently ignored in most conservation plans. Less known and represented in natural history collections than other groups such as birds, mammals or flowering plants, the large contribution of inconspicuous taxonomic groups to diversity is difficult to assess, and thus commonly operationalized using diversity measures of these other groups as surrogates. However, these better-known taxonomic groups are poor surrogates for highly diverse but less showy or studied taxa. Including inconspicuous species groups, such as bryophytes, representativeness in systematic conservation planning assessments would lead to more robust conservation measures. From a conservation perspective, rare species deserve priority attention as they are at a high risk of extinction. However, because of their own nature, many rare species of unseen biodiversity groups suffer from a lack of information on environmental requirements or their distribution. Species Distribution Models (SDMs), which allow to quantify the statistical relationships between species observations and environmental conditions from known locations, can provide useful tools for assessing ecological preferences of rare species or predicting their distributions. More precisely, SDM-based predictions are achieved by using the relevant environmental conditions as proxies of species occurrence. However, the ability of traditional SDMs to predict rare species has been strongly limited by the number of occurrences available, with increases in prediction accuracy with increased sample size. Furthermore, modeling species with low prevalence often results in a high predictors/occurrences ratio, which can lead to model overfitting and reduced applicability to new data. Fortunately, recent advances in modeling techniques and approaches such as Ensembles of Small Models (ESMs) have been shown to provide robust predictions for rare plants. ESMs are ensembles of bivariate models generated from all pairwise predictor combinations from a larger set of predictors. ESMs can produce more accurate predictions than traditional SDMs and reduce model overfitting for rare species. In parallel, remote sensing (RS) offers a powerful tool to derive and integrate environmental information into SDMs and generate predictions on species distribution over large areas. Although a considerable number of studies have successfully integrated RS predictors into SDMs, no study has generated ESMs using only RS predictors, nor has used this approach to generate SDMs of inconspicuous organisms such as bryophytes, much less of their rare species. In this paper we use RS-derived predictors in an ESMs framework to produce predictive models of rare bryophyte species in Eastern Canadian boreal forests. Bryophyte rare species were selected based on their prevalence in the study area (\<30 occurrences;). This rare species selection approach was chosen because of the lack of knowledge on bryophytes related to their distribution, ecological preferences and abundance in the region, which make it difficult to apply more informative approaches such as multicriteria rare species classification methods. In fact, the most complete rare bryophyte species list published to date for the region used species’ prevalence as the only criterion for rare species classification. It should be noted that rare bryophytes from were not targeted here as their low prevalence (≤5 occurrences) greatly restricts the development of SDMs. We hypothesize that high ESMs-based prediction accuracy can be achieved for rare bryophyte species despite their low number of occurrences. Our specific objectives are to assess i) if there is a relationship between the number of occurrences and the predictive performance of ESMs, ii) if the predictive performance of models varies by the modeled bryophyte guild (mosses, liverworts and sphagna), and iii) if there is a spatial relationship between the richness patterns of rare bryophyte species and overall bryophyte species both for bryophytes as a whole and at the guild level. A total of 52 rare bryophyte species were targeted in the present study, including 33 mosses, 14 liverworts and 5 sphagna. # 2. Materials and methods ## 2.1 Bryophyte field data set We used a 389-plot database of presences-only including the field data from three studies previously conducted in our study area, which integrated young, mature and old-growth forests and both recent fires and cut-blocks. The study area of 72,292 km² is located in the southwest of the Nord-du-Québec administrative region of western Quebec (48° 51’ to 50° 42’N and 74° 31’ to 79° 26’W;), within the Black spruce–feathermoss forest bioclimatic domain. Natural dynamics of these forests are primarily driven by stand-replacing fires, whose cycle has been estimated at 398 years after 1920. The region is characterized by a flat topography, dominance of poorly drained clay soils and a moderately humid and cold climate (927.8 mm annual precipitation and 1.0°C annual mean temperature). These conditions favor the accumulation of organic layer between fires, which is known as the paludification process. Bryophytes were collected following a “floristic habitat sampling” method, which consists in collecting all bryophytes found in all microhabitats within 5 x 10 m plots. Rare bryophyte species were selected based on their prevalence within the study area (\<30 occurrences). From an initial set of 214 species, 142 rare species were pre-selected, and among them, only those with a minimum of 5 occurrences were retained for modeling, since meaningful predictions can be achieved at this sample size. A total of 52 rare bryophyte species (33 mosses, 14 liverworts and 5 sphagna;) were finally selected for modeling (species occurrence coordinates are shown). ## 2.2 Remote sensing environmental predictors The selection of RS-derived predictors was carried out based on their sensitivity to environmental factors known to influence bryophyte distribution, namely topography, canopy cover and structure, and vegetation and soil moisture. Climatic variables were not included due to their coarse spatial resolution (≥ 1 km) and low spatial variability across the study area (annual mean temperature and total precipitation with an approximate variability range of 1°C and 150 mm respectively), which could lead us to overestimate the distribution of rare species. In addition, the climatic variability that could be integrated into the individual models of our rare species would be even more limited by the low number of available occurrences. It should be noted that climate variables also present lower reliability compared to RS variables at the scale of our study. This is because climatic variables are based on interpolation methods with high uncertainty, especially in northern latitudes where weather stations are scarce, while RS information is spatially continuous by nature. Therefore, we selected RS variables showing higher variability across the study area and capable of detecting changes in local conditions more closely related to bryophyte occurrence. RS-derived environmental data were acquired using Google Earth Engine (GEE). The initial set of 6 predictors included topographic position index (TPI), 2-band enhanced vegetation index (EVI2), normalized difference water index (NDWI1), vegetation continuous fields (VCF), PALSAR HV/HH polarization index (PALSAR_HVHH), and bare soil index (BSI; see for predictor descriptions). TPI was derived from the Shuttle Radar Topography Mission (SRTM) digital elevation model in ArcGIS v.10.5 using an annulus neighborhood with inner and outer radius of 15 and 20 pixels, respectively. EVI2, NDWI1, and BSI predictors were derived from Sentinel-2 spectral bands. For each band, a mosaic was built from the images available for the summer season (July 1-August 31) between 2015–2019 to ensure homogeneity in the reflectance values. Cloudy pixels were masked in all selected images using the Sentinel-2 QA60 band, which allows to identify pixels with opaque clouds and cirrus clouds. Mosaics were performed by applying the median of the overlapping pixel values. We chose EVI2 instead of EVI since EVI2 does not require the blue band, which is sensitive to the presence of residual clouds and aerosols. VCF represents percent tree cover at 30 m resolution, after rescaling the 250 m MODIS VCF Tree Cover layer using circa-2010 and 2015 Landsat images and incorporating the MODIS Cropland Layer to improve accuracy in agricultural areas (<https://catalog.data.gov/dataset/global-forest-cover- change-tree-cover-multi-year-global-30m-v003>). The VCF predictor presented pixels (0.1% of the total) with missing values in the study area. PALSAR_HVHH was calculated as the ratio of HV-polarized to HH-polarized L-bands from the Advanced Land Observing Satellite (ALOS) Phased Arrayed L-band Synthetic Aperture Radar (SAR). HV-polarized and HH-polarized L-bands were averaged from yearly mosaics between 2015 and 2017. All predictors were generated and standardized at a 30 m spatial resolution (see for original spatial resolutions). Pearson correlation coefficient was used to identify pairs of highly correlated predictors (\|r\|)\> 0.7) from a set of 10,000 random background points. Only the NDWI1-BSI predictor pair showed a high correlation (r = -0.87). We retained NDWI1 which is sensitive to vegetation and soil moisture, since bryophytes are poikilohydric organisms whose distribution is highly dependent on available moisture. This resulted in a final set of 5 uncorrelated predictors to run the models. ## 2.3 Modeling approach: Ensembles of small models ESMs based on bivariate models were developed to spatially predict 52 rare bryophyte species (5–29 occurrences) using two modeling machine-learning techniques: Maxent and Random Forest (RF). Both Maxent and RF techniques can provide robust predictions when few occurrences are available. Maxent estimates the probability distribution for a given species by finding the probability distribution of maximum entropy according to a set of constraints representing the input known locations. RF uses a bootstrap aggregation technique to provide mean predictions from a multitude of independent decision trees built from randomly selected subsamples from the training dataset. A random subset of candidate predictors is assessed to split each node of each individual tree, selecting the predictor that provides the most information in each case. ESMs were generated in R v.3.6.3 using the *biomod2* package v.3.4.6. As we used presence-only data, 10,000 background points were randomly generated within the study area and used as pseudo-absences for all species. Presences and pseudo- absences were weighted equally for training the ESMs. The pairwise combinations of our 5 final predictors resulted in 10 candidate bivariate models per modeling technique (Maxent and RF) for each species. We used default settings of the *biomod2* package for computing Maxent and RF models. Predictive performance of each bivariate model was assessed via 10-fold cross-validation procedure, using 80% of the data to train the model and 20% for its validation. While we acknowledge that validation would be optimal using an external dataset, this is hardly available when dealing with rare species. The Somers’ D metric was used to identify and select bivariate models better than random (Somers’ D score \> 0, i.e. AUC \> 0.5). Maxent-ESMs and RF-ESMs were then performed using a weighted mean of predicted probabilities from their corresponding retained bivariate models based on their Somers’ D scores. The contribution of each bivariate model was thus proportional to its predictive accuracy. The final ESMs selected for each species was generated by weighted averaging predictions from Maxent-ESMs and RF-ESMs. Predictive performance of final ESMs was evaluated using the area under the receiver operating characteristic curve (AUC), and the true skills statistic (TSS). AUC is not dependent on a threshold and ranges from 0.5 for an uninformative model to 1 for a perfect fit model, while TSS ranges from -1 to 1 and was chosen instead of kappa because it is not affected by prevalence. Since AUC and TSS values were highly correlated (Pearson r \> 0.95), the results and discussion on models’ overall predictive performance will be based on the AUC statistics, following and. The statistic sensitivity was also calculated, which allows the assessment of the proportion of actual presences correctly predicted. We computed sensitivity for those species whose final ESMs were better than random (AUC \> 0.5). Besides of the continuous models (values 0–1000), we generate binary models (presence/absence) using the maximum training sensitivity plus specificity threshold, or TSS optimum (predictive mapping of the distribution of the target species is available in). Finally, we mapped the richness patterns (species number) for total rare bryophyte species, as well as for rare species by guild, by stacking their binary predictions (presence/absence). Missing values associated with the predictions of the three species that included the VCF predictor in their final models were classified as absences before richness computation. We then compared the spatial richness patterns obtained here for rare species with those obtained recently for overall bryophyte species in a smaller region (28,436 km²) but fully included in our study area at the same spatial resolution (30 m). The comparison was performed for bryophytes as a whole (i.e. rare bryophyte richness versus overall bryophyte richness), and between homologous bryophyte guild pairs. This spatial correspondence analysis was carried out using Lee’s L statistic through the *lee* function from the *spdep* package v.1.1–5.. Lee’s L statistic, in contrast to non-spatial bivariate association measures such as Pearson’s correlation coefficient, integrates and corrects for the spatial autocorrelation of each variable when computing the pixel-to-pixel spatial correlation. Due to the high computational requirements to carry out this analysis, the 30 m pixels were previously averaged into 300 m pixels through the *aggregate* function of the *raster* package v.3.4–5. Outputs of *lee* function were centered at 0 and re- scaled to -1 and 1 to facilitate the interpretation of the results by subtracting the overall mean and dividing by the maximum value. We then calculated, for each pixel, the quantile associated with its Lee’s L value using a Monte Carlo test with 999 simulations in order to identify significant positive (quantile \>0.975) or negative (quantile \<0.025) spatial associations. ## 2.4 Species traits characterization Species traits can influence the accuracy and therefore the ability of SDMs to predict their occurrence. We evaluated the relationship between ESMs’ model performance, as measured by AUC, and rare species traits, namely substrate preference (six categories), reproduction mode (three categories), and spore size (maximum and minimum;), as well as their interactions. This assessment was performed using a multiple linear regression through the *lm* function from the *stats* package v.3.6.3. Relationships were considered significant at α = 0.05. # 3. Results ## 3.1 ESMs’ predictive performance versus number of occurrences and bryophyte guilds RS-based ESMs provided poor to excellent predictive accuracy for 38 of the 52 modeled rare species, with AUC values ranging from 0.551 to 0.979 and a mean AUC (mAUC) of 0.795 ± 0.132. Of these 38 species, 19 species were predicted with AUC values greater than 0.8, confirming our hypothesis that high ESMs-based prediction accuracy can be achieved for rare bryophyte species despite their low number of occurrences (\<30). Sensitivity for these 38 species ranged from 0.8 to 1 with an average of 0.959 ± 0.063, indicating that actual presences were usually accurately predicted. Only predictions for 14 species were not better than random (AUC ≤ 0.5). Regarding our first specific objective, a negative correlation (Pearson r = -0.34) was found between the number of occurrences of the 52 target species and the predictive accuracy as measured by AUC. This negative correlation was also observed at the guild level. To accomplish our second specific objective, we grouped the 52 modeled species by guild and found that predictive accuracy was similar for mosses (mAUC = 0. 715 ± 0.167) and liverworts (mAUC = 0.735 ± 0.185), and lower for sphagna (0.663 ± 0.208). No significant relationships were found between ESMs’ performance and rare species traits (or their interactions). ## 3.2 Richness patterns of rare bryophyte species Predictive mapping of richness patterns of total rare bryophyte species and rare species at the guild level (mosses, liverworts and sphagna) are presented in. Predicted richness values ranged from 0 to 30, 21, 9, and 3 species, respectively. The richness pattern of total rare bryophytes was largely structured by the similar richness patterns observed for rare mosses and liverworts, with high richness values mostly found towards the center and southwest of the study area. Conversely, rare sphagna species were concentrated in very specific areas mainly towards the north of the study area with two additional spots towards the southeast. Regarding our third specific objective, the Lee’s L statistic identified areas of significant positive and negative spatial association between rare and overall species richness for the four homologous bryophyte group pairs. Large areas in which the spatial association between the two types of richness was not significant were also consistently observed across pairs. # 4. Discussion Boreal regions are large areas lacking sharp environmental contrasts, as shown by the low variability of our predictors, and thus a habitat where obtaining high-performance SDMs can be challenging. Despite this, our ESMs provided reasonably accurate predictions for rare bryophytes using only 5 uncorrelated RS predictors. Specifically, RS-based ESMs provided poor to excellent predictive accuracy for 73% of the target species despite their very low number of occurrences. Indeed, 16 species with less than 10 occurrences showed an AUC \> 0.7. In addition, the computation of the metric sensitivity allowed us to independently show the ability of our ESMs to accurately predict known presences, with high values for the 38 species modeled better than random. Therefore, the combination of RS data at 30 m spatial resolution and ESMs proved to be a powerful approach to predict the distribution of rare bryophyte species in Eastern Canadian boreal forests. The negative relationship found between models’ predictive performance and the number of occurrences of all bryophytes, as well as at the guild level, illustrated the suitability of ESMs for predicting the distribution of very rare bryophyte species regardless of guild. This result agrees with those obtained in, who showed a higher predictive performance of ESMs for the rarest vascular plants. Regarding bryophyte species by guild, we consider that the lower overall predictive performance obtained for sphagna species compared to that of mosses and liverworts may be an artifact resulting from the low number of rare sphagna species modeled (n = 5). In fact, the occurrences of two of these five sphagna species were successfully predicted (AUC values of 0.76 and 0.97). However, we do not exclude the possibility that some ecologically meaningful variables that describe the habitat of these species, such as drainage class, were missing from our models. In general, our results show that the development of SDMs from RS data allows not only to make predictions of rare species distribution at spatial scales relevant to ecological planning, but also to do so at a level of detail (30 m resolution) that can not be achieved using the traditionally used climatic variables at coarse resolutions (≥ 1 km). This is particularly important for inconspicuous species such as bryophytes, which interact with their environment at more local scales and for which the use of coarse resolutions can result in a critical lose of information. Likewise, SDMs developed at coarse resolutions can overestimate species distribution and greatly limits the practical utility of derived predictions to subsequently detect species in the field. On the other hand, the wide variety of potentially relevant predictors for rare plants that can be derived from RS (related to vegetation, humidity, forest structure, topography, etc.), can allow a more realistic approach to the environment- species relationship, which can be particularly useful for species with complex ecological niches. Thus, our methodology can play an important role in filling existing knowledge gaps on bryophyte distribution ranges, as well as their ecological preferences, in largely unexplored regions such as boreal forests. The Identification of diversity hotspots has been one of the most used criteria in biodiversity conservation planning in order to locate areas of biological and ecological interest that should be prioritized by decision makers. Conservation measures targeting these areas will be more effective if multiple components of biodiversity are spatially concentrated. Specifically, both species richness and the presence of rare species have frequently been cited as the main criteria to select areas for conservation, while many rare species might not be represented in species-rich areas. Our study however revealed a spatial concordance between the richness of overall bryophyte species and that of their rare taxa in different regions of the study area. While more bryophyte biodiversity components could be subsequently evaluated, this result have important implications for Canadian conservation planning. We consider that the identification of areas harboring high level of both overall and rare bryophyte species diversity, as well as the development of informative tools that serve these purposes, is a significant and necessary step to promote the systematic integration of these species into conservation plans and programs. Likewise, conservation planning targeting bryophytes and other inconspicuous taxa could further benefit from individual SDMs-based predictions as a basis for assessing their representation in nature reserve networks, to quantify the impact of land use changes on their distribution ranges, to inform assessments of their conservation status, and to identify suitable areas for their recovery or reintroduction. # 5. Conclusions Our work demonstrates the ability for RS data to characterize the habitat of rare bryophyte species and predict their distribution patterns across the landscape. This study also reaffirms the effectiveness of ESMs in estimating rare plant distributions, and highlights, for the first time, the suitability of this modeling approach for making predictions of inconspicuous rare species. We consider that our methods and results provide an important advance in the application of techniques focused on the study of bryophytes, with potential valuable applications for their management and conservation. In fact, although our study focuses on a particular taxonomic group, the combined use of ESMs and RS would lend useful results for other overlooked inconspicuous taxa lacking information on distribution, which would facilitate their integration in systematic conservation planning. # Supporting information We thank Marion Barbé, Chafi Chaieb and Joëlle Castonguay for sharing their bryophyte field data, and Julie Arseneault for her help in identifying samples. We thank Rubén G. Mateo for his suggestions and assistance in the preliminary analyzes of the data. We also thank the anonymous reviewers for the constructive comments and suggestions provided on our manuscript. This research was funded by Environmental Damages Fund, Environment and Climate Change Canada. [^1]: The authors have declared that no competing interests exist.

# Introduction Multiple system atrophy (MSA) denotes an adult-onset neurodegenerative disorder of relentless progression and unknown aetiology that is clinically characterized by the variable combination of autonomic failure, levodopa-unresponsive parkinsonism, cerebellar ataxia, and pyramidal signs. MSA affects men and woman equally, usually starting in the sixth decade and progresses rapidly with death occurring after an average of nine years. Pathological features cover selective neuronal cell loss and gliosis in the basal ganglia, cerebellum, pontine and inferior olivary nuclei, pyramidal tract, intermediolateral cell column and Onuf's nucleus. Morphologically, MSA is considered a primary oligodendrogliopathy based on the cellular hallmark, the glial cytoplasmic inclusions (GCIs). GCIs contain primarily α-synuclein (αSYN) and hence link MSA with other α-synucleinopathies, such as Parkinson's disease (PD) and dementia with Lewy Bodies (DLB). Still the underlying mechanism of αSYN aggregates, which appear to play a fundamental role in disease pathogenesis, remains to be determined. However, several molecular and cellular changes, including oxidative stress, mitochondrial dysfunction and apoptotic processes might be involved in neuronal degeneration. Microglial activation has been reported to parallel the neuronal multisystem degeneration in MSA, suggesting neuroinflammation as a key pathogenic mechanism comparable to findings in PD. During the last years, studies analysing polymorphism of genes involved in inflammatory processes, such as interleukin-1alpha (IL-α), interleukin-1beta (IL-1β), interleukin-8, intercellular adhesion molecule-1 and tumor necrosis factor showed elevated MSA risk. These studies point towards a possible role of neuroinflammation in MSA pathogenesis. At present, MSA therapy is only symptomatic and mainly targets parkinsonism and autonomic failure as there is no drug treatment that provides MSA patients with consistent long-term benefits. Neuroprotective or regenerative strategies, including neurotransplantation, appear to be an alternative therapeutic approach for managing MSA patients. Experimentally, different cell types for neural restoration in MSA have been tried. E13 whole ganglionic eminence grafts survive and exert functional benefit in toxin-based MSA models. Moreover, survival, integration and functional benefit of E13 ventral mesencephalic (VM) grafts in toxin-based MSA models has been demonstrated. A particular type of stem cells which is considered clinically more attractive, ethically less problematic and exhibiting immunological properties that make them superior over other cell types are mesenchymal stem cells (MSCs). First described by Friedenstein and colleagues, as a population of bone marrow (BM) cells, also known as fibroblast-colony-forming cells, which adhere to cell culture plastic surfaces, these cells were shown to differentiate into many mesodermal derivatives, such as adipocytes, osteocytes and chondrocytes, *in vitro* and *in vivo* when exposed to appropriate stimuli. The MSCs' ability to differentiate into neural-like and glial-like cells could be shown, albeit *in vitro* only. Based on this and similar results, subsequent studies have been initiated and designed in order to prove these cells' potential to support neuroregeneration and also to provoke their immunomodulatory properties in regions, which are actually void of MSC (reviewed). This particular body of literature is vastly growing, yet lacking strong *in vivo* evidence which demonstrates that MSCs, unless they are ectopically placed there or infused in large numbers, are indeed capable of bringing forth neuro-ectodermal derivatives. In light of this, many findings and interpretations remain elusive. Long-term clinical and radiological effects of MSCs in patients with MSA have been described by Lee and co-workers in 2008. In an open-label study design, the neurological deficits in 11 patients with the cerebellar type of MSA (MSA-C), who received consecutively intra-arterial and three repeated intravenous injections for three months, were compared with non-treated MSA patients, demonstrating a delay in progression of neurological deficits after MSC therapy. A recent study by the same group investigated successful neuroprotective and immunomodulatory effects of human MSCs in a double-toxin induced animal model of MSA-P. However this double-toxin induced model solely represents striatonigral- like pathology, without reproducing oligodendroglial inclusion pathology, mediating secondary neuronal multisystem degeneration. At present to our knowledge, there is no experimental evidence for the neuroregenerative potential of MSCs in transgenic mice, overexpressing oligodendroglial αSYN, mimicking important aspects of MSA, such as neuronal loss linked to MSA-like progressive autonomic failure, cerebellar ataxia and parkinsonism, GCI pathology, astrogliosis and microglial activation. For this reason we applied murine MSCs intravenously in aged (PLP)-αSYN transgenic mice and analysed possible neuroprotective effects and the capacity of modulating neuroinflammation. # Methods ## Animals In the present study homozygous (PLP)-α-SYN mice at the age of 18 months were used. The animal study was designed compliant with the Austrian guidelines for the care and use of laboratory animals and all experiments were approved by the Federal Ministry for Education, Science and Research of Austria with the reference number do. ZI. 5004. Animals were housed at the Animal Facility of the Innsbruck Medical University under a 12-hour light/dark cycle with food and water available *ad libitum*. ## Isolation of GFP MSCs MSCs were obtained from C57BL/6-Tg(UBC-GFP)30Scha/J mice (Charles River, Germany), 6–8 weeks old, expressing the enhanced green fluorescent protein (GFP) gene under the human ubiquitin C promoter. Primary GFP mMSC cultures were isolated according to established protocols. Briefly, tibia and femur were treated with collagenase (Sigma, St. Louis, MO, USA) for 2 h, 37°C, 20% O₂, 5% CO₂. Thereafter, fragmented bones were centrifuged and cell fractions were loaded on a Ficoll-Paque Plus gradient (Amersham Biosciences, Piscataway, NJ, USA) to harvest cells from the interphase, followed by a washing step. The isolated bone marrow cells were seeded for expansion in complete isolation medium constituted of RPMI-1640 (Gibco, Invitrogen, Carlsbad, Ca, USA) supplemented with 20% fetal bovine serum (Invitrogen), 100 units/ml penicillin and 100 µg/ml streptomycin (Invitrogen). After 24 hours non-adhering cells were removed by extensive washing with Dulbecco's phosphate buffered saline (DPBS Invitrogen). The attached cells were cultured until confluent and subsequently subcultured at low density (50 cells/cm²) with complete expansion medium consistent of Iscove's Modified Dulbecco's Medium (IMDM, Invitrogen) supplemented with 20% fetal bovine serum (Invitrogen), 100 units/ml penicillin and 100 µg/ml streptomycin (Invitrogen). Medium exchange was performed twice weekly. ## Characterization of GFP MSCs by flow cytometry GFP MSCs were washed with DPBS, harvested with 0.25% trypsin and 1 mM EDTA (Invitrogen) for five minutes at 37°C, divided into round-bottom polystyrene tubes and incubated with pooled mouse IgG (Sigma Aldrich, St. Louis, MO, USA) for 15 minutes at room temperature (RT). Subsequently, cells were labelled with phycoerythrin (PE)- or peridinin-chlorophyll protein complex (PerCP)-conjugated monoclonal antibodies (mABs) specific for CD29, CD11b, CD105, CD34, CD117 (c-kit), CD44, Ly6A/E (Sca-1) (all Biolegend, San Diego, CA, USA) as wells as SSEA-4 (R&D Systems, Mineapolis, MN, USA), MHC Class II (I-A/I-E), MHC Class I (H2D) and CD45 (all Becton Dickinson Biosciences, San Jose, CA, USA) for 30 minutes at 4°C in the dark. PE- as well as PerCP-conjugated isotype-matched mABs were used as negative controls. To asses cell viability Via**-**Probe™ Cell Viability solution (Becton Dickinson Biosciences, San Jose, CA, USA) was added shortly before flow cytometric analysis. Two additional washes were performed and cell surface antigen expression was analyzed on a FACScan using CellQuest^TM software (both BD Biosciences, San Jose, CA, USA) with 10,000 events recorded for each sample. ## In vitro differentiation of GFP MSCs Assessing the potential of isolated cells to differentiate into osteogenic and adipogenic lineages was performed as previously described. Briefly osteogenic differentiation was induced culturing MSCs in 6-well culture plates (TPP, Trasadingen, Switzerland) in IMDM medium containing 10% FBS, 100 units/ml penicillin and 100 µg/ml streptomycin and supplemented with 50 µM ascorbate 2-phosphate, 10 mM β-glycerol phosphate and 100 nM dexamethasone (all from Sigma Aldrich, St. Louis, MO, USA). Medium was changed twice a week for a period of 2–3 weeks. To observe calcium deposition, cultures were washed with PBS, fixed with 4% paraformaldehyde (PFA, Sigma Aldrich, St. Louis, MO, USA) for ten minutes and stained with Alizarin Red, pH 4.1, for ten minutes on a rotating platform. Cultures were rinsed two or three times with PBS to reduce non- specific staining. Adipogenic differentiation was induced after growing MSCs as a monolayer and allowing them to become confluent. Complete medium was exchanged to adipogenic induction medium consisting of IMDM medium containing 10% FBS, 100 units/ml penicillin and 100 µg/ml streptomycin, 1 µM dexamethasone and 0.5 mM methyl- isobutylxanthine, 10 µg/ml insulin and 100 µM indomethacin (all from Sigma Aldrich, St. Louis, MO, USA). Cells were incubated in this medium 48–72 hours and then adipogenic maintenance medium containing 10 µg/ml insulin and 10% FBS in IMDM was applied for 24 hours. Cells were then again treated for 48–72 hours with adipogenic induction medium followed by a period of 24 hours in maintenance medium and a third treatment with induction medium. Finally, cultures were kept for one week in adipogenic maintenance medium. Cells were fixed in 4% PFA and lipid droplet staining was performed using Oil Red O (Sigma Aldrich, St. Louis, MO, USA). ## Cell transplantation For MSC transplantation, two groups of (PLP)-α-SYN mice were included in the study, one group termed (PLP)-α-SYN+MSC (n = 12) receiving 500,000 cells in 150 µl of saline through the tail vein, and the control group termed (PLP)-α-SYN (n = 6) sham injected with an equal amount of saline only. Survival in the (PLP)-α-SYN+MSC group was around 60%. ## Behaviour To determine the efficacy of intravenously transplanted MSCs with respect to a potential amelioration of motor dysfunction in the (PLP)-α-SYN mouse modelling MSA, the following motor function assessment was carried out: beam walking test and stride length analysis with DigiGait. ### Beam walking test Fine motor coordination and balance capabilities of mice were assessed by the beam walking test. The beams consisted of long stripes of wood (each measuring 70 cm) with square cross sections of 0.9 cm and 1.6 cm, horizontally placed 50 cm above the bench surface. The mice were encouraged to walk a distance of 50 cm. For training, three daily sessions of three trials (9 crossings) were performed using the 1.6 cm square large beam. Mice were then tested 1 week, 2 weeks and 4 weeks post transplantation (p.t.) using the 0.9 cm square beam. Mice were allowed up to 60 seconds to traverse the beam. The latency to traverse the beam and the number of times the hind feet slipped off, over the given distance of 50 cm, were recorded for three consecutive runs. Analysis of each session was based on the mean score of the three trials. ### Stride length analysis with DigiGait The DigiGait System (DigiGait Imaging System, Mouse Specifics, Boston, MA, USA) is a non-invasive method for quantitatively compare gait dynamics. Each mouse was placed on a transparent belt of a treadmill enclosed by a plastic scaffold. The speed of the treadmill was set to 20 cm/s for all experimental groups. The ventral side of the mice as they walk was imaged by a high-speed camera, which captured the dynamics of the paws and corresponding limbs as they approach and move away from the belt. A special software, DigiGait Imaging System, Mouse Specifics, Boston, MA, USA) automatically calculated stride length and other spatial and temporal gait indices for each limb. DigiGait analysis was performed at 1 week, 2 and 4 weeks p.t. ## Tissue processing Four weeks after MSC treatment, animals were transcardially perfused with PBS under deep thiopental anaesthesia. Brains were removed and cut to separate the hemispheres. One hemisphere was put into 4% PFA overnight and cryoprotected with 20% sucrose. Brains were slowly frozen and kept at −80°C for further processing. The other hemisphere was cut to obtain midbrain-brainstem tissue, put into a cryovial (Nunc, Rochester, New York, USA) and frozen in liquid nitrogen and stored at −80°C until use. A modified RIPA buffer was used to homogenize brain tissue of each preparation. Brain homogenates were centrifuged at 16,000 g for ten minutes at 4°C and supernatants were stored at −80°C until further processing. ## Immunohistochemistry Six series of 40 µm sections throughout the whole hemisphere were cut on a cryostat (Leica, Nussloch, Germany). One series was directly mounted on gelatine-coated slides and used for cresyl violet (Nissl) staining. Immunolabelling was performed on free floating sections using the following antibodies: rat anti-mouse CD11b (1∶150; AbD Serotec, Oxford, UK), polyclonal rabbit anti-green fluorescence protein (1∶1000; GFP, Abcam, Cambridge, UK), polyclonal rabbit CD3 (1∶7500, Abcam, Cambrige, UK) and monoclonal mouse anti- tyrosine hydroxilase (TH, 1∶1000, Sigma, St. Louis, MO, USA). Secondary antibodies were biotinylated anti-rat IgG, biotinylated anti-rabbit IgG, anti- mouse IgG (1∶200, all Vector Laboratories, Burlingame, CA), Alexa-fluor 488-conjugated goat anti-rabbit or Alexa-fluor 594-conjugated goat anti-rat (both 1∶500, Molecular Probes, Leiden), respectively. Endogenous peroxidase activity was quenched in H₂O_2. After normal serum blocking, sections were incubated with the primary antibody overnight at 4°C, followed by incubation in biotinylated secondary antibody. After incubation in Vectastain ABC reagent (Vectastain ABC kit, Vector Laboratories, Burlingame, CA), the immunohistochemical reaction was developed with 3,3′-diaminobenzidine (DAB) and sections were mounted onto gelatine-coated slides, counterstained with cresyl violet or Mayer's haematoxylin solution, dehydrated and coverslipped with Entellan. Immunofluorescence staining performed for tracing GFP positive MSCs was carried out by normal serum blocking and overnight incubation with the anti-GFP primary antibody followed by incubation with the respective secondary antibody, and counterstaining of the nucleus by 4′, 6-Diamidin-2′-phenylindoldihydrochlorid (DAPI, Sigma Aldrich, St. Louis, MO, USA). For GFP staining, brain sections from C57BL/6-Tg(UBC-GFP)30Scha/J mice were used as positive controls and tissue sections from (PLP)-αSYN transgenic animals as negative controls. Double-immunofluorescence staining for GFP and CD11b was performed as described above; dilution of the commercial rat anti- mouse CD11b antibody was 1∶50. ## Quantification of α-synuclein concentration in brain lysates Brain lysates were analysed for αSYN concentration using an α-synuclein immunoassay kit (Invitrogen, Carlsbad, CA, USA), following manufacturer's instructions. ELISA plates were analysed with a multi-well plate reader at 450 nm (Beckman Coulter, Brea, CA, USA). ## Quantification of cytokine concentrations in brain lysates Brain lysates were analysed using the mouse Th1/Th2 10-plex kit, MCP-1 and TGF-β1 (Flow Cytomix, Bender MedSystems, Vienna, Austria) according to the manufacturer's instructions. Data were acquired using a FACScan (BD Biosciences, San Jose, CA, USA) with 1500 events recorded for each sample and further analysed by the Flow Cytomix Software version 2.3 (Bender MedSystems, Vienna, Austria). ## Microscopy and image analysis technique Cell culture microscopy was performed using a Leica DMI 4000B microscope and Application Suite V3.1 (Leica, Wetzlar, Germany). Fluorescent histological sections were analysed with the aid of an ApoTome® microscope and AxioVision Software (both Carl Zeiss Microimaging GmbH, Jena, Germany). All morphometric analysis was done in a blinded way applying a computer-assisted image analysis system (Nikon E-800 microscope, CCD video camera, Optronics MicroFire, Goleta, USA; Stereo Investigator Software, MicroBrightField Europe e.K., Magdeburg, Germany). The optical fractionator method, was used to estimate the total number of neurons and microglia in the substantia nigra pars compacta (SNc). ## Statistics All data are given as means ± standard error of the mean (SEM). Behavioural data were compared by two-way analysis of variance (ANOVA) for time and treatment effects followed by a *post hoc* Bonferroni test (corrected for multiple comparisons). Data from cytokine measurements were subjected to two-tailed unpaired Student's *t-test* with regard to treatment. Data obtained from image analysis technique were analysed by unpaired Student's *t-test*. Correlations between cytokines and the number of TH⁺ neurons were performed with the Pearson correlation analysis. All statistical analyses were performed with GraphPad Prism 5 Software (GraphPad Software Inc., San Diego, CA, USA). A p-value of p\<0.05 was considered significant. # Results ## Isolation and characterization of murine GFP MSCs MSCs of GFP transgenic mice were isolated from tibia and femur and kept in culture for several passages. For characterization, differentiation and transplantation, cells at passage 8 were used (**, A**). Before intravenous application the cells were characterised, flow cytometry analysis confirmed that the cells at the stage of transplantation were positive for GFP (96.65%), CD29 (98.72%), CD44 (99.51%), CD105 (99.43%), MHC Class I (H2D, 48.23%), Sca-1 (99.48%), SSEA-4 (95.91%) and had a low number of CD11b (13.28%), CD34 (13.10%), CD45 (1.72%), CD117 (c-kit, 15.76%) and MHC Class II (I-A/I-E, 11.84%). At the stage of transplantation 99.92% of all cells were viable, as revealed by Via Probe staining. Multilineage potential was demonstrated by differentiation into adipocytes, indicated by Oil-Red O staining (**, B**) as well as calcium deposits indicating osteogenic lineage differentiation when stained with Alizarin Red (**, C**). ## Cell transplantation and tracing of GFP MSCs Two groups of aged (PLP)-α-SYN mice were included in the study, one group receiving 500,000 cells/150 µl of saline through the tail vein designated (PLP)-αSYN+MSC, and one group serving as controls with an equal amount of saline only into the tail vein termed (PLP)-αSYN. Immunofluorescence with a polyclonal rabbit anti-GFP antibody was performed on PFA fixed MSCs in culture to ensure tracing of GFP MSCs. Single engrafted donor cells were detected four weeks post MSC injection in the (PLP)-αSYN+MSC group whereas sections from PLP-αSYN mice served as negative controls. To control the efficiency of the antibody in fixed tissue, brain sections from (PLP)-αSYN control group served as negative control while brain sections from C57BL/6-Tg(UBC-GFP)30Scha/J served as positive control. In face of potential contaminations of myeloid cells (CD11b) in the MSC culture, double staining with GFP and CD11b **(**) was performed in sections of the (PLP)-αSYN+MSC group to prove the phenotype of GFP positive cells being MSCs and not myeloid cells. ## Behaviour The efficacy of intravenously transplanted MSCs to restore motor function in aged (PLP)-αSYN mice versus controls, was measured with the beam walking test, and stride length was analysed with the DigiGait system. The beam walking test determines fine motor coordination and balance capabilities. Traversing the beam was performed 1 week, 2 and 4 weeks p.t. Over the period of 4 weeks, no significant improvement in the time traversing the beam (p\>0.05) or in the number of sideslips (p\>0.05) was detected. We analysed the stride length on both hindlimbs (left, right) with the DigiGait System in the transplant and control group. Previous results from our research group have demonstrated that (PLP)-αSYN transgenic mice show shortening of hindlimb stride length associated with TH⁺ cell loss in the SNc. In the current study we analysed the stride length of the left and right hindlimbs in the transplant and control group at 1 week, 2 week and 4 weeks p.t. Stride length in animals treated with i.v. MSCs was not significantly altered from stride length in the control group (for both hindlimbs p\>0.05). Both tests demonstrate that i.v. MSC treatment in aged (PLP)-αSYN mice has not induced changes in motor behaviour. ## Neuroprotective effect on TH⁺ neurons in the SNc after MSC treatment Reduction of TH-immunoreactive neurons was previously reported in the SNc of (PLP)-αSYN transgenic mice suggesting that the presence of αSYN in oligodendrocytes induces dopaminergic neuron loss. We performed TH staining and stereological counting in the SNc in the MSC treatment and control group **** demonstrating a significant recovery of the total number of TH⁺ neurons in the MSC treated group ((PLP)-αSYN+MSC 4747±356.8 vs. (PLP)-αSYN 3510±368.8; p = 0.036) 4 weeks after transplantation. ## αSYN concentration in midbrain-brainstem lysates In previous studies, research on the (PLP)-α-SYN mouse model has demonstrated that pathological αSYN accumulation promotes degeneration of neurons in the SNc, locus coeruleus, nucleus ambiguous, laterodorsal tegmental nucleus, pedunculopontine nucleus and Onuf's nucleus, similar to findings in MSA patients (reviewed). With a αSYN immunoassay we investigate whether MSC treatment had an effect on αSYN concentration in midbrain-brainstem lysates. We chose midbrain- brainstem samples since the dissected area includes the affected nuclei. Statistical analysis by unpaired Student's *t-test* did not show differences on αSYN concentration in MSC treated (n = 7) versus control animals (n = 6) ((PLP)-α-SYN+MSC 1.899±0.1032 vs. (PLP)-α-SYN 1.692±0.1322, p = 0.237). ## MSCs influence cytokine levels in midbrain-brainstem lysates IL-1α, IL-2, IL-4, IL-5, IL-6, IL-10, IL-17, GM-CSF, INFγ, MCP-1, TGF-β1 and TNFα were assessed in midbrain-brainstem lysates of (PLP)-αSYN+MSC versus (PLP)-αSYN animals. Four weeks after intravenous MSC application, we observed a significant downregulation of IL-1α (p = 0.0014), IL-2 (p = 0.019), IL-10 (p = 0.046), IL-17 (p = 0.056), GM-CSF (p = 0.029), TGF-β1 (p = 0.001) and TNFα (p = 0.001) in midbrain-brainstem lysates of MSC treated animals versus control group, whereas no significant difference was found for IL-4 (p = 0.16), MCP-1 (p = 0.18), IL-5 (0.072), IL-6 (p = 0.18) and INFγ (p = 0.31). Due to a significant reduction of the lymphocytic signalling molecules IL-2 and IL-17, we performed a staining for the T-cell marker CD3 in (PLP)-αSYN+MSC and (PLP)-αSYN animals. We could detect single CD3⁺ cells throughout different brain regions. Moreover, we analysed microglia as an additional source of cytokines. Previous work of our group has shown that microglial activation in the (PLP)-αSYN mouse is present in the SNc and mediates neurodegeneration. Hence counting CD11b-immunoreactive cells in the SNc as a marker for microglia was performed (**, B–C**). Statistical analysis with two-tailed unpaired Student's *t-test*, revealed no significant difference in the number of CD11b⁺ cells in transplanted animals (n = 7) versus the control group (n = 6) ((PLP)-αSYN+MSC 6588±242 vs. (PLP)-αSYN 6880±529.3; p = 0.6075) (**, A**). Finally, in order to evaluate whether the cytokines in midbrain-brainstem lysates (IL-1α, IL-2, IL-4, IL-5, IL-6, IL-10, IL-17, GM-CSF, INFγ, MCP-1, TGF-β1 and TNFα) contribute to TH⁺ neuronal rescue we established a Pearson correlation between the cytokines and TH⁺ neurons. We could demonstrate significant inverse correlations for IL-1α (Pearson r = −0.697, p = 0.0081), IL-2 (Pearson r = −0.823, p = 0.0005), TGF-β1 (Pearson r = −0.647, p = 0.0169) and TNFα (Pearson r = −0.0575, p = 0.0397). # Discussion In the present study we aimed to determine whether intravenous application of murine MSCs in an aged (PLP)-αSYN MSA mouse model ameliorates behavioural deficits and exerts neuroprotective and immunomodulatory properties. To date, to our knowledge there is no experimental evidence of a MSC therapeutic approach in transgenic MSA models. Recently, human MSCs have been demonstrated to protect against loss of neurons in SN and striatum in an animal model of double toxin- induced MSA-P. A few years ago, in an open-label study design, disease modifying effects of intra-arterial and intravenous injected MSCs in eleven patients with MSA-C, have been reported. Together with *in vitro* and *in vivo* findings in PD and other neurodegenerative disorders, MSCs seem to be an attractive and feasible therapeutic intervention,. As a matter of fact, experimental results are often difficult to compare, in lieu of standards for MSC isolation, cultivation and in vivo application. Furthermore there is no unique marker to identify MSCs. In general, MSCs are characterised upon expression of a group of surface receptors and upon their multilineage potential. Prior to transplantation we have characterized GFP MSCs, based on a set of criteria proposed by the International Society for Cellular Thearpy, which includes expression and lack of surface markers and differentiation potential into adipocytes and osteocytes. In the (PLP)-αSYN+MSC transplant group the survival rate after intravenous MSC infusion was low, probably due to the high age of the animals. Further we presume that due to their size MSCs got trapped within the pulmonary capillaries, causing pulmonary and hemodynamic alterations. On the other hand numerous animal studies and clinical trials have reported favourable outcomes following systemic infusion of MSCs. Since we have only encountered a few single donor engrafted GFP⁺ MSCs four weeks after transplantation, the use of GFP as experimental tool to examine the survival and fate in future studies is in question, since there are a lot of inconsistent results in the literature, on tracking and determining cell fate of MSCs using GFP as a reporter in transplantation studies. In addition, we cannot exclude a detrimental effect of the GFP protein on MSC survival and therefore preventing proper integration into sites of neurodegeneration. On the other hand, a number of reports state that MSCs exert effects on tissue repair despite exhibiting low and/or transient levels of engraftment. This foreshadows a novel concept of tissue repair relying on secretion of trophic factors and/or crosstalk with the microenvironment rather than MSC transdifferentation. The (PLP)-αSYN transgenic mouse model has been widely characterized in terms of effects of αSYN overexpression on neurodegeneration and motor activity. Furthermore, this animal model has been successfully applied in neuroprotective studies as a preclinical rationale for phase II clinical trials. We performed the beamwalking test to define fine motor coordination and balance capabilities as well as stride length with a digital system, since shortening of the stride length has been reported in the (PLP)-αSYN transgenic mouse. In a recent study, intravenous human MSCs have ameliorated behavioural deficits in a double-toxin induced mouse model of MSA-P. Our experimental data show no significant effects of intravenously applied MSCs to alleviate behavioural failure. On the other hand, the study with double-toxin induced striatonigral degeneration has been performed in mice lacking αSYN expression in oligodendroglia and thus replicating solely striatonigral-like pathology without reproducing other cardinal features of MSA. It remains elusive whether oligodendroglial αSYN exerts a deleterious effect on transplanted MSCs, however in a recent study addressing the fate of embryonic striatal grafts in presence of oligodendroglial αSYN inclusions, disturbed dopaminergic re-innervation and reduced p-zone volume of the grafts in the MSA mouse model has been attributed to effects of host αSYN pathology. For our pilot study, we have chosen transgenic animals at the age of 18 months, overexpressing αSYN under control of the PLP promoter, since GCIs are the hallmark of MSA and MSA is a late onset disease. Previous studies have clearly demonstrated that in this MSA animal model αSYN overexpression leads to neurodegeneration, resembling human neuropathology. At the age of 18 months, neurodegeneration due to αSYN overexpression is at a much more progressed stage, impairing motor activity drastically as shown in MSA mouse models overexpressing αSYN under control of oligodendroglial promoters. Currently we are examining this issue more closely in the (PLP)-αSYN mouse. Nevertheless, the absent behavioural improvement after MSC treatment, leads to the conclusion, that MSC treatment at later disease stages does not induce the desired effect of ameliorating behavioural deficits. Since one pathological feature in MSA patients covers selective neuronal loss in the SNc, and the (PLP)-αSYN mouse model replicates this feature, we evaluated whether MSC treatment had an effect on number of TH⁺ neurons. There was a subtle but significant recovery of numbers of dopaminergic neurons in the MSC transplant group compared to transgenic controls. We further investigated putative factors that may contribute to this “rescue” of dopaminergic neurons after MSC treatment. We analysed whether TH recovery is caused by the decrease of αSYN concentration in midbrain-brainstem lysates. αSYN is known to be a key factor involved in oligodendroglial and neuronal loss in MSA patients and in the (PLP)-αSYN transgenic animal model. Recently genetic variants in the αSYN gene *SNCA* have been associated with an increased risk in developing MSA. However, αSYN concentration in the midbrain-brainstem region was not significantly altered between (PLP)-αSYN+MSC treatment compared to (PLP)-αSYN control group. This finding highlights the concept of tissue repair of MSCs by releasing anti-inflammatory and trophic molecules. Neuroinflammation has been widely regarded as a possible key player in progressing disease pathogenesis in various neurodegenerative diseases. In PD patients as well as PD animal models, neuroinflammation in terms of microglial activation has been observed. Lately emerging evidence for the presence of T-lymphocytes in the midbrain of PD patients suggests that a potential role of infiltrated peripheral cells is related to PD pathogenesis. In a recent study infiltration of T-cells into the brain actively participated in dopaminergic neuron degeneration in the SNc. Additionally, overexpression of human αSYN in mouse SN neurons, induced by an adeno-associated viral vector, has led to activation of microglia, production of inflammatory cytokines and stimulated the adaptive immune response. In our experiment we analysed twelve cytokines, IL-1α, IL-2, IL-4, IL-5, IL-6, IL-10, IL-17, GM-CSF, INFγ, MCP-1, TGF-β1 and TNFα in midbrain-brainstem lysates and found significant downregulation of IL-1α, IL-2, IL-10, IL-17, GM-CSF, TGF-β1 and TNFα four weeks after intravenous MSC application in the treatment group. Additionally we encountered CD3⁺ T-cells throughout the brain in (PLP)-αSYN and (PLP)-αSYN+MSC treated animals. After MSC treatment, however, the T-cell specific cytokines IL-2 and IL-17 were significantly downregulated. Furthermore, TH neuronal “rescue” was inversely correlated with IL-2, indicating that MSC treatment influenced pathogenic T-cell response in (PLP)-αSYN mice. Similar effects have been widely investigated for multiple sclerosis and experimental autoimmune encephalitis and support the role of MSC treatment by modulation of T-cell response. Finally we analyzed the influence of MSC treatment on microglial activation, since microglia have been reported to parallel the neuronal multisystem degeneration in MSA and mediates dopaminergic neuronal loss related to oligodendroglial α-synucleinopathy in the (PLP)-αSYN mouse. We quantified the number CD11b⁺ microglial cells in the SNc, yet could not demonstrate a significant difference in microglial cell number after MSC treatment. However, this finding cannot exclude a modulatory effect on microglial activation status. Furthermore, the major proinflammatory cytokine TNFα and IL-1α, also released by activated microglia and astroglia was significantly decreased in brain lysates of the MSC treatment group and inversely correlate with the number of TH neurons indicating that suppression of microglial activation and astrogliosis may contribute to dopaminergic neuronal survival. Similar results, demonstrating decreased activation of astrocytes and microglia by human MSCs in a mouse model of MSA-P have been recently reported and are in good concordance with our findings that MSCs exert modulatory effects on neuroinflammation and promote survival of dopaminergic neurons in the SNc. In summary, our study describes the first experimental attempt using MSCs as a therapeutic intervention in an aged transgenic mouse model of MSA featuring oligodendroglial α-synucleinopathy. We have demonstrated that intravenous application of MSCs leads to a rescue of dopaminergic neurons in the SNc. Furthermore we could demonstrate a profound immunomodulatory effect after MSC treatment, resulting in downregulation of various proinflammatory cytokines, which are linked to microglial activation, astrogliosis and mediation of adaptive immunity. However, MSC treatment did not alter behavioural deficits in aged transgenic MSA mice. Our data have potential implications for MSCs as a future stem cell source in MSA therapies. Nevertheless, further experimental studies on the efficacy of MSCs as disease modifying candidates in MSA as well as different routes of application have to be performed. Prior to embarking on further human trials, preclinical studies are necessary since they will reveal if all that glitters experimentally is truly clinical gold. # Supporting Information The authors are grateful for technical assistance by Monika Hainzer. Sincere thanks to Maria Auer for assistance with the ApoTome. [^1]: Conceived and designed the experiments: SS MR. Performed the experiments: SS AJ. Analyzed the data: SS NS MR. Wrote the paper: SS. Revised the manuscript for important intellectual content: NS GL MR GKW. [^2]: The authors have declared that no competing interests exist.