Datasets:

suolyer

/

pile_pubmed-central

like 1

1. Introduction {#sec1-ijms-20-03201} =============== Plants are continuously subjected to a broad range of biotic and abiotic stresses that negatively affect plant productivity. Therefore, revealing the relevant mechanisms of plant responses to stress is of great importance. Available evidence suggests that reactive oxygen species (ROS) production in plants is a common response to almost all environmental challenges \[[@B1-ijms-20-03201],[@B2-ijms-20-03201],[@B3-ijms-20-03201]\]. ROS work as highly controlled signaling species that are able to deliver different environmental cues to the plant cell transcriptional machinery \[[@B4-ijms-20-03201],[@B5-ijms-20-03201],[@B6-ijms-20-03201],[@B7-ijms-20-03201],[@B8-ijms-20-03201],[@B9-ijms-20-03201]\]. This signaling function of ROS is becoming more operative through their interactions with other various signaling components including phytohormones, G proteins, calcium ions and mitogen-activated protein kinases (MAPKs) \[[@B7-ijms-20-03201]\]. Furthermore, ROS are reported to mediate cellular signals through direct interaction with transcription factors, including heat shock factors (HSFs). For example, HSFA2 and HSFA4A were reported to be induced by oxidative stress and involved in H~2~O~2~ sensing \[[@B10-ijms-20-03201],[@B11-ijms-20-03201]\]. H~2~O~2~, the most widespread cellular ROS, is an uncharged molecule that defuses readily through biological membranes \[[@B12-ijms-20-03201],[@B13-ijms-20-03201]\] and thus can efficiently deliver intracellular systemic signals \[[@B7-ijms-20-03201]\]. Indeed, H~2~O~2~ has the ability to modulate the expressional behavior of many genes in various organisms. The expression of about one-third of *Saccharomyces cerevisiae* genes was shown to be modulated upon exposure to H~2~O~2~ \[[@B14-ijms-20-03201]\]. In Arabidopsis, 175 out of 11,000 non-redundant expressed sequence tags were found to be regulated by H~2~O~2~ treatment \[[@B15-ijms-20-03201]\]. Similarly, production of H~2~O~2~ in chloroplasts of Arabidopsis plants treated with methyl viologen led to the differential expression of a large number of genes, with heat shock proteins (HSPs) being the most abundant group between the upregulated genes \[[@B16-ijms-20-03201]\]. Various HSPs are produced in plants under environmental stress conditions such as high temperature, drought, salinity, osmotic, cold and oxidative stresses \[[@B17-ijms-20-03201],[@B18-ijms-20-03201],[@B19-ijms-20-03201],[@B20-ijms-20-03201],[@B21-ijms-20-03201],[@B22-ijms-20-03201],[@B23-ijms-20-03201]\]. It is widely accepted that HSPs play a crucial role in protecting cell structures against stress in plants and other organisms \[[@B24-ijms-20-03201],[@B25-ijms-20-03201],[@B26-ijms-20-03201],[@B27-ijms-20-03201],[@B28-ijms-20-03201],[@B29-ijms-20-03201]\]. Small HSPs (sHSPs) are the most abundant HSPs that are produced universally in prokaryotic and eukaryotic cells upon heat stress \[[@B30-ijms-20-03201],[@B31-ijms-20-03201]\]. In Arabidopsis, sHSP genes are not expressed in unstressed tissues but show high levels of gene expression during heat stress and other environmental challenges \[[@B25-ijms-20-03201],[@B32-ijms-20-03201],[@B33-ijms-20-03201]\]. A number of sHSPs were highly induced on exposure of Arabidopsis plants to heat, osmotic and salinity; with the induction of these genes were more pronounced under a combined treatment of these three stresses \[[@B34-ijms-20-03201]\]. High levels of sHSP gene expression and protein accumulation upon environmental stresses support the hypothesis that these proteins play an important role in stress tolerance \[[@B35-ijms-20-03201],[@B36-ijms-20-03201]\]. For cell survival under stress conditions, the maintenance of proteins in their functional conformations and the prevention of the aggregation of non-native proteins are particularly important. In this regard, HSPs are involved in stabilizing proteins and membranes, and thus they can assist protein refolding under stress conditions \[[@B24-ijms-20-03201],[@B37-ijms-20-03201],[@B38-ijms-20-03201],[@B39-ijms-20-03201]\]. It was reported that sHSPs represent the first line of defense in the cell to prevent protein misfolding \[[@B40-ijms-20-03201]\]. Certain sHSPs, such as HSP18.1 isolated from pea (*Pisum sativum*) and HSP16.6 from *Synechocystis* sp. PCC6803, were shown to bind to unfolded proteins in vitro and this enables further refolding by HSP70/HSP100 complexes \[[@B41-ijms-20-03201]\]. In addition to their chaperone function, it is also suggested that sHSPs modulate fluidity and composition of cell membranes \[[@B42-ijms-20-03201]\]. sHSPs have roles in protection against oxidative stress. For example, overexpression of chloroplastic sHSPs in tomato (*HSP21*) and tobacco (*HSP26)* provided evidence that these sHSP protect photosystem II from oxidative stress \[[@B43-ijms-20-03201],[@B44-ijms-20-03201]\]. The overexpression of an sHSP (*LimHSP16.45*) from the David lily (*Lilium davidii*) in Arabidopsis led to enhanced cell viability under high temperatures, salinity, and oxidative stress \[[@B45-ijms-20-03201]\]. In addition, *LimHSP16.45* overexpressing plants showed greater activity of superoxide dismutase and catalase than control plants, suggesting that LimHSP16.45 can protect plants against abiotic stresses by enhancing enzymes that scavenge ROS, in addition to their roles in preventing irreversible protein aggregation \[[@B45-ijms-20-03201]\]. In previous work, we found that the expression level of the Arabidopsis sHSP gene, *HSP17.4CI*, was highly induced by heat, salt and drought \[[@B34-ijms-20-03201],[@B46-ijms-20-03201]\]. This finding attracted our attention to further study the dynamic changes in expression patterns of this gene under a wide range of stresses. We were particularly interested in finding out the specific patterns of *HSP17.4CI* expression that would shed light on the modes by which different environmental challenges modulate gene expression. Here, we investigated the expression patterns of *HSP17.4CI* under various abiotic, biotic and oxidative stresses, as well as after treatment with stress-related phytohormones. The expression of *HSP17.4CI* was significantly enhanced by stresses that involve ROS production such as heat, cold, salt, drought, high light, the biotrophic pathogen *Pseudomonas syringae,* in addition to phytohormones such as abscisic acid (ABA) and salicylic acid (SA) that are involved in mediating signals of these stresses. Furthermore, by using mutant plants that are deficient in catalase 2 activity and consequently accumulate endogenous H~2~O~2~, we have supported the proposal that the induction of *HSP17.4CI* is modulated by ROS, where its expression was significantly induced in these *cat2* loss-of-function mutants. 2. Results {#sec2-ijms-20-03201} ========== 2.1. The Expression of HSP17.4CI is Induced by Various Abiotic Stresses {#sec2dot1-ijms-20-03201} ----------------------------------------------------------------------- To systematically profile the expression pattern of the Arabidopsis *HSP17.4CI*, encoding a cytosolic sHSP, under various stress treatments, we first studied the effect of different abiotic stresses (including heat, cold, salt, drought and high light stress treatments) on the expression pattern of the At*HSP17.4CI* gene. Various growth media used here and sampling time points for each treatment were selected based on previous work \[e.g., 46, and references therein\]. Our results show that exposure to a heat dose of 45 °C for 2 h, dramatically induced *HSP17.4CI* transcript levels ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}A). The effect of prolonged heat shock at 45 °C for 4 or 6 h on *HSP17.4CI* expression was similar to that observed at 2 h. These results show that *HSP17.4CI* has very low basal expression levels in unstressed plants, but is strongly induced by heat stress. This finding is consistent with previous reports indicating that under normal growth conditions, transcripts of most sHSPs cannot be detected in vegetative tissues, but are rapidly produced in response to heat \[[@B35-ijms-20-03201]\]. The present results also show that cold treatment at 0 °C for 3 h upregulated *HSP17.4CI* expression by more than 4-fold compared to control, with no additional significant increase in the expression level after prolonged cold exposures of 6 or 12 h ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}B). Under salt stress, *HSP17.4CI* was induced by more than 3-fold after 2 h of treatment and 7-fold after 6 h. After 10 h of salt treatment no additional increase in *HSP17.4CI* transcript abundance was detected ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}C). Results presented in [Figure 1](#ijms-20-03201-f001){ref-type="fig"}D show that drought treatment led to upregulation of *HSP17.4CI* by about 2-fold after 1 h and by more than 4-fold at 2 h after the treatment. However, there was no detectable increase compared to control after shorter period of 30 min. Furthermore, our results show that *HSP17.4CI* was induced by high-light stress ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}E). Compared to the control, transcript levels of *HSP17.4CI* increased 10 times after 6 h, and 15 times after 12 h of high light exposure at 800 µE ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}E). Altogether, these data indicate that the expression level of the *HSP17.4CI* gene is significantly induced by various abiotic stress treatments. 2.2. HSP17.4CI is Responsive to Biotrophic but Not to Necrotrophic Pathogens {#sec2dot2-ijms-20-03201} ---------------------------------------------------------------------------- In addition to their roles in plant tolerance to abiotic stresses, HSPs were reported to play a role in plant defense against biotic stresses \[[@B47-ijms-20-03201]\]. Here, we studied the effect of different pathogens on the expression of the Arabidopsis *HSP17.4CI*. Our study included the biotrophic pathogen *Pseudomonas syringae*, and the necrotrophic pathogens *Alternaria brassicicola* and *Fusarium oxysporum.* We observed no changes in the expression levels of *HSP17.4CI* at 6 h after inoculation of Arabidopsis plants with *P. syringae* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}A). However, after 24 h from inoculation we observed an increase of *HSP17.4CI* expression by more than 8-fold compared to the mock-treated control. After 48 h from inoculation, the expression level of *HSP17.4CI* was lower than that after 24 hours, but still 4-times higher than the corresponding mock treated control, suggesting a relatively early role for *HSP17.4CI* during defense against this biotrophic pathogen. In contrast, no changes in the expression level of *HSP17.4CI* were found after infection with the necrotrophic pathogens *A. brassicicola* or *F. oxysporum* at any of the indicated time points ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}B,C). 2.3. HSP17.4CI is Upregulated by Plant Hormones ABA and SA but Not by MJ {#sec2dot3-ijms-20-03201} ------------------------------------------------------------------------ Plant responses to different abiotic and biotic environmental stresses are coordinated by plant hormones such as ABA, SA, and MJ. To investigate the involvement of these plant hormones in the modulation of the expression of *HSP17.4CI*, the transcript level of this gene was monitored in Arabidopsis plants after hormone treatment. Our results show that the *HSP17.4CI* expression level was significantly increased at 1, 4 and 24 h after ABA treatment, reaching more than 4-fold increase after 24 h compared to the control ([Figure 3](#ijms-20-03201-f003){ref-type="fig"}A). Considering the suggestion that ABA regulates plant responses to various abiotic stresses, the induction of *HSP17.4CI* by ABA is in line with *HSP17.4CI* induction by the applied stresses shown in [Figure 1](#ijms-20-03201-f001){ref-type="fig"}. The plant hormone SA is known to be involved in plant growth and development and plays a role during defense against biotrophic pathogens \[[@B48-ijms-20-03201]\]. To investigate the possible involvement of SA in the induction of *HSP17.4CI* by the biotrophic pathogen *P. syringe* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}A), we analyzed *HSP17.4CI* expression in plants treated with SA. Our results show that treating Arabidopsis seedlings with SA significantly induced *HSP17.4CI* at all of the examined time points ([Figure 3](#ijms-20-03201-f003){ref-type="fig"}B). These results are consistent with the induction of this sHSP by the biotrophic pathogen, *P. syringae* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}A) that activates SA signaling \[[@B49-ijms-20-03201]\]. In contrast to ABA and SA, MJ treatment showed no significant effect on *HSP17.4CI* expression in the treated Arabidopsis plants at any of the investigated time points ([Figure 3](#ijms-20-03201-f003){ref-type="fig"}C). This result is in line with the finding that the expression level of this gene was not affected by the necrotrophic pathogens *A. brassicicola* and *F. oxysporum* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}B,C), to which the MJ signaling pathway is known to be involved in plant defense \[[@B49-ijms-20-03201]\]. 2.4. HSP17.4CI is Strongly Upregulated by Oxidative Stress Treatments {#sec2dot4-ijms-20-03201} --------------------------------------------------------------------- A general role of sHSPs in oxidative stress tolerance was presumed by observations that production and accumulation of plant sHSPs increase in response to H~2~O~2~ treatments in vitro \[[@B35-ijms-20-03201],[@B50-ijms-20-03201]\]. We found that the expression level of *HSP17.4CI* was strongly induced in Arabidopsis plants treated with H~2~O~2~ ([Figure 4](#ijms-20-03201-f004){ref-type="fig"}A). After only 3 h of H~2~O~2~ treatment, we observed 130-fold higher expression levels than those of the control samples. Similarly, as shown in [Figure 4](#ijms-20-03201-f004){ref-type="fig"}B, transcript levels of *HSP17.4CI* were strongly increased after treatment with methyl viologen, a herbicide that produces superoxide radicals which are converted into H~2~O~2~ by plant superoxide dismutase in chloroplasts. We found that after 2 h of treatment with methyl viologen, the expression level of *HSP17.4CI* was 21-times higher than the untreated control, reaching a maximum level of 71-times higher than the control plants after 4 h ([Figure 4](#ijms-20-03201-f004){ref-type="fig"}B). 2.5. HSP17.4CI Expression is Enhanced in Mutant Plants that Accumulate Endogenous H~2~O~2~ {#sec2dot5-ijms-20-03201} ------------------------------------------------------------------------------------------ To further examine the hypothesis that the observed induction of *HSP17.4CI* under various stress conditions may be mediated by the production of ROS, we used a photorespiratory mutant deficient in catalase 2 activity (*cat2-2*), which accumulate H~2~O~2~ under normal air conditions \[[@B6-ijms-20-03201],[@B51-ijms-20-03201]\]. We found that the expression level of *HSP17.4CI* in the *cat2-2* mutant seedlings was significantly higher than that of the wild type grown in the same conditions ([Figure 5](#ijms-20-03201-f005){ref-type="fig"}A). As expected, we found that the H~2~O~2~ content in *cat2-2* seedlings was significantly higher than that in the wild type ([Figure 5](#ijms-20-03201-f005){ref-type="fig"}B). These data reinforce the hypothesis that the induction of *HSP17.4CI* under various abiotic and biotic stresses would be at least partly mediated by ROS. 2.6. Responsiveness of the Arabidopsis sHSPs genes to stress factors {#sec2dot6-ijms-20-03201} -------------------------------------------------------------------- To investigate whether the expression profiles monitored in the current study for the cytosolic sHSP, *HSP17.4CI*, may be shared by other sHSPs, we first cross-examined the Arabidopsis genome for genes annotated as sHSPs \[accessed in 34\]. A number of 36 genes were found to be annotated as sHSPs or HSP20-like chaperones ([Table S1](#app1-ijms-20-03201){ref-type="app"}). We considered genome-wide expression data for a number of different stress treatments including heat, drought and salinity \[[@B42-ijms-20-03201]\], as well as oxidative stresses imposed by H~2~O~2~ \[[@B52-ijms-20-03201]\] and methyl viologen \[[@B16-ijms-20-03201]\]. As plants in their natural habitats are usually subjected to combined stresses, and plant response to these combinations cannot be predicted form their responses to individual ones, we also included expression data of *HSP17.4CI* and other sHSPs under a combined treatment of heat, osmotic and salinity \[[@B34-ijms-20-03201]\]. In this study \[[@B34-ijms-20-03201]\], *HSP17.4* was reported to be highly induced by combined stress treatment of heat, drought and salinity. Out of the 36 sHSPs, we found that the expression of 19 genes was differentially expressed by at least on stress treatment ([Table 1](#ijms-20-03201-t001){ref-type="table"}). Only, *HSP17.6C* showed similar expression pattern to *HSP17.4CI* under the considered abiotic and oxidative stress treatments. With exception of drought stress, another three genes (*HSP17.6A*, *HSP17.6* and *HSP23.5*) were induced by heat, salinity and combined treatments, in addition to oxidative stresses imposed by H~2~O~2~ and methyl viologen ([Table 1](#ijms-20-03201-t001){ref-type="table"}). Furthermore, another four genes (*HSP17.6B*, *HSP17.6A*, *HSP17.4B* and *HSP15.7*) were induced only by heat and oxidative stresses ([Table 1](#ijms-20-03201-t001){ref-type="table"}). Notably, two HSP20-like chaperone genes were repressed by multiple (AT4G21870) or heat (AT1G76770) stresses, but not responsive to any of the other treatments. To find any correlation between subcellular localization of various sHSP genes and their expression patters, we allocated these genes to their subcellular sites ([Table 1](#ijms-20-03201-t001){ref-type="table"} and [Table S1](#app1-ijms-20-03201){ref-type="app"}). Data showed that the sHSP genes that showed similar responsiveness to the expression of *HSP17.4CI* were localized to cytoplasm except *HSP23.5* (mitochondrial) and *HSP15.7* (peroxisomal). 3. Discussion {#sec3-ijms-20-03201} ============= ROS produced during adaptation to the majority of relatively mild environmental stresses can act as highly controlled signaling molecules with the ability to deliver different environmental cues to the transcriptional machinery \[[@B1-ijms-20-03201],[@B9-ijms-20-03201],[@B55-ijms-20-03201]\]. The induction of HSPs is one important strategy that plants use to protect themselves against different stresses, including oxidative stress \[[@B21-ijms-20-03201],[@B24-ijms-20-03201],[@B32-ijms-20-03201],[@B39-ijms-20-03201],[@B56-ijms-20-03201]\]. In view of the proposed protective role of the Arabidopsis cytosolic class I sHSP, *HSP17CI*, against diverse stress conditions, the aim of the current study was to systematically investigate the expression patterns of *HSP17.4CI* under a wide range of abiotic, biotic and oxidative stress treatments and to investigate whether the expression of *HSP17.4CI* gene under different stress conditions could be linked to the production of ROS. Our results showed that after application of various abiotic stresses, including heat, cold, salinity, drought and high light, the expression level of *HSP17.4CI* significantly raised compared to controls ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}). The rapid induction of *HSP17.4CI,* by these treatments suggests that this gene may act in a common protection mechanism against the damaging effects of these stresses. In Arabidopsis, the overexpression of the *Medicago sativa MsHSP17.7* increased the tolerance of the transgenic lines to heat shock, high salinity, oxidative and drought stresses \[[@B21-ijms-20-03201]\]; similarly, the overexpression of two rice sHSPs (*OsSHSP1* and *OsSHSP2*) led to higher germination rates compared to wild-type plants under salt treatment \[[@B57-ijms-20-03201]\]. HSPs were reported to be involved in plant defense responses following biotic stress \[[@B47-ijms-20-03201],[@B58-ijms-20-03201],[@B59-ijms-20-03201]\]. Here, we studied the effect of infection with different pathogens on *HSP17.4CI*. Our data show that the inoculation of Arabidopsis plants with *P. syringae* led to an increase of *HSP17.4CI* expression ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}A), suggesting the role of *HSP17.4CI* during defense against this biotrophic pathogen. In contrast, no significant change in the expression level of *HSP17.4CI* was noticed after infection with the necrotrophic pathogens, *A. brassicicola* or *F. oxysporum* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}B,C). Based on the mode of infection of the biotrophic pathogens that feed on living plant cells and the necrotrophic pathogens that feed on dead cells, plants defend themselves against biotrophic pathogens by rapid induction of an oxidative burst (excessive production of ROS) that leads to hypersensitive response and programmed cell death, while in the case of necrotrophic pathogens, the infected plants respond by avoiding the production of ROS \[[@B49-ijms-20-03201]\]. Thus, the reported upregulation of *HSP17.4CI* by biotrophic but not necrotrophic pathogens suggests that the expression of this sHSP may be at least partially mediated by ROS. Phytohormones such as ABA, SA, and MJ orchestrate plant responses to different abiotic and biotic environmental stresses \[[@B60-ijms-20-03201]\]. We found that Arabidopsis seedlings treated with ABA have enhanced relative transcription levels of *HSP17.4CI* compared with the untreated control ([Figure 3](#ijms-20-03201-f003){ref-type="fig"}A). Considering the fact that ABA mediates plant responses to various abiotic stresses, especially drought stress, the induction of *HSP17.4CI* expression by ABA is consistent with the induction of expression observed after drought stress treatment ([Figure 1](#ijms-20-03201-f001){ref-type="fig"}D). This suggests that the upregulation of expression of this sHSP by drought could be mediated by ABA signaling, which could integrate a cross-talk with ROS \[[@B61-ijms-20-03201],[@B62-ijms-20-03201]\]. Furthermore, we found that the expression of *HSP17.4CI* rapidly increased after treatment with SA ([Figure 3](#ijms-20-03201-f003){ref-type="fig"}B). These results are in line with the induction of expression of *HSP17.4CI* by infection with the biotrophic pathogen *P. syringae* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}A), which activates SA signaling \[[@B49-ijms-20-03201]\]. MJ treatment showed no effect on the expression level of *HSP17.4CI* in Arabidopsis seedlings ([Figure 3](#ijms-20-03201-f003){ref-type="fig"}C). This result is consistent with the finding that the expression level of *HSP17.4CI* gene was not affected by *A. brassicicola* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}B) and *F. oxysporum* ([Figure 2](#ijms-20-03201-f002){ref-type="fig"}C), where MJ signaling pathway is known to be involved in plant defense against these pathogens \[[@B49-ijms-20-03201]\]. The modulation of expression of *HSP17.4CI* observed after treatment with phytohormones, and abiotic and biotic stresses suggests that environmental stresses might influence the expression of *HSP17.4CI* through a cross-talk between hormonal and ROS signaling. Indeed, almost all environmental stresses enhance ROS production \[[@B63-ijms-20-03201],[@B64-ijms-20-03201]\]. The participation of ROS generated during the applied abiotic and biotic stresses in the induction of *HSP17.4CI* ([Figure 1](#ijms-20-03201-f001){ref-type="fig"} and [Figure 2](#ijms-20-03201-f002){ref-type="fig"}) is supported by the finding that treatment with cellular oxidants, such as H~2~O~2~ and methyl viologen, also enhanced the expression level of *HSP17.4CI* ([Figure 4](#ijms-20-03201-f004){ref-type="fig"}). These results are in line with the knowledge that expression of various plant sHSPs is increased in response to externally applied oxidative stress \[[@B16-ijms-20-03201],[@B35-ijms-20-03201]\]. Moreover, our results showed that expression of *HSP17.4CI* is also enhanced in catalase loss-of-function plants which accumulate metabolically-produced H~2~O~2~ after their transfer from non-photorespiratory to photorespiratory conditions ([Figure 5](#ijms-20-03201-f005){ref-type="fig"}). Reviewing the expression profiles of sHSPs in Arabidopsis genome indicated that only one cytosolic sHSP (*HSP17.6C*) showed identical expression pattern to *HSP17.4CI* under different stress treatments ([Table 1](#ijms-20-03201-t001){ref-type="table"}). Both genes were reported to be induced by heat, drought, salinity and their combination \[[@B42-ijms-20-03201]\], as well as by oxidative stresses imposed by H~2~O~2~ \[[@B52-ijms-20-03201]\] or methyl viologen \[[@B16-ijms-20-03201]\]. Another three sHSP genes (*HSP17.6A*, *HSP17.6* and *HSP23.5*) showed relatively similar expression behaviors under abiotic and oxidative stress treatments ([Table 1](#ijms-20-03201-t001){ref-type="table"}). Genes that are regulated by oxidative stress share regulatory elements in their promotor sequences that enable concerted changes in gene expression and downstream signaling events \[[@B10-ijms-20-03201]\]. A small number of newly found motifs in promotor sequences is associated with upregulation of the respective genes upon oxidative stress \[[@B11-ijms-20-03201],[@B16-ijms-20-03201]\]. Two of these motifs, Motif 1 (\[A/T\]\[A/T\]TGGGCCT\[T/A\]AA) and Motif 5 (GAA\[A/C\]\[G/C\]TT\[C/G\]\[C/A\]AGA), are present in the promoter sequence of *HSP17.4CI* ([Figure 6](#ijms-20-03201-f006){ref-type="fig"}). Motif 5 is present in a direct tandem repeat, 22 nucleotide upstream of the 5' untranslated region (UTR) of *HSP17.4CI*. Motif 1 was found to be more upstream, 219 nucleotide from the start of the 5'UTR. Two members of the heat shock transcription factor family, HSFA2 and HSFA4A, are induced by oxidative stress and suggested to be involved in H~2~O~2~ sensing \[[@B10-ijms-20-03201],[@B11-ijms-20-03201]\]. Interestingly, the finding that Motif 1 is shared by HSFA2 suggest that *HSP17.4CI* may be involved in an oxidative stress signaling pathway that uses a specific subset of HSFs and HSPs to integrate oxidative stress in the flexible gene network that controls the plant's response to various environmental stress conditions. In conclusion, the rapid and strong expression of *HSP17.4CI* under a myriad of stress conditions that involve ROS production, suggests that this gene might play an essential role during oxidative stress mitigation. We propose that *HSP17.4CI* can be used as molecular marker for oxidative stress in Arabidopsis and that its orthologs, upon further future studies, could represent candidate target genes for engineering stress tolerant crop plants. 4. Materials and Methods {#sec4-ijms-20-03201} ======================== 4.1. Plant Materials and Growth Conditions {#sec4dot1-ijms-20-03201} ------------------------------------------ *Arabidopsis thaliana* ecotype Columbia (Col-0, wild type) was used in all experiments. Seeds were stratified at 4 °C for 2 days and then transferred to the growth cabinets. For plate-grown Arabidopsis plants, seeds were surface-sterilized and sown on 1x MS (Murashige and Skoog) plates \[[@B65-ijms-20-03201]\]. Plants were grown in cabinets at 24 °C and 15 h photoperiod at 150 µmol photons.m^−2^.s^−1^. Fourteen-day-old Seedlings were exposed for treatments. For soil-grown plants, Arabidopsis seeds were sown on soil (University of California mix) and left to grow in a growth chamber at 24 °C and 8 h photoperiod (150 µmol photons.m^−2^.s^−1^). After around 10 days, seedlings were transplanted to new soil. Four-week-old seedlings were exposed to different treatments. For control samples, plants were mock-treated/inoculated. The treatments were started 1 h after the light turned on to allow direct comparisons between treatments. The growth media and the collection time points for each treatment were chosen based on previous work \[e.g., 46, and references therein\]. For experiments done on soil grown plants, the above ground plant parts were used for the subsequent analyses. In all measurements, three biological replicates for each treatment were used. 4.2. Abiotic Stress Treatments {#sec4dot2-ijms-20-03201} ------------------------------ Heat, cold, salt and water stress treatments were carried out on 14-day-old MS plate-grown seedlings. For heat treatment, plates were moved to an incubator at 45 °C (light intensity was 75 µmol photons.m^−2^.s^−1^). Cold treatment was conducted by incubating the plates on ice in a cold room (temperature was about 0 °C) for the indicated time points. Salt stress was applied by transferring the seedlings from the MS medium to Petri dishes containing water. After 1 h recovery, a NaCl solution was added to make a final concentration of 350 mM. Control samples were left without salt. For drought treatment, seedlings were moved from the MS plates to dry filter paper, whereas control seedlings were moved to distilled-water-wetted filter paper. High light stress was imposed by subjecting 4 week-old soil-grown plants to 800 µmol photons.m^−2^.s^−1^. 4.3. Biotic Stress Treatments {#sec4dot3-ijms-20-03201} ----------------------------- All pathogen treatments were performed on 4 week-old soil-grown plants. The pathogens used were *Pseudomonas syringae, Alternaria brassicicola,* and *Fusarium oxysporum.* For the details of *Pseudomonas* inoculations see \[[@B46-ijms-20-03201]\]. *Alternaria* inoculations were carried out as follows: *A. brassicicola* (isolate UQ4273) was grown on agar plates containing V8 vegetable juice (Campbell Soup Co., Camden, NJ). Drops of 5 µL size of a spore suspension (5x10^5^ spores/mL in water) were pipetted onto three to four leaves per plant (one drop per leaf). The plants were then placed in a 20 L-container with a clear polystyrene dome and kept at high humidity. Mock-inoculated control plants were treated the same way with water instead \[[@B66-ijms-20-03201]\]. *Fusarium* inoculations were carried out as follows: *F. oxysporum* was grown on ½ PDB (Potato Dextrose Broth) agar plates at 28°C for one week in the dark. An agar plug was removed from the plate and added to ½ PDB (Potato Dextrose Broth) liquid medium. The inoculum was kept in a shaker for 2 d at 28 °C before draining the culture with a sieve. Using a hemocytometer, the spore number was determined, and then the culture was diluted with distilled water to 10^6^ spores/mL. After carefully removing plants from soil, roots were rinsed with distilled water then dipped in the *Fusarium* spore solution for about 1 min. The inoculated plants were immediately returned to soil and covered with a transparent plastic cover. For the mock-inoculated control roots were dipped in sterilized water instead. 4.4. Plant Hormone Treatments {#sec4dot4-ijms-20-03201} ----------------------------- For all hormone treatments, 4-week-old soil-grown plants were used. Solutions of 400 µM ABA or 4 mM SA in 1% ethanol was used for plant spraying. Control plants were sprayed with 1% ethanol solution. In hormone and mock treatments, the pH was adjusted to 5. MJ treatment was applied by taping a piece of cotton containing 100 µL of a 0.8% MJ solution onto the inside wall of a 20 L-container and wrapped in a transparent plastic bag. For mock treatments a piece of cotton containing 100 µL ethanol was used \[[@B66-ijms-20-03201]\]. 4.5. Oxidative Stress Treatments {#sec4dot5-ijms-20-03201} -------------------------------- Four-week-old soil-grown plants were sprayed with freshly-prepared solutions of H~2~O~2~ (500 mM) or methyl viologen (30 µM), a herbicide that produces superoxide radicals which are converted into H~2~O~2~ by superoxide dismutase in chloroplasts. Both chemicals were dissolved in water. Control plants were sprayed with water. 4.6. Growth Conditions of Cat2-2 Plants {#sec4dot6-ijms-20-03201} --------------------------------------- Arabidopsis wild type and *cat2-2* \[[@B6-ijms-20-03201]\] seeds were sown on MS plates as described above. After two days at 0 °C, the plates were transferred to a growth cabinet with high CO~2~ level (3000 ppm), at a light intensity of 80 µmol photons.m^−2^.s^−1^, long day conditions (16 h light/8 h dark) and 22 °C day/18 °C night temperatures. After two weeks of growth under the above-mentioned non-photorespiratory conditions, we induced the metabolic formation of H~2~O~2~ by transferring the plates 2 hours after the light switch on to a growth chamber with normal air (CO~2~ \~380 ppm), while keeping all other parameters unchanged. After 8 hours under photorespiratory conditions plants were collected in liquid N~2~ and stored at −80 °C for subsequent analysis. 4.7. H~2~O~2~ Quantification {#sec4dot7-ijms-20-03201} ---------------------------- The optimized Amplex Red-based quantitation method by \[[@B67-ijms-20-03201]\] was applied for H~2~O~2~ estimation. Harvested seedlings (50 mg FW) were ground in liquid N~2~ and re-suspended in 250 µL 50 mM sodium phosphate buffer. The re-suspended tissue was vortexed for 10 s and shaken continuously at room temperature for 30 min. The samples were centrifuged for 5 min at 12,000 rpm at room temperature, and the supernatant was transferred to new tubes. The collected supernatant was re-centrifuged for additional 2 min. The supernatant was transferred to new tubes and kept on ice until analysis (on same day). H~2~O~2~ quantification was conducted according to the manufacturer protocol of Amplex^®^ Red Hydrogen Peroxide/Peroxidase Assay Kit. 4.8. RNA Isolation and cDNA Synthesis {#sec4dot8-ijms-20-03201} ------------------------------------- After the indicated times for each treatment, the whole rosettes were collected in liquid nitrogen and stored at −80 °C. After grinding in liquid nitrogen, 70 mg plant tissue was used for RNA extraction using the Promega Kit (SV Total RNA Isolation System, spin protocol). The integrity of RNA was tested by gel electrophoresis and quantified by NanoDrop spectrophotometer (ND-1000). For cDNA synthesis, the same amount of RNA (from 1000 to 2000 ng) was used and reverse transcriptase system (SuperScript III, Invitrogen) was used. For primer design, the Primer Express 2.0 software (Applied Biosystems) was used. The primers were selected to amplify segments of cDNA of about 100--150 bp and close to the 3′ prime end of the gene. The primer pair used for *HSP17.4CI* (At3g46230) was as follows; forward primer: TCATGAGGAGGTTTCGGTTGC; reverse primer: CTCTCCTGAACTTTCGGCACC. As an internal control, the following *ACTIN* primer mixture was used: Forward universal *ACTIN* AGTGGTCGTACAACCGGTATTGT, *ACTIN2* (At3g18780) reverse GATGGCATGAGGAAGAGAGAAAC, *ACTIN7* (At5g09810) reverse GAGGAAGAGCATTCCCCTCGTA, and *ACTIN8* (At1g49240) reverse GAGGATAGCATGTGGAACTGAGAA. 4.9. Real-time RT-PCR {#sec4dot9-ijms-20-03201} --------------------- For quantitative RT-PCR analysis, 6 µL-reactions consisted of 3µL SYBR Green master mix reagent (Applied Biosystems), 2 µL primer mixture (forward and reverse, 1 µM each) and 1 µl cDNA (containing the equivalent of 10 ng RNA) were used. Quantitative RT-PCR was performed with an ABI PRISM 7900 Sequence Detection System (SDS) (Applied Biosystems) using SYBR Green to monitor the real-time synthesis of double-stranded DNA. The following thermal cycles were applied; Stage 1: 95 °C 10 min, stage 2: 45 cycles of 95 °C 15 s and 60 °C 1 min, stage 3: one cycle of 95°C 2 min, 60°C 15 s, 95°C 15 s. SDS 2.2.2 software (Applied Biosystems) was used to produce amplification plots and dissociation curves of the PCR results. For all amplification plots the Rn (normalized reporter signal) was set to 0.2 to get the threshold cycle (C~t~) values. Linear regression analysis (LinReg PCR software) has been applied to calculate the primer efficiency (E value) of each primer pair. The average of primer efficiency for each primer pair in all samples was calculated after excluding the primer efficiencies with R square less than 0.998. The relative expression of each gene (compared to *ACTIN*) was calculated using the following equation: E~S~\^^(-CtS)^/E~A~\^^(-CtA)^, where E~S~ is the average primer efficiency of the primer pair for this gene, E~A~ is that for *ACTIN*, C~tS~ is the threshold cycle for the studied gene and C~tA~ is that for *ACTIN* \[[@B68-ijms-20-03201]\]. Standard deviation has been calculated from three replicates of each treatment. 4.10. Statistical Analysis {#sec4dot10-ijms-20-03201} -------------------------- Before performing analysis of variance (ANOVA), the data was tested for its normality of distribution and homogeneity of variance, and whether log-transformation was necessary. The significance of variation in the expression data under various treatments was assessed using one-way ANOVA. The significant differences between the means were identified using Tukey´s HSD test at *p* \< 0.05. For *HSP17.4CI* expression in *cat2-2* plants and H~2~O~2~ quantification experiments, the Student's t-test was used to evaluate the difference between genotypes (wild type (WT) and *cat2-2*). All statistical analyses were performed using SPSS 15.0 software 9 (SPSS Inc., **2006**, Chicago, IL, USA). We thank Ebrehem Eid for his help performing the statistical analysis. Supplementary materials can be found at <https://www.mdpi.com/1422-0067/20/13/3201/s1>. ###### Click here for additional data file. Conceptualization, N.S., K.K., V.G.M. and P.M.S.; Formal analysis, N.S. and M.H.; Funding acquisition, V.G.M. and P.M.S.; Investigation, N.S. and K.K.; Methodology, N.S. and M.H.; Resources, K.K., V.G.M. and P.M.S.; Writing -- original draft, N.S.; Writing -- review & editing, K.K., M.H., V.G.M. and P.M.S. This work was funded by the Australian Research Council (DP140103363), grants of the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany´s Excellence Strategy -- EXC 2048/1 -- Project ID: 390686111 and EXC 1028 to VGM, and the Egyptian Ministry of Higher Education. The authors declare no conflict of interest. {#ijms-20-03201-f001} {#ijms-20-03201-f002} {#ijms-20-03201-f003} {#ijms-20-03201-f004} {#ijms-20-03201-f005} {#ijms-20-03201-f006} ijms-20-03201-t001_Table 1 ###### List of Arabidopsis genes annotated as small heat shock proteins (sHSPs) or HSP20-like chaperones and showed expression responsiveness to stress treatments. Gene Locus Gene Name Expression Levels (Log FC) under Stress Treatments Localization ------------ ---------------------- ---------------------------------------------------- -------------- ------ ------- ------ ------ --------------- AT3G46230 *HSP17.4CI* 6.12 1.50 2.00 7.16 3.52 4.28 Cytoplasm AT1G53540 *HSP17.6C* 9.08 1.50 3.14 10.18 4.08 3.07 Cytoplasm AT5G12030 *HSP17.6A* 8.72 \- 2.89 9.21 3.70 5.21 Cytoplasm AT5G12020 *HSP17.6* 7.53 \- 1.97 8.22 4.89 4.04 Cytoplasm AT5G51440 *HSP23.5* 6.34 \- 1.50 6.73 3.64 2.02 Mitochondrion AT2G29500 *HSP17.6B* 5.76 \- \- 7.23 4.23 4.48 Cytoplasm AT1G59860 *HSP17.6A* 4.52 \- \- 5.06 4.73 4.63 Cytoplasm AT1G54050 *HSP17.4B* 2.44 \- \- 3.03 3.64 4.08 Cytoplasm AT5G37670 *HSP15.7* 2.71 \- \- 2.95 2.85 1.85 Peroxisome AT4G10250 *HSP22.0* 8.33 \- \- 10.17 1.53 \- ER AT4G27670 *HSP21* 8.02 \- \- 10.04 1.61 \- Chloroplast AT4G25200 *HSP23.6* 11.23 \- 2.73 12.12 \- \- Mitochondrion AT1G07400 *HSP17.8* 5.96 \- 1.50 6.91 \- \- Cytoplasm AT2G19310 *HSP18.5* 3.47 \- \- 3.37 \- \- Cytoplasm AT1G52560 *HSP26.5* 2.40 \- \- 5.12 \- \- Mitochondrion AT5G59720 *HSP18.2* 1.99 \- \- 3.98 \- \- Cytoplasm AT4G16550 HSP20-like chaperone 1.60 2.17 1.50 1.82 \- \- Unknown AT4G21870 HSP20-like chaperone \- \- \- −2.32 \- \- Cytoplasm AT1G76770 HSP20-like chaperone −1.79 \- \- \- \- \- Cytoplasm Data for heat (35 °C, 4 h), drought (imposed by mannitol, 200 mM, 16 h), salt (NaCl, 150 mM, 16 h) and multiple (a combination of the previous three stresses), was adopted from our previous work \[[@B42-ijms-20-03201]\]. For H~2~O~2~ (20 mM, 1 h) and MV (methyl viologen, 50 µM, 2 h), data was quoted from \[[@B52-ijms-20-03201]\] and \[[@B16-ijms-20-03201]\], respectively. The gene names and their subcellular locations were quoted from Uniport \[[@B53-ijms-20-03201]\] and The Arabidopsis Information Resource (TAIR) \[[@B54-ijms-20-03201]\] websites. Expression values with Log FC (fold change) of 1.5 or more compared to controls were considered. All shown values were significantly different at *p* \< 0.05. The hyphen "- "means not responsive. ER: endoplasmic reticulum. Genes on the list were ordered where genes induced by both abiotic and oxidative stresses come first, then by the level of induction by heat, with the investigated gene, *HSP17.4CI*, on the top.

{#sp1 .18}

**Funding information** Engineering and Physical Sciences Research Council, Grant/Award Number: EP/R013764/1 1. INTRODUCTION {#amp210060-sec-0001} =============== Vaccines are considered the most effective form of healthcare intervention^\[^ [^1^](#amp210060-bib-0001){ref-type="ref"}, [^2^](#amp210060-bib-0002){ref-type="ref"} ^\]^ and offer the promise of overcoming the current Covid‐19 pandemic caused by the SARS‐CoV‐2 virus. One of the most promising technologies for the development of an effective SARS‐CoV‐2 vaccine are considered to be RNA and viral vector‐based platforms.^\[^ [^3^](#amp210060-bib-0003){ref-type="ref"}, [^4^](#amp210060-bib-0004){ref-type="ref"}, [^5^](#amp210060-bib-0005){ref-type="ref"} ^\]^ The RNA vaccine platform uses the natural cellular protein expression pathway, based on the central dogma of molecular biology, in which genetic information encoded in DNA is transcribed into messenger RNA (mRNA) and translated into protein. This way, RNA vaccinology works by outsourcing the production of the vaccine protein antigen to the cells of the human body, based on the information in the RNA sequence.^\[^ [^6^](#amp210060-bib-0006){ref-type="ref"}, [^7^](#amp210060-bib-0007){ref-type="ref"}, [^8^](#amp210060-bib-0008){ref-type="ref"}, [^9^](#amp210060-bib-0009){ref-type="ref"} ^\]^ For this, the RNA vaccine is commonly injected into the muscle using predominantly a liposome‐based formulation, known as lipid nanoparticle, or a polycation‐based formulation.^\[^ [^7^](#amp210060-bib-0007){ref-type="ref"}, [^9^](#amp210060-bib-0009){ref-type="ref"} ^\]^ Once inside the cells, the ribosomes produce the protein encoded by the RNA sequence, which for the Covid‐19 vaccine is the spike protein on the SARS‐CoV‐2 virus surface. The produced protein antigen (expressed as spike protein trimer) then induces the immune response likely required to gain immunity against the virus. There are two main types of RNA vaccines, mRNA^\[^ [^7^](#amp210060-bib-0007){ref-type="ref"} ^\]^ and self‐amplifying RNA (saRNA) vaccines.^\[^ [^8^](#amp210060-bib-0008){ref-type="ref"}, [^10^](#amp210060-bib-0010){ref-type="ref"} ^\]^ As their name implies, saRNA vaccines, replicate inside cells by encoding a viral RNA replication machinery in the saRNA strand and expressing this inside the human cells.^\[^ [^8^](#amp210060-bib-0008){ref-type="ref"}, [^10^](#amp210060-bib-0010){ref-type="ref"} ^\]^ This way, lower amounts of RNA are required per vaccine dose, potentially providing substantial cost benefits and higher productivity, in terms of doses per liter of bioreaction, compared to non‐replicating mRNA vaccines.^\[^ [^10^](#amp210060-bib-0010){ref-type="ref"}, [^11^](#amp210060-bib-0011){ref-type="ref"} ^\]^ On the other hand, mRNA vaccines are clinically more developed and widely tested compared to saRNA vaccines. Viral vector‐based vaccines, such as Adenovirus vector vaccines, also utilize the cells of the human body to synthesize the target antigen, however they deliver a DNA payload,^\[^ [^12^](#amp210060-bib-0012){ref-type="ref"} ^\]^ which is first transcribed into an mRNA and then translated into the spike protein in the case of the Covid‐19 vaccine. With clinical trials of RNA vaccines currently ongoing, herein we conduct a techno‐economic analysis of RNA vaccine manufacturing and present the advantages of this platform with respect to development speed, manufacturing footprint and vaccine cost. 2. ASSESSMENT OF THE RNA VACCINE PLATFORM {#amp210060-sec-0002} ========================================= 2.1. Development timeline {#amp210060-sec-0003} ------------------------- Conventional vaccine development takes on average 8 to 14 years and costs 0.55 to 1 billion USD,^\[^ [^13^](#amp210060-bib-0013){ref-type="ref"}, [^14^](#amp210060-bib-0014){ref-type="ref"}, [^15^](#amp210060-bib-0015){ref-type="ref"}, [^16^](#amp210060-bib-0016){ref-type="ref"}, [^17^](#amp210060-bib-0017){ref-type="ref"}, [^18^](#amp210060-bib-0018){ref-type="ref"}, [^19^](#amp210060-bib-0019){ref-type="ref"}, [^20^](#amp210060-bib-0020){ref-type="ref"}, [^21^](#amp210060-bib-0021){ref-type="ref"}, [^22^](#amp210060-bib-0022){ref-type="ref"} ^\]^ as illustrated in Figure [1](#amp210060-fig-0001){ref-type="fig"}. Fast‐tracked regulatory processes implemented for emergency response to pandemics can cut the duration of pre‐clinical and clinical development to 0.8 to 1.5 years if patient recruitment and testing can be carried out rapidly.^\[^ [^21^](#amp210060-bib-0021){ref-type="ref"}, [^23^](#amp210060-bib-0023){ref-type="ref"} ^\]^ Emerging platform technologies, such as the RNA platform, promote pre‐clinical development at unprecedented speeds.^\[^ [^24^](#amp210060-bib-0024){ref-type="ref"}, [^25^](#amp210060-bib-0025){ref-type="ref"} ^\]^ For example, the Shattock group at Imperial College London generated a prototype saRNA vaccine candidate 2 weeks after the selection of the genetic sequence of the spike protein from the SARS‐CoV‐2 virus,^\[^ [^3^](#amp210060-bib-0003){ref-type="ref"}, [^26^](#amp210060-bib-0026){ref-type="ref"} ^\]^ and US company Moderna, Inc. went from genetic sequence information of the clinical material manufacturing to human testing in only 42 days.^\[^ [^22^](#amp210060-bib-0022){ref-type="ref"} ^\]^ These speeds to clinical material production provide huge advantage to clinical investigation of multiple vaccine candidates in face of a pandemic. During clinical development, the highest costs and longest durations are encountered in phase III clinical trials and the highest failure rates tend to occur in phase II clinical trials where the efficacy of the vaccines is assessed,^\[^ [^14^](#amp210060-bib-0014){ref-type="ref"}, [^20^](#amp210060-bib-0020){ref-type="ref"}, [^27^](#amp210060-bib-0027){ref-type="ref"}, [^28^](#amp210060-bib-0028){ref-type="ref"} ^\]^ cf. Figure [1A](#amp210060-fig-0001){ref-type="fig"}. {ref-type="ref"}, [^14^](#amp210060-bib-0014){ref-type="ref"}, [^15^](#amp210060-bib-0015){ref-type="ref"}, [^16^](#amp210060-bib-0016){ref-type="ref"}, [^17^](#amp210060-bib-0017){ref-type="ref"}, [^18^](#amp210060-bib-0018){ref-type="ref"}, [^19^](#amp210060-bib-0019){ref-type="ref"}, [^20^](#amp210060-bib-0020){ref-type="ref"}, [^21^](#amp210060-bib-0021){ref-type="ref"}, [^22^](#amp210060-bib-0022){ref-type="ref"} ^\]^ Once the platform is developed and used to produce a licensed product, the costs and failure rates for developing further products using the same platform technology would drop substantially. The highest costs and longest development timelines are encountered in phase III clinical trials and the highest failure rates tend to occur in phase II clinical trials. B, Process development and facility construction timelines and cost estimates for conventional and emerging platform technologies with drug substance annual production capacities of tens to hundreds of million doses.^\[^ [^14^](#amp210060-bib-0014){ref-type="ref"}, [^19^](#amp210060-bib-0019){ref-type="ref"}, [^40^](#amp210060-bib-0040){ref-type="ref"}, [^41^](#amp210060-bib-0041){ref-type="ref"}, [^42^](#amp210060-bib-0042){ref-type="ref"}, [^59^](#amp210060-bib-0059){ref-type="ref"} ^\]^ Process development and facility design is usually initiated during pre‐clinical and clinical testing and investments are usually made as failure risks reduce during clinical development. C, Comparison of overall vaccine production rates for conventional and new platform technologies, considering the development and testing phases presented in parts A and B above. Once fully developed and validated, the new vaccine platform technologies will produce vaccines within weeks to months after antigen identification, which is at least 10‐fold faster than conventional technologies](AMP2-2-e10060-g001){#amp210060-fig-0001} Once the cGMP platform production process for RNA vaccines is developed, RNA vaccines can be produced substantially faster compared to conventional expression systems. For example, in the case of inactivated or live‐attenuated viral (such as the PiCoVacc SARS‐CoV‐2 virus vaccine candidate being produced in Vero Cells), or recombinant protein vaccine candidate production, product‐specific manufacturing processes have to be developed and ideally optimized, validated and approved for cGMP production, which can take a substantial amount of time. Additionally, the agility of the RNA platform, which is agnostic to the disease target, means that multiple iterations and vaccine variants can be rapidly produced and tested without the need for process modification or re‐validation. The high productivity of the RNA platform as expressed per unit volume of process and per unit time is also considerably higher than the aforementioned conventional expression systems. This makes the production of higher volumes required for later phase trials and for large scale production substantially easier. The time and resource gains to be made with the RNA platform are expected to become more apparent in the future when this platform is fully developed and RNA vaccines gain regulatory approval. 2.2. Manufacturing process and footprint {#amp210060-sec-0004} ---------------------------------------- Following successful clinical trials and demonstration of efficacy, the next challenge becomes the manufacturing of the vaccine at the quality standards, scale and rate required for meeting global demand. This is particularly cumbersome in the case of pandemic‐response manufacturing, when several billion doses of vaccines need to be manufactured within months under current Good Manufacturing Practices (cGMP), and ideally at low enough cost to allow affordability and mass immunization globally. Small‐scale cGMP compliant RNA vaccine production processes have been already developed and GMP grade RNA vaccine candidates have already been produced for clinical trials.^\[^ [^29^](#amp210060-bib-0029){ref-type="ref"}, [^30^](#amp210060-bib-0030){ref-type="ref"}, [^31^](#amp210060-bib-0031){ref-type="ref"}, [^32^](#amp210060-bib-0032){ref-type="ref"}, [^33^](#amp210060-bib-0033){ref-type="ref"} ^\]^ In the light of the current COVID‐19 pandemic, several companies and consortia are scaling up RNA vaccine production to the billion dose annual production scale.^\[^ [^34^](#amp210060-bib-0034){ref-type="ref"}, [^35^](#amp210060-bib-0035){ref-type="ref"}, [^36^](#amp210060-bib-0036){ref-type="ref"}, [^37^](#amp210060-bib-0037){ref-type="ref"}, [^38^](#amp210060-bib-0038){ref-type="ref"}, [^39^](#amp210060-bib-0039){ref-type="ref"} ^\]^ For rapid response manufacturing, optimal utilization of existing facilities is crucial, however if the demand cannot be met, the construction of additional facilities will be urgently required. Crucially, manufacturing a new vaccine of conventional format (eg, inactivated viral vaccines) against a new disease, such as Covid‐19, by re‐deploying existing large‐scale facilities would also adversely affect the supply of other medicines. With the construction of such conventional facilities requiring around several years and hundreds of million US dollars^\[^ [^14^](#amp210060-bib-0014){ref-type="ref"}, [^19^](#amp210060-bib-0019){ref-type="ref"}, [^40^](#amp210060-bib-0040){ref-type="ref"}, [^41^](#amp210060-bib-0041){ref-type="ref"}, [^42^](#amp210060-bib-0042){ref-type="ref"}, [^43^](#amp210060-bib-0043){ref-type="ref"}, [^44^](#amp210060-bib-0044){ref-type="ref"} ^\]^ (Figure [1B](#amp210060-fig-0001){ref-type="fig"} **)**, certain emerging platform technologies present considerable advantages for rapid‐response manufacturing.^\[^ [^24^](#amp210060-bib-0024){ref-type="ref"} ^\]^ We have built a production process model for RNA vaccine manufacturing in SuperPro Designer (Intelligen, Inc.) based on state‐of‐the‐art processes, implemented in industry for RNA synthesis utilizing DNA‐templated enzymatic RNA synthesis via the in vitro transcription reaction catalysed by the T7 RNA polymerase enzyme,^\[^ [^7^](#amp210060-bib-0007){ref-type="ref"}, [^45^](#amp210060-bib-0045){ref-type="ref"}, [^46^](#amp210060-bib-0046){ref-type="ref"} ^\]^ cf. the process diagram in Figure [2](#amp210060-fig-0002){ref-type="fig"}. The saRNA vaccine drug substance encoding for the spike protein from the SARS‐CoV‐2 has a size of around 10 k bases (kb) and a molecular mass of around 5 Mega Dalton (MDa), which is much larger when compared to the mRNA vaccine drug substance of the same spike‐encoding protein that has a size of around 3 to 4 kb, corresponding to 1.5 to 2 MDa.^\[^ [^5^](#amp210060-bib-0005){ref-type="ref"}, [^47^](#amp210060-bib-0047){ref-type="ref"} ^\]^ The molecular masses of saRNA and mRNA are an order of magnitude larger than the T7 RNA polymerase enzyme and therefore size‐based separation of the T7 RNA polymerase from the RNA seems feasible. After RNA synthesis, downstream purification can be achieved via a series of tangential flow filtration (TFF) steps, and purities of 90% to 99.9% and yields of 90% to 95% have been reported.^\[^ [^48^](#amp210060-bib-0048){ref-type="ref"} ^\]^ TFF can also be complemented by chromatographic purification techniques, such as hydroxyapatite chromatography, oligo(dT) chromatography, ion exchange chromatography and core bead flow‐through chromatography (eg, Capto Core 700 beads from Cytiva, Danaher Corporation, formerly GE Healthcare Life Sciences).^\[^ [^48^](#amp210060-bib-0048){ref-type="ref"}, [^49^](#amp210060-bib-0049){ref-type="ref"} ^\]^ The produced RNA drug substance is then formulated into lipid nanoparticles^\[^ [^3^](#amp210060-bib-0003){ref-type="ref"}, [^9^](#amp210060-bib-0009){ref-type="ref"}, [^50^](#amp210060-bib-0050){ref-type="ref"} ^\]^ to complete the production of the LNP‐encapsulated/formulated RNA. The overall LNP‐encapsulation process and formulation is being independent of the RNA sequence and, thus the targeted disease vaccine drug product. Once the formulated RNA is ready, it enters the fill‐to‐finish process where in the required dose (plus overage) are filled into glass vials to produce the final vaccine drug product. {ref-type="ref"}, [^46^](#amp210060-bib-0046){ref-type="ref"}, [^48^](#amp210060-bib-0048){ref-type="ref"}, [^49^](#amp210060-bib-0049){ref-type="ref"}, [^60^](#amp210060-bib-0060){ref-type="ref"} ^\]^](AMP2-2-e10060-g002){#amp210060-fig-0002} Based on our techno‐economic assessment, the RNA vaccine production process can be two to three orders of magnitude smaller than conventional vaccine production processes in terms of facility scale, and can be constructed in less than half the time with 1/20 to 1/35 of the upfront capital investment, as shown in Figure [1B](#amp210060-fig-0001){ref-type="fig"}. It therefore presents a strong advantage of requiring small‐scale, high‐capacity facilities, which can be constructed more rapidly and could make wide use of single‐use disposable equipment. Due to its small scale, the RNA vaccine drug substance production process could be placed in a small part of an existing conventional vaccine facility, for example in a room, and still produce more doses worth of drug substance than the entire original conventional vaccine production facility. To rapidly establish such an RNA vaccine drug substance production line, off‐the‐shelf single‐use equipment can be used to build the entire process. Once such a process is established and validated based on readily available single‐use equipment, the technology can be transferred to other facilities for scaling out purposes, thereby reducing process and quality control design and development timelines and streamlining validation and start‐up activities. Moreover, the RNA vaccine platform technology offers the flexibility of producing a very large range of vaccine products using the same production process, quality control system and facility, rapidly and at high capacity. Therefore, the production of new vaccines can be achieved around 10× faster compared to conventional vaccine production technologies, as shown in Figure [1C](#amp210060-fig-0001){ref-type="fig"}. In such a scenario, the cost of an RNA vaccine drug substance manufacturing facility, besides scale also depends on the grade of the clean rooms or required containment level. If the RNA production can take place in a closed system,^\[^ [^48^](#amp210060-bib-0048){ref-type="ref"} ^\]^ then lower grade facilities and rooms can even be used, which would cost substantially less to construct, operate and maintain compared to high grade clean room containing facilities, of course, following the appropriate regulatory and compliance guidelines.^\[^ [^49^](#amp210060-bib-0049){ref-type="ref"} ^\]^ 2.3. Process performance and costs {#amp210060-sec-0005} ---------------------------------- Based on our process‐cost modeling in SuperPro Designer (the model, relevant assumptions and simulation results report are available on GitHub at <https://github.com/ZKis-ZK/RNA-vaccine-drug-substance-production-techno-economic-modelling>), the saRNA platform can produce over 1 billion vaccine doses worth of drug substance per year at a small process scale corresponding to 5 L bioreactor working volume in a correspondingly small facility footprint that would cost around 20 million USD to construct, equip, validate and start up. The annual operating costs are estimated to be over 100 million USD due to the high cost of raw materials involved in RNA synthesis and LNP production. This 100 million USD/year estimate includes material and consumable costs, labor costs, facility‐dependent costs, quality control and quality assurance costs, and waste disposal costs. Out of these, the 5′ cap analogue raw material is the major cost component, accounting for over 50% of the total operating costs. The highest‐efficiency 5′ cap analogues are CleanCap AG and CleanCapAU (TriLink Biotechnologies, Inc.) for mRNA and saRNA vaccines, respectively.^\[^ [^51^](#amp210060-bib-0051){ref-type="ref"} ^\]^ The 5′ cap structure is crucial for avoiding degradation of the RNA by the innate immune mechanisms, which associate non‐capped RNA with foreign (eg, viral) material and ensures protein antigen expression from the RNA polymer.^\[^ [^52^](#amp210060-bib-0052){ref-type="ref"} ^\]^ The production of 5′ cap analogues is currently being scaled up to ensure availability for billion‐dose scale RNA vaccine production.^\[^ [^53^](#amp210060-bib-0053){ref-type="ref"} ^\]^ We therefore do not envisage the availability of 5′ cap analogues to be a bottleneck, particularly for saRNA production, although raw material shortages cannot be ruled out during pandemic‐response mass vaccine manufacture. The amount of RNA drug substance per dose is estimated to be in the range of 0.1 to 10 μg/dose^\[^ [^3^](#amp210060-bib-0003){ref-type="ref"}, [^10^](#amp210060-bib-0010){ref-type="ref"}, [^11^](#amp210060-bib-0011){ref-type="ref"} ^\]^ and 25 to 250 μg/dose^\[^ [^3^](#amp210060-bib-0003){ref-type="ref"}, [^54^](#amp210060-bib-0054){ref-type="ref"} ^\]^ for saRNA and mRNA vaccines, respectively. The actual amount will be determined during clinical trials. Given this difference and range in the amount of RNA drug substance per dose, the price per dose, production amounts and production rates will vary accordingly. This is in line with the original intended purpose of these vaccines: mRNA vaccines were originally developed as anti‐cancer vaccines without prioritizing ultra‐low cost per dose.^\[^ [^5^](#amp210060-bib-0005){ref-type="ref"}, [^6^](#amp210060-bib-0006){ref-type="ref"}, [^10^](#amp210060-bib-0010){ref-type="ref"} ^\]^ On the other hand, saRNA vaccines have been developed for infectious diseases and rapid pandemic response, considering the purchasing power of developing countries and thus aiming to minimize cost per dose.^\[^ [^5^](#amp210060-bib-0005){ref-type="ref"}, [^6^](#amp210060-bib-0006){ref-type="ref"}, [^10^](#amp210060-bib-0010){ref-type="ref"} ^\]^ Based on our cost modeling, a cost per dose of below 1 USD/dose, including options for fill‐to‐finish into multidose vials, appears achievable for saRNA vaccines. This cost could be well above 1 USD/dose for mRNA vaccines. The drug substance cost per dose and productivity, in terms of doses worth of drug substance produced per unit time, has a linear dependence on the drug substance amount per dose. This way, the drug substance cost per dose can decrease and productivity can increase by two orders of magnitude in case of moving from 10 to 0.1 μg/dose for saRNA vaccines. As listed in Table [1](#amp210060-tbl-0001){ref-type="table"}, besides the RNA amount per dose, the cost per dose and production process performance for both mRNA and saRNA vaccines will depend on the process scale, production titre, cost of the 5′ cap analogue or the 5′ capping approach used (co‐transcriptional vs enzymatic post‐transcriptional),^\[^ [^7^](#amp210060-bib-0007){ref-type="ref"} ^\]^ efficiency and cost of downstream purification methods, the facility‐related costs and the possibility of recycling high value materials for producing a subsequent batch of the same product. The time required to produce a batch of RNA drug substance is around 11 hours, if production batches are scheduled such that the end of a batch overlaps with the beginning of the subsequent batch in order to increase the utilization of the production line. Continuous RNA synthesis production processes whereby the RNA product is continuously exiting the bioreactor and the high value raw materials are kept in the bioreactor also offer a substantial material cost reduction potential.^\[^ [^55^](#amp210060-bib-0055){ref-type="ref"} ^\]^ Besides the manufacturing costs, the final sale price of the vaccine is also expected to include R&D costs, costs of clinical trials, marketing and supply chain costs and a profit margin. ###### Parameters influencing RNA vaccine production performance and cost +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Parameter | Range | Unit | Influencing and determining factors | Reference | +:=================================================================+:=================+:============+:==============================================================+:============================================================================================================================================================================+ | RNA amount per dose | 0.1‐10 for saRNA | μg/dose | Clinical trials | [3](#amp210060-bib-0003){ref-type="ref"}, [10](#amp210060-bib-0010){ref-type="ref"}, [11](#amp210060-bib-0011){ref-type="ref"}, [54](#amp210060-bib-0054){ref-type="ref"} α | | | | | | | | | 25‐250 for mRNA | | | | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Process scale | 0.5‐50 | L | Demand, scale‐up optimization | [46](#amp210060-bib-0046){ref-type="ref"}, [49](#amp210060-bib-0049){ref-type="ref"}, [55](#amp210060-bib-0055){ref-type="ref"}, [60](#amp210060-bib-0060){ref-type="ref"} | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Production titres | 1.5‐7 | g/L | Reaction optimization, process development | [46](#amp210060-bib-0046){ref-type="ref"}, [55](#amp210060-bib-0055){ref-type="ref"} | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | 5′ Cap analogue cost | 2500‐10 000 | USD/g | Scale and supplier purchase price | [53](#amp210060-bib-0053){ref-type="ref"} | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Downstream purification losses | 20‐50 | \% | Type of unit operations, process development | [48](#amp210060-bib-0048){ref-type="ref"}, [49](#amp210060-bib-0049){ref-type="ref"}, [60](#amp210060-bib-0060){ref-type="ref"} β | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Raw material recycling[^a^](#amp210060-note-0002){ref-type="fn"} | 0‐8 | Fold | Stability of the materials, regulatory approval | α | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Capital investment costs | 10‐40 | Million USD | Production scale, grade and containment level of the facility | β | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ | Vial or container cost | 0.1‐0.6 | USD/dose | Number of doses per container or vial | [56](#amp210060-bib-0056){ref-type="ref"}, [61](#amp210060-bib-0061){ref-type="ref"} | +------------------------------------------------------------------+------------------+-------------+---------------------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------+ *Note*: α---Assumed by the authors; β---calculated using SuperPro Designer V10 (Intelligen, Inc.). Recycling or re‐use of the materials for producing multiple batches of the same product. For this, high cost raw material (eg, the 5′ cap analogue and enzymes) can be separated from the RNA product using TFF and fed back into the RNA synthesis bioreactor. 3. CHALLENGES AHEAD AND POTENTIAL SOLUTIONS {#amp210060-sec-0006} =========================================== The biggest challenge for saRNA and mRNA vaccines currently is to demonstrate efficacy against target disease in clinical trials, especially the Phase II/III efficacy trials, where the highest proportion of vaccines tend to fail. However, the success rate during clinical trials can be increased and the clinical development timelines can in principle be reduced, because the generic feature of the RNA platform allows for the rapid production and development of many vaccine variants, allowing for rapid iterations during clinical development. Additionally, there are numerous intellectual property challenges to be solved, including the ones related to LNP‐based formulation. However, it seems that there are no technical or scalability bottleneck related to current formulation processes when considering high‐volume pandemic‐response production. On the other hand, formulation raw material availability can become a bottleneck during high‐volume pandemic‐response vaccine production, however, their production is also being scaled up to meet global demand. Pandemic‐induced transportation disruptions can limit the availability of raw materials, impacting specialty materials more substantially due to the lower number of suppliers. To mitigate this, distributed or outsourced manufacturing of raw materials across several continents could offer a solution. The stability of the vaccine can also influence its global distribution and availability. This way, vaccines which are stable and require distribution at −70°C, would limit distribution in low and middle income countries due to the lack of the appropriate cold chain infrastructure, would limit the use of multidose vials, and would increase the cost per dose due to cold chain distribution costs and because of the single dose vial (or low dose number) format. To overcome this issue, vaccine developers are evaluating formulations with higher thermostability or lyophilized formulations. Finally, there is a need for additional manufacturing capacity for pandemic‐response production. However, the RNA vaccine production processes are extremely productive, especially for saRNA vaccines due to lower expected amount per dose. With the annual production of saRNA drug substance for over 1 billion doses using a 5 L bioreactor working volume, the bottleneck is expected to be the fill‐to‐finish process that may not be able to fill billions of vials with formulated RNA, at room temperature, particularly when the formulated RNA may not be stable at that condition for a prolonged period of time. To address this, 200 dose bags are being evaluated by CEPI, which can be filled at a rate of 3 million vaccine doses during an 8 to 10 hours manufacturing work shift.^\[^ [^56^](#amp210060-bib-0056){ref-type="ref"}, [^57^](#amp210060-bib-0057){ref-type="ref"} ^\]^ This will potentially add complexity to the final vaccine administration step in clinic where the pharmacists and medical professionals will need familiarity with the new requirements of this technology. Once the saRNA and mRNA platforms are fully developed, stability issues resolved, manufacturing and operational processes addressed and pandemic‐ready, the regulatory processes could, in principle, be accelerated to rapidly respond to future epidemics and pandemics by applying a Quality by Design (QbD) framework aided by computational modeling, cf. Figure [3](#amp210060-fig-0003){ref-type="fig"}. QbD is a systematic approach to pharmaceutical product development that begins with pre‐defined objectives and integrates product and process understanding, identification of critical quality attributes (CQAs) and critical process parameters (CPPs), product quality risk management and development of robust control strategy to ensure the quality of the product.^\[^ [^58^](#amp210060-bib-0058){ref-type="ref"} ^\]^ For accelerated development, this framework would incorporate bioprocess modeling, soft sensing and advanced process control in order to consistently ensure the quality of the manufactured drug substance and drug product. The RNA platform technologies can be supplemented with the above mentioned QbD framework, which could be applied universally and a priori to assure the quality and speed during development of any RNA vaccine production process, independently of the RNA sequence and disease target. {ref-type="ref"} ^\]^](AMP2-2-e10060-g003){#amp210060-fig-0003} To develop such a QbD framework, the iterative cycle described in Figure [3](#amp210060-fig-0003){ref-type="fig"} can be employed. For this the Quality Target Product Profile (QTPP) can be determined based on the optimal profile of the vaccine to meet patient needs and provide highest level of protection. Then the critical quality attributes (CQAs) of the product and their ranges can be determined based on the QTPP, by using a risk assessment scoring. Next, critical process parameter (CPP) ranges can be defined based on the CQAs and on the understanding of the production process. Next mathematical relations between CPPs and CQAs can be established, thereby obtaining a mathematical model of the vaccine production process. This model can be data‐driven, statistical, mechanistic or hybrid and, in all cases, it can be further calibrated and then validated with experimental data. Mechanistic models or mechanistic components of a hybrid model usually tend to provide more predictive power than data‐driven mathematical descriptions. By running the validated mathematical model, the ranges of CPPs which yield the desired CQAs can be determined and, based on these, the design space can be established. Within the design space, a sub‐space, called the normal operating range (NOR), can be defined that offers the flexibility of modifying and optimizing operating parameters in the GMP production process, rather than "freezing" the cGMP process. The model can be simplified and adapted for advanced automation, based on model predictive control, which uses real‐time data from the production processes. By coupling such a modeling‐aided QbD framework to the saRNA or mRNA platforms, the quality aspects of RNA‐product can be harmonized, and regulatory processes could, in principle, be accelerated. In addition, this modeling‐aided QbD framework can also ensure better management of product quality risks during scale‐up and subsequent manufacturing. During a pandemic, manufacturing and supply chain challenges can occur due to lockdowns, closures of upstream manufacturing facilities (eg, raw material, consumable and single‐use equipment manufacturing), reduction of labor force due to health issues caused by the outbreak, travel and transportation restrictions and due to contamination risks of input materials. These threats can cause more severe sourcing disruptions in case of centralized manufacturing and supply chains, due to the reliance on a small number of key manufacturing facilities and supply chain routes. This can be partially addressed by maintaining adequate stock levels and more appropriately addressed by the implementation of distributed, that is, decentralized, manufacturing and supply chains. This way, the number of facilities and supply routes could increase, therefore reducing the risk of lack of raw materials and consumables at a single location (although we note that the decentralized system creates more complex inbound material supply chains), and ultimately increasing the probability of the sustained vaccine supply. 4. CONCLUSIONS {#amp210060-sec-0007} ============== To address pandemics, such as the Covid‐19, several billion doses of vaccines are needed within months. Emerging outbreak‐response platform technologies, especially the RNA platform, appear capable of addressing this extremely challenging task, due to their high productivity at low manufacturing footprint and its ability to" release" millions of doses through rapid quality control testing. Specifically, a facility with a single 5 L bioreactor working volume can be sufficient to produce an estimated 1 billion vaccine doses per year at a drug product cost of below 1 USD/dose. This further increases the possibility of distributed manufacturing and thus contributing to vaccine supply sustainability. Given that RNA vaccine production processes are two to three orders of magnitude smaller than conventional vaccine production processes, they can be built in less than half the time with 1/20 to 1/35 of the upfront capital costs compared to conventional vaccine production processes. Inclusion of single‐use technologies in RNA vaccine manufacturing can further accelerate this timeline. Once the RNA production platform is established, the overall clinical development process could, in principle, be further accelerated by the utilization of a computational model‐aided QbD platform, which would complement the platform production, independent of the RNA sequence. This would enable the development of a platform process agnostic to the infectious disease target, which can be rapidly deployed to both develop candidate vaccines for testing and large‐scale manufacture. Despite the numerous production and affordability advantages, the RNA platform still remains unproven with no commercial vaccine developed using this "disruptive" technology. As multiple organizations exploit this technology to develop vaccines against SARS‐CoV‐2, the benefits and practical limitations of this technology will be tested, providing lessons for further iteration and improvements. This research is funded by the Department of Health and Social Care using UK Aid funding and is managed by the Engineering and Physical Sciences Research Council (EPSRC, Grant No. EP/R013764/1). The views expressed in this publication are those of the author(s) and not necessarily those of the Department of Health and Social Care.

{#sp1 .38} {#sp2 .39} {#sp3 .40}

Abbreviations ============= CCL5 : C--C chemokine ligand 5 CCR5 : C--C chemokine receptor 5 CRC : colorectal cancer GranB : Granzyme B LM : liver metastasis MSI : microsatellite instability OS : overall survival PD-L1 : programmed death-ligand 1 TAM : tumor-associated macrophages TIL : tumor-infiltrating lymphocytes Introduction {#s0001} ============ Metastatic colorectal cancer (CRC) accounts for more than 55.000 deaths per year in the United States alone.[@cit0001] Around 15% of the patients with liver metastases (LM) can be cured by surgical resection of the metastases.[@cit0002] Immune system interactions with the primary tumor and the metastases have an important influence on the clinical course of patients. Immune cells infiltrating or surrounding colorectal primary tumors[@cit0003] and especially tumor-infiltrating lymphocytes (TILs) represent a major prognostic factor.[@cit0004] On a broader scale, high densities of TILs were shown to be correlated with an improved survival in patients with primary CRC as well as LM.[@cit0006] Nevertheless, tumor heterogeneity in many cancers represents an obstacle in detailed analyses of T cell distributions, whereas colorectal cancer liver metastases (CRC-LM) are rather homogenous with a clear border between liver and tumor tissue allowing more precise analyses of spatial distributions.[@cit0015] Especially, T cells and their relation to tumor tissue are gaining increased attention, as the majority of the T cells can be found in the stromal part of primary CRC without direct contact to tumor epithelium.[@cit0017] Evaluation of LM revealed heterogeneous immune cell infiltration into the metastasis center dependent of histopathological tumor conditions like necrosis or fibrosis. Previous analyses, however, did not reveal a relationship of these lymphocytic infiltrates and the clinical course.[@cit0017] In contrast, in the invasive margin, immune cell infiltrations showed remarkably strong variations between different patients and allowed prediction of responses to chemotherapy.[@cit0013] While the only few T cells in contact with the tumor or within the epithelial compartment are generally regarded as being engaged in direct antitumor activity, more distant T cells might be blocked by immunosuppressive factors and exploited by the tumor.[@cit0018] It is therefore of high interest to analyze differences in this distribution. The spatial distribution of T cells at the invasive margin of colorectal cancer LM has not been investigated in detail. The aim of this study was to explore the pattern of detailed T cell distribution in the invasive margin of CRC-LM patients using automated high-throughput image quantification analyses on whole slide sections samples. This quantitative approach allows the identification of distribution patterns in 10 µm steps within 100 µm distance from the tumor border and represents a high-resolution map of tumor infiltrating cells. By using serial sections, T cell distance analysis not only to the tumor tissue but also to the existing barriers like immunosuppressive cells is possible. Surface tissue areas of 36 patients were analyzed, allowing robust statistical evaluation. Clustering analyses revealed distinct distribution patterns that could be associated with improved overall survival (OS). Results {#s0002} ======= The classification of the 10 µm distance classes within the 100 µm area between the tumor and adjacent liver as outlined in [Fig. 1](#f0001){ref-type="fig"} and the following cell quantification analyses could be performed successfully for all 36 patients. Thereby, bar plots for all distance classes were generated. This revealed distinct patterns that were detectable within each patient. For CD3, T cell quantification representative results of different patients are shown in Fig. S1. Apparently, CD3 T cells are located in all distance classes with maximum densities at different distances. Figure 1.(A) Representative invasive margin of a CRC liver metastasis with annotations (CD3 positive lymphocytes appear dark brown). (B) The invasive margin of liver metastases is analyzed separately, encompassing 100 µm into normal adjacent liver from the tumor epithelium. Shaded areas highlight distinct distance classes of 10 µm in relation to the tumor epithelium. As shown in [Fig. 2](#f0002){ref-type="fig"}, the distribution of CD3 cell densities obviously varied between the different distance classes (Kruskal--Wallis test, *p* = 0.0001). The statistically significant differences between adjacent distance classes are marked by asterisks (post-hoc Mann--Whitney U tests). It was observed that in the 20--30 µm distance class median T cell densities significantly decreased below 100 cells/mm^2^, whereas in the distance classes closer than 20 µm and further away than 30 µm from the tumor epithelium median densities higher than 150 cells/mm^2^ were found. To identify patterns of T cell densities, unsupervised clustering was performed. To avoid possible bias due to different lengths of the invasive margin between different patients, we randomly selected one single field from each patient for further pattern analysis. The clustering analysis revealed distinct groups as shown in [Fig. 3](#f0003){ref-type="fig"} (non-informative distance classes were removed stepwise, informative distance classes shown). The adjacent distance classes 10--20 µm and 20--30 µm clustered, whereas the most proximal distance classes (\< 10 µm) and the classes with the highest distance from the tumor epithelium (90--100 µm) were found to be least related to the 20--30 µm distance classes. Furthermore, clustering of patients according to their T cell densities within single distance classes was observed. Survival analysis comparing patients with high and low CD3 T cell densities within each distance class resulted in a significant prolongation of OS for patients with high CD3 cell numbers in the most proximal distance class (\< 10 µm), while in the 20--30 µm region a significant prolongation of survival for patients with low T cell numbers was observed ([Fig. 4](#f0004){ref-type="fig"}). No significant differences between high and low T cell infiltrates were found in all other distance classes. Figure 2.Boxplots of identified CD3 T cell densities in different distance classes of 36 patients. Statistically significant differences between adjacent distance classes are marked with asterisks (*p* values given for Kruskal--Wallis with post-hoc Mann--Whitney U tests). Figure 3.Heatmap based on unsupervised clustering encompassing randomly selected single fields from all 36 patients. Color code indicates clustering of patients according to their relative T cell densities within single distance classes. Figure 4.Kaplan--Meier plots for estimated overall survival probabilities of CD3 T cell high and CD3 T cell low patient groups within all distance classes. Graphs indicate cumulative survival (*y*-axis) and survival in months (*x*-axis). *p* values for statistically significant differences between groups are shown (Breslow test). Cutoff values are taken from previous works.[@cit0006] To get a more detailed picture of the CD3 cell distribution pattern, we did a similar distribution and survival analysis (only CD8^+^) in representative patient samples for CD8^+^ (n = 10), Granzyme B (n = 6), CD163 (n = 4) and CD20 (n = 4) as shown in [Fig. 5](#f0005){ref-type="fig"} (and Fig. S2). Especially, GranB was found to significantly "decrease" at the distance of 20--30 µm (Wilcoxon signed-rank test, *p* = 0.03) following a significant increase further away from the tumor (*p* = 0.03). In line with this, also CD8^+^ cells slightly but significantly decreased in the 20--30 µm region (*p* = 0.049). CD8^+^ cells, however, were rather evenly distributed among the whole area analyzed with a median cell number of ∼87 cells/mm^2^. Furthermore, survival analysis showed that high CD8^+^ infiltrates in the \>10 µm region significantly elongated OS similar to CD3 ([Fig. 6](#f0006){ref-type="fig"}). Nevertheless, no significant differences between CD8^+^ high and low infiltration were observed in the other distance classes. In contrast to the T cells, CD163 macrophages showed a trend to increase in the 20--30 µm distance class ([Fig. 5](#f0005){ref-type="fig"}). While in the other distance classes CD163 cells were rather evenly distributed with a median of ∼200 cells/mm^2^, the distribution of CD20 cells was found to be more heterogenous with least cells at the tumor border, and slightly increasing densities further away ([Fig. 5](#f0005){ref-type="fig"} and Fig. S2). From all cell types analyzed, GranB and CD20 cells appeared to be the smallest populations present in the tissues with median cell counts of 9 and 16 cells/mm^2^, respectively. In Figs. S3 and 4, representative images of CD3, CD8^+^, GranB, CD163 and CD20 immunostainings are shown. Figure 5.Boxplots of identified CD8^+^, GranB, CD163 and CD20 cell densities in different distance classes. Statistically significant differences between adjacent distance classes are marked with asterisks (Wilcoxon signed-rank test; blue *p* values given for paired *t*-tests; patient numbers for CD8^+^ n = 10, GranB n = 6, CD163 n = 4 and CD20 n = 4). Figure 6.Kaplan--Meier plots for estimated overall survival probabilities of CD8^+^ high and low patient groups within the first three distance classes. Graphs indicate cumulative survival (*y*-axis) and survival in months (*x*-axis). *p* values for statistically significant differences between groups are shown (Breslow test). Cutoff values are taken from previous works.[@cit0006] In the course of this analysis, we further wanted to understand whether CD163 cells achieve immunosuppressive effects against T cells. We hypothesized that both cell types might be in close contact with each other and therefore analyzed the localization of T cells in relation to CD163 macrophages using serial section analysis ([Fig. 7](#f0007){ref-type="fig"}). We identified that 52--81% (median 68%) of CD3 lymphocytes within the tumor microenvironment were in close contact (\< 10 µm) with CD163 macrophages. In addition, immunofluorescent double staining of CD163 and programmed death-ligand 1 (PD-L1) showed that the majority of CD163 macrophages were PD-L1 positive ([Fig. 8](#f0008){ref-type="fig"}). Figure 7.Processing of serial section images for the quantification of distances between CD3 T cells and CD163 macrophages. Figure 8.Immunofluorescent double staining for CD163 (red) and PD-L1 (green) at the invasive margin. Dashed lines separate the tumor (top left) from the adjacent liver. DAPI was used as counterstain. Discussion {#s0003} ========== Image analysis is a powerful tool to analyze spatial distributions as well as patterns of cells in the tumor microenvironment[@cit0025] and allows in combination with virtual microscopy the automated analysis of large tissue areas. Among the inflammatory cells in the composition of the tumor stroma and epithelium, T cells play a crucial role in regulating tumor growth and progression. It was shown that the invasive margin and its immune cell composition are predictive for the clinical course of cancer,[@cit0006] and that immunophenotypes and distribution of those can change under therapy.[@cit0028] Whereas the sheer quantity of T cells is associated with clinical outcome,[@cit0006] it has not yet been established whether there are any significant differences in the spatial distribution of T cells at the invasive margin. For this reason, we have investigated the detailed localization of CD3ε+ T cells in the invasive margin of colorectal cancer LM. CD3ε is a pan-T cell marker and shows great heterogeneity in distribution intratumorally and at the invasive margin.[@cit0017] The interface between metastatic liver lesions and adjacent liver allows the precise evaluation of T cell localization, while in the primary tumor the heterogeneity is much more pronounced and is therefore difficult for this analysis.[@cit0015] The analyzed regions within 100 µm distance from the tumor show a significant "drop" of T cell density in the region of 20--30 µm from tumor epithelium, whereas higher T cell densities are found in close proximity and more far away from the tumor as schematically shown in [Fig. 9](#f0009){ref-type="fig"}. Similar examination of CD8^+^ cells revealed that this CD3 T cells subset is likewise affected by showing a slight but significant "drop" at 20--30 µm from the tumor. The ratio between CD8^+^ and CD4^+^ cells varied across distance classes. In general, the CD8^+^ cell numbers observed among the complete analyzed regions from 0 to 100 µm were ∼45% (87 cells/mm^2^) of the CD3 cell numbers (195 cells/mm^2^) leading to the assumption that the other major CD3 cell subtype (CD4^+^) is evenly present in this area. This is, however, not the case at the location of the T cell "drop" in the 20--30 µm region. Here, a median of ∼90 CD3 cells/mm^2^ and ∼67 CD8^+^ cells/mm^2^ was counted, meaning that CD4^+^ cells are more affected to decrease in this region. Interestingly, when examining spatial cytotoxic activity by GranB immunostaining it was likewise found to be significantly decreased at the 20--30 µm distance. As natural killer cells have been shown to be very rare in CRC-LM, cytotoxic CD8^+^ T cells are supposed to be the main source of GranB.[@cit0030] These findings strikingly confirm not only a reduction of cell quantities but also a reduction of cytotoxic T cell function at the 20--30 µm region. Figure 9.Overview of the observed T cell densities (gray cells) in relation to the distance to the tumor epithelium (red cells). Those new observations prompted us to further investigate the possible factors involved in this heterogenous T cell distribution patterns. As regulatory T cell numbers are generally low in CRC-LM (3.5 cells/mm^2^),[@cit0013] we decided to repeat our spatial distribution analysis to examine the localization of CD20 B cells and CD163 macrophages. This latter tumor-associated macrophage (TAM) population, which is characterized by a tumor-promoting and immunosuppressive phenotype and associated with poor clinical outcome in many different tumors,[@cit0031] we found highly enriched in CRC-LM tissues (∼200 cells/mm^2^). In our exploratory analyses, we noticed an increase (*p* = 0.068) at the 20--30 µm region, whereas no relevant differences between the other examined distances classes were observed. This opposite distribution to the T cells leads to the assumption that CD163 cells fill the gap between the tumor distant and tumor proximal T cells. Our data suggests that CD163 macrophages are forming a physical barrier which only a specific T cell subset can overcome. Even more important, this barrier could have functional impact, because it also marks the border, where cytotoxic GranB significantly decreased at the tumor margin. Indeed, TAM have been shown to directly suppress T cell responses through immune inhibitory checkpoint molecules like PD-L1.[@cit0034] Interestingly, these PD-L1 positive TAM were also found to be in close contact with T cells. This data goes in line with our own results showing a high percentage (68%) of T cells being in direct contact with CD163 macrophages, and furthermore, the majority of CD163 cells in the invasive margin were found to express PD-L1. Our data, therefore, suggests itself that TAM directly suppress T cells and consequently GranB production via immune checkpoints. The fact that in the invasive margin of CRC-LM 98% of all CD3 T cells express the checkpoint receptor PD-1 further encourages this assumption.[@cit0019] In contrast to CD163, the distribution of CD20 B cells is very heterogenous among the distance classes with lower cell densities at the tumor border and increased densities further away (*p* = 0.062). The role of tumor-infiltrating CD20 B cells is controversial. Some studies report about a protective role against tumors,[@cit0036] whereas others connect B cells with enhanced tumor growth and suppression of antitumor immunity.[@cit0036] Notably, B cells can polarize macrophages toward immunosuppressive phenotypes.[@cit0041] Nevertheless, B cell numbers appeared to be very low with ∼16 cells/mm^2^ among the complete tissues analyzed and especially compared with CD163 (∼200 cells/mm^2^) a minor role for CD20 B cells is presumed. Besides immunosuppressive cells, the distribution of the T cells in the tissues could also be accounted for other factors. Chemotactic fields, which might be generated by tumor, endothelial cells or even macrophages, could induce the observed T cell distribution patterns.[@cit0019] Another explanation could be cell-matrix interactions.[@cit0043] Physical properties of the extracellular matrix can influence directional cell movement and might govern the positioning of T cells in the direct vicinity of tumor cells. In addition, CD31, which is a marker for angiogenesis and expressed on endothelial cells, also has a functional role in leukocyte migration.[@cit0045] We analyzed two CRC-LM patient tissues for detailed distribution of CD31 (Fig. S5). Nevertheless, CD31 area was generally very high and showed strong heterogeneity among patients (∼700--8,000 mm^2^). In addition to that, CD31 can be expressed by leukocytes including T cells and thereby even having inhibitory effects on T cell activation[@cit0046] underscoring its highly diverse role in tumors. In this study, we performed a high-resolution analysis of different immune cell subsets. Consequently, the detailed functional mechanisms and cell interactions involved in the detected spatial T cell distribution patterns need to be investigated in future studies. The clear limitation of our study is the low sample number for some markers that limited our ability to identify statistically significant differences between the distance classes. Whereas we could not see distinct patterns limited to single patients (data not shown), the clustering analysis revealed that specific groups of patients show patterns of CD3 T cell aggregates, which were limited to a given region (e.g., 0--10 µm or 20--30 µm, [Fig. 3](#f0003){ref-type="fig"}). Indeed, OS analysis showed that patients with high T cell densities (CD3 as well as CD8^+^) within the \>10 µm distance class had a survival benefit compared with patients with low T cells in the aforementioned distance ([Figs. 4--6](#f0004 f0005 f0006){ref-type="fig"}). Interestingly, the opposite effect was found at the 20--30 µm distance from the tumor. Here, low CD3 T cell numbers had positive effects on patient survival. Consequently, despite the physical or functional T cells barrier, the observed pattern of T cell distribution at 20--30 µm distance obviously is beneficial for patients (under palliative therapy). When considering this observation from the other side, high T cell numbers unexpectedly might have a tumor promoting effect. Our own previous work illustrated the exploitation of CCL5-producing T cells in the invasive margin of CRC-LM patients by supporting CCL5 receptor (CCR5)-expressing tumor cells.[@cit0019] Furthermore, blockade of CCR5 led to repolarization of TAM resulting in beneficial clinical responses. These findings might explain, why especially in patients with low T cells in the 20--30 µm region a better survival is observed than in patients with high T cell numbers, and furthermore, it shows that reeducated macrophages, which were found to be increased at the 20--30 µm distance and which are in fact generally high in the invasive margin, can render antitumor effects under therapy. In addition to that, our results indicate that T cell induced antitumor effects can be conducted by T cells that have engaged the tumor margin at the 0--10 µm distance class, which is indeed also true for the cytotoxic CD8^+^ subtype, while T cells further away remain blocked. The contrary outcome of our detailed distribution analysis that both localization-dependent high as well as low T cell numbers have clinical benefit might explain startling observations for similar tumors, where in one study high[@cit0048] and in another study low T cell densities[@cit0049] indicate prognostic relevance. Our finding might further elucidate related controversial observations, where studies identify high densities of immunosuppressive cell populations being associated with favorable prognosis.[@cit0031] The reasons for the controversial observation of local T cell decrease being associated with favorable prognosis and if the observed patterns change under therapy are currently investigated. The paradoxical observation of localization-dependent prognostic relevance of T cell densities is only decipherable by detailed spatial analyses. Therefore, the analysis of spatial distribution patterns at higher resolution, and especially the distance classes \<10 and 20--30 µm alone, could serve as more precise prognostic markers for responses to chemotherapy and OS compared with cell densities measured on a broader scale. In summary, we present here for the first time detailed data on the local spatial distribution of immune cells in colorectal cancer LM. Thereby, distinct patterns of immune cells reveal a physical or functional T cell barrier at 20--30 µm distance from the tumor epithelium with decreased T cell and cytotoxic Gran B densities as well as increased CD163 macrophages. These specific distribution patterns with either high T cells at the direct tumor border or low T cells at the 20--30 µm distance are associated with improved OS. This detailed analysis of spatial profiles within the microenvironment, therefore, represents new insights into T cell distribution, function as well as influence of T cell localization on clinical responses and introduces possible spatial immunosuppressive hurdles that need to be overcome by successful immunotherapies. We suggest the analysis of spatial profiles within the \< 10 and 20--30 µm distance from the tumor epithelium as new predictive biomarker for survival and response to therapies. Materials and methods {#s0004} ===================== Tissue samples {#s0004-0001} -------------- The study was approved by the Medical Ethics Committee of the University of Heidelberg. Tumor samples were used after receiving a signed informed consent from all 36 patients. Patients were included in this study when diagnosed with stage UICC IV CRC after metastatic disease in a surgically resected liver metastasis was histologically proven. None of the patients received neoadjuvant treatment. Samples had to contain (a) metastasis, (b) invasive margin and (c) adjacent liver ([Fig. 1A](#f0001){ref-type="fig"}). Patients received standard of care palliative therapies and follow-up was obtained as described previously.[@cit0013] All samples were screened for microsatellite instability (MSI) using established protocols.[@cit0021] No MSI sample was included in this analysis. Clinical data as well as detailed pathological reports showed no evidence for the presence of ethyltoxic liver damage, chronic inflammation due to ongoing infections (e.g., hepatitis) or other chronic inflammatory conditions. Immunohistochemistry and immunofluorescence staining {#s0004-0002} ---------------------------------------------------- Tissue sections (4 µm) were prepared from formalin-fixed, paraffin-embedded material for sequent immunohistochemically detection of T cells (CD3^+^, CD8^+^), Granzyme B (GranB^+^), B cells (CD20^+^) and TAM (CD163+). All processing steps were performed with a fully automated staining system (Leica BOND-MAX™, Leica Microsystems, Germany) to achieve maximum reproducibility. Briefly, after deparaffinization and rehydration, slides were boiled for 20 min in either 10 mM citrate buffer (pH 6) for CD3, CD8^+^ and CD20 or 1 mM EDTA buffer (pH8) for CD163 and GranB to retrieve the antigens. The endogenous peroxidase activity was blocked by incubation with 3--4% (v/v) hydrogen peroxide for 10 min. The sections were blocked with 10% normal goat serum. Mouse monoclonal antibodies recognizing human CD3ε (1:50 dilution, clone PS1, Acris \#DM112--05), CD8^+^ (1:50 dilution, clone 4B11, Novocastra \#NCL-CD8--4B11), CD20 (1:50 dilution, clone L26, Novocastra \#NCL-L-CD20-L26), GranB (1:20 dilution, clone 11F1, Novocastra \#NCL-Gran-B) and human CD163 (1:500 dilution, clone EDHu-1, AbDSerotec \#MCA1853) were applied as primary antibodies for 30 min at room temperature. Slides were then incubated with the secondary antibodies and subsequent visualization was performed according to the manufacturer\'s instructions (Bond Polymer Refine Detection Kit, Leica \#DS9800). Antigen detection was performed by a color reaction with 3,3-di-amino-benzidine (DAB). The sections were counterstained with hematoxylin and mounted with Aquatex (VWR \#1085620050). Controls without primary antibody and isotype controls were used for all antibodies. Positive controls for the presence of T cells and macrophages consisted of adjacent normal tissue. For immunofluorescence double staining of CD163 (1:1000 dilution, clone EPR19518, abcam \#ab182422) and PD-L1 (1:100 dilution, clone 29E.2A3, BioLegend \#329710) deparaffinization and rehydration of slides was performed following 20 min in 1 mM EDTA buffer (pH8) to retrieve the antigens. After incubation of the first primary antibodies overnight at 4 °C, secondary antibodies (Alexa Fluor 594, Life Technologies \#A21207; Alexa Fluor 488, Life Technologies \#A-11029) were applied for 1 h. DAPI (AppliChem \#1001) was used for counterstain and sections were mounted using Vectashield (Vector \#H-1000). Evaluation of immunostainings and tissue classification {#s0004-0003} ------------------------------------------------------- Using the NDP Nanozoomer system (Hamamatsu Photonics, Japan) complete microscopic images of full tissue sections were automatically obtained. Cell quantification and distance analysis were performed with the VisioMorph software (Visiopharm, Denmark). Image segmentation and subsequent classification according to color and morphology was established on a separate training set with manually counted cell numbers previously.[@cit0022] This approach in combination with serial cuts of tissue blocks allows robust and reproducible large scale histological evaluations with high precision across complete (adjacent) sections. 1 mm^2^ regions of the invasive margin ([Fig. 1A](#f0001){ref-type="fig"}) were generated in an automated process, visually controlled and adapted correspondingly. Then, the positions of the T cells in relation to the tumor epithelium were identified in consecutive 10 µm steps ([Fig. 1B](#f0001){ref-type="fig"}). Distances to tumor epithelium were grouped into distinct classes (distance classes), e.g., "below 10 µm," "10--20 µm," "20--30 µm" and so on. So for each tissue section, T cell quantities in 10 distance classes within 100 µm from the tumor epithelium into the normal adjacent liver were established. Immune cells and other markers were quantified in this process by spectral and morphological criteria as described before[@cit0015] in combination with aggregation of data from the spatial registry. To evaluate the distance between CD163 macrophages and T cells within the tumor microenvironment, serial sections were stained with the respective markers and after scanning of the sections, these were computationally analyzed using VisioMorph software ([Fig. 7](#f0007){ref-type="fig"}). Statistical analyses {#s0004-0004} -------------------- Statistical analyses were performed using the SPSS 18.0 and 22 packages (IBM) as well as GraphPad Prism version 5. To discriminate CD3 cell counts between distance classes Kruskal--Wallis tests with post-hoc Mann--Whitney U tests were performed. Results with two-tailed *p* values \<0.01 were judged to be statistically significant. Boxplots depict the minimum, first quartile, median, third quartile and maximum. Measurements with values between 1.5 and 3 lengths of the box away from the upper and lower border of the box were classified as outliers (white circles). The length of the box equals the interquartile range. To discriminate cell counts between single distance classes for CD8^+^ and GranB Wilcoxon signed-rank and for CD20 and CD163 paired *t*-tests were performed. Results with two-tailed *p* values \<0.05 were judged to be statistically significant. Unsupervised hierarchical clustering analysis using Euclidean distance was performed with the R software package.[@cit0023] OS curves were computed using the Kaplan--Meier estimator (Breslow test) and significance was assumed for *p* values of 0.05 or less. Cutoff values were adapted (quantile-quantile class) from previous work,[@cit0013] which is in line with findings from primary CRC tumors.[@cit0024] Supplementary Material ====================== ###### KONI_A\_1286436_s02.docx Disclosure of potential conflicts of interest ============================================= No potential conflicts of interest were disclosed. Acknowledgments =============== The authors thank the NCT tissue bank for their support. Funding ======= This work was funded by the Helmholtz Alliance on Immunotherapy of Cancer (NH) and through a grant from the Cancer Research Institute (DJ). The study sponsors had not role in the study design, in the collection, analysis and interpretation of data; in the writing of the manuscript; and in the decision to submit the manuscript for publication. [^1]: Supplemental data for this article can be accessed on the [publisher\'s website](http://dx.doi.org/10.1080/2162402X.2017.1286436).

Introduction {#S1} ============ Systemic thrombolysis with the recombinant tissue plasminogen activator (rtPA) is the only approved treatment of acute ischemic stroke. It aims at rapid recanalization of the occluded cerebral artery affording neurological recovery.^[@R1]^ In recent years it was shown that early recanalization results in better neurological outcome and less lesion growth than lacking recanalization.^[@R2]--[@R4]^ However, the severity of the ischemic attack prior to thrombolytic treatment and a poor status of arterial collaterals have been described to predict a poor recovery.^[@R5],[@R6]^ On average, thrombolysis has been found to enhance the proportion of patients with a favorable outcome after ischemic stroke and to downsize the proportion of patients with severe neurological deficits.^[@R7]--[@R9]^ In fact, a large recent meta-analysis of different thrombolysis studies revealed that the application of alteplase rendered some 42 percent of patients with an excellent outcome \[modified Rankin Scale (mRS) of 0 and 1\] on day 90 as compared with 30 percent of patients receiving placebo.^[@R10]^ Prompted by the results of platelet receptor inhibition in the management of patients with acute myocardial ischemia we have introduced a regime for systemic thrombolysis composed of an intravenous bolus of low dose rtPA in combination with a subsequent infusion of the nonpeptide GPIIb/IIIa platelet receptor antagonist tirofiban.^[@R11],[@R12]^ The rational for a low dose of rtPA was to avoid intra- and extracranial hemorrhage, since symptomatic and fatal hemorrhages are a severe complication of a high dose of rtPA.^[@R13]^ Further, GPIIb/IIIa receptor antagonists selectively inhibit the platelet integrin IIb III fibrinogen receptor and thereby inhibit platelet aggregation.^[@R14],[@R15]^ Since rtPA is known to induce a hypercoagulation following thrombolysis,^[@R16]^ tirofiban was expected to antagonize the hypercoagulation following rtPA administration and thereby help to maintain the cerebral blood vessels patent after thrombolysis. Likewise, it might help to disaggregate secondary thrombi resulting from the large thrombus in the initially occluded cerebral artery. We have shown previously that the combined use of low dose rtPA and tirofiban is safe and effective.^[@R12],[@R17]^ Thus, we used this treatment as first line regime. In this prospective mono-center, non-randomized study we analyzed the clinical data of the consecutive patients treated with a low bolus of rtPA and a subsequent infusion of tirofiban in our hospital in the years 2005 through 2007. We investigated four questions: i) what is the proportion of patients who benefit from this treatment? ii) what is the proportion of patients who do not benefit from this treatment? iii) what is the proportion of patients who expire after this stroke treatment? iv) what are predictors for poor or good recovery? Materials and Methods {#S2} ===================== Subjects -------- All patients who were treated with systemic thrombolysis on the Stroke Unit of our institution between 1 January 2005 and 31 December 2007 were included. Criteria for inclusion into this study were: an acute ischemic brain infarct,documentation of a causal cerebral artery occlusion before thrombolysis using computed tomographic angiography (CTA) or magnetic resonance angiography (MRA),systemic thrombolysis with rtPA and tirofiban,clinical evaluation on admission and at discharge from the Stroke Unit using the stroke scale of the National Institutes of Health (NIHSS), the mRS and Barthel index (BI),^[@R18]--[@R21]^follow-up telephone questionnaire of the BI and mRS, which was performed after at least 130 days (median 487 days, range 131--1056 days). Patients were subjected either to computed tomography including CTA or multiparametric magnetic resonance imaging including a MRA depending on availability prior to thrombolysis. Thrombolysis was performed in each patient with an intravenous bolus of 20 mg rtPA within 3 h after stroke onset followed immediately by an intravenous infusion of tirofiban.^[@R12],[@R17]^ Tirofiban was given in a body-weight adjusted dosage starting with a bolus of 0.4 µg/ kg body weight/min for 30 min followed by continuous infusion of 0.1 µg/ kg body weight/min for 48 h. In addition, patients were treated according to their individual requirements. The treatment procedure and the study were approved by the Ethics Committee of the Heinrich-Heine-University Düsseldorf, Germany. The patients gave informed consent. Statistical analysis -------------------- Statistical analysis was done using SPSS 10.1 for Windows (release 2005). Group comparisons were done using the two-tailed t-test. Evaluation of the neurological deficit and the impairment was assessed with the non-parametric Mann-Whitney test. Group comparison of the abnormalities in the cerebral arterial circulation was performed with the distribution free Wilcoxon\'s rank test. Multiple regression analyses were performed to single out the variables that accounted for the effect of thrombolysis. Results {#S3} ======= Of 2241 patients treated on the Stroke Unit of this institution in 2005 through 2007, 252 patients (11%) were treated with systemic thrombolysis using intravenous rtPA and tirofiban. Fifteen patients were subjected to body weight adjusted rtPA or received a catheter intervention due to a stroke in the vertebrobasilar territory. The 192 patients (70±13 years, 50 percent males) matching the inclusion criteria of this study were severely impaired at stroke onset having a median NIHSS of 10 (range 2--27). 81% of the patients had a middle cerebral artery (MCA) territory infarct. There were only few patients with a tandem occlusion of the internal carotid artery (ICA) and the MCA. On average the patients improved by 7 points on the NIHSS (P\<0.0001) until discharge from the Stroke Unit ([Table 1](#T1){ref-type="table"}). The profound improvement was also apparent in the mRS, and BI ([Table 1](#T1){ref-type="table"}). The majority of the patients with a tandem occlusion of the ICA and MCA deceased. In contrast, the more peripheral the MCA occlusions occurred, the more patients survived ([Figure 1](#F1){ref-type="fig"}). There was no difference among the subgroups concerning age, gender or vascular risk factors, apart from atrial fibrillation being more frequent in tandem occlusions of the ICA and MCA or at the bifurcation as compared to a MCA-branch (P\<0.001). Twenty-three patients died with 6 patients due to an intracranial haemorrhage (3.1%) and 10 patients suffering malignant brain infarction. After discharge eleven further patients deceased such that 34 patients (18%) deceased within 100 days after stroke ([Figure 2](#F2){ref-type="fig"}). Death was predicted by older age (76±10 years, P\<0.05) and a more severe affection (mRS 5, P\<0.0001). Also, the patients more frequently had atrial fibrillation (P\<0.03) than the surviving patients. But none of the other vascular risk factors was predicting or different between the deceased or surviving patients ([Table 1](#T1){ref-type="table"}). The surviving patients improved further (P\<0.0001) leaving 48% of study cohort with a mRS of 0 and 1 at follow-up, while 18 % of the patients were still severely affected having a mRS of 4 or 5 ([Figure 3](#F3){ref-type="fig"}). Table 1Characteristics of the 192 patients.Age (years ± SD)70±13Male/Female (n)95/97NIHSS, admission (median, range)10 (2--27)BI, admission (median, range)30 (0--100)MRS, admission (median, range)4 (0--5)NIHSS, discharge (median, range)3 (0--23)[\*](#TF1-1){ref-type="table-fn"}BI, at discharge (median, range)80 (0--100)[\*](#TF1-1){ref-type="table-fn"}MRS, at discharge (median, range)3 (0--6)[\*](#TF1-1){ref-type="table-fn"}Arterial hypertension (%)85Diabetes mellitus (%)20Atrial fibrillation (%)36Coronary artery disease (%)26Current smoking (%)22[^3][^4] Figure 1Influence of the location of the occlusion of the middle cerebral artery on the rate of survival. The numbers in each column indicate the number of patients. Figure 2Distribution of dead patients following acute stroke. The majority of patients expired within the first 100 days. Note that 23 patients already died on the Stroke Unit (open column). Figure 3Neurological impairment of the 192 patients as assessed with the modified Rankin Scale on admission before thrombolysis (A) and at follow up (F). Note that half of the patients had no only or little impairment at follow up.^[@R20]^ Discussion {#S4} ========== This study revealed that the combined use of an intravenous bolus of low dose rtPA and a subsequent body-weight adjusted infusion of tirofiban had an optimal outcome (mRS of 0 and 1) at follow-up in about 48 percent of our patients. In comparison to the large clinical trials of systemic application of rtPA this was clearly better than in the placebo group and similar to the thrombolysis group.^[@R10]^ Note, that the patients in this study had a median NIHSS at treatment onset of 10 being as severely affected as the patients in the ECASS 3 study.^[@R1]^ Importantly, the patients who benefitted from our treatment regime were younger and less severely affected at stroke onset than those who remained severely impaired or deceased. This was the case both at discharge from the Stroke Unit and for follow-up up to 100 days after stroke. These data accord well with recent reports about intravenous thrombolysis with 0.9 mg rtPA per kilogram body weight in the Safe Implementation of Thrombolysis in Stroke Monitoring Study,^[@R22]^ the Virtual International Stroke Trial Archive,^[@R23]^ and the Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution.^[@R24]^ These and our data accord with the notion that age and initial NIHSS impact on clinical outcome independent of treatment. Our rational for the combined use of low-dose rtPA and a subsequent infusion of tirofiban was our attempt to minimize the risk of cerebral hemorrhage on one hand and to counteract arterial reocclusion after stroke and thrombolysis.^[@R16]^ Thrombus formation involves the aggregation of activated platelets and the bridging between platelets by fibrinogen via the reconfigured GPIIb/IIIa platelet receptor.^[@R25],[@R26]^ The GPIIb/IIIa receptor is a prototypic integrin which is exclusively expressed on the membranes of platelets and megakaryocytes. In experimental stroke models, GPIIb/IIIa antagonists given as only drug or in combination with rtPA have been shown to be safe and effective, since the treated animals had reduced infarct volumes.^[@R27],[@R28]^ Clinical studies suggest that tirofiban can disintegrate acute thrombi in the large internal carotid artery and suppress microemboli as detected with transcranial Doppler sonography.^[@R29]^ Moreover, in small prospective series we were able to show that a low-dose bolus of 20 mg rtPA and body-weight adjusted infusion of tirofiban resulted in a profound clinical recovery along with recanalization and a decrease in infarct volume.^[@R12],[@R17]^ Similarly, the monoclonocal antibody abciximab that also targets the platelet GPIIb/IIIa receptors was shown to be effective when combined with rtPA in the interventional treatment of vertebrobasilar stroke.^[@R30]^ In contrast, abciximab was found to result in an exaggeration of fatal hemorrhages and not to improve outcome in stroke of the anterior circulation.^[@R31]^ Since the non-peptide tirofiban was found to be safe in stroke,^[@R18]^ it is possible that the monoclonal antibody abciximab blocked the GPIIb/IIIa receptors of all circulating platelets irreversibly precipitating hemorrhagic complications. In our patients atrial fibrillation was more frequent in the more severely affected patients with more proximal MCA occlusions than in the less severely affected patients. These data highlight the impact of cardioembolic stroke for disability and dependence particularly in the elderly. In contrast, the vascular risk factors were not predictive for poor outcome or discriminating between good recovery and poor recovery or death. It is possible that larger cardioembolic thrombi would occlude larger more proximal arteries including the distal ICA and that the thrombi may be more resistant to thrombolysis in older than in younger patients. Both factors would prevent rapid recanalization, early reperfusion and neurological recovery.^[@R1]--[@R4]^ Since rtPA is beneficial also in the elderly,^[@R32],[@R33]^ it should not be withheld from them in acute stroke. But oral anticoagulation because of absolute arrhythmia related to atrial fibrillation is of even greater importance. In accordance with other recent studies we found that the rate of survival was better the more distal the occlusion of the MCA was.^[@R34]--[@R37]^ Similar results were obtained concerning the neurological disability and recovery.^[@R17],[@R38]^ Most likely, this was due to smaller infarcts in more distal cerebral artery occlusions that have been found to exhibit a better recanalization rate.^[@R34]--[@R39]^ It is tempting to speculate that mechanical devices may be employed in patients with proximal MCA occlusions. The benefit of tirofiban in such a situation is that it can be combined with local thrombolysis enhancing its effect beyond the narrow time window of thrombolysis. Thus, it may be of interest in the context of interventional approaches for the treatment of acute stroke. The rate of fatal hemorrhage in our study was in the range of thrombolysis with 0.9 mg rtPA per kilogram body weight.^[@R34]--[@R42]^ This accords with our recent observation that tirofiban does not increase the bleeding rate in acute stroke.^[@R18]^ It should be noted that in that former study tirofiban was administered as a single drug or in addition to aspirin, while in this study tirofiban was combined with rtPA even in patients receiving clopidogrel hydrogensulfate prior to their present stroke. Conclusions {#S5} =========== Here, we have shown that the combined use of an intravenous bolus of 20 mg rtPA and a subsequent body-weight adjusted infusion of tirofiban had an efficacy and safety similarly to systemic thrombolysis with body-weight adjusted rtPA. Our study, however, has limitations. First, we did not perform a direct comparison with the standard thrombolytic treatment with body-weight adjusted rtPA. As evident from this study and calculated from the known effect size, a double-blind, randomized, multi-center trial would require at least 5000 patients per treatment arm. Also, since rtPA has been approved as standard care for acute stroke, a placebo control group was ethically not acceptable. Second, we did not asses the rate of early recanalization in these patients. This will be the topic of a subsequent study. Third, since it is difficult to follow-up patients for face-to-face interviews, we chose to perform the follow-up study using telephone interviews. It has been found previously that both the BI and mRS can be used reliably in this way.^[@R43],[@R44]^ Fourth, we did not monitor depression or dementia that both have been reported to affect post-stroke recovery and death.^[@R45]^ [^1]: Funding: the study was supported by the Competence Net Stroke of the BMBF. [^2]: Conflict of interests: the authors report no potential conflict of interests. [^3]: Significant change from admission to discharge (P\<0.0001, Wilcoxon rank test). [^4]: NIHSS, Stroke Scale of the National Institute of Health; MRS, modified Rankin Scale; BI, Barthel index.

Introduction {#s1} ============ The combination of irinotecan and temozolomide has shown activity against many solid tumors including neuroblastoma, Ewing sarcoma, and rhabdomyosarcoma \[[@B1],[@B2]\]. There are both preclinical and clinical evidence of synergy between these two agents, and this may be schedule dependent \[[@B3],[@B4]\]. The non-overlapping dose limiting toxicities of these two agents, diarrhea (irinotecan) and myelosuppression (temozolomide) make this combination attractive. In addition, irinotecan and vincristine have shown synergistic activity in patients with rhabdomyosarcoma \[[@B5]\]. Based on preclinical data, irinotecan was initially administered as a protracted regimen (five consecutive days in two weeks) \[[@B4],[@B6]\]. Subsequently, studies have shown that there was no difference in efficacy between irinotecan administered as protracted regimen or as shortened regimen over five days \[[@B5],[@B7]\]. The Children's Oncology Group (COG) has studied the combination of vincristine, oral irinotecan and temozolomide in the phase I setting, and demonstrated its feasibility and safety \[[@B7]\]. Combining newer targeted agents to this backbone may provide additional anti-tumor activity. Angiogenesis is the hallmark of tumor development and metastases. Bevacizumab is a humanized monoclonal neutralizing antibody against vascular endothelial growth factor \[[@B8]\]. Bevacizumab is approved in adults for use in colorectal, renal, non-small cell lung cancer and glioblastoma \[[@B9]--[@B11]\]. Bevacizumab has shown activity in preclinical models of pediatric cancers \[[@B12]--[@B14]\]. Bevacizumab at a dose of 15 mg/kg administered every two weeks was well tolerated in a phase I study conducted by COG \[[@B15]\]. Even though no objective responses were seen in that study, responses were observed when bevacizumab was used in combination with irinotecan in children with low and high-grade glioma \[[@B16],[@B17]\]. Anecdotal reports and case series of combination of bevacizumab, irinotecan and temozolomide have been published, but this combination has not been systematically studied in children \[[@B18]\]. The maximum tolerated dose of irinotecan administered intravenously over five days in combination with temozolomide is yet to be defined. A previous study of irinotecan and temozolomide performed in neuroblastoma patients used a lower threshold for platelets \[[@B1]\]. We conducted a phase I study of escalating doses of irinotecan together with standard doses of vincristine, temozolomide and bevacizumab (VITB) in patients with relapsed or refractory solid tumors. Materials and Methods {#s2} ===================== The protocol for this trial and supporting TREND checklist are available as supporting information; see Checklist S1 and Protocol S1. Ethics Statement {#s2.1} ---------------- This research was approved by the institutional review board at Children's Hospital, Los Angeles. Written informed consent was obtained from patients. In case of minors, written informed consent was obtained from parents/legal guardians. Patient Eligibility {#s2.2} ------------------- Patients \>1 and \<21 years of age with histologically confirmed solid tumor without known effective therapy, body weight ≥ 10 kilograms, a Karnofsky or Lansky performance score of \> 50%, and with an expected life expectancy of \> 8 weeks were eligible. Patients must have recovered from acute toxic effects of prior therapy; and must not have received 1) myelosuppressive therapy within two weeks (four weeks if nitrosourea); 2) biological agent within one week; 3) small-port palliative radiotherapy within 2 weeks; 3) total body, craniospinal or hemi-pelvic radiation within 6 months; 4) autologous stem cell transplant within six months; 5) hematopoietic growth factors within 1 week. Organ function requirements were as follows: peripheral blood neutrophil count (ANC) ≥ 750/µL, platelet count ≥ 75,000/µL (transfusion-independent), Hemoglobin ≥ 8 g/dL; ANC ≥ 500/µL, platelet count ≥ 50,000/µL (transfusion-independent) in patients with bone marrow involvement; normal serum creatinine for age or a glomerular filtration rate of [\>]{.ul} 70ml/min/m^2^; urine protein creatinine ratio \<0.5; left ventricular shortening fraction of ≥ 28% by echocardiogram or ejection fraction of ≥ 55% by MUGA scan; total bilirubin level ≤ 1.5 x upper limit of normal for age, alanine aminotransferase ≤ 5 x upper limit of normal for age. Patients who underwent a major surgical procedure within 28 days or a minor surgical procedure within seven days were ineligible. Patients with a history of deep vein thrombosis within three months, cerebrovascular accident within six months, or with uncontrolled hypertension were excluded. Patients who had received irinotecan or bevacizumab previously were ineligible. Patients who had received temozolomide were ineligible except for patients with central nervous system tumors. Pregnant or breastfeeding females, patients with uncontrolled infection were excluded. The study was registered in ClinicalTrials.gov (NCT00993044). Study Design {#s2.3} ------------ Each cycle was 21 days and included: vincristine 1.5 mg/m^2^ (2 mg maximum total dose) intravenously on days 1 and 8; bevacizumab 15mg/kg intravenously on day 1 administered over 90 minutes during cycle 1, over 60 minutes in cycle 2 and over 30 minutes from cycle 3 onward provided there was no reaction during infusion; temozolomide 100mg/m^2^ orally on days 1-5; irinotecan intravenously on days 1-5 approximately 60 minutes after temozolomide intake. Patients received granulocyte colony stimulating factor (G-CSF) 5 µg/kg/day subcutaneously starting on day 6 and continued until ANC\> 2000/µL, or one dose of PEG filgrastrim 100 µg/kg. Patients received a maximum of 12 cycles in the absence of disease progression. Irinotecan was administered intravenously on days 1-5 of each 21 day cycle. Two dose levels were planned: dose level 1, 30 mg/m^2^ (maximum 60 mg/day); dose level 2, 50 mg/m^2^ (maximum 100mg/day), with the option of two de-escalation dose levels: dose level -1, 20 mg/m^2^ (maximum 40 mg/day) and dose level 1.5, 40 mg/m^2^ (maximum 80 mg/day). In the event of diarrhea within 12 hours of irinotecan administration, patients received atropine 0.01mg/kg intravenously. If diarrhea occurred after 12 hours, or after administration of atropine, oral loperamide was administered. If the patient developed ≥ grade 3 diarrhea, cefixime or cefpodoxime was started 5 days before administration of irinotecan on subsequent cycles and continued for a minimum of 24 hours after the last dose of irinotecan. A minimum of three evaluable patients were treated at each dose level. In the absence of dose limiting toxicity (DLT), patients were enrolled at the next dose level. If 1 of 3 patients had a DLT, the cohort was expanded to include 6 patients. If ≥ 2 patients experience DLT, maximum tolerated dose (MTD) was exceeded and further enrollment at that dose level was stopped. The MTD was the highest dose level at which ≤ 1 of 6 patients experienced a DLT. If the MTD was exceeded at the dose level 1, then the subsequent cohort of patients were to be treated at dose level -1. If the MTD was exceeded at the dose level 2, then the subsequent cohort of patients were to be treated at dose level 1.5. Patients with DLT who had no evidence of progressive disease were allowed to continue on protocol therapy at the lower dose level as long as all toxicities returned to baseline. Patient Evaluation {#s2.4} ------------------ A medical history, physical examination, renal and liver function tests and serum electrolytes were obtained prior to study enrollment, weekly during the first cycle of treatment and prior to each cycle thereafter. Complete blood counts (CBC) were obtained prior to study enrollment and twice weekly during each cycle. Coagulation studies and urine protein creatinine ratio were obtained prior to each cycle. Echocardiogram was obtained prior to study enrollment, following cycle 2, and at the end of treatment. Tibial growth plate evaluation was performed by plain radiograph of the right or left knee prior to enrollment. If the growth plate was open, further evaluations of the same knee were performed at the end of cycles 2, 5, 8 and 12. Response Evaluation Criteria in Solid Tumors (RECIST) were used to assess tumor response at the end of cycles 2, 5, 8 and 12. Responses were required to be sustained for a minimum of two consecutive imaging evaluations. Toxicities were graded according to NCI Common Terminology Criteria for Adverse Events (CTCAE), Version 4.0. Hematological dose limiting toxicity was defined as grade 4 neutropenia or grade 4 thrombocytopenia lasting \>14 days or myelosuppression which caused delay in chemotherapy for \>14 days. Any grade 3 or grade 4 non-hematological toxicity was considered a dose limiting toxicity with the specific exclusion of 1) grade 3 or 4 nausea and vomiting controlled by anti-emetics; 2) grade 3 transaminase elevation which returned to ≤ grade 1 before next cycle; 3) grade 3 diarrhea lasting \< 3 days; 4) grade 3 infection or fever; 5) grade 3 electrolyte abnormalities that improved to ≤ grade 2 within seven days; 6) grade 3 catheter related venous thrombosis; 7) vincristine related neuropathy. In addition, grade 2 arterial thrombosis, and grade 2 pulmonary or CNS hemorrhage were considered dose limiting. Only, toxicities during the first 2 cycles were used to determine the MTD. Results {#s3} ======= Patient characteristics {#s3.1} ----------------------- Thirteen patients were prospectively enrolled between December 2009 and August 2012 ([Table 1](#tab1){ref-type="table"} [Figure 1)](#pone-0068416-g001){ref-type="fig"}. One patient developed fever on the second day of cycle 1 and did not receive further protocol therapy. This patient was removed from the study and deemed ineligible for toxicity evaluation since the fever was unrelated to therapy. Twelve eligible patients received 87 cycles of therapy (47 on dose level 1, 40 on dose level 2). Four patients had previously received high-dose chemotherapy followed by autologous bone marrow transplant. ###### Patient Characteristics (n=12). **Category** **Subcategory** **No. of Patients (%)** -------------------------- ---------------------------------------------- ------------------------- **Age at enrollment, y** Median 11 Range 3.9-19.4 **Gender** Male 8 (66.7) Female 4 (33.3) **Diagnosis** Wilms tumor 3 (25) Osteosarcoma 2 (16.7) Hepatocellular carcinoma 2 (16.7) Ewing sarcoma 1 (8.3) Medulloblastoma 1 (8.3) Liposarcoma 1 (8.3) Synovial Sarcoma 1 (8.3) Angiosarcoma 1 (8.3) **Prior therapy** No. of chemotherapy regimens, Median (range) 2 (1-4) Radiotherapy 7 Autologous bone marrow transplant 4 {#pone-0068416-g001} Toxicity {#s3.2} -------- The first patient enrolled on dose level 1 had Noonan syndrome and recurrent synovial sarcoma. This patient developed dose limiting hyperbilirubinemia during cycle 1 and was removed from protocol therapy after two cycles due to non-compliance. None of the additional five patients enrolled on dose level 1 developed dose limiting toxicity. Dose limiting colitis developed in one of the first three patients enrolled on dose level 2. Three additional patients enrolled at this level did not have a DLT. Therefore dose level 2 (irinotecan 50 mg/m^2^) was determined to be the MTD. Toxicities which developed during the first two cycles were used to determine the MTD in an attempt to capture possible delayed toxicity due to bevacizumab. The hematological toxicities are summarized in [Table 2](#tab2){ref-type="table"}. None of the hematological toxicities met the study definition of DLT. Due to the potential for bleeding when bevacizumab is administered, platelet transfusion was administered for platelet count less than 20,000/µL. In spite of this requirement, five patients on this study did not require platelet or packed red blood cell transfusion throughout their therapy (median 5 cycles, total 28 cycles). Therapy was administered in the ambulatory setting, and hospitalization for fever and neutropenia occurred only in three cycles. The frequency of grade 3 or 4 non-hematological toxicity was low ([Table 3)](#tab3){ref-type="table"}. Two patients developed grade 2 hypertension; one was transient and related to pain. In the other patient, hypertension developed during cycle 11 and was well controlled with low-dose single agent anti-hypertensive treatment. This patient was able to continue protocol therapy without interruption. Severe proteinuria requiring discontinuing treatment was not observed. Mild (grade 1) abnormalities in coagulation studies were detected in 14 cycles. There were no episodes of thrombosis, and intermittent grade 1 epistaxis was seen in four patients. Growth plate changes were not seen in serial knee radiographs obtained in seven patients (median of 5 cycles) whose growth plate was open at study enrollment. Mild diarrhea (grade 1 and 2) occurred in 60 cycles, and was well controlled with loperamide. ###### Hematological Toxicities. **Toxicity** **Cycle 1 (N=12)** **Cycles 2-12 (N=75)** ------------------ -------------------- ------------------------ --- ---- ---- ---- Anemia 3 2 28 19 Leukopenia 2 11 7 2 Neutropenia 1 1 4 5 7 11 Thrombocytopenia 1 2 9 6 8 ###### Non-hematological Toxicities. **Toxicity** **Cycle 1 (N=12)** **Cycles 2-12 (N=75)** -------------------- -------------------- ------------------------ --- --- ALT 1 Colitis 1\* 1 Diarrhea 1 Fever Neutropenia 2 1 Hyperbilirubinemia 1\* 1 Hypoalbuminemia 1 Hypocalcemia 1 Hypokalemia 1 1 Hyponatremia 1 1 Neuropathy 1 \* Dose limiting toxicity Antitumor activity {#s3.3} ------------------ Complete response was achieved in a patient with Wilms tumor after 8 cycles ([Table 4)](#tab4){ref-type="table"}. In another patient with relapsed Wilms tumor, a lung nodule persisted despite a dramatic response initially. This nodule was resected after 8 cycles, and did not have histological evidence of viable tumor. Therefore, this patient was determined to have a complete response. Partial responses were seen in patients with Wilms tumor, medulloblastoma and hepatocellular carcinoma. The patient with medulloblastoma was removed from protocol therapy by the primary oncologist after 4 cycles to administer radiation therapy. One patient was removed from protocol therapy after cycle 1 due to progressive disease. Another patient with an angiosarcoma of the heart and a 8 cm x 3 cm chest wall mass showed a dramatic response to the first two cycles with greater than 50% reduction in tumor size on imaging. The start of his third cycle was delayed due to fever by one week. He developed progressive disease during this period and was removed from protocol therapy. Five patients completed all 12 cycles of chemotherapy. ###### Dose-limiting Toxicity and Response. **Tumor Type** **Dose Level** **No. of courses** **Best Response** **Dose limiting toxicity** **Reason for coming off protocol therapy** -------------------------- ---------------- -------------------- ------------------- ---------------------------- -------------------------------------------- Synovial Sarcoma 1 2 SD Hyperbilirubinemia Toxicity/non compliance Osteosarcoma 1 5 SD PD Hepatocellular carcinoma 1 12 SD End of therapy Liposarcoma 1 12 SD End of therapy Medulloblastoma 1 4 PR Physician preference Wilms Tumor 1 12 CR End of therapy Hepatocellular carcinoma 2 12 PR End of therapy Angiosarcoma 2 2 PD PD Osteosarcoma 2 1 PD Colitis PD Wilms Tumor 2 8 PR PD Wilms Tumor 2 12 CR End of therapy Ewing Sarcoma 2 5 SD PD SD, stable disease; PD, progressive disease; PR, partial response; CR, complete response Discussion {#s4} ========== Irinotecan has been administered using various schedules in both adults and children \[[@B6],[@B19]--[@B22]\]. Preclinical studies in pediatric tumors by Houghton et al, showed that a protracted schedule of irinotecan given daily for 5 days per week for 2 consecutive weeks resulted in greater response rates when compared to the same dose administered over 5 days \[[@B23]\]. Based on this, irinotecan was initially administered on a protracted schedule of 5 consecutive days for 2 weeks in pediatric studies \[[@B4],[@B6]\]. A subsequent randomized phase II study of protracted versus 5 day schedule of irinotecan did not show any difference in response rates in rhabdomyosarcoma patients \[[@B5]\]. Another trial comparing the two regimens of oral irinotecan given with vincristine and temozolomide reported higher frequency of dose limiting toxicity in the protracted regimen \[[@B7]\]. Since the 5 day schedule is more convenient for patients, we used this schedule in our trial. The MTD for irinotecan administered as a single agent in a 5 day regimen ranges from 39-50 mg/m^2^ depending on the number of previous treatment regimens received \[[@B19]\]. Myelosuppression was dose limiting in heavily pretreated patients while diarrhea was dose limiting in less heavily pretreated patients. Irinotecan 50mg/m^2^ and temozolomide 150 mg/m^2^ administered over 5 days every 3-4 weeks has been studied in neuroblastoma patients, but this study used a lower platelet count threshold of 30,000/µL for administering subsequent cycles. Therefore, we decided to study escalating dose levels of irinotecan. Overall this regimen was tolerated well. There was no delay in therapy due to hematological toxicity. Similar to other studies with this backbone, the number of patients requiring platelet or blood transfusions was low \[[@B4]\]. Based on our experience in this study, routine use of myeloid growth factors may not be needed with this regimen. Even though we did not use prophylactic antibiotics, diarrhea was well controlled with loperamide, and only one patient developed grade 3 diarrhea. Majority of grade 3 and 4 toxicities described in [Table 3](#tab3){ref-type="table"} occurred in one patient with Noonan syndrome. We do not know if Noonan syndrome predisposed this patient to have more toxicity. The dose limiting hyperbilirubinemia is most likely attributable to irinotecan. Hyperbilirubinemia has been reported with the use of irinotecan in both single agent and combination pediatric studies \[[@B7],[@B19]\]. Known serious adverse effects of bevacizumab including severe hemorrhage, gastrointestinal perforation, arterial thromboembolism, posterior leukoencephalopathy and cardiac side effects were not seen. The number of cycles administered in this study may have been too few to detect these rare side effects that are reported in adult studies. Central nervous system hemorrhage, and transient leukoencephalopathy have been reported in children who received bevacizumab \[[@B16],[@B24]\]. The patient who developed hypertension requiring antihypertensive treatment in our study, had a history of bilateral nephron sparing surgery, which may have contributed to developing hypertension. Due to reversible physeal dysplasia seen in juvenile monkeys following bevacizumab administration \[[@B25]\], we performed serial imaging of growth plates in seven patients who had open growth plates. We did not detect any growth plate expansion. Even though this study was performed primarily to study toxicity, the antitumor activity of this combination is encouraging. Five of the 12 patients (42%) had objective responses. Two other patients had stable disease through 12 cycles. This combination showed significant activity in all 3 patients enrolled with relapsed Wilms tumor. All three patients were heavily pretreated and had a history of previous autologous bone marrow transplant and lung irradiation. This was unexpected as 17 patients with Wilms tumor who had received either single agent irinotecan or irinotecan and temozolomide in previous studies did not show a response \[[@B4],[@B19]--[@B21]\]. Blockade of vascular endothelial growth factor has been shown to cause regression of Wilms tumor in preclinical studies \[[@B26],[@B27]\]. Addition of bevacizumab to the chemotherapy backbone may explain the activity observed in Wilms tumor in our study. The MTD was irinotecan 50 mg/m^2^ on days 1-5 administered with vincristine 1.5 mg/m^2^ on days 1 and 8, temozolomide 100 mg/m^2^ and, bevacizumab 15mg/kg on day 1 every 21 days. This combination was tolerable and showed significant antitumor activity. This study supports additional investigation of this combination, particularly in patients with Wilms tumor. Our study can also serve as a template for adding other targeted therapies to the vincristine, irinotecan and temozolomide chemotherapy backbone. Supporting Information ====================== ###### TREND checklist for nonrandomized controlled trails. (DOC) (PDF) ###### Click here for additional data file. ###### Study Protocol. (DOC) (PDF) ###### Click here for additional data file. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RV LM RS MM. Performed the experiments: RV LM MM WM TD. Analyzed the data: RV LM RS. Wrote the manuscript: RV LM.

INTRODUCTION ============ Connective tissue disease (CTD) refers to disorders characterized by autoimmune-mediated damage associated with circulating auto-antibodies \[[@b1-kjim-2016-212],[@b2-kjim-2016-212]\] that target various body organs and cause symptomatic presentation. Diagnosis and treatment of CTD has been the focus of a significant amount of research. Pathological mechanisms associated with interstitial changes in lung connective tissue show similar histological, radiological, and clinical characteristics as idiopathic interstitial pneumonia. The overall incidence of CTD-associated interstitial lung disease (CTD-ILD) is 15% \[[@b3-kjim-2016-212]\]. Treatments for CTD-ILD are similar to treatments for CTD \[[@b4-kjim-2016-212]\], although there is no clear protocol in cases where the patient is not responsive to first-line CTD treatments. Additionally, treatment can be difficult if the etiology is unclear, and it is not always evident when a change in the current treatment may be required. There are several issues associated with the management of CTD-ILD. Evidence-based pharmacological treatments for CTD-ILD are not readily available, with the exception of ILD-associated systemic sclerosis (SSc). ILDs show a wide spectrum of histological manifestations \[[@b1-kjim-2016-212],[@b5-kjim-2016-212]\], ranging from fibrosis to inflammation, complicating the process of diagnosis and treatment. Optimal timing of treatment intervention has not been established. Finally, biomarker levels and the severity of extrathoracic and intrathoracic activity can be inconsistent. Due to the lack of large population studies and proven therapeutic drugs for CTD-ILD, there is no global guideline for treatment. This paper reviews the current approaches used in the treatment of CTD-ILD based on the type of CTD. DEFINITION OF CONNECTIVE TISSUE DISEASE-ASSOCIATED INTERSTITIAL LUNG DISEASE ============================================================================ CTD can affect the chest wall, pleura, vasculature, airways, and parenchyma \[[@b6-kjim-2016-212]\]. ILD is one of the most common and clinically significant manifestations of CTD \[[@b7-kjim-2016-212]\]. A large percentage of patients with CTD-ILD have a typically progressive and irreversible course that is characterized by inflammation or fibrosis of the lung parenchyma \[[@b6-kjim-2016-212]\]. A definitive diagnosis of CTD-ILD is difficult due to variations in pathological presentation and clinical findings ([Table 1](#t1-kjim-2016-212){ref-type="table"}). CTD-ILD is defined as ILD within the setting of well-defined CTD \[[@b8-kjim-2016-212],[@b9-kjim-2016-212]\], which involves cases of CTD that affect the lung parenchyma and present with respiratory symptoms. In addition to the possibility of being diagnosed with another presentation, new-onset ILD has no identifiable underlying cause and is regarded as idiopathic, although, in the process of differentiating the cause, the etiology of ILD is assumed to be an autoimmune process \[[@b10-kjim-2016-212]\]. The current definition fails to satisfy the rheumatological criteria for a definitive diagnosis of CTD \[[@b11-kjim-2016-212]\], and fails to affirm the findings of serological tests. CTD-ILD is also known by the term "interstitial pneumonia with autoimmune features (IPAF)" \[[@b12-kjim-2016-212]\]. It is difficult to provide a differential diagnosis of CTD-ILD with regard to drug toxicity and infection. A combination of careful physical examination and clinical expertise is required of the physician to accurately determine the CTD-ILD status of a patient. PREVALENCE OF CTD AND CTD-ILD ============================= The incidence of ILD from CTD increased from 4.46 per 100,000 person-years in 1995 to 2000 to 12.32 per 100,000 person-years in 2001 to 2005 \[[@b13-kjim-2016-212]\]. ILD can be identified in all types of CTD and is particularly common in rheumatoid arthritis (RA), SSc, and polymyositis and dermatomyositis (PM/DM) \[[@b1-kjim-2016-212]\]. Although the prevalence of all CTDs is not well known, RA is the most common CTD, affecting 0.5% to 2% of the general population in the USA ([Table 1](#t1-kjim-2016-212){ref-type="table"}) \[[@b14-kjim-2016-212]\]. The prevalence of RA in Korea was 0.27 % in 2008. The incidence of RA in 2008 was estimated at 42 per 100,000 in the general population of South Korea \[[@b15-kjim-2016-212]\]. The prevalence of SSc is 26 per 100,000 people in the USA \[[@b16-kjim-2016-212]\]. Sjögren's syndrome (SS) has a prevalence of about 3% in people over the age of 50 years \[[@b17-kjim-2016-212]\]. The prevalence of systemic lupus erythematosus (SLE) in the USA is estimated at 15 to 50 per 100,000 persons \[[@b18-kjim-2016-212]\]. According to the Korean nationwide data, the prevalence of SLE in Korea is 18.8 to 21.7 per 100,000 people \[[@b19-kjim-2016-212]\]. The prevalence of mixed connective tissue disease (MCTD) was 3.8 per 100,000 adults in Norway \[[@b20-kjim-2016-212]\]. PM and DM are very rare \[[@b18-kjim-2016-212]\]. The overall incidence of CTD-ILD is estimated at 15% ([Table 1](#t1-kjim-2016-212){ref-type="table"}) \[[@b3-kjim-2016-212],[@b21-kjim-2016-212]-[@b29-kjim-2016-212]\]. Results of a previous study indicated that the radiographic prevalence rate of subclinical ILD was 33% to 57% in CTD patients \[[@b30-kjim-2016-212]\]. Radiologic prevalence on high-resolution computerized tomography (HRCT) was 19% in RA \[[@b31-kjim-2016-212]\], 23% to 38% in PM/DM \[[@b3-kjim-2016-212]\], 30% in SLE \[[@b21-kjim-2016-212]\], 52% to 85% in MCTD \[[@b20-kjim-2016-212],[@b32-kjim-2016-212]\], 75% in SS \[[@b33-kjim-2016-212]\], and 70% to 90% in SSc \[[@b21-kjim-2016-212]\]. EVIDENCE FOR TREATMENT EFFICACY =============================== Treatment decisions should be determined by comparing the reversible treatment effects with the adverse drug reactions. Sufficient evidence exists only for the efficacy of cyclophosphamide (CYC), azathioprine, or rituximab in ILD patients with SSc (SSc-ILD). Systemic sclerosis ------------------ CYC is the only immunosuppressant drug that has been studied in a randomized, controlled trial in CTD-ILD patients. In the Scleroderma Lung Study I, oral CYC at a dosage of 2 mg/kg/day was administered for 1 year to patients, and resulted in significant improvements to forced vital capacity (FVC; 2.53%, *p* = 0.03). No significant differences in serious adverse events were reported compared to the placebo group \[[@b34-kjim-2016-212]\]. A randomized, controlled trial in Belgium compared oral CYC (2 mg/kg daily for 12 months followed by 1 mg/kg daily) against oral azathioprine (2.5 mg/kg daily for 12 months and then maintained on 2 mg/kg daily). After 12 months of treatment, FVC and diffusing capacity of the lung for carbon monoxide (D~Lco~) did not change in the CYC group, but showed a statistically significant worsening effect in the azathioprine group \[[@b35-kjim-2016-212]\]. In the Scleroderma Lung Study II, SSc patients with symptomatic ILD were randomized to the oral CYC group (2 mg/kg/day for 1 year followed by placebo) or the mycophenolate mofetil (MMF) group (up to 3,000 mg for 2 years) to assess its efficacy and safety. The FVC, transition dyspnea index (TDI), and modified Rodnan skin thickness scoring (MRSS) improved in both groups. FVC improvement was comparable in the two treatment groups, but there was a greater trend towards improvement of TDI and MRSS in the CYC group compared with the MMF group. The MMF group recorded fewer adverse events such as leukopenia, thrombocytopenia or incidences of weight loss \[[@b36-kjim-2016-212]\]. Adding rituximab to standard medication may improve FVC in SSc-ILD. According to a small, randomized trial (n = 14) \[[@b37-kjim-2016-212]\], there was a significant improvement in FVC in the rituximab group compared with the control group after 1 year (*p* = 0.0018). Low-dose corticosteroids and CYC (600 mg/m^2^) followed by maintenance with azathioprine did not show a significant improvement in FVC when compared against the placebo \[[@b38-kjim-2016-212]\]. We recommend oral CYC at a dosage of 2 mg/kg/day for 1 year, or MMF up to 1,500 mg twice daily for 2 years, as the first-line treatment for SSc-ILD. Adding rituximab to previous immunosuppressant drugs may be an effective therapy for SSc-ILD patients. Rheumatoid arthritis -------------------- Patients with RA can be treated with disease-modifying antirheumatic drugs (DMARD) in addition to other medications \[[@b39-kjim-2016-212]\]. In DMARD-naive RA patients, DMARD monotherapy is recommended. Methotrexate (MTX) is the preferred treatment, but sulfasalazine, hydroxychloroquine, or leflunomide may also be recommended. If disease activity remains moderate or high despite the use of DMARD, the American College of Rheumatology guidelines recommend either a combination of traditional DMARD usage, or the addition of a tumor necrosis factor (TNF) inhibitor as an adjunct therapy \[[@b40-kjim-2016-212]\]. RA-ILD patients lack specific adjunctive treatment options in addition to their traditional treatments. Although a high dose of prednisone has been used as a first-line treatment option in patients with RA-ILD \[[@b41-kjim-2016-212]\], there is insufficient evidence to support its efficacy and safety \[[@b42-kjim-2016-212]\]. Moreover, clinicians may hesitate to commence treatment in RA-ILD patients because DMARD and newer biologic agents may exacerbate ILD and induce opportunistic infection. Rituximab is effective and tolerated when added to MTX therapy in patients with active RA \[[@b43-kjim-2016-212]\]. Large, randomized, controlled trials evaluating the safety of rituximab in RA-ILD patients are limited. An open label pilot study with RA-ILD patients showed that FVC remained stable in most patients treated with rituximab in combination with MTX at week 48 \[[@b44-kjim-2016-212]\]. However, the study had a very small cohort of patients. MMF may stabilize or slightly improve lung volumes in patients with RA-ILD \[[@b45-kjim-2016-212]\]. Although MMF is safe and allows for a reduction in prednisone dosage, there is insufficient evidence to support treatment of RA-ILD with MMF owing to the small scale of the prospective cohort study. The stabilizing effects of MMF may be maintained over a median of 2.5 years of follow-up \[[@b46-kjim-2016-212]\]. Sjögren's syndrome ------------------ The majority of SS-ILD patients receive glucocorticoid therapy, antimalarials or immunosuppressive treatment \[[@b47-kjim-2016-212]\]. Corticosteroids administered with hydroxychloroquine, azathioprine or CYC resulted in improved FVC and D~Lco~ in SS-ILD patients with histologically proven nonspecific interstitial pneumonia \[[@b48-kjim-2016-212],[@b49-kjim-2016-212]\]. Patients treated with azathioprine or azathioprine-prednisone also show a significant improvement in FVC after at least 6 months of treatment compared with non-treated patients \[[@b50-kjim-2016-212]\]. There are numerous effects of rituximab on serum B cell biomarker changes in patients with systemic complications of SS. However, there are few studies assessing the effects of rituximab on lung function in SS-ILD \[[@b51-kjim-2016-212]-[@b53-kjim-2016-212]\]. Mixed connective tissue disease ------------------------------- Most patients diagnosed with MCTD were extremely responsive to corticosteroid therapy. Some patients who were non-responsive ("non-responders") may progress to a stage of severe organ involvement \[[@b54-kjim-2016-212]\]. Conventional therapies for MCTD-ILD include a combination of corticosteroids with steroid-sparing agents \[[@b1-kjim-2016-212]\]. Corticosteroids, CYC, hydroxychloroquine, MTX and different types of vasodilators have also been used \[[@b55-kjim-2016-212]\]. Polymyositis and dermatomyositis -------------------------------- A monthly intravenous pulse of CYC improved symptoms, vital capacity, and HRCT in progressive interstitial pneumonia in PM/DM in an open-label study \[[@b56-kjim-2016-212]\]. Azathioprine and MMF were also associated with the stabilization of pulmonary physiology, improved dyspnea, and a reduction in steroid dosage. There were no significant differences in these outcomes among the patients who received CYC, azathioprine, and MMF treatments \[[@b57-kjim-2016-212]\]. A systematic review of CYC usage in idiopathic inflammatory myopathies (IIM) and IIM-ILD showed improvements in vital capacity or FVC (42/59) and HRCT findings (40/52) for the majority of patients \[[@b58-kjim-2016-212]\]. In contrast, non-responders reported a worsening of their condition despite treatment with a high dose of prednisolone in combination with CYC pulses, cyclosporin A, or azathioprine as an adjunct therapy \[[@b59-kjim-2016-212],[@b60-kjim-2016-212]\]. Even early recognition of acute interstitial pneumonia followed by an immediate course of intensified immunosuppressants had limited efficacy \[[@b61-kjim-2016-212]\]. Systemic lupus erythematosus ---------------------------- The European League Against Rheumatism (EULAR) recommends hydroxychloroquine or chloroquine as the first-line treatment option in patients with SLE \[[@b62-kjim-2016-212]\]. If indicated, initial non-steroidal anti-inflammatory drugs and/or glucocorticoids are recommended. Immunosuppressive agents are used in the treatment of severe refractory cases \[[@b63-kjim-2016-212]\]. The treatment for SLE-ILD typically includes corticosteroids along with steroid-sparing agents. The chosen treatment strategy is based on expert opinion. Highdose corticosteroids and a steroid-sparing agent, often CYC, are initial treatment options for severe SLE-ILD. Corticosteroids with either azathioprine or MMF have been used for mild to moderate cases \[[@b64-kjim-2016-212]\]. Systemic corticosteroids may stabilize inspiratory vital capacity and D~Lco~ in the majority of patients with SLE-ILD, although evidence for its efficacy is limited \[[@b65-kjim-2016-212]\]. OPTIMAL TIMING FOR TREATMENT ============================ Indications for treatment with CTD-ILD patients ----------------------------------------------- Although there is no consensus for the optimal timing and duration of treatment for CTD-ILD, clinically significant (severe, extensive, or progressive) CTD-ILD is commonly treated with immunomodulatory agents ([Table 2](#t2-kjim-2016-212){ref-type="table"}) \[[@b66-kjim-2016-212]\]. What is the 'optimal' timing for treatment in patients with SSc-ILD? -------------------------------------------------------------------- Patients with CTD may experience involvement of the lung parenchyma. Early detection of CTD-ILD is possible through close monitoring for the detection of newly formed lesions on HRCT and a decline in pulmonary function. Serial evaluation by history taking, physical examination, chest X-ray, pulmonary function test, and HRCT are helpful in the early recognition of CTD-ILD, allowing for an adaptive change in medications upon progression ([Table 2](#t2-kjim-2016-212){ref-type="table"}) \[[@b21-kjim-2016-212],[@b67-kjim-2016-212]\]. Is it beneficial to start treatment 'early'? -------------------------------------------- A higher disease extent (reticular changes in HRCT \> 20%) or lower FVC (FVC \< 70%) are considered risk factors predisposing patients to poor survival outcomes \[[@b68-kjim-2016-212]\]. An analysis of the Scleroderma Lung Study data \[[@b69-kjim-2016-212]\] showed that patients with less severe baseline HRCT findings may have relatively stable lung function over time, and treatment with CYC does little to improve FVC changes in those patients. Severe HRCT findings are an independent predictive factor of responsiveness to CYC. In patients with an HRCT disease extent of \> 50%, the average CYC treatment effect was a 9.81% difference in FVC at 18 months compared to the placebo group (*p* \< 0.001) \[[@b69-kjim-2016-212]\]. Therefore, patients with severe baseline HRCT findings may be responsive to CYC treatment. Optimal timing to commence treatment in patients with SSc-ILD is recommended at an HRCT disease extent of \> 20% or a FVC below 70%. The expected therapeutic yield should be greater than the risk of adverse effects associated with immunosuppression, such as opportunistic infections and drug toxicities. PROMISING TREATMENTS FOR CTD-ILD ================================ Calcineurin inhibitors in DM/PM-ILD: tacrolimus ----------------------------------------------- A retrospective study in Japanese patients with PM/DM-ILD showed significant improvements in eventfree survival in patients who received the conventional therapy of prednisone with intravenous CYC or cyclosporin and tacrolimus compared to those who received conventional therapy alone \[[@b70-kjim-2016-212]\]. Imatinib in SSc-ILD ------------------- Open-label pilot studies in SSc-ILD patients demonstrate stabilization of lung function. However, the use of imatinib is associated with high rates of patient withdrawal due to the severity of adverse complications, which include gastrointestinal distress, rash, and renal dysfunction. As many as 40% of enrolled patients withdrew in one study investigating the use of imatinib at a dose of up to 600 mg/day in 20 patients with scleroderma-ILD \[[@b71-kjim-2016-212],[@b72-kjim-2016-212]\]. Target tumor necrosis factor: infliximab and etanercept ------------------------------------------------------- Infliximab and etanercept are often regarded as suitable options for patients with RA-ILD. However, patients with RA-ILD may develop new ILD associated with the use of etanercept \[[@b73-kjim-2016-212]\]. Between 1990 and 2010, there were 122 reported cases of new-onset or exacerbated ILD that developed as a secondary event to the administration of biologic therapies (etanercept in 58 cases and infliximab in 56 cases) \[[@b74-kjim-2016-212]\]. These agents should be used with caution in patients with CTD-ILD \[[@b7-kjim-2016-212],[@b75-kjim-2016-212]\]. Targeting antigen-presenting cells: abatacept --------------------------------------------- A case series with abatacept in patients who developed ILD or whose ILD deteriorated while on anti-TNF-α therapy had no reports of exacerbation or new-onset ILD following administration of abatacept \[[@b76-kjim-2016-212]\]. Targeting interleukin 6 receptor: tocilizumab --------------------------------------------- Other biologic agents inhibiting the interleukin 6 receptor, such as tocilizumab, may have a role in the future treatment of CTD-ILD. A case report in which tocilizumab was used for an RA patient with a history of MTX-associated pneumonitis, with previous MTX and leflunomide treatment, showed an improvement in the disease activity score and pulmonary functions following tocilizumab treatment \[[@b77-kjim-2016-212]\]. Pirfenidone, nintedanib ----------------------- Improvements in the vital capacity of patients with scleroderma-ILD following pirfenidone usage have been reported \[[@b78-kjim-2016-212]\]. An open label, randomized, parallel group study for the safety and tolerability of pirfenidone in scleroderma-ILD has been completed recently and similar studies are planned for nintedanib for CTD-ILD patients \[[@b7-kjim-2016-212]\]. We expect that these drugs will have a similar function in CTD-ILD patients to that they would in in idiopathic pulmonary fibrosis. Immunoglobulins --------------- Myositis associated with SSc with early-stage ILD has been treated with intravenous immunoglobulins (2 g/kg/month) and azathioprine (150 mg/day). Ground glass opacities, septal thickenings, and lung function improved following treatment \[[@b79-kjim-2016-212]\]. Stem cell transplantation ------------------------- Autologous hematopoietic stem cell transplantation (HSCT) has been studied in patients with severe refractory scleroderma \[[@b80-kjim-2016-212]\]. An open label parallel group study comparing the safety and efficacy of HSCT treatment and an intravenous pulse of CYC for 12 months showed that HSCT was associated with an increased treatment-related mortality (eight treatment-related deaths) following treatment in 156 patients with early diffuse scleroderma. However, patients undergoing HSCT showed a significant improvement in FVC and total lung capacity. HSCT treatment in scleroderma shows promise with respect to long term survival benefits in patients with SSc-ILD \[[@b81-kjim-2016-212]\]. Lung transplantation -------------------- Lung transplantation is considered the final option in the management of CTD-ILD \[[@b82-kjim-2016-212]\]. Between 1995 and 2010, fewer than 2% of all lung transplants worldwide were given to patients with CTD-ILD \[[@b83-kjim-2016-212]\]. According to a Nationwide Cohort Study in the United States \[[@b84-kjim-2016-212]\], SSc patients are considered to be an at-risk group with a high 1-year postoperative mortality rate. This risk is similar to that observed in patients with pulmonary arterial hypertension (PAH): the mortality rates are 21.37, 19.04, and 17.82 per 100 person-years for SSc, PAH, and ILD patients, respectively. Therefore, SSc-ILD may be considered an acceptable indication for lung transplantation. MULTIDISCIPLINARY DISCUSSION ============================ Even in cases of well-defined CTD, it is difficult to assess disease progression and how this may impact the tailoring of treatment options. It is important to delineate between the progression of CTD-ILD and other health effects such as respiratory infection, hypersensitivity, drug-associated pneumonitis, environmental exposure, malignancy, or aspiration-induced lung injury \[[@b85-kjim-2016-212]\]. If a definitive diagnosis is not determined in the initial clinical assessment, patients may be diagnosed with unclassifiable ILD (U-ILD). Subsequently, the patients are assessed by a rheumatologist and are classified among the three series of U-ILD: undifferentiated connective tissue disease, IPAF, and U-ILD \[[@b8-kjim-2016-212]\]. Treatment and management options available for CTD-ILD can be improved using a multidisciplinary approach that includes rheumatological collaboration. Multidisciplinary teams should consider intrathoracic and extrathoracic disease activity. Similar to the approach taken with the diagnosis of idiopathic interstitial pneumonia \[[@b86-kjim-2016-212]\], a clinical-radiological-pathological team approach can be used to improve the decision-making processes involved in CTD-ILD \[[@b8-kjim-2016-212],[@b12-kjim-2016-212],[@b87-kjim-2016-212]\]. CONCLUSIONS =========== CTD-ILD is typically defined as a progressive lung parenchymal manifestation of CTD. Although there is no consensus regarding the optimal timing and duration for treatment of CTD-ILD, clinically severe, extensive, or progressive CTD-ILD is commonly treated with immunomodulatory agents. Multidisciplinary approaches that involve pulmonologists, rheumatologists, radiologists, and pathologists in treatment decisions are necessary for optimal outcomes. No potential conflict of interest relevant to this article was reported. This work was supported by Soonchunhyang University Research Fund. ###### Clinical features of patients with CTD-ILD Type of connective tissue disease ------------------------------------------------------------- ------------------------------------------------------------------------------- ------------------------------------------------------------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------- CTD prevalence 26^[a](#tfn1-kjim-2016-212){ref-type="table-fn"}^ 0.5%--2% of the general population \[[@b14-kjim-2016-212]\] 3% in \> age of 50 years 3.8^[a](#tfn1-kjim-2016-212){ref-type="table-fn"}^ Unknown 15%--50^[a](#tfn1-kjim-2016-212){ref-type="table-fn"}^ Clinical ILD^[b](#tfn2-kjim-2016-212){ref-type="table-fn"}^ ≤ 45% \[[@b21-kjim-2016-212]\]^[c](#tfn3-kjim-2016-212){ref-type="table-fn"}^ 7.7% \[[@b22-kjim-2016-212]\]^[d](#tfn4-kjim-2016-212){ref-type="table-fn"}^ 11%--15% \[[@b23-kjim-2016-212]\]^[e](#tfn5-kjim-2016-212){ref-type="table-fn"}^ \[[@b24-kjim-2016-212]\]^[f](#tfn6-kjim-2016-212){ref-type="table-fn"}^ 54% \[[@b25-kjim-2016-212]\]^[g](#tfn7-kjim-2016-212){ref-type="table-fn"}^ 15%--78% \[[@b21-kjim-2016-212],[@b26-kjim-2016-212]-[@b28-kjim-2016-212]\]^[h](#tfn8-kjim-2016-212){ref-type="table-fn"}^ 11% \[[@b29-kjim-2016-212]\]^[i](#tfn9-kjim-2016-212){ref-type="table-fn"}^ HRCT abnormality 70%--90% \[[@b21-kjim-2016-212]\] 19% \[[@b31-kjim-2016-212]\] 75% of asymptomatic patients \[[@b33-kjim-2016-212]\] 52%--85% \[[@b20-kjim-2016-212],[@b32-kjim-2016-212]\] 23%--61% \[[@b3-kjim-2016-212],[@b27-kjim-2016-212]\] ≤ 30% \[[@b21-kjim-2016-212]\] Common ILD type NSIP UIP NSIP NSIP NSIP NSIP UIP NSIP LIP, OP, UIP, DIP OP, UIP, DAD, LIP LIP, OP, UIP OP, DIP AIP, DIP ILD-cause mortality Unknown 10%--20% \[[@b42-kjim-2016-212]\] 5-Year survival: 84% \[[@b49-kjim-2016-212]\] Unknown From subclinical to rapidly progressive and fatal \[[@b61-kjim-2016-212]\] 50% \[[@b63-kjim-2016-212]\] CTD, connective tissue disease; ILD, interstitial lung disease; SSc, systemic sclerosis; RA, rheumatoid arthritis; SS, Sjogren's syndrome; MCTD, mixed connective tissue disease; PM, polymyositis; DM, dermatomyositis; SLE, systemic lupus erythematosus; HRCT, high-resolution computerized tomography; NSIP, nonspecific interstitial pneumonia; UIP, usual interstitial pneumonia; OP, organizing pneumonia; DIP, desquamative interstitial pneumonia; LIP, lymphocytic interstitial pneumonia; DAD, diffuse alveolar damage; AIP, acute interstitial pneumonitis. Number of cases per 100,000 persons. Clinical ILD is defined as patients who had radiological abnormality and respiratory symptoms and/or impaired lung function related to CTD-ILD in this study. Definition of the CTD-ILD from each study were as follows: Defined by moderate-to-severe restriction on pulmonary function test (PFT) or moderate-to-severe lung involvement on HRCT \[[@b21-kjim-2016-212]\]. Probable ILD is defined as radiological report containing terms such as "pulmonary fibrosis," "fibrotic changes," "fibrosis," "RA-lung," "fibrosing alveolitis," and presence of nonspecific abnormalities that can be observed in ILD; Treating physician's diagnosis of "pulmonary fibrosis," "RA-lung," "fibrosing alveolitis," or other terms in the medical record consistent with ILD \[[@b22-kjim-2016-212]\]. Radiological abnormality with impaired pulmonary function \[[@b23-kjim-2016-212]\]. Presence of pulmonary signs/symptoms, and/or impaired PFT and pathological HRCT findings \[[@b24-kjim-2016-212]\]. Abnormal HRCT findings and symptomatic/clinical findings \[[@b25-kjim-2016-212]\]. Findings on radiographic examination and/or PFT compatible with ILD \[[@b28-kjim-2016-212]\]. The presence of reticular or interstitial opacities or honeycombing on chest imaging \[[@b29-kjim-2016-212]\]. ###### Treatment strategies according to the type of CTD Type of connective tissue disease --------------------------------------------------------------- -------------------------------------------------------------------------------------- --------------------------------------------------------- -------------------------------------------- ---------------------------------------- -------------------------- ------------------------------------------------- First line treat ment choice for limited CTD Digital vasculopathy: nifedipine/iloprost \[[@b4-kjim-2016-212]\] DMARD Stimulation of salivary secretions (ssica) Corticosteroid Corticosteroid + AZA/MMF Hydroxychloroquine or chloroquine Antimalarial agents (extraglandular) First choice for clinical CTD-ILD CYC 2 mg/kg/day or MMF up to 3,000 mg \[[@b36-kjim-2016-212]\] High-dose PD Glucocorticoid, antimalarial agent Corticosteroids + cytotoxic drug (CYC) CYC Corticosteroid + AZA/MMF MMF AZA AZA+ PD MMF Follow-up interval with PFT/D~Lco~, chest X-ray or HRCT Check PFT/D~Lco~: every 6--12 mo Initiation of MTX: check chest radiograph within 1 year NA NA Check PFT NA After progression: every 3--4 mo \[[@b67-kjim-2016-212]\] Stable disease: every 6 mo RA-ILD: PFT/D~Lco~ with HRCT 3--6 mo \[[@b21-kjim-2016-212]\] Progression: every 3--4 mo \[[@b21-kjim-2016-212]\] Refractory CTD-ILD Add rituximab (375 mg/m^2^) at 4-week interval for 24 weeks \[[@b37-kjim-2016-212]\] Rituximab NA NA High-dose PDL + CYC High-dose steroid + steroid-sparing agent (CYC) Rescue therapy Lung transplantation \[[@b81-kjim-2016-212]\] Lung transplantation NA NA NA NA CTD, connective tissue disease; SSc, systemic sclerosis; RA, rheumatoid arthritis; SS, Sjogren's syndrome; MCTD, mixed connective tissue disease; PM, polymyositis; DM, dermatomyositis; SLE, systemic lupus erythematosus; DMARD, disease-modifying antirheumatic drug; AZA, azathioprine; MMF, mycophenolate mofetil; ILD, interstitial lung disease; CYC, cyclophosphamide; PD, prednisone; PFT, pulmonary function test; D~Lco~, diffusing capacity of the lung for carbon monoxide; HRCT, high-resolution computerized tomography; MTX, methotrexate; NA, not applicable; PDL, prednisolone.

1. Introduction {#sec1-sensors-19-04304} =============== Sensing and quantifying damage plays a critical role in the process of structural health monitoring, which aims to detect structural damage and provide early warnings when a possible risk of failure is detected. Many structural health monitoring systems employ accelerometers, displacement sensors, or piezoelectric sensors located at selected locations to monitor changes in the structure's deformation, natural frequencies, and modal shapes \[[@B1-sensors-19-04304],[@B2-sensors-19-04304]\]. These systems then evaluate possible failure modes, damage levels, and locations. While accelerometers are typically employed for beam-column-based structures such as buildings, these are not the optimal sensors for structures whose failure modes are insensitive to the structure's vibration characteristics. For some types of structures such as dams, tunnels, and reinforced concrete vessels, or shear-critical components such as reinforced concrete (RC) walls, the detection and evaluation of cracks is a relatively practical approach for safety assessment and monitoring. Several structural damage indices have been proposed. Park et al. \[[@B3-sensors-19-04304]\] proposed a damage index for a structural system according to its largest system displacement, ultimate displacement, accumulated strain energy, cyclic loading effect, and system yield force and displacement. Based on the calculated damage index, the structural system can be classified into one of the following damage levels: slight, minor, moderate, severely damaged, and collapsed. Roufaiel and Meyer \[[@B4-sensors-19-04304]\] proposed a damage index that uses the initial stiffness, current stiffness, and failure stiffness. Powell and Allahabadi \[[@B5-sensors-19-04304]\] proposed an index based on the current displacement, yield displacement, and ultimate displacement. These damage indices consider a structure as a single-degree-of-freedom system to simplify damage level estimations. However, in practical applications, these damage indices are difficult to use, as the stiffness and the displacement of a structure is sometimes difficult to measure for real, multiple degrees of freedom, and partially damaged structures. Detailed structural performance and safety may require advanced structural analyses based on finite element analysis tools \[[@B6-sensors-19-04304],[@B7-sensors-19-04304]\] or structural experiments \[[@B8-sensors-19-04304],[@B9-sensors-19-04304]\] which are specific to a certain type of structure. For the purpose of structural health monitoring, the displacements of certain locations can be monitored by pre-installed displacement devices; however, current stiffness and other structural properties are difficult to accurately measure or estimate. Alternatively, for easy to implement and quick structural safety assessments of reinforced concrete (RC) structures, several evaluation methods have been proposed that instead consider the surface cracks of concrete structures. The Japan Building Disaster Prevention Association (JBDPA) provides a guide based on the visible cracks in the concrete surface of beams, columns, or walls, and categorizes damage into five classes according to the maximum opening width of the cracks \[[@B10-sensors-19-04304]\]. According to the JBDPA criterion, structures with a maximum crack width larger than 0.2 mm, 1 mm, and 2 mm are categorized as showing light damage, moderate damage, and heavy damage classes, respectively. The International Atomic Energy Agency (IAEA) uses a more conservative standard that categorizes cracks with an opening width larger than 0.2 mm and 1 mm as moderate and severe damage, respectively \[[@B11-sensors-19-04304]\]. The bridge inspector's reference manual, published by US Department of Transportation \[[@B12-sensors-19-04304]\], categorizes cracks into structural cracks, flexural cracks on a tee beam, shear cracks on a slab, temperature cracks, shrinkage cracks, longitudinal cracks, etc. For surface damage detection and evaluation, image-based measurement is an automatic and cost-efficient method in terms of hardware cost. As the aforementioned structural health monitoring or damage detection methods have different features, advantages, and limits, no single method can be used to replace another, nor can it be used as the sole means of structural health monitoring or damage detection. Image-based measurements, and their potential for damage detection, are not intended to replace any of the aforementioned methods. Instead, the image-based method aims to provide an area-based measurement method to measure or monitor cracks \[[@B13-sensors-19-04304]\], strain fields \[[@B14-sensors-19-04304],[@B15-sensors-19-04304]\], multi-axial displacement \[[@B16-sensors-19-04304]\], or structural vibrations \[[@B17-sensors-19-04304]\], where technology for conventional displacement measurements is inadequate \[[@B18-sensors-19-04304]\]. The hardware cost may be relatively low \[[@B19-sensors-19-04304]\], and may even employ existing surveillance cameras in the structure, thus eliminating the need to install additional cameras \[[@B20-sensors-19-04304]\]. With recent dramatic improvements in digital image processing techniques, image analysis algorithms, accuracies, reliability, and computing speed have improved as well; thus, image measurement has a strong potential for practical structural health monitoring applications \[[@B21-sensors-19-04304]\]. This work develops an image analysis-based damage indexing method following a previously developed image-based crack measurement method. This method is tested using two cyclic tests of RC containment vessels \[[@B22-sensors-19-04304]\]. The vessels are shear critical with a large number of shear cracks induced by only a small displacement. A fractal dimension method \[[@B23-sensors-19-04304]\] is modified and employed in this work to quantify the number of cracks. Based on the number of cracks, as well as their opening widths and orientations, a method for calculating damage indices is proposed. This method modifies the previous image analysis method \[[@B24-sensors-19-04304]\], such that concrete surface crack orientations can be determined automatically. In addition, the fractal dimension crack analysis method \[[@B23-sensors-19-04304]\] is modified so that the damage index can be separated into a shear damage index and a flexural damage index to distinguish between the different types of failure. The combination of these methods will make it possible to carry out structural health monitoring in an automatic manner in practical applications in the future. This paper further demonstrates the image measurement and damage indices calculation procedure based on the aforementioned RC containment vessel experiments. 2. Image Measurement of Cracks on Concrete Surfaces {#sec2-sensors-19-04304} =================================================== Image-based monitoring and damage identification consists of two major procedures: image measurements and damage quantification. Image measurements analyze the image(s) of the measurement regions of interest and provide details, such as locations, lengths, opening widths, sliding displacements, and the orientation of the cracks. The damage evaluation procedure estimates the damage level or index of the measurement region according to the analyzed results from the image measurement. Many image measurement algorithms and methods have been proposed to detect cracks on measurement regions, such as on concrete surfaces or pavements. These methods can be classified into two groups: (1) edge detection-based methods, and (2) displacement field-based methods. Edge detection-based methods are capable of finding cracks that appear as dark lines in an image. The cracks need to be of sufficient width to appear as dark lines, which is theoretically the width of a pixel. Edge detection methods \[[@B25-sensors-19-04304],[@B26-sensors-19-04304],[@B27-sensors-19-04304],[@B28-sensors-19-04304]\] or machine learning methods \[[@B29-sensors-19-04304],[@B30-sensors-19-04304],[@B31-sensors-19-04304],[@B32-sensors-19-04304]\] are typically employed to identify the locations or widths of cracks. A review of crack detection methods can be found in \[[@B33-sensors-19-04304]\]. Alternatively, the displacement field-based method identifies cracks according to the displacement field of the measurement region, where the displacement field is analyzed by image analysis techniques \[[@B24-sensors-19-04304],[@B34-sensors-19-04304]\]. Due to the high precision of image-based displacement field measurements, displacement field-based methods are capable of detecting thin cracks with widths of much less than one pixel. Yang et al. \[[@B34-sensors-19-04304]\] detected cracks as thin as 0.2 pixels in photos in an outdoor experiment where images contained environmental light noise. The same image analysis technique detected thin cracks whose width was equivalent to 0.03 pixels in photos in a structural laboratory \[[@B24-sensors-19-04304]\]. This type of method estimates the cracks' opening widths, sliding displacements, and orientations, according to the change in the displacement field between each set of photos taken before and after cracks occurred, respectively. Thus, the first set of photos is used as a reference for the displacement field. Compared with edge detection-based methods, displacement field-based crack detection methods are suitable for thin crack detection, monitoring the early stages of crack development, or monitoring large regions where pixels are relatively coarsened. However, it should be mentioned that most edge detection-based methods used are tailored for inspection, rather than health monitoring. They are more suitable for that purpose than displacement field methods. In addition, displacement field methods tend to be more computationally expensive. This work employs a displacement field-based method for crack measurement. However, this does not mean that edge detection-based methods cannot be applied to the damage evaluation method proposed in this work. The displacement field-based method is employed here because it is capable of detecting thin cracks that occur in the early stages of structural damage. In addition, the image measurement software, ImPro Stereo, is publicly available on the internet \[[@B35-sensors-19-04304]\], and is further integrated with the damage evaluation computer codes developed in this work. The displacement field-based method for crack measurement includes five main steps: camera calibration, measurement region positioning, metric rectification, displacement field analysis, and crack analysis. A detailed procedure can be found in \[[@B34-sensors-19-04304],[@B36-sensors-19-04304]\]. This work only focuses on the analysis results as related to the follow-up damage evaluation procedure which is proposed herein. Camera calibration is the process of finding the intrinsic and extrinsic parameters of the camera. The intrinsic parameters are essentially its optical properties, such as the fields of view and optical distortion coefficients. The extrinsic parameters describe the camera position and its orientation. Typically, the camera calibration process is only carried out once, by taking more than 10 pairs of photos of calibration objects (such as a chessboard of known size) during camera installation (see [Figure 1](#sensors-19-04304-f001){ref-type="fig"}). Measurement region positioning tracks the updated position of the measurement by precisely tracking the 3D positions of the control points that are used to define the measurement surface. Defining an ideal planer rectangle measurement requires at least three control points, while a cylindrical measurement region requires at least four, as shown in [Figure 2](#sensors-19-04304-f002){ref-type="fig"}. The positions of control points P1 to P4 describe the movement and deformation of the overall measurement region. Details of the process can be found in \[[@B24-sensors-19-04304]\]. The image rectification process generates a rectangular image that represents the image pattern on the measurement region. The perspective and lens distortion effects are removed during this process. The metric rectified image can be seen as an expanded planer surface of the measurement region so that the ratio of a pixel to its physical length is constant over the entire measurement region; thus, it is essentially an image that represents the unfolded plane from the measurement region. The constant pixel-to-physical length ratio is an important property for the subsequent displacement field-based crack analysis. The rectified image is generated pixel-by-pixel, while the image intensity of each pixel is estimated by mathematically projecting a 3D point onto the surface to its image position in the photo according to the intrinsic and extrinsic parameters of the camera. Its image intensity is acquired through the numerical interpolation of neighboring pixels, as shown in [Figure 3](#sensors-19-04304-f003){ref-type="fig"}. The displacement fields of the measurement region can be estimated by comparing the initial and current rectified images (see [Figure 4](#sensors-19-04304-f004){ref-type="fig"}a,b) using an object tracking method, such as template matching, digital image correlation, an enhanced correlation coefficient, or the optical flow method. Details of the process can be found in \[[@B24-sensors-19-04304]\]. The example presented in [Figure 4](#sensors-19-04304-f004){ref-type="fig"} was obtained from an experiment that had a measurement region of approximate dimensions of 1.4 m × 0.9 m. Each rectified image in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}a is approximately 2400 × 1600 pixels. The displacement field in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}b is a vector field with 90 × 60 cells, that is, each cell is represented by a sub-image with a size of 27 × 27 pixels (rounded from 2400 / 90 = 26.67). The crack opening in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}c is a scalar field with the same refinement. The refinement is assigned by users, and should be tuned according to the image quality of photos when this method is being applied in practical applications. The displacement field of the rectified images is obtained by optical flow analysis \[[@B37-sensors-19-04304]\]. The resolution of the rectified images and the refinement of the displacement and crack fields are adjusted by the user, and typically depend on the resolution and quality of the experimental photos. Crack analysis converts a displacement field to a crack distribution. Crack analysis is suitable for thin cracks that are too thin to display as a dark line in photos, thus requiring the use of the displacement field to estimate the crack's opening width. Each cell of the crack opening width $c_{o}$ (see [Figure 4](#sensors-19-04304-f004){ref-type="fig"}c) and crack sliding displacement $c_{s}$ of any arbitrary cell in the grid is estimated according to the displacement of its four neighboring cells. Crack sliding is the relative displacement of part A with respect to part B, i.e., parallel to the crack orientation. By using the formulation presented in \[[@B24-sensors-19-04304]\], as shown in Equations (1)--(3), the crack distribution can be estimated by a displacement field. The crack analysis method is only suitable for brittle materials such as concrete, as it assumes that the deformation in the displacement field is mainly caused by cracks, rather than strains \[[@B34-sensors-19-04304]\]. In addition, since the image is the appearance of the material surface, it does not represent the crack opening or sliding under beneath the surface; these are the limitations of this method. The crack distribution is a field of crack opening widths, sliding displacements, and crack orientations. It is discretized to a grid with the same grid density as the displacement. Each cell of the crack opening width $c_{o}$ and crack sliding displacement $c_{s}$ of any arbitrary cell in the grid can be calculated by Equations (1)--(3). $$\begin{pmatrix} c_{o} \\ c_{s} \\ \end{pmatrix} = \begin{pmatrix} {\cos\theta} & {\sin\theta} \\ {- \sin\theta} & {\cos\theta} \\ \end{pmatrix}\left( {\mathbf{u}_{\mathbf{A}} - \mathbf{u}_{\mathbf{B}}} \right)$$ where $$\mathbf{u}_{\mathbf{A}} = \left\{ \begin{matrix} {\frac{\mathbf{u}_{\mathbf{U}} \cdot \left| {\cos\theta} \right| + \mathbf{u}_{\mathbf{L}} \cdot \left| {\sin\theta} \right|}{\left| {\cos\theta} \right| + \left| {\sin\theta} \right|},~{if}~0 \leq \theta < 0.5\pi} \\ {\frac{\mathbf{u}_{\mathbf{D}} \cdot \left| {\cos\theta} \right| + \mathbf{u}_{\mathbf{L}} \cdot \left| {\sin\theta} \right|}{\left| {\cos\theta} \right| + \left| {\sin\theta} \right|},~{if}~0.5\pi \leq \theta < \pi} \\ \end{matrix} \right.$$ $$\mathbf{u}_{\mathbf{B}} = \left\{ \begin{matrix} {\frac{\mathbf{u}_{\mathbf{D}} \cdot \left| {\cos\theta} \right| + \mathbf{u}_{\mathbf{R}} \cdot \left| {\sin\theta} \right|}{\left| {\cos\theta} \right| + \left| {\sin\theta} \right|},~{if}~0 \leq \theta < 0.5\pi} \\ {\frac{\mathbf{u}_{\mathbf{U}} \cdot \left| {\cos\theta} \right| + \mathbf{u}_{\mathbf{R}} \cdot \left| {\sin\theta} \right|}{\left| {\cos\theta} \right| + \left| {\sin\theta} \right|},~{if}~0.5\pi \leq \theta < \pi} \\ \end{matrix} \right.$$ $\mathbf{u}_{\mathbf{U}}$, $\mathbf{u}_{\mathbf{D}}$, $\mathbf{u}_{\mathbf{L}}$, and $\mathbf{u}_{\mathbf{R}}$ are the displacement vectors of the upper, lower, left, and right neighboring cells of any arbitrary cell in the displacement field, respectively (see [Figure 5](#sensors-19-04304-f005){ref-type="fig"}). The orientation of the crack of the analyzed cell is determined by iteratively testing $\theta$ within 0 and 180 degrees with a step of 15 degrees (i.e., 0, 15, 30, 45, ..., 165 degrees). To be conservative, the $\theta$ which leads to the largest crack opening is selected in this method. If there is no crack on the cell, $c_{s}$ and $c_{o}$ would be very small compared with those with cracks. Small values of $c_{s}$ and $c_{o}$ are caused by either noise, image analysis errors, or relatively small strains, and are ignored in the crack analysis. [Figure 4](#sensors-19-04304-f004){ref-type="fig"}c demonstrates the discretized grid of a crack pattern estimated from its displacement in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}b. It should be noted that the size scale in [Figure 5](#sensors-19-04304-f005){ref-type="fig"} is only for demonstration. A crack is typically much thinner than the size of a cell. The cracks shown in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}c are actually as thin as 0.02--0.2 mm, i.e., much thinner than the size of a cell in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}b,c. In [Figure 4](#sensors-19-04304-f004){ref-type="fig"}c, the size of a cell is equivalent to a 27 × 27-pixel sub-image. While a 0.02-mm crack can be recognized by the naked eye at a close distance when inspecting damage in structural experiments, it cannot be recognized by most of the edge detection-based methods, as the crack is typically too thin to appear as a dark line in photos. In addition, human inspection is not practical for automatic structural health monitoring. 3. Damage Indices based on Image Analysis of Cracks {#sec3-sensors-19-04304} =================================================== The quantification of cracks in this work is based on a Fractal Analysis of Cracks (FAC) \[[@B23-sensors-19-04304]\]. The quantification of the total length of cracks within a measurement region can be scale dependent; the smaller the scale and the more refined the crack pattern, the more likely it is that a longer total length of cracks would be measured. A typical scale-dependent example is the measurement of a coastline, which depends on the measurement scale. This method aims to quantify the number of cracks in a more objective and scale-invariant manner, rather than directly measuring the total lengths of cracks. The FAC method adopts a fractal analysis as a benchmark method to quantify a crack by estimating its fractal dimension. While mathematically, a line is one-dimensional and a filled rectangle is two-dimensional, the dimensions of a crack distribution over a measurement region are typically a real number between 1 and 2, and do not need to be an integer. The FAC method quantifies a crack by its fractal dimension. The details of FAC can be found in \[[@B23-sensors-19-04304]\]. The crack analysis method proposed in this paper modifies the FAC method. The main modifications made in this work include the following: (1)The crack data for FAC is based on a hand-sketched crack pattern. The crack data for the modified FAC is based on an image analyzed crack pattern.(2)The modified FAC is capable of differentiating between the damage induced by shear cracks and that of flexural cracks according to crack orientation. In this work, the crack orientation is automatically determined by finding the orientation that results in the largest crack opening. In this work, a framework for determining the damage indices by image analysis is proposed. In this framework, the damage indices include a flexural damage index *d~F~* and a shear damage index *d~S~*. The modified FAC method to determine these damage indices is composed of seven steps. All steps have been implemented in a public software implementation developed by the authors \[[@B35-sensors-19-04304]\]. Analyze the crack opening pattern (see [Figure 6](#sensors-19-04304-f006){ref-type="fig"}a) using the image analysis approach described above, as shown in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}. In this step, the crack opening field $c_{o}$ is generated.Define a threshold of crack opening width, such as 0.05 mm, and convert the crack opening pattern to a binary crack pattern (see [Figure 6](#sensors-19-04304-f006){ref-type="fig"}b). The crack opening width threshold is subjective and must be determined on the basis of the actual situation. While the image analysis method in this work is capable of observing cracks as thin as 0.02 mm (see cracks shown in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}c, while some of the shown cracks are as thin as thin as 0.02 mm), a threshold of 0.05 mm was chosen in this work as it is the minimum crack width in a typical crack width ruler.Analyze the fractal dimension by the FAC method. The FAC method is a multi-level discretization of the binary crack pattern. In each level, the crack pattern is discretized into a mesh composed of many square cells, with the number of cells that contain cracks then being counted (*N*). The width of each cell is ***ε***. At each level, *log*(1/***ε***) and *log*(*N*) can be calculated, as shown in [Figure 6](#sensors-19-04304-f006){ref-type="fig"}c. Further details of calculating the fractal dimension can be found in \[[@B26-sensors-19-04304]\]. Note that, typically, the actual meshes in FAC analyses are more refined, and the number of discretization levels is greater (e.g., 4 levels or higher) than as shown in [Figure 6](#sensors-19-04304-f006){ref-type="fig"}.By applying multi-level mesh refinements (i.e., different sizes of ***ε***), *log*(*N*) versus *log*(1/***ε***) can be plotted on a 2D plot. The fractal dimension $f$ of the crack pattern is the slope of the line found by linear regression. Since the dimension of surface crack *f* is between 1 (that is, an ideal line) and 2 (a filled area), the damage index is estimated by *f* − 1 in the FAC method. A damage index $d$, defined by Equation (4), is calculated, with a value between zero and one (see [Figure 6](#sensors-19-04304-f006){ref-type="fig"}d). $$d = f - 1$$According to the crack orientation of each crack field cell, separate the crack opening field into a shear crack opening field and a flexural crack opening field, as shown in [Figure 6](#sensors-19-04304-f006){ref-type="fig"}e,f. The crack orientation is the angle of the crack. A crack orientation of zero degree means a horizontal crack; An orientation of 45 or 135 degrees means a diagonal crack. The range of the angle is from 0 to 180 degrees. The crack orientation of each crack field cell was calculated during the crack image analysis, as shown in [Figure 4](#sensors-19-04304-f004){ref-type="fig"}. In this work, the horizontal cracks, whose orientation is between 0 to 22.5 degrees or 157.5 to 180 degrees, are classified into flexural cracks and are assigned to the flexural crack opening field, while the remaining cracks are assigned to the shear crack opening field.Separately calculate the total crack areas in the flexural crack opening field $A_{F}$ and the shear crack opening field $A_{S}$. Since the crack opening field represents the crack opening widths, $A_{F}$ is the summation of all values in the flexural crack opening field multiplied by the width of each cell. $A_{S}$ is calculated in the same manner.Calculate the flexural damage index $d_{F}$ and a shear damage index $d_{S}$ using Equations (5) and (6). $$d_{F} = d \cdot \frac{A_{F}}{A_{S} + A_{F}}$$ $$d_{S} = d \cdot \frac{A_{S}}{A_{S} + A_{F}}$$ In most RC structures or components, crack orientation is a typical factor used to classify a crack as either flexural or shear. For RC columns or components that are subjected to bending and horizontal shear forces, horizontal cracks are typically classified as flexural, while the remaining cracks are classified as shear. This classification method is followed here. Furthermore, since the displacement field-based image analysis method provides not only the positions, opening widths, and sliding displacements of cracks, but also their orientations, it is practical to classify cracks according to their orientations. It should be noted that the classification of flexural and shear cracks by orientation is one of several classification methods, and is not necessarily applicable to all structure types. More details can be found in \[[@B10-sensors-19-04304],[@B12-sensors-19-04304]\]. The proposed method not only integrates the previous crack image analysis \[[@B24-sensors-19-04304]\] and FAC methods \[[@B23-sensors-19-04304]\], but also makes some modifications. While the previous crack image analysis method requires analyzers to assign a crack orientation, the proposed method determines the crack orientation of each analyzed cell by finding the orientation that leads to the largest opening crack. While this is a conservative way to estimate crack orientation and opening width, it makes this method automatic, and does not require the orientation to be input manually. In addition, while the FAC method was originally designed for manually plotted cracks, this method uses automatically analyzed crack data for the FAC method. In the proposed method, the analyzed damage index is further separated into shear and flexural parts, providing more information on the failure mode for further safety evaluation. The integration of these methods and modifications makes it possible to carry out structural health monitoring based on crack information in practical applications. 4. Experiments {#sec4-sensors-19-04304} ============== The proposed image-based shear and flexural damage indices were tested using two RC structural experiments \[[@B22-sensors-19-04304]\]. The specimens were reduced-scale RC containment vessels (RCCVs), i.e., relatively short and wide tubular structures. They are denoted as RCCV \#1 ([Figure 7](#sensors-19-04304-f007){ref-type="fig"}a) and RCCV \#2 ([Figure 7](#sensors-19-04304-f007){ref-type="fig"}b), respectively. The specimens were identical in terms of geometry. The specimens were subjected to a constant vertical force of 160 kN, and a cyclic horizontal displacement history imposed through hydraulic controlled actuators, as shown in [Figure 7](#sensors-19-04304-f007){ref-type="fig"}c. The outer and inner diameters were 2500 mm and 2200 mm, respectively. The height of the structures was 2250 mm. The concrete strengths of the two specimens were 37.0 and 43.4 MPa, respectively. The yields and ultimate strength of steel rebars were 379 MPa and 572 MPa, respectively. Four cameras were set up to take photos of the measurement regions, as shown in [Figure 7](#sensors-19-04304-f007){ref-type="fig"}d. The photos from the two northern cameras were used in this work. The two RCCVs had slightly different rebar designs. Four cylindrical layers of rebars were constructed in the concrete tubular structures. Each layer contained up to 90 rebars. The steel ratio of RCCV \#1 was 0.02 with reinforcement extending into the top and bottom for strong interfaces between the roof, the specimen, and the foundations. RCCV \#2 had gradually increasing vertical steel ratios $\rho_{v}$ near the top and bottom, as shown in [Figure 8](#sensors-19-04304-f008){ref-type="fig"}. The increased vertical steel reinforcement in RCCV \#2 was designed to prevent sliding shear failure at the boundaries between the tubular structures and the top/bottom of the RC blocks, which occurred in the RCCV \#1 test. Four cameras were set up in both experiments; two were positioned to the north side and two to the south, as shown in [Figure 7](#sensors-19-04304-f007){ref-type="fig"}d. Two cameras were set up for each image measurement region, because stereo image analysis was employed, as described in the previous section. The measurement regions were painted with randomly striped patterns that provided image features for the displacement fields. The lightening conditions at the top and button regions of the specimens were not as good as those in the middle regions. In addition, the middle regions had better focal conditions in the experiments. [Figure 9](#sensors-19-04304-f009){ref-type="fig"} shows the initial photos taken by the north cameras in both experiments. The experimental results show that the shear strength of the RCCV \#2 was slightly higher than that of the RCCV \#1 (see [Figure 10](#sensors-19-04304-f010){ref-type="fig"}a,b). The shear strengths of RCCV \#1 and RCCV \#2 were 5805 kN and 5580 kN, respectively. In addition, RCCV \#1 and RCCV \#2 had different ductilities. While both vessels reached their shear strengths for a displacement cycle of 16.9 mm (i.e., a drift ratio of 0.75% with respect to the specimen height of 2250 mm), RCCV \#1 rapidly lost its shear strength after the 16.9 mm displacement cycle. In contract, RCCV \#2 retained its shear capacity to 22.5 mm (i.e., a drift ratio of 1%), which was significantly higher because of the increased reinforcement at the top and bottom, as shown in [Figure 10](#sensors-19-04304-f010){ref-type="fig"}. The hysteresis loops of these specimens (see [Figure 10](#sensors-19-04304-f010){ref-type="fig"}c,d) show that the tangential stiffness did not significantly change until the cyclic displacements reached +/−3 mm. Details of the experimental results and explanations can be found in \[[@B22-sensors-19-04304]\]. There were 163 and 1399 pairs of photos taken by the north cameras in the RCCV \#1 and RCCV \#2 experiments, respectively. Each pair of photos included a photo taken by the left camera and a photo taken by the right camera. The cameras were Canon EOS 5D Mark III with photo resolution of 3840 × 5760 pixels. Measurement regions were illuminated using a 100 W light-emitting-diode (LED). [Figure 11](#sensors-19-04304-f011){ref-type="fig"} shows several north left camera photos of RCCV \#1 and RCCV \#2. The u in [Figure 11](#sensors-19-04304-f011){ref-type="fig"} is the horizontal displacement at the top of the specimen. The displacements are so minor that the deformations are difficult to visually recognize in the figure. Since the RCCVs are shear-critical structures, a small displacement can cause significant shear failure. In addition to the experimental facilities and measurement devices, such as the load cells, the major way that we could observe the damage and the failure of the structure was to inspect the cracks on the surface. Diagonal (45-degree) shear cracks appeared on the north and south sides of the specimens, while the horizontal flexural cracks appeared at the top and bottom on the east and west sides. These cracks could be observed by human eyes only when we paused the testing, allowing people to get closer to the specimen to inspect the cracks. Details of the comparison of the manually plotted cracks and image analyzed cracks can be found in \[[@B24-sensors-19-04304]\]. While both specimens underwent shear failures, different shear failure modes were observed for each vessel. RCCV \#1 had a sliding shear mode at the top of the specimen, as shown in [Figure 12](#sensors-19-04304-f012){ref-type="fig"}a. A horizontal crack occurred at the top, where the shear stiffness dramatically changes, typically inducing a stress concentration. The red lines in [Figure 12](#sensors-19-04304-f012){ref-type="fig"} represent the locations of the cracks. Sliding shear did not occur in RCCV \#2 due to the gradual change in rebar density (the steel ratio was from 2% to 4%). RCCV \#2 had a web shear failure in which the major shear crack passed through the specimen at a diagonal (45-degree) angle, as shown in [Figure 12](#sensors-19-04304-f012){ref-type="fig"}b. The crack patterns of the experimental photos, as shown in [Figure 11](#sensors-19-04304-f011){ref-type="fig"}, can be obtained by displacement field-based crack analysis. By using the displacement-based analysis, cracks as thin as 0.03 mm (approximately 0.06 pixels wide in the photos) that appeared at the very beginning of the failure could be detected. The crack patterns of the selected displacement peaks are shown in [Figure 13](#sensors-19-04304-f013){ref-type="fig"}. The crack patterns were analyzed and presented in a field discretized with a grid containing 90 × 60 cells. The size of each cell is equivalent to a sub-image with 27 × 27 pixels. In both cases, from the beginning of the tests, the cracks were distributed over almost the entire measurement region. The widths of the cracks then gradually increased from 0.03 mm (for the 2.3-mm displacement cycle) to up to 0.4 mm (for the 11.3-mm displacement cycle). The proposed crack-based damage indices are calculated on the basis of the crack pattern obtained by the displacement field-based analysis (see [Figure 14](#sensors-19-04304-f014){ref-type="fig"}). In both experiments, the shear damage increased from 0 to approximately 0.75 for the displacement cycle of 8.4 mm (i.e., drift ratio of 0.375%), and did not significantly increase after that. The shear damage indices present a warning index that is capable of capturing the early stages of shear failure. Since both RCCVs \#1 and \#2 are shear critical, the cracks were mostly either at 45 degrees or 135 degrees, or typical shear cracks, with relatively fewer horizontal cracks observed in the measurement regions. This work examined the linear regression plots of several selected actuator control steps when analyzing the fractal dimension (as shown in [Figure 6](#sensors-19-04304-f006){ref-type="fig"}d). The plots showed that these points were very close to the line, and that the residual values were small. A selected plot of the linear regression of each specimen is shown in [Figure 15](#sensors-19-04304-f015){ref-type="fig"}. The crack pattern is a grid containing 90 × 60 cells, and is converted to different refinement of meshes with ***ε*** of 1, 2, 4, 8, 16, 32, 64, and 128 (while the most refined one is slightly more refined than the crack pattern), seven points were calculated in each of the fractal analyses. The computing speed of the proposed method is great enough for static structural health monitoring, but still not sufficient for non-stop real-time dynamic analysis. For each step of the analysis, including image rectification, displacement field analysis, crack opening and orientation analysis, fractal analysis of cracks, and damage indices calculation, it takes about 40 seconds of computing time using a laptop equipped with an Intel i5-7300HQ 2.5 GHz processor and 32GB main memory. Sufficient computing speed may allow us to carry out automatic, non-stop crack detection and health monitoring with a sampling rate of 0.025 Hz, that is, once or twice per minute. It is still insufficient for detecting dynamic responses during a vibration event such as an earthquake, which typically requires a sampling rate of 200 Hz to 1000 Hz. To achieve non-stop dynamic analysis for structural health monitoring, this method requires not only a significant improvement in camera and computing hardware, but also further optimization of the algorithms and programming code. 5. Conclusions {#sec5-sensors-19-04304} ============== This work proposed a damage indexing method based on crack image analysis, with the aim of indicating the early stage failure of shear critical RC structures. This method is based on a displacement field-based crack image analysis method, which is capable of detecting early stage, thin cracks on concrete surfaces. It is especially practical when displacement sensors and load cells are not applicable in real structures. Early stage, thin cracks can be detected when they are as thin as 0.03 mm, which is considerably thinner than the width of a pixel in a digital photo, and cannot be visually seen as a dark line. Based on the crack image analysis, a previously proposed fractal analysis of cracks was employed to estimate the overall damage index. According to the crack orientations, this method separates the fractal analysis damage index into a shear damage index and a flexural damage index to distinguish between the different types of failure. The software implementation method is publicly available. The results of two RCCV experiments were used to verify the proposed damage indexing method. Since both RCCV specimens were shear critical structures, the analyzed damage indices showed that the shear cracks dominated the major failure. The flexural crack indices were relatively low throughout the experiments. In both experiments, the shear damage indices reached a relatively high value (i.e., 0.7) at a displacement of only 8.4 mm on the top of the specimen (i.e., a drift ratio of 0.375%). Earlier damage could be detected when the displacement was only 3.4 mm (i.e., a drift ratio of 0.15%) or even earlier, while the stiffness was still unchanged. This indicates that the crack image analysis-based damage indexing method is capable of indicating early stage failure in shear critical structures. While this method estimates the damage indices of a structure, damage indices obtained from different types of structures are not comparable. The safety of a structure depends on many factors, including complicated design details such as the design of ties and stirrups, which are not visually observable. A non-ductile structure having a lower damage index does not mean it is safer than a ductile structure with a higher damage index. The practical health monitoring application of this method to other structures still requires sufficient experiments and investigations based upon the specific structure type. The authors would like to acknowledge the National Center for Research on Earthquake Engineering in Taiwan for providing partial measurement resources and data of the shake table tests presented in this paper. We would like to thank Uni-edit ([www.uni-edit.net](www.uni-edit.net)) for editing and proofreading this manuscript. Data curation, C.-H.C.; Funding acquisition, C.-l.W.; Investigation, Y.-S.Y. & C.-l.W.; Methodology, Y.-S.Y. & C.-H.C.; Writing---original draft, Y.-S.Y.; Writing---review & editing, Y.-S.Y. This work is partially funded by the Ministry of Science and Technology \[MOST 107-2625-M-027-003 and MOST 108-2625-027-001\] and the National Center for Research on Earthquakes in Taiwan. The authors declare no conflict of interest. {#sensors-19-04304-f001} {#sensors-19-04304-f002} {#sensors-19-04304-f003} {#sensors-19-04304-f004} {#sensors-19-04304-f005} {#sensors-19-04304-f006} {#sensors-19-04304-f007} {#sensors-19-04304-f008} {#sensors-19-04304-f009} {#sensors-19-04304-f010} {#sensors-19-04304-f011} {#sensors-19-04304-f012} {#sensors-19-04304-f013} {#sensors-19-04304-f014} {#sensors-19-04304-f015}

Introduction {#s1} ============ Oral cancer is among the ten most common forms of cancer in the world [@pone.0070883-Parkin1]. A trend of rising incidence has been noted on a global scale in Western countries as well as Asian countries such as Taiwan [@pone.0070883-Mignogna1], [@pone.0070883-Yang1]. Among all cancers in Taiwanese males, oral cancer which contributed 70% of head and neck cancer has been ranked fourth in incidence and mortality since 1995. It is an obvious consequence that the treatment of oral cancer makes an increasing economic burden [@pone.0070883-Lee1], [@pone.0070883-Lang1]. Previous reports revealed excess cancer mortality in psychiatric patients, and several possible mechanisms were proposed [@pone.0070883-Lawrence1]--[@pone.0070883-Tran1]. Patients with mental illness are associated with medical comorbidity, unemployment, living alone and low socioeconomic status [@pone.0070883-Jeste1], [@pone.0070883-Agerbo1]. Due to the problems of cognitive, affective, and social manifestations, it may be hard to get complete informed consent from mentally ill patients [@pone.0070883-Druss1]. Reduced access to general medical care and physician discretion for ordering examination may decrease the probability of full evaluation of patients with mental illness [@pone.0070883-Kisely1], [@pone.0070883-Druss2]. High incidence of postoperative complications in patients with mental illness may further compromise the survival rates [@pone.0070883-Copeland1]. However, there is no large-scale studies which explored the receipt of medical care and survival rate among oral cancer patients with mental illness. Although the implementation of the National Health Insurance in Taiwan in 1995 may reduce the financial barriers to access medical care for all diseases, there may be some disparities in treatment modality and long-term survival for oral cancer between patients with and without mental illness. We hypothesize here that oral cancer patients with mental illness tend to have higher risk of mortality, compared with those without mental illness. To address this hypothesis, we evaluate the association between mental illness and 5-year survival rates among oral cancer patients through the National Health Insurance Research Database (NHIRD) in Taiwan. Using a population-based data allows us to trace all medical service utilization history among oral cancer patients and measure the relationship of mental illness with oral cancer survival rates. Materials and Methods {#s2} ===================== Ethics Statement {#s2a} ---------------- This study was approved by the Institutional Review Board of Buddhist Dalin Tzu Chi General Hospital, Taiwan. Review board requirements for written informed consent were waived because all personal identifying information was removed from the dataset prior to analysis. Database {#s2b} -------- The data for this study were collected from Taiwan's NHIRD for the years 2002 to 2006. This dataset is organized and managed by Taiwan's National Health Research Institutes but collected by Taiwan's National Health Insurance Program, which has been in place in Taiwan since 1995. The program covers approximately 99% of the residents in Taiwan and has contracts with 97% of the medical providers there [@pone.0070883-NHI1]. To verify accuracy of diagnoses, Taiwan's Bureau of National Health Insurance randomly reviews the charts of one per 100 ambulatory and one per 20 inpatient claims [@pone.0070883-Tseng1], [@pone.0070883-Bureau1]. Our cohort study consisted of Taiwan's incidental oral cancer patients (*International Classification of Diseases, Ninth Revision, Clinical Modification* \[ICD-9-CM\] codes 140.0--145.9; malignant salivary gland tumor \[ICD code 142\] was excluded) who had undergone treatment from 2002 and 2006. A total of 16687 patients with oral cancer with treatment were identified. In addition, cancer staging was not known in our database and neither was smoking status. Independent Variables {#s2c} --------------------- Admission ICD9-CM diagnosis identified patients with or without coexisting mental illness deemed current and ongoing at the time of initial diagnosis: (1) no mental illness, and (2) mental illness with a diagnosis of schizophrenia (ICD9-CM codes 295.00--295.99) or major affective disorder (ICD9-CM codes 296.00--296.99), or other major mental illness (ICD9-CM codes 290.00--294.99, 297.00--319.99). Other patient characteristics contained age, gender, and severity of comorbid disease, socioeconomic status, geographic region, and urbanization of residence. The NHIRD didn't have information on cancer staging. The disease severity of each patient was based on the Charlson Comorbidity Index score, which is widely used in recent years for risk adjustment in administrative claims data sets. We used a modified Charlson Comorbidity Index score, which is calculated as the sum of weighted scores based on the relative mortality risk of 19 conditions [@pone.0070883-Deyo1]. This study used labor enrollee category (EC) as a proxy measure of socioeconomic status (SES), which is an important prognostic factor for cancer [@pone.0070883-Braaten1], [@pone.0070883-Kwok1]. The oral cancer patients were classified into four groups: EC1 (civil servants, full-time, or regularly paid personnel with a government affiliation), EC2 (employees of privately owned institutions), EC3 (self-employed individuals, other employees, and members of the farmers' or fishermen's association), and EC4 (veterans, jobless families, and substitute service draftees) [@pone.0070883-Chen1]. These patients were then further classified into three subgroups: EC1--2 (high SES), EC3 (moderate SES), and EC4 (low SES). Geographical regions and urbanization were included. The level of urbanization was determined by population density, percentage of residents with college or higher education, percentage of residents over 65 years of age, percentage of residents who were agriculture workers, and the number of physicians per 100,000 people. The level of urbanization was assigned as urban, suburban, and rural areas [@pone.0070883-LiuCYHung1]. The hospitals were categorized by ownership (public, non-for-profit or for-profit), and hospital accreditation level (medical center, regional or district hospital). Dependent Variables {#s2d} ------------------- The key dependent variable of interest was the 5-year survival of the patients. Rates of undergoing treatment were also studied. Statistical Analysis {#s2e} -------------------- The SAS statistical package (version 9.2; SAS Institute, Inc., Cary, N.C.) and SPSS (version 15, SPSS Inc., Chicago, IL, USA) were used to analyze this data. A two-sided value of *P*\<0.05 was used to determine statistical significance. Pearson's chi-square tests were used to explore the differences between categorical variables. The cumulative 5-year survival rates and the survival curves were constructed with Kaplan-Meier method and compared by the log-rank test. Survival was measured from the time of the oral cancer diagnosed by using overall mortality as event variables. These patients were tracked for 5 years and further linked to the administrative data for the period 2002--2006 to estimate overall survival, with cases censored for patients who drew back guarantees from the National Health Insurance Program or were still robust without defined events at end of follow-up. A multiple logistic regression model was constructed to estimate adjusted odds ratios for differences of treatment type between oral cancer patients with and without mental illness. The Cox proportional regression model was used to evaluate the effect of mental illness on survival rates after adjusting patients' characteristics (age, gender, Charlson Comorbidity Index Score, urbanization, area of residence and enrollee category), treatment modality (surgery with or without adjuvant therapy, and radiotherapy/chemotherapy/chemoradiotherapy) and hospital characteristics (ownership and accreditation level). Results {#s3} ======= Of 16687 patients, 206 (1.2%) had a secondary diagnosis of a mental illness; 174(84%) patients with other major mental disorders and 32(16%) patients with schizophrenia or other major affective disorders. The median follow-up time was 18.1 months. [Table 1](#pone-0070883-t001){ref-type="table"} shows the characteristics of the study sample according to the presence or absence of mental illness in oral cancer patients. Patients with mental illness were more likely to be younger and high socioeconomic status (*P*\<0.001, and = 0.002; respectively). Oral cancer patients with mental illness were significantly less likely to be admitted to medical center (*P* = 0.004). 10.1371/journal.pone.0070883.t001 ###### Characteristics of oral cancer, 2002--2006 (*n* = 16687). {#pone-0070883-t001-1} Characteristics With mental illness (*n* = 206) Without mental illness (*n* = 16481) *P* value ---------------------------------- --------------------------------- -------------------------------------- ----------- ------ --------- ***Patient characteristics*** Age \<0.001 ≦40 49 23.8 2429 14.7 41--50 76 36.9 5253 31.9 51--60 46 22.3 4686 28.4 61--70 20 9.7 2749 16.7 ≧71 15 7.3 1364 8.3 Gender Male 184 89.3 15149 91.9 0.175 Female 22 10.7 1332 8.1 Charlson Comorbidity Index Score 0.342 ≦3 146 70.9 12164 73.8 \>3 60 29.1 4317 26.2 Geographic region 0.322 Northern 62 30.1 5432 33.0 Central 48 23.3 3657 22.1 Southern 83 40.3 6741 40.9 Eastern 13 6.3 651 4.0 Urbanization level 0.207 Urban 36 17.5 3722 22.6 Suburban 97 47.1 7433 45.1 Rural 73 35.4 5326 32.3 Enrollee category 0.002 1--2 76 36.9 4450 27.0 3 93 45.1 7821 47.5 4 37 18.0 4210 25.5 ***Hospital characteristics*** Hospital ownership 0.296 Public 56 27.2 4786 29.0 Non-for-profit 98 47.6 8266 50.2 For-profit 52 25.5 3429 20.8 Hospital accreditation level 0.004 Medical center 146 70.9 13154 79.8 Regional hospital 56 27.2 3004 18.2 District hospital 4 1.9 323 2.0 Values are given as number (percentage). [Figure 1](#pone-0070883-g001){ref-type="fig"} depicts the distribution of treatment modality between the two groups. After adjusting for age and gender, patients with mental illness were substantially less likely to undergo surgery with or without adjuvant therapy (odds ratio \[OR\] = 0.45; 95% confidence interval \[CI\] = 0.33--0.60; *P*\<0.001; [Table 2](#pone-0070883-t002){ref-type="table"}). After adjusting patients' characteristics (age, gender, Charlson Comorbidity Index Score, urbanization, area of residence and enrollee category) and hospital characteristics (ownership and accreditation level), patients with mental illness remained less likely to receive surgery with or without adjuvant therapy (OR = 0.47; 95% CI = 0.34--0.65; *P*\<0.001). {#pone-0070883-g001} 10.1371/journal.pone.0070883.t002 ###### Use of treatment in individuals with and without mental illness (*n* = 16687). {#pone-0070883-t002-2} Characteristics Model A Model B[\*](#nt103){ref-type="table-fn"} Model C[\*\*](#nt104){ref-type="table-fn"} ------------------------------------------ --------- ------------------------------------------ -------------------------------------------- ------ ------------ --------- ------ ------------ --------- Surgery with or without adjuvant therapy Without mental illness 1 1 1 Mental illness 0.47 0.35--0.63 \<0.001 0.45 0.33--0.60 \<0.001 0.47 0.34--0.65 \<0.001 Age 0.99 0.99 0.98--0.99 \<0.001 Gender Male 1 1 Female 1.63 1.39--1.93 \<0.001 1.61 1.35--1.91 \<0.001 Charlson Comorbidity Index Score 0.52 0.48--0.57 \<0.001 Geographic region Northern 1 Central 1.27 1.11--1.45 0.001 Southern 0.92 0.82--1.03 0.162 Eastern 0.59 0.47--0.73 \<0.001 Urbanization level Urban 1 Suburban 0.93 0.83--1.04 0.214 Rural 0.88 0.78--1.02 0.081 Enrollee category 1--2 1 3 1.06 0.95--1.19 0.291 4 0.75 0.67--0.85 \<0.001 ***Hospital characteristics*** Hospital ownership Public 1 Non-for-profit 0.82 0.73--0.92 0.001 For-Profit 0.19 0.17--0.22 \<0.001 Hospital accreditation level Medical center 1 Regional hospital 0.80 0.72--0.89 \<0.001 District hospital 0.21 0.16--0.27 \<0.001 95% CI, 95% confidence interval. Adjusted for patient age, and gender, Adjusted for patient age, gender, Charlson Comorbidity Index Score, geographic region, urbanization level, enrollee category, hospital ownership, and accreditation level. The 5-year survival rates, by presence or absence of mental illness, were illustrated in [Figure 2](#pone-0070883-g002){ref-type="fig"}. The 5-year survival rates were 50.5%, and 68.1% for patients with mental illness and without mental illness (*P*\<0.001). In Cox proportional hazards models adjusted for patients' characteristics and hospital characteristics, patients with mental illness conferred a significant increased risk for death (hazard ratios \[HR\] = 1.83; 95% CI, 1.50--2.23; *P*\<0.001; [Table 3](#pone-0070883-t003){ref-type="table"}). In models that additionally adjusted for treatment modality, patients with mental illness remained conferred a 1.58 times risk of death, compared to those without mental illness (HR = 1.58; 95% CI, 1.30--1.93; *P*\<0.001). {#pone-0070883-g002} 10.1371/journal.pone.0070883.t003 ###### Hazard ratios for 5-year overall mortality in oral cancer patients with and without mental illness (*n* = 16687). {#pone-0070883-t003-3} Characteristics Model A Model B[\*](#nt106){ref-type="table-fn"} Model C[\*\*](#nt107){ref-type="table-fn"} --------------------------------------------- --------- ------------------------------------------ -------------------------------------------- ------ ------------ --------- ------ ------------ --------- Oral cancer patients Without mental illness 1 1 1 Mental illness 1.86 1.53--2.27 \<0.001 1.83 1.50--2.23 \<0.001 1.58 1.30--1.93 \<0.001 Age 1.01 1.01--1.01 \<0.001 1.01 1.00--1.01 \<0.001 Gender Male 1 1 Female 0.82 0.73--0.91 \<0.001 0.87 0.79--0.97 0.013 Charlson Comorbidity Index Score 2.26 2.14--2.39 \<0.001 2.00 1.89--2.12 \<0.001 Geographic region Northern 1 1 Central 1.00 0.92--1.08 0.930 1.01 0.93--1.10 0.76 Southern 0.92 0.85--0.99 0.019 0.89 0.83--0.96 0.002 Eastern 1.10 0.95--1.28 0.197 0.99 0.85--1.15 0.884 Urbanization level Urban 1 1 Suburban 1.05 0.97--1.13 0.224 1.05 0.97--1.13 0.227 Rural 1.14 1.05--1.25 0.002 1.15 1.06--1.25 0.001 Enrollee category 1--2 1 1 3 1.10 1.02--1.83 0.013 1.13 1.05--1.21 0.002 4 1.43 1.32--1.54 \<0.001 1.38 1.28--1.49 \<0.001 ***Hospital characteristics*** Hospital ownership Public 1 1 Non-for-profit 1.00 0.93--1.07 0.979 0.97 0.91--1.04 0.40 For-profit 1.43 1.32--1.54 \<0.001 1.07 0.98--1.16 0.13 Hospital accreditation level 2.68 2.52--2.85 Medical center 1 1 Regional hospital 1.23 1.14--1.32 \<0.001 1.20 1.12--1.29 \<0.001 District hospital 0.99 0.82--1.19 0.887 0.70 0.58--0.85 \<0.001 Treatment modality Surgery with or without adjuvant therapy 1 Radiotherapy/Chemotherapy/Chemoradiotherapy 2.68 2.52--2.85 \<0.001 95% CI, 95% confidence interval. Adjusted for patient age, gender, Charlson Comorbidity Index Score, geographic region, urbanization level, enrollee category, hospital ownership, and accreditation level. Adjusted for patient age, gender, Charlson Comorbidity Index Score, geographic region, urbanization level, enrollee category, hospital ownership, accreditation level, and treatment modality. Discussion {#s4} ========== This population-based study found that oral cancer patients with mental illness had a grave prognosis. Oral cancer patients with mental illness conferred a 1.58-times risk of mortality, compared with those without mental illness after adjusting for patient characteristics (age, gender, Charlson Comorbidity Index Score, urbanization, area of residence and enrollee category), hospital characteristics (ownership and accreditation level), and treatment modality. Patients with mental illness were substantially less likely to undergo surgery with or without adjuvant therapy. There was no significant difference in utilization of radiotherapy or chemotherapy or chemoradiotherapy in oral cancer patients with or without mental illness. There are several possible mechanisms for poor prognosis in oral cancer patients with mental illness. Previous studies have explored that patients with mental illness were medically sicker and older, compared with patients without mental illness. Besides, oral cancer patients with severe comorbidity were associated with poor prognosis [@pone.0070883-Jeste1], [@pone.0070883-Lin1]. In our study, mentally ill patients were more likely to be younger and from higher socioeconomic status. There was no significant difference in comorbidity severity between patients with mental illness and those without mental illness. Nevertheless, oral cancer patients with mental illness were prone to visit regional and district hospitals, which may lack the adequate facility, such as experienced head and neck surgeons, plastic surgeons, intensive care team, radiation oncologist, hematology oncologist and linear accelerator to treat oral cancer. Oral cancer patients with mental illness were less likely to receive surgery with or without adjuvant therapy and the differences in rates of receipt of medical care between the two groups remained statistically significant in multivariate analyses that adjusted for patient characteristics, comorbidity score and hospital characteristics. Druss et al. used a large national survey in United States found that patients with mental illness tended to face particular difficulties in obtaining and maintaining both health insurance and needed medical care [@pone.0070883-Druss2]. Although financial barriers to access medical care had been reduced after the implementation of the National Health Insurance in Taiwan in 1995, medical treatment modality remained different in mentally ill patients with either oral cancer or other disease [@pone.0070883-Tsay1]. This disparity should be of concern. Moreover, mentally ill patients are more likely to undergo postoperative complications, such as deep vein thrombosis, sepsis and respiratory failure [@pone.0070883-Kudoh1]. These complications may delay the beginning of adjuvant radiotherapy or chemotherapy and cause poor prognosis for survival rates [@pone.0070883-Chong1]. Furthermore, deteriorating psychiatric status was found more frequently in patients with mental disease [@pone.0070883-Kudoh2]. Postoperative confusion, for example, may make the postoperative care more complicated and decrease the physicians' intention to perform the surgery for oral cancer patients with mental illness. Physician slant may exist when dealing with oral cancer patients with mental illness. One study demonstrated potential racial bias in influencing the decision of physicians to refer patients for cardiac catheterization [@pone.0070883-Schulman1]. This phenomenon may be observed among patients with mental illness in physical care [@pone.0070883-Penn1], [@pone.0070883-Hinshaw1]. All the above reports indicated that potential physician bias may contribute to the difference of surgery rates in oral cancer treatment. The reasons for excess mortality in oral cancer patients with mental illness may be associated with advanced disease and poor treatment compliance with therapy. Cognitive deficits, disorganized thinking, and impaired ability to communicate important medical symptoms may contribute to presentation at a later stage of disease [@pone.0070883-Lawrence1], [@pone.0070883-Cooke1]. However, this phenomenon can't be validated in our series due to lack of cancer staging in National Health Insurance Dataset. As regards interruption of the treatment course, such as radiotherapy or chemotherapy, may lower survival rates among mentally ill patients. Patients who had major deficiencies in treatment plans incurred a doubled risk of mortality compared with those who initially complied with treatment protocol [@pone.0070883-Peters1], [@pone.0070883-Kubicek1]. This study adds several novel points on the receipt of medical care and survival rates in oral cancer patients with or without mental illness. Using a population-based dataset enables us to trace all medical claims during the 5-year follow-up period. Previous reports presented the excess cancer mortality in psychiatric patients. However, the difference of medical care was not analyzed during these studies [@pone.0070883-Lawrence1], [@pone.0070883-Kisely1]. Our study revealed the decreased access to medical treatment, especially surgery with or without adjuvant therapy among oral cancer patients with mental illness. This partly explained the poor survival rate in mentally ill patients. However, significant risk of death in patients with mental illness remained unaccounted for even in our fully adjusted model. Some unobservable variables, such as social isolation, and lack of family support may compromise the survival rates in patients with mental illness. There are several limitations in this study. First of all, it is lack of access to detailed information from the insurance claims database with regard to oral cancer stage and pattern of relapse. These may be the important variables for increased mortality rate among cancer patients with mental illness. Delayed diagnosis or lack of access to screen in mentally ill patients might lead to more advanced staging at diagnosis and reduce possibility of surgical intervention after diagnosis. Further study is indicated using cancer registry data with more details on staging. Second, the database lacks information of lifestyle factors such as dietary habits, alcohol or tobacco use, which may be risk factors and prognostic factors for oral cancer [@pone.0070883-Siahpush1]. Third, instead of cancer-specific survival rate, the overall survival rate was used, because it was not possible to determine cause-specific mortality based on this registry data. Mentally ill patients often have medical comorbidities and the direct cause of death may be another disease which susceptible to malignant neoplasm. Fourth, postoperative complication or treatment-related complications were not included. Fifth, the detail receipts of anti-psychotic drugs were not available. Given the magnitude and statistical significance of the observed effects in this study, these limitations are unlikely to compromise the results. In summary, oral cancer patients with mental illness were substantially less likely to undergo surgery with or without adjuvant therapy than those without mental illness. Patients with mental illness conferred a poor prognosis in multivariate analysis. Oral cancer treatment for the mentally ill patients is a complex issue and deserves more concern. Meanwhile, there should be efforts to reduce disparities in physical health including oral mucosal screening during routine dental evaluations or physical examination. This study is based in part on data from the National Health Insurance Research Database provided by the Bureau of National Health Insurance, Department of Health and managed by the National Health Research Institutes (Registered number 99018 and 101115). The interpretation and conclusions contained herein do not represent those of the Bureau of National Health Insurance, Department of Health, or National Health Research Institutes. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TSC SJH LFC CCL. Analyzed the data: TSC SJH CCL. Wrote the paper: TSC SJH YCS LFC HCH MSL CHL PC CCL.

1. Introduction {#sec1-ijerph-17-05347} =============== The frequency of earthquakes in Korea has historically been low, and their size has been relatively small. Therefore, up until 2016, the contents of Korea's disaster education program did not include how to respond to earthquakes, because people viewed the country as being safe from such incidents. However, two seismic, unprecedented events then took place in Korea. In 2016, an earthquake of Magnitude 5.8 (M 5.8) occurred in Gyeongju, which was followed by the Pohang earthquake of M 5.4 about one year later, on 15 November 2017 (see [Table 1](#ijerph-17-05347-t001){ref-type="table"}). The distance between the epicenter of these two earthquakes was about 40 km. [Figure 1](#ijerph-17-05347-f001){ref-type="fig"} shows the location of seismic events in Pohang and Gyeongju. Although the magnitude of the Gyeongju earthquake was larger than the one in Pohang, the damage inflicted on Pohang was greater \[[@B1-ijerph-17-05347]\]. The seismic waves of the Pohang earthquake shocked the entire population of Korea, causing the Korean government to postpone the National College Scholastic Aptitude Test, which had been scheduled for November 16. Following this decision, Korean schools felt the need to offer earthquake response education to students, and the first steps toward offering earthquake disaster education were put in place \[[@B2-ijerph-17-05347]\]. For example, Pohang City Hall planned a disaster education program designed to ensure the safety of students by training them protective action for when an earthquake occurs. The effects of disasters can be substantially reduced if people are well-informed about how to respond to them. Disaster education plays an important role in improving students' disaster awareness and preparedness level \[[@B3-ijerph-17-05347]\]. Furthermore, as the Sendai Framework for Disaster Risk Reduction \[[@B4-ijerph-17-05347]\] stated that children and youth can be the "agents of change" in transforming and improving the safety of society. The participants of disaster education programs tend to be associated with realistic risk perceptions, increased disaster mitigation knowledge, and increased discussions among family members about disasters \[[@B5-ijerph-17-05347]\]. In the Republic of Korea, schools and public institutions have provided many different forms of disaster educations. They offer knowledge of hazards, causes of injury, protective actions, and mitigation actions, and these activities can help people improve resilience, survival, and mental trauma recovery. Despite these efforts, many children and teens still either do not know how to respond to disasters \[[@B6-ijerph-17-05347]\] or fail to take any preparatory measures \[[@B7-ijerph-17-05347]\]. However, there have not been many studies that have evaluated disaster education programs or their effectiveness \[[@B8-ijerph-17-05347]\]. In particular, it is difficult to find research evaluating the effects of disaster experiences on disaster education. Having disaster experience is one of the key factors affecting an individual's disaster preparedness behaviors, as people who have experienced disasters either directly or indirectly face different levels of intensity and damage compared to those who have not. In the case of children and teens, it is more important to investigate their emotional reactions to disaster experiences, as such feelings linger in their minds for a long time and affect their subsequent behavior \[[@B9-ijerph-17-05347]\]. Therefore, when determining the effectiveness of disaster education, we need to consider people's disaster experiences and their emotions. The purpose of this study is to analyze the earthquake experiences of children and teens and evaluate the effects of their experiences on disaster education. Educational effectiveness in this study will be measured using the students' knowledge level regarding earthquake drills and their levels of training satisfaction. 2. Literature Review {#sec2-ijerph-17-05347} ==================== 2.1. Disaster Education for School Children and Teens {#sec2dot1-ijerph-17-05347} ----------------------------------------------------- Experiencing a disaster when you have an immature psyche can be more stressful, leaning toward having a negative effect on a young person's growth. Although the disaster may have occurred a long time ago, children can still remember such events vividly \[[@B10-ijerph-17-05347]\]. Accordingly, disaster education for children and teens is important to improve their resilience to disasters \[[@B11-ijerph-17-05347],[@B12-ijerph-17-05347],[@B13-ijerph-17-05347]\]. Various forms of disaster education were conducted: lectures \[[@B14-ijerph-17-05347],[@B15-ijerph-17-05347],[@B16-ijerph-17-05347]\], drills \[[@B17-ijerph-17-05347]\], discussions with teachers and friends \[[@B13-ijerph-17-05347],[@B18-ijerph-17-05347]\], reading and listening \[[@B19-ijerph-17-05347]\], using hypermedia systems \[[@B20-ijerph-17-05347]\], games \[[@B21-ijerph-17-05347],[@B22-ijerph-17-05347]\], after-school activities \[[@B12-ijerph-17-05347]\], and on-line intercultural projects \[[@B23-ijerph-17-05347]\]. Disaster education can be conducted across several levels. The individual level tends to improve the awareness of the disaster risk, and the community level, including home-based education, increases the adjustment and preparedness activities \[[@B13-ijerph-17-05347]\]. However, the most common and important type is school-based education \[[@B24-ijerph-17-05347],[@B25-ijerph-17-05347]\]. Schools, as information communicators, can effectively enhance students' safety knowledge, help with planning for a disaster situation, and structuralize mitigation \[[@B26-ijerph-17-05347]\]. 2.2. Assessment of Disaster Education {#sec2dot2-ijerph-17-05347} ------------------------------------- To provide effective disaster education for children and teens, it is important to determine how the effectiveness of education can be measured. Usually, the measurement indices are devised by methodologies such as surveys, interviews, and observations. These indices include fear and risk perception \[[@B27-ijerph-17-05347],[@B28-ijerph-17-05347]\], disaster risk knowledge, disaster risk reduction skills \[[@B3-ijerph-17-05347],[@B17-ijerph-17-05347],[@B18-ijerph-17-05347],[@B29-ijerph-17-05347]\], attitude to disaster \[[@B23-ijerph-17-05347]\], confidence to respond against disasters \[[@B12-ijerph-17-05347]\], other preparedness activities \[[@B13-ijerph-17-05347],[@B28-ijerph-17-05347]\], and student satisfaction \[[@B24-ijerph-17-05347]\]. In the case of applying new teaching methods, the methods itself were mainly used as factor \[[@B20-ijerph-17-05347],[@B22-ijerph-17-05347]\]. Among all these factors, in general, students' direct assessment of disaster prevention programs consists of learning satisfaction and effectiveness \[[@B30-ijerph-17-05347]\]. Johnson (2014) found that many studies used knowledge-based outcome indicators, such as "knowledge of hazard risks" and "knowledge of protective actions during an emergency" \[[@B8-ijerph-17-05347]\], and pointed out the necessity of developing an evaluation index. However, there are not many studies about individual factors that can increase the effectiveness of disaster education, given the importance of individual attributes \[[@B19-ijerph-17-05347]\]. This may be a reason why many youths do not know what to do in an emergency, although institutions have offered disaster education \[[@B6-ijerph-17-05347]\]. 2.3. Disaster Experience and Preparedness {#sec2dot3-ijerph-17-05347} ----------------------------------------- Previous studies examined disaster experience as a factor that motivates people to prepare for future incidents \[[@B6-ijerph-17-05347],[@B31-ijerph-17-05347]\]. More specifically, the physical, financial, and emotional damages inflicted by earthquakes, encourage people to prepare for the reoccurrence of such events \[[@B32-ijerph-17-05347]\]. The number of earthquakes experienced \[[@B33-ijerph-17-05347]\], the combination of the level of fear and the amount of damage suffered by homeowners \[[@B34-ijerph-17-05347]\], proximity to the earthquake epicenter, earthquake-related experiences, fear, and levels of pre-earthquake preparedness \[[@B32-ijerph-17-05347],[@B33-ijerph-17-05347]\] are factors that have been identified for motivating preparedness. Disaster experience is a major factor in promoting preparedness \[[@B35-ijerph-17-05347],[@B36-ijerph-17-05347]\]. This means that victims who suffer from a disaster are better prepared to cope with future disasters. Furthermore, disaster experience also strengthens an individual's ability to perceive disasters or risks, as well as their understanding of the actual situation \[[@B6-ijerph-17-05347],[@B36-ijerph-17-05347]\]. Therefore, it is important to analyze disaster experience and examine its impact on preparedness. However, such an experience can also have negative effects on levels of preparedness. For example, experiencing a moderate disaster or less traumatic exposure can cause "a false sense of confidence" or normalcy bias \[[@B35-ijerph-17-05347],[@B37-ijerph-17-05347]\]. When the disaster is not catastrophic, people may go through minor seismic experiences and eventually consider themselves to be safe from potential earthquake risks. This may result in people being reluctant to engage in disaster preparedness activities in advance \[[@B38-ijerph-17-05347]\]. Several attempts have been made to interpret or define disaster experiences, which can be measured using their intensities and the level of physical, emotional, and financial damage inflicted \[[@B32-ijerph-17-05347]\]. Becker et al. focused on the emotional experiences, which were classified into the following categories: direct experience, indirect experience, vicarious experience, and life experience \[[@B35-ijerph-17-05347]\]. In this study, the earthquake experiences of children and teens were analyzed using both their emotional and cognitive response. The cognitive response factor includes the participants' perceived Modified Mercalli Intensity Scale degree. The measurement of the Modified Mercalli Intensity (MMI scale) is a mechanism of intelligent judgement and observation that can be determined by the observation of each case, such as the building or geologic condition \[[@B39-ijerph-17-05347]\]. In contrast, an individual's emotional response can be independent of their cognitive understanding \[[@B34-ijerph-17-05347],[@B40-ijerph-17-05347]\], and is more powerful than a cognitive judgment \[[@B41-ijerph-17-05347]\]. For example, the emotional response following flooding encourages people to prepare more thoroughly \[[@B34-ijerph-17-05347]\]. 2.4. Disaster and Emotions {#sec2dot4-ijerph-17-05347} -------------------------- In disaster situations, negative emotions constitute the principal response to risks and threats \[[@B9-ijerph-17-05347]\]. However, people do not experience purely negative emotions in disaster situations, and their reactions may depend on both the scale of the damage they faced and personal characteristics. In particular, each age group has a different reaction to a disaster, and we need to pay greater attention to children due to how non-typical their reactions are \[[@B42-ijerph-17-05347]\]. Therefore, it is necessary to comprehensively examine children and teens' emotional reactions to disaster experiences \[[@B43-ijerph-17-05347]\]. The common emotional reactions to a disaster are fear, anger, guilt, and sadness \[[@B42-ijerph-17-05347]\]. To understand emotions comprehensively, previous research has studied the classification of emotions. Ekman et al. (1980) categorized emotional experiences through facial expressions in order to develop a classification system that is simply divided into positive or negative. In his paper, the primary emotions are disgust, surprise, sadness, fear, pain, and arousal \[[@B44-ijerph-17-05347]\]. Shaver et al. (1987), meanwhile, suggest that the basic emotions are fear, joy, surprise, sadness, anger, and love \[[@B45-ijerph-17-05347]\], while Parrott (2001) classified these emotions as love, joy, surprise, anger, sadness, and fear \[[@B46-ijerph-17-05347]\]. Russell (1980) proposed the "Circumplex Model." This model has two dimensions: the degree of activation and the degree of pleasure. For example, surprise and fear are activated emotions, while anger, disgust, and sadness constitute unpleasant emotions \[[@B47-ijerph-17-05347]\]. Emotions represent the strongest system for motivating an individual's behavior \[[@B48-ijerph-17-05347]\], with each emotion triggering a specific behavior or tendency \[[@B49-ijerph-17-05347]\]. Fear and worry tend to induce evacuation and preparedness to avoid danger, while joy is related to encouraging advances and protecting a person's life. Anger makes people want to destroy or punish the cause, while sadness has a propensity to make people give up their willingness to act and seek comfort by accepting their loss \[[@B9-ijerph-17-05347]\]. Emotion in particular can affect an individual's motivation for learning preparedness and survival skills. Arousing fears of health risks has a positive impact on the attitude toward acceptance of recommendations. People who are in a high state of fear take the health recommendations more seriously and intend on following it \[[@B50-ijerph-17-05347]\]. In neuroscience, the learning ability of those who have fearful faces are better than those who have neutral faces \[[@B51-ijerph-17-05347]\]. 3. The School Disaster Educational Program for Children and Teens in Pohang City {#sec3-ijerph-17-05347} ================================================================================ In 2018, Pohang City Hall's "Earthquake Damage Settlement Association" and "Disaster Prevention Policy" department established a joint plan to provide a customized earthquake disaster educational program for children and teens. The purpose of the program, which was based on the earthquake response manuals of the Korean government, was to ensure the safety of school students from future earthquakes. As part of the plan, the "2018 school earthquake disaster education program in Pohang" was conducted over a period of four months (August to November) in 2018. The content was taught by professors and graduate students from the Ulsan National Institute of Science and Technology (UNIST), who also conducted surveys regarding the program's effectiveness. Fourteen schools (six middle and eight high schools) in Pohang city participated, and the number of students who joined the program was about 5000. [Figure 2](#ijerph-17-05347-f002){ref-type="fig"} shows the schools that participated in the program. All of the schools are located within 24 km of the epicenter of the Pohang earthquake in November 2017. The training content consisted of three parts. The first part focused on earthquake science, including the phenomena of earthquakes and seismic waves. When describing seismic waves, students were taught the difference between P waves and S waves using experiments and video materials. The second part concerned earthquake-resistant and earthquake-proof construction, with key elements and goals regarding the seismic design explained. In the final section, the general protective action and strategy regarding earthquakes was introduced, as well as memorization skills designed to help students remember key actions easily. 4. Methods {#sec4-ijerph-17-05347} ========== 4.1. Research Design {#sec4dot1-ijerph-17-05347} -------------------- This study examined factors relating to the disaster experience increase the effectiveness of earthquake response education for the students who experienced the Pohang earthquake. The experience was measured by two responses to the earthquake: cognitive response and emotional response. The cognitive response was determined by the perceived scale of the earthquake. The emotional response was measured by five emotions: fear, joy, surprise, sadness, and anger. The variables for the effectiveness of education were measured by the knowledge level of the earthquake protection drills and the level of satisfaction of the education program. The structural equation model (SEM) was conducted to find the statistically significant factors from students' previous disaster experience that impacted educational effectiveness. The hypothesis of this study is that cognitive response and two emotional responses (using the activated emotional and unpleasant emotional response categories developed by Russell \[[@B47-ijerph-17-05347]\]), stemming from students' previous disaster experiences, will have an impact on educational effectiveness. [Figure 3](#ijerph-17-05347-f003){ref-type="fig"} shows the hypotheses of this paper. *Cognitive response will positively influence educational effectiveness*. *Activated emotional response will positively influence educational effectiveness*. *Unpleasant emotional reaction will negatively influence educational effectiveness*. Two responses to the earthquake were used both to analyze the previous disaster experience and identify the factors influencing educational effectiveness. The latent endogenous variable was educational effectiveness, which was measured through a combination of knowledge level and level of satisfaction. The latent exogenous variables were cognitive response, activated emotional response, and unpleasant emotional response. The observed exogenous variables for emotional response included five emotions (fear, joy, surprise, sadness, and anger), which were classified into two latent variables. The activated emotional response was measured using fear, joy, and surprise, while the unpleasant emotional response was measured using sadness and anger. 4.2. Data Collection {#sec4dot2-ijerph-17-05347} -------------------- The survey data were collected as a part of the "2018 school earthquake disaster education program in Pohang" developed by Pohang City Hall. The survey was conducted from August to November (four months) in 2018, about a year after the Pohang earthquake. The respondents were about 5000 students from six middle schools and eight high schools in Pohang. Prior to the survey, the earthquake response education was conducted by members of UNIST. The data of 3316 respondents were analyzed after eliminating insincere answers and missing values from among the initial \~5000 questionnaires. [Table 2](#ijerph-17-05347-t002){ref-type="table"} shows the demographic information of the target respondents. In both middle and high schools, the proportion of women was higher. The questionnaire has five parts: location when the earthquake occurred, perceived scale, emotional response, training assessment, and demographic characteristics. 4.3. Survey Questionnaire {#sec4dot3-ijerph-17-05347} ------------------------- ### 4.3.1. Earthquake Intensity {#sec4dot3dot1-ijerph-17-05347} The question designed to gauge students' cognitive response concerned how they perceived or observed their surroundings and intensity during the shock ("When the earthquake occurred, what happened around you? Choose one."). The statement about magnitude was based on the Modified Mercalli Intensity scale, developed by Harry O. Wood and Frank Neumann. Several additional features were also included in the survey, such as the middle size (**M**5.4) earthquake and the level of damage. The scales are "1 = Not felt," "2 = Felt only by a few persons at rest," "3 = Felt quite noticeably. Windows and furniture shook," "4 = Flowerpots or furniture fell down or windows shook violently," "5 = Some windows broken or walls cracked," "6 = Walls or buildings collapsed" \[[@B39-ijerph-17-05347]\]. ### 4.3.2. Emotional Response {#sec4dot3dot2-ijerph-17-05347} The questions concerned five different emotions: fear, joy, surprise, sadness, and anger ("When you experienced an earthquake at first, you could feel various emotions. How did you feel about the earthquake in Pohang?"). These items were measured using a five-point Likert-type scale ("1 = not at all," "2 = hardly," "3 = somewhat," "4 = fairly," and "5 = very."). ### 4.3.3. Knowledge Level {#sec4dot3dot3-ijerph-17-05347} The respondents answered six right or wrong questions concerning earthquake response actions. To measure the educational effectiveness, the six questions were based on the content of the education program ("Here are six yes or no questions about earthquake response actions. If the statement is right, you should choose yes. Check your answer"). The six questions were as follows: (1) "If you are indoors, right away, you must try to get under a desk or table to protect your head and torso" (correct answer: yes); (2) "During the shaking, you must evacuate outside of buildings immediately" (no); (3) "During the shaking, you must use an elevator to evacuate the building as fast as possible" (no); (4) "When the earthquake occurs, you must evacuate to open space, such as a park or a ground" (yes); (5) "If you are driving, you must use a vehicle to move to a safe place" (no); and (6) "If you are outside, you must stay next to the walls" (no). The questions and their contexts are based on the manual of the Ministry of the Interior and Safety. ### 4.3.4. Level of Satisfaction {#sec4dot3dot4-ijerph-17-05347} The level of educational effectiveness was assessed by the students. They answered questions regarding their level of satisfaction with the training ("I am satisfied with this earthquake response training.") using a five-point Likert-type scale ("1 = strongly disagree," "2 = disagree," "3 = neither agree nor disagree," "4 = agree," and "5 = strongly agree."). 4.4. Data Analysis {#sec4dot4-ijerph-17-05347} ------------------ The analysis was conducted using the IBM SPSS Statistics (IBM, Armonk, NY, USA) and The R Project for Statistical Computing (The R Foundation for Statistical Computing, Vienna, Austria). The basic statistical characteristics and correlations among the variables were found through frequency analysis, correlation analysis, and regression analysis. The Structural Equation Model was applied to evaluate the causal relationship between earthquake experience and training effectiveness. The five emotional response variables were converted into two factors through Principal Component Analysis (PCA). The internal consistency, Construct Reliability, and Average Variance Extracted were then determined to confirm the validity of the PCA results. 5. Results {#sec5-ijerph-17-05347} ========== 5.1. Response to Past Eartqhuake Experiences {#sec5dot1-ijerph-17-05347} -------------------------------------------- All of the respondents to this survey had experienced the Pohang earthquake. The result shows that the respondents perceived an intensity of 3.7 MMI (Mean (M) = 3.71, Standard Deviation (SD) = 0.89) on average. The perceived MMI tended to decrease as the distance from the epicenter increases (see [Figure 4](#ijerph-17-05347-f004){ref-type="fig"}). When compared by gender, female students (M = 3.75, SD = 0.91) tended to recognize a stronger scale than male students (M = 3.61, SD = 0.85), and there is a statistically significant difference between male and female students (t-test = −4.166, *p* = *0*.000). According to the ANOVA test results, there were no significant differences between male high vs. male middle school students (*p* = *0*.807). The same is also true of female high vs. female middle school students (*p* = *0*.912). In contrast, the survey results show that the strongest emotional response during the Pohang earthquake was surprise (M = 4.33, SD = 0.89), while the weakest emotion was anger (M = 2.00, SD = 1.11). The mean values of the other emotions are fear (M = 3.91, SD = 1.09), sadness (M = 2.53, SD = 1.26), and joy (M = 2.17, SD = 1.18). According to the result of the independent group t-test between emotional responses and gender, there are significant differences (*p*-value = 0.000), and joy and other emotions show an opposite tendency. While most emotional responses are more strongly experienced by females, males rated joy higher than females. The correlation between respondents' emotional responses and proximity to the epicenter is significant. The nearer the location of the respondent, the stronger the level of emotion other than joy. By contrast, the correlation coefficient between the level of joy and distance is positive (corr. = 0.096, *p* \< 0.01). Age has a statistically significant relationship with joy (corr. = −0.042, *p* \< 0.05) and anger (corr. = 0.082, *p* \< 0.01). The Pohang earthquake was not as big an earthquake compared with other significant examples, and some students thought it was an interesting experience. As joy is not a common response during a disaster, this shows unique characteristics that differ from other emotions. More specifically, students who were male, younger, and far away from the epicenter felt higher levels of joy. The other reactions, including anger, had contrasting tendencies. These results show that the emotional response to disasters varied according to age, gender, and location during the event. Exploratory factor analysis was used to reduce the dimensions of the five emotional responses and examine the classification hypothesis (see [Table 3](#ijerph-17-05347-t003){ref-type="table"}). It analyzed two latent factors: activated emotion and unpleasant emotion. The five emotional responses were reduced to two components by applying principal component analysis (PCA). The activated emotional factor includes fear, joy, and surprise, while the unpleasant emotional factor covers sadness and anger. The reliability analyses used Cronbach's alpha and AVE to validate the reliability of the model. The internal consistency (Cronbach's α) of each factor was 0.692 and 0.693. The AVE value of each factor was 0.544 and 0.771. 5.2. Effect of Disaster Education {#sec5dot2-ijerph-17-05347} --------------------------------- Respondents completed six questionnaires in order to evaluate their level of satisfaction regarding the training. The quizzes consisted of yes/no type questions. For each problem, one point was given for the correct answer and zero for the wrong answer, meaning that the lowest score is zero and the highest score is six. The average score of respondents was 5.16 (SD = 1.04) out of 6.00. Satisfaction levels regarding the earthquake response training were measured using a five-point Likert-type scale, from "1 = not at all" to "5 = strongly agree." The mean value was 4.61 (SD = 0.71) out of 5.00. 5.3. Factors Affecting the Level of Preparedness {#sec5dot3-ijerph-17-05347} ------------------------------------------------ The SEM (Structural Equation Model) was formed on the hypothesis that the factors relating to earthquake experience will enhance the effectiveness of the emergency training. [Figure 5](#ijerph-17-05347-f005){ref-type="fig"} shows the causal diagram for the model with an assessment of standardized (unstandardized) path loadings, β. As we can see in [Table 4](#ijerph-17-05347-t004){ref-type="table"}, the model is reasonably consistent with the data, as the overall model fit indices are within the recommended values. The only exceptions to this are the indices $\chi^{2}$ significance and $\chi^{2}$/d.f. \[[@B52-ijerph-17-05347],[@B53-ijerph-17-05347],[@B54-ijerph-17-05347]\]. However, the $\chi^{2}$ significance and $\chi^{2}$/d.f. indices tend to increase in line with the sample size. As the sample size of this study is very large compared to typical samples (N = 200--300) of SEM, we will accept this model considering the overall significance of the test results across the entire model \[[@B54-ijerph-17-05347]\]. [Table 5](#ijerph-17-05347-t005){ref-type="table"} shows the results of hypothesis testing. The standardized coefficient of H2 was 0.134 (*p* = 0.000), while for H3 it was −0.033 (*p* = 0.03). Therefore, we can accept hypotheses H2 and H3: namely, that the activated (unpleasant) emotional response will positively (negatively) influence educational effectiveness. However, the standardized estimate of H1 was only 0.004 (*p* = 0.721). As this was not a statistically significant result, we rejected the H1 hypothesis, which stated that the cognitive response (perceived MMI) would positively influence educational effectiveness. This result shows that an individual's emotional response is more relevant to the effectiveness of disaster training than the recognized scale by witnessing or experiencing such incidents. More specifically, activated emotions (fear, surprise, and joy) have the greatest influence among the factors, while unpleasant emotions (anger and sadness) have a statistically significant but opposite influence. In other words, sadness and anger demotivate students to learn about the disaster. 6. Discussion {#sec6-ijerph-17-05347} ============= The experiences of the Pohang earthquake of M 5.4 in 2017 were analyzed and divided into cognitive responses and emotional responses. The cognitive responses of the students did not impact educational effectiveness significantly. This factor was measured using the earthquake intensity that the students perceived, which was measured to be around 3.7 MMI on average. The measured intensity result differed according to the individual's proximity to the epicenter; students who were further away from the epicenter tended to experience weaker shaking. The reason why the effects of cognitive responses are not statistically significant may be due to the characteristics of the Pohang earthquake. Although its size was M 5.4, the second strongest event recorded in South Korea since 1978, it was only a mid- or minor-sized earthquake compared to the large magnitude earthquakes witnessed in Japan or other countries around the world. Consequently, in spite of the respondents' surprise and fear, which were based on the unfamiliarity of experiencing earthquake, the physical damage inflicted by the earthquake was not very serious. Consequently, students might think that they do not need additional preparation because there was no notable damage. Students felt various emotional reactions, including both negative and positive responses. The two emotional responses, activated emotional response and unpleasant emotional response, were statistically significant in terms of educational effectiveness (*p* \< 0.01), with activated emotional responses identified as most significant factor. Students who felt surprise and fear, but not joy, tended to have more effective educational experiences. On the other hand, the unpleasant emotional reactions, such as anger and sadness, had a negative effect on educational effectiveness. These results imply that students' emotional responses are more important than their cognitive responses in terms of providing an effective disaster education program. This means that even though an earthquake may be small in terms of magnitude or physical damage caused, we still need to provide immediate disaster education to children and teens when they are surprised and afraid of future disasters. As the results of the analysis indicate, if children and teens are emotionally stimulated, their motivation to learn can be increased and can consequently increase the effectiveness of disaster education. Previous research supports the idea that "the fear-arousing communications" have a positive effect on behavioral change, such as recommendation acceptance \[[@B50-ijerph-17-05347]\]. Therefore, it is very important for school agencies to provide disaster education to the children and teens who are emotionally stimulated due to disasters. If these education programs are conducted right after a disaster has occurred, it would be more effective in successfully enhancing students' knowledge of disasters and protective action, substantially contributing to reducing the damage caused by future disasters. The emotional response, however, has both a positive and a negative impact on the effectiveness of disaster educations. If children and teens feel severe sadness and anger during a disaster, this can decrease the effectiveness of disaster education. In such cases, students feel helpless and tend to give up the willingness to learn helpful information designed to protect them from future disasters \[[@B9-ijerph-17-05347]\]. The most representative example of these negative emotional impacts can be found in some tragic cases relating to the loss of parents due to large disasters. In these cases, we need to provide post-traumatic stress disorder education, rather than the usual forms of disaster education. 7. Conclusions {#sec7-ijerph-17-05347} ============== The purpose of this study was to examine the factors relating to disaster experiences that impact the effectiveness of disaster education on school students (children and teens). As shown in the results, the "2018 school earthquake disaster education program in Pohang", developed by Pohang City Hall, was meaningful. The M 5.4 Pohang earthquake in 2017 was a mid-sized earthquake compared with larger incidents, such as the M 8.0 Sichuan earthquake in China in 2008 or the M 9.0 Tohoku earthquake in Japan. However, the Pohang earthquake was the second largest earthquake in Korean history, and almost all people in Pohang were shocked by it because it was so unfamiliar. In particular, the surprise and fear of children and teens in Pohang city was severe. These activated emotional states motivated students in Pohang to learn about disaster science and how to take protective actions effectively. Although the education program started nine months after the earthquake, the vivid emotions of surprise and fear still lingered in the hearts of children and teens in Pohang city. Educational institutions need to clearly understand the importance of students' emotional conditions when planning a disaster education program. Furthermore, it is also important to change students' perceptions regarding the controllability of the disaster. The message that preparedness can reduce the resulting damage may induce students to prepare more thoroughly \[[@B7-ijerph-17-05347]\], especially when the children and teens in question are emotionally stimulated. Conceptualization, J.-B.C.; methodology, J.-B.C.; software, D.-H.Y.; validation, J.-B.C.; formal analysis, D.-H.Y.; investigation, D.-H.Y. and D.-H.I.; resources, J.-B.C.; data curation, D.-H.Y. and J.-B.C.; writing---original draft preparation, D.-H.Y.; writing---review and editing, D.-H.Y. and J.-B.C.; visualization, D.-H.Y.; supervision, J.-B.C.; project administration, J.-B.C.; funding acquisition, J.-B.C. All authors have read and agreed to the published version of the manuscript. This research was supported by Human Resource Development Project in Disaster Management and Fundamental Technology Development Program for Extreme Disaster Response (2020-MOIS31-013) funded by Ministry of Interior and Safety (MOIS, Korea). The authors declare no conflict of interest. {#ijerph-17-05347-f001} {#ijerph-17-05347-f002} {#ijerph-17-05347-f003} {#ijerph-17-05347-f004} {#ijerph-17-05347-f005} ijerph-17-05347-t001_Table 1 ###### The Pohang earthquake and the Gyeongju earthquake \[[@B1-ijerph-17-05347]\]. Earthquake Date Time (UTC \*) Location of epicenter Depth (km) Magnitude --------------------- ------------------- --------------- ----------------------- ------------ ----------- ----- Pohang earthquake 15 November 2017 05:29:31 36.11° 129.37° 7 5.4 Gyeongju earthquake 12 September 2016 11:32:54 35.76° 129.19° 15 5.8 \* Universal Time Coordinated. ijerph-17-05347-t002_Table 2 ###### The demographic information of subjects. Total Number Percentage (%) ------------------- ------------- ---------------- ------- *n* = 3316 100 Gender Male 1053 31.76 Female 2263 68.24 School High school 1799 54.25 Middle school 1517 45.75 The year of birth 1999 3 0.09 2000 191 5.76 2001 839 25.30 2002 766 23.10 2003 526 15.86 2004 457 13.78 2005 531 16.01 2006 3 0.09 ijerph-17-05347-t003_Table 3 ###### Result of Principal Component Analysis of Emotional Response. Emotional Response Factor1 Factor2 Communality -------------------- --------- --------- ------------- Surprise 0.850 0.041 0.724 Fear 0.839 0.263 0.773 Joy −0.622 −0.174 0.417 Anger 0.033 0.920 0.848 Sadness 0.400 0.769 0.750 ijerph-17-05347-t004_Table 4 ###### Model Fit Indices with Recommended Values (*n* = 3316). Statistic Recommended Value Obtained Value ------------------------- ------------------- ---------------- $\chi^{2}$ \- 198.605 d.f. \- 15 $\chi^{2}$ significance *p* \> 0.05 0.000 $\chi^{2}$/d.f. \<5.0 13.240 RMR \<0.05 0.036 RMSEA \<0.08 0.061 TLI \>0.9 0.929 NFI \>0.9 0.959 ijerph-17-05347-t005_Table 5 ###### Hypothesis Testing Results and Unstandardized Path Loadings (*n* = 3316). --------------------------------------------------------------------------------------------------------------------------------------- Estimate Standardized Error *z*-Value *p*-Value Standardized\ Value --------------- ------------------------------------------- -------------- -------------------- ----------- ----------- --------------- Hypothesis 1: Perceived Earthquake Scale → Preparedness 0.004 0.011 0.357 0.721 0.014 Hypothesis 2: Emotional Response 01 → Preparedness 0.134 \*\*\* 0.024 5.583 0.000 0.520 Hypothesis 3: Emotional Response 02 → Preparedness −0.033 \*\* 0.011 −2.922 0.003 −0.163 --------------------------------------------------------------------------------------------------------------------------------------- \*\* *p* \< 0.01, \*\*\* *p* \< 0.001.

Thalassemias are common monogenic disorders caused by partial or complete reduction synthesis of one or more globin chains.^([@R1])^ The normal concentrations of fetal hemoglobin (Hb F) in adults without Hb alterations range from 0% to 1%.^([@R2])^ It is known that stimulation of Hb F production is beneficial to homozygous beta-thalassemia individuals^([@R3])^ and that the *Xmn*I polymorphism may be related to increases.^([@R4])^ The objectives of this study were to evaluate the frequency of the *Xmn*I polymorphism in heterozygous beta-thalassemia subjects and in individuals without Hb alterations, to estimate the polymorphism frequency related with beta thalassemia mutations and to correlate the presence of the *Xmn*I polymorphism with Hb F levels. A total of 325 peripheral blood samples from control (n=169) and beta thalassemia trait individuals (n=156) were submitted to classical tests for hemoglobinopathies.^([@R5])^ The presence or absence of the *Xmn*I polymorphism was analyzed in both groups by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP).^([@R6])^ Statistical analysis employed the Statistica 7.0 computer program (Statsoft Inc.) with significance being set for a p-value \< 0.05. The *Xmn*I polymorphism was observed in 36.5% of heterozygous beta-thalassemia patients and in 41.4% of control subjects. There was no statistically significant difference between the two groups (p-value = 0.32). There were significantly higher concentrations of Hb F (p-value = 0.01) in individuals with the polymorphism compared to those without ([Table 1](#T1){ref-type="table"}). ###### Fetal hemoglobin concentrations in patients with heterozygous beta-thalassemia *Xmn* I Polymorphism ---------------------------------- ------------ --------------------------------------------- --------------------------------------------- Fetal hemoglobin concentration % 3.75 ±2.61 2.58 ±2.59[\*](#T2-fn){ref-type="table-fn"} 1.55 ±1.98[\*](#T2-fn){ref-type="table-fn"} Data presented as means ± standard deviation \* Statistically significant difference between heterozygous and wild type homozygous individuals (Mann-Whitney test, p-value = 0.011) The CD39 was found in 60.25% of cases, corroborating other published results that this is the most common mutation in Southeastern and Southern Brazil.^([@R7])^Additionally, the IVS-I-110 (25.64%) and IVS-I-6 (5.12%) were found as were other unidentified mutations (8.9%). There was no statistical difference between the presence of the *Xmn*I polymorphism and beta thalassemia mutations (p = 0.99). In conclusion, the *Xmn*I polymorphism influences Hb F concentrations in patients with the beta-thalassemia trait. The presence of the polymorphic site showed no difference between heterozygous beta-thalassemia carriers and control subjects. The average levels of Hb F in individuals with heterozygous beta-thalassemia and with the *Xmn*I polymorphism were higher than normal, showing the influence of this site on the gene expression of γ-globin. Conflict-of-interest disclosure: The authors declare no competing financial interest

The incidence of childhood cancer in Europe is estimated to increase around 1% per year, with the strongest increase of 2.1% seen in children \< 1 year of age for the period 1978--1997. Better diagnostic tools and recent evolution of our genetic background can only partly explain this increase; therefore, changes in our environment are the most probable causes of the observed increase (for review, see [@r28]). Among these environmental factors, diet, including maternal diet during pregnancy, might play a key role. In this context, the European Union Framework Programme NewGeneris ([@r42]) aims to explore the possible role of *in utero* and maternal exposure to genotoxic compounds from diet and environment as a possible risk factor for the development of cancer during childhood, using biomarkers as a main research tool ([@r36]). The influence of environmental stress on children's health can be assessed by biomonitoring studies in maternal and umbilical cord blood using validated biomarkers of exposure, early genetic effects, and genetic susceptibility, in combination with new approaches, for example, (epi)genomics to facilitate mechanistic interpretations (for review, see [@r55]; [@r57]). The goal of the present study, which is being conducted as part of the NewGeneris project, is to assess early genetic effects in newborns, and hence their potential risk to develop childhood cancer, using a well-validated biomarker of cancer risk in adults, the cytokinesis block micronucleus (CBMN) assay. Micronuclei (MN) are small extranuclear bodies containing genetic material, such as acentric chromosome/chromatid fragments or whole chromosomes/chromatids, that form during cell division. They are not included in the daughter nuclei but are encapsulated into a separate, smaller nucleus, a micronucleus. MN formation occurs as a result of both direct and indirect DNA damage and can be used to classify chemicals into clastogens (which induce chromosome breakage) or aneugens (which induce chromosome loss) (for review, see [@r13]; [@r29]; [@r35]). MN frequencies in peripheral blood T lymphocytes have been extensively used as a highly reproducible and reliable biomarker of chromosomal damage, genomic stability, and cancer risk in adults ([@r4]; [@r15], [@r14]; for review, see [@r35]). In performing the assay, cells are cultured and stimulated to undergo cell division. Use of cytochalasin B, a compound that prevents cytokinesis but not nuclear division, allows discrimination between cells that did not divide \[mononucleated cells (MONO)\] *in vitro* and those that divided one \[binucleated cells (BNs)\] or more times (polynucleated cells) ([@r18]; [@r30]; [@r31]). MN frequencies in MONO cells may give an estimation of the genome instability accumulated over many years in stem cells and circulating T lymphocytes, thus before the blood was sampled, whereas MN frequencies in BN cells additionally provide a measure of the lesions that have accumulated in the DNA or in key proteins since the cells last replicated *in vivo* ([@r30]; [@r31]). Taking into account the numbers of MONO, BN, and polynucleated cells, the cytokinesis block proliferation index (CBPI) can be calculated to provide an estimate of the *in vitro* division rate. Age and sex are the major contributors to differences in MN frequencies in adult populations not exposed to occupational or environmental toxicants (for review, see [@r16]). In the present study of a well-documented NewGeneris cohort of mother--newborn pairs (the Rhea cohort from Greece--Crete), we used the CBMN assay to determine MN frequencies in both MONO and BN T lymphocytes and to derive the CBPI in 251 newborns and 223 mothers, including 182 mother--child pairs. We hypothesized that gestational factors and delivery type would influence MN levels in newborns, in addition to maternal smoking and the child's age and sex. Materials and Methods ===================== *Subject recruitment.* The recruitment of the mother--child pairs and preparation of the MN slides were performed at the University of Crete. The Rhea mother--child cohort is a study of pregnant women (Greek and immigrants) who are residents of the prefecture of Heraklion in Crete, Greece ([@r5]). Female residents (Greek and immigrants) who were pregnant during a 12-month period starting in February 2007 were contacted and asked to participate in the study. The first contact was made at the time of the first major ultrasound examination, around the 12th week of gestation. The inclusion criteria for study participants were as follows: residents in the study areas, pregnant women \> 16 years of age, first visits to hospitals or private clinics at the time of the first major ultrasound examination at the 10th--13th week of gestation, and no communication handicap. The study was approved by the ethical committee of the University Hospital in Heraklion, Crete, Greece, and all participants provided written, informed consent after complete description of the study. Structured questionnaires and medical records were used to obtain information on several factors, including maternal age, prepregnancy body mass index (BMI), lifestyle (tobacco smoking, alcohol consumption, and fruit intake), delivery type, type of anesthesia, supplement intake, birth weight, child sex, gestational age (GA), and singleton versus twin status. GA was based on the interval between the last menstrual period and the date of delivery of the baby for 84% of the subjects. When the menstrual estimate of GA was inconsistent by ≥ 7 days with the ultrasound measurement taken in the first trimester of pregnancy, a quadratic regression formula describing the relationship between crown--rump length and GA was used instead ([@r56]). *Blood collection.* Peripheral blood samples from the mothers and umbilical cord blood samples from the children were collected in heparinized tubes (BD Vacutainer, Plymouth, UK) immediately after the delivery. To prevent clotting, 0.5 mL extra heparin (Leo Pharmaceutical Products, Ballerup, Denmark) was added to the tubes used to collect umbilical cord blood. After collection, the samples were kept at 4°C and processed within 24 hr ([@r7]). *Folate and vitamin B~12~.* Plasma vitamin B~12~ and erythrocyte folate concentrations were measured in subgroups of 88 and 42 mothers, respectively, and 79 and 33 children, respectively (including 76 and 30 mother--child pairs, respectively) at the National Cancer Institute (Genoa, Italy). In vitro *CBMN assay.* The CBMN assay was carried out according to the standardized protocol developed by us for the semiautomated image analysis system ([@r8]). Sampling and making of cultures occurred in Crete. The protocol included preparation of whole-blood cultures (5 mL) within 24 hr after collection and cultivation at 37°C. Umbilical cord blood was diluted (1:3) with phosphate-buffered saline before culture preparation. After 44 hr of phytohemagglutinin A 16 (Remel, Kent, UK) stimulation, cytochalasin B (Sigma, Steinheim, Germany) was added at a final concentration of 6 μg/mL. At 72 hr, the whole-blood cultures were harvested and subjected to a cold hypotonic treatment, using 90 mM KCl (Fisher Bioreagents, Pittsburgh, PA, USA) for umbilical cord blood and 110 mM KCl for venous blood. After fixation according to the protocol, cell suspensions were dropped onto clean slides. Duplicate cultures and two slides per culture were prepared per donor. *Staining.* After slide preparation, slides were sent to Vrije Universiteit Brussel, where staining and MN analysis occurred. Slides were stained for 20 min with freshly prepared 5% Giemsa in Sorensen buffer (pH 6.8; Prosan, Merelbeke, Belgium), which was filtered twice through Whatman 41 filters (Whatman International Ltd., Maidstone, UK). *Slides scoring and interactive validation.* The automated scoring procedure followed by visual validation of selected MN cells was carried out by the same researcher (K.V.L.), using the PathFinder platform installed by Imstar (version 6; Paris, France) at the Laboratory of Cell Genetics from Vrije Universiteit Brussel, consisting of a PathFinder CELLSCAN capture station and two PathFinder MN analysis workstations. At the end of the processing step, the cells containing detected MN are presented one by one on the computer screen for visual confirmation or rejection by the scorer, according to Human MicroNucleus project (HUMN) scoring criteria ([@r17]). After validation, the total numbers of mono- and binucleated T lymphocytes with MN (MNMONO and MNBN, respectively) and without MN, and the CBPI, were recorded in a data file. CBPI was calculated as (number mononucleate cells + 2*x* number binucleate cells + 3*x* number polynucleate cells) ÷ total number of cells. *Statistics.* Consistent with Organisation for Economic Co-operation and Development (OECD) guideline T487 (OECD 2010), data were included only for subjects with at least 1,000 BN lymphocytes counted (except for two mothers and two children, where 985, 986, 995, and 997 binucleates were counted). Normality was evaluated by the Kolmogorov--Smirnov goodness-of-fit test, and because not all data were normally distributed, differences between groups were tested using the nonparametric Mann--Whitney *U*-test. Associations between genotoxicity parameters and demographic characteristics were tested using Spearman's rho correlation analysis. Multivariable linear regression analysis with backward selection was used to identify maternal and child factors that were significant predictors of the different genotoxicity parameters (MNBN, MNMONO, and CBPI). The backward selection was done by eliminating variables that were not significantly associated with the outcome (*p* \< 0.05). Each initial model included maternal age, birth weight, sex of the child (0 = boy, 1 = girl), prepregnancy BMI, GA, delivery type (0 = vaginal, 1 = cesarean delivery), smoking at the 30th week of pregnancy (0 = no, 1 = yes), and the use of antioxidant supplements during pregnancy (0 = no, 1 = yes). The models were restricted to observations with complete data for all variables. For the analysis of all newborns, preterm status (0 = term birth, 1 = GA \< 37 weeks) was added as an independent variable. Additional models of newborn data from mother--child pairs included maternal genotoxicity parameters as initial predictors. Twin births (four pairs, eight children in total) were modeled as individual observations, without accounting for correlated observations. To reduce the number of independent variables and limit multiple testing, highly intercorrelated (Spearman's rho correlation) independent variables (vitamin B~12~ and folate concentrations) were excluded from the regression model. The level of significance was set at *p* \< 0.05 for all statistical analyses. We used SPSS (version 16.0; SPSS Inc., Chicago, IL, USA) to analyze the data. Results ======= *Study population.* Demographic data were available for most of the subjects ([Table 1](#t1){ref-type="table"}). The mean age of the mothers was 29 years and ranged between 20 and 41 years. Most of the women were of Greek origin (86.2%), and their prepregnancy BMI ranged between 14 and 47. Almost all of the women took diet supplements during pregnancy (94.4%). Before becoming pregnant, 43.8% of the mothers smoked; 37.3% were still smoking at the first week of the pregnancy, and 20.1% at the 30th week. Concerning alcohol consumption, 22.6% declared to have consumed alcohol during pregnancy, resulting in a mean ± SD of 8.25 ± 31.96 g/day. Half of the deliveries (50.9%) occurred naturally, and 49.5% were without any anesthesia. Most of the deliveries with anesthesia occurred using general (39%) or spinal (42%) anesthesia. Only 19% were with epidural anesthesia. As far as the newborns are concerned, GA ranged between 33 and 41 weeks, with a mean age of 38 ± 1.38 weeks. Only 8.9% of the newborns were \< 37 weeks (preterm). The average birth weight of the newborns was 3,194 ± 439.31 g, and 51.9% of the newborns were male ([Table 1](#t1){ref-type="table"}). Median folate and vitamin B~12~ concentrations ([Table 2](#t2){ref-type="table"}) measured in 42 and 88 mothers, respectively, and 33 and 79 children, respectively, were consistent with values reported in the literature ([@r2]). ###### Clinical and lifestyle characteristics of the total mother--newborn population. Characteristic *n* Percent Mean ± SD Range ---------------------------------------------- ----- --------- ----------- ---------------- -- ------------------ -- -------------- Mothers Age (years) 201 --- 29.26 ± 4.85 20--41 BMI before pregnancy 188 --- 24.79 ± 5.91 14.68--47.46 Origin (Greece) 217 86.2 --- --- Smoking During last 3 months before pregnancy (yes) 208 43.8 --- --- During first weeks of pregnancy (yes) 209 37.3 --- --- At 12th week of pregnancy (yes) 209 19.6 --- --- At 30th week of pregnancy (yes) 214 20.1 --- --- Delivery type (vaginal) 222 50.9 --- --- Anesthesia (yes) 212 50.5 --- --- Type of anesthesia (spinal/epidural/general) 105 42.0/19.0/39.0 --- --- Supplements during pregnancy (yes) 196 94.4 --- --- Total fruit intake (g/day) 165 --- 410.40 ± 264.17 0.00--1,142 Alcohol (yes) 164 22.6 --- --- Total alcohol intake (g/day) 164 --- 8.25 ± 31.96 0.00--330.00 Wine (g/day) 164 --- 3.09 ± 11.97 0.00--106.00 Beer (g/day) 164 --- 5.13 ± 29.28 0.00--330.00 Newborns GA (weeks) 241 --- 38.33 ± 1.38 33--41 Birth weight (g) 238 --- 3194.89 ± 439.31 1,860--4,300 Sex (male) 241 51.9 --- --- Twin (yes) 241 4.1 --- --- GA \< 37 weeks (yes) 246 8.9 --- --- ###### Biomarker distribution measured in umbilical cord blood and maternal blood from the total mother--newborn population. Mothers Newborns *p*-Value (mothers vs. newborns) --------------------------------------------------------------------------------------- ---------- ---------------------------------- -- ----------------- -- -------------------------- -- ----- -- ----------------- -- ------------------------- -- ---------- Total population MNBN/1,000 BN 223 2.85 ± 2.18 2.36 (1.40--3.85) 251 1.86 ± 1.59 1.53 (0.77--2.47) \< 0.001 MNMONO/1,000 MONO 223 0.72 ± 1.04 0.42 (0.00--1.01) 251 0.62 ± 0.71 0.44 (0.00--0.97) 0.770 CBPI 223 1.64 ± 0.25 1.70 (1.47--1.83) 251 1.56 ± 0.22 1.58 (1.44--1.71) \< 0.001 Folate (ng/mL) 42 851.40 ± 365.32 794.20 (525.25--1119.53) 33 766.65 ± 272.63 714.20 (580.35--917.95) 0.445 Vitamin B~12~ (pg/mL) 88 234.67 ± 115.97 204.70 (152.28--300.28) 79 275.50 ± 156.33 256.60 (163.80--321.30) 0.130 Paired population MNBN/1,000 BN 180 2.95 ± 2.27 2.50 (1.48--3.88) 182 1.77 ± 1.41 1.55 (0.77--2.30) \< 0.001 MNMONO/1,000 MONO 180 0.73 ± 1.10 0.40 (0.00--1.00) 182 0.67 ± 0.74 0.50 (0.00--1.02) 0.313 CBPI 180 1.66 ± 0.24 1.72 (1.53--1.83) 182 1.59 ± 0.20 1.61 (1.48--1.74) \< 0.001 Folate (ng/mL) 30 880.50 ± 376.57 814.70 (525.50--1222.00) 30 770.53 ± 286.06 676.10 (577.80--929.63) 0.327 Vitamin B~12~ (pg/mL) 76 234.53 ± 117.79 203.05 (152.93--291.68) 76 275.62 ± 155.29 258.90 (164.63--319.70) 0.129 Differences between mothers and newborns were analyzed with a Mann--Whitney *U*-test. *Biomarker distribution.* Genotoxicity parameters were available for 251 newborns and 223 mothers, including 182 mother--child pairs. Average total numbers of BN and MONO T lymphocytes were 3,500 (range, 985--17,459) and 3,616 (range, 319--18,929), respectively. In the total population ([Table 2](#t2){ref-type="table"}), median MNBN frequencies were significantly higher in mothers (2.36 per 1,000 BN cells) than in newborns (1.53 per 1,000; *p* \< 0.001, Mann--Whitney). Median values were similar when restricted to data from participants in mother--child pairs ([Table 2](#t2){ref-type="table"}). Median CBPI values in maternal samples (1.70) were statistically significantly higher than those in cord blood (1.58; *p* \< 0.001). Median MNMONO frequencies were not significantly different between maternal and child samples (total population: 0.42 and 0.44 per 1,000, respectively; paired samples: 0.50 and 0.40 per 1,000, respectively). Genotoxicity parameters did not differ significantly between preterm and full-term newborns (data not shown). In newborns, MNMONO and MNBN frequencies were positively correlated with MNBN (*r* = 0.346) (data not shown). Within the pairs (data not shown), we found a significant positive correlation between the number of MNMONO/1,000 MONO from newborns and mothers (*r* = 0.263). *Multivariable analysis.* None of the variables included in the initial model (maternal age, birth weight, child sex, maternal BMI, GA, delivery type, preterm status, smoking, and supplement intake) were significant predictors of MNBN and CBPI among the 173 children with complete data for all potential predictors ([Table 3](#t3){ref-type="table"}). However, MNMONO frequency was significantly inversely associated with maternal age, GA and preterm status and positively associated with maternal prepregnancy BMI. ###### Significant predictors from multiple regression analysis with backward selection of preterm status (total newborns only), maternal age, birth weight, child sex, mother BMI, GA, delivery type, smoking, and supplement intake on genotoxicity biomarkers in newborns and mothers. Variable Partial *r*^2^ Slope *R*^2^ *p*-Value -------------------------------- ---------------- --------------------------- -------- ----------- -- ------- -- --------- Newborns (*n* = 173) MNBN/1,000 BN No significant predictors MNMONO/1,000 MONO Mother's age 0.026 --0.025 0.036 Mother's BMI 0.049 0.029 0.004 GA 0.058 --0.183 0.002 Preterm 0.026 --0.532 0.038 0.108 0.001\* CBPI No significant predictors Full-term newborns (*n* = 156) MNBN/1,000 BN No significant predictors MNMONO/1,000 MONO Child's sex 0.026 0.228 0.048 Mother's BMI 0.069 0.033 0.001 GA 0.074 --0.199 0.001 0.120 0.000\* CBPI No significant predictors Mothers (*n* = 163) MNBN/1,000 BN Child's sex 0.026 --0.677 0.041 0.026 0.041\* MNMONO/1,000 MONO Mother's age 0.027 0.037 0.036 Smoking 0.018 --0.289 0.091 0.044 0.028\* CBPI Mother's BMI 0.144 0.008 0.022 Delivery type 0.055 0.117 0.003 0.076 0.002\* \**p* \< 0.05. To reduce potential bias due to pregnancy complications associated with preterm delivery, we also conducted multivariable linear regression analysis using data from full-term births only (*n* = 156). As for all births combined, no variables significantly predicted MNBN frequency or CBPI ([Table 3](#t3){ref-type="table"}). GA was associated with significantly decreased MNMONO frequency, whereas maternal BMI was associated with increased MNMONO frequency. In addition, MNMONO frequency was significantly higher in female than in male infants. We confirmed the results in a paired sample subset (data not shown). Next, we performed multivariable regression analysis for the full-term infants with paired samples from their mothers to estimate associations with maternal genotoxicity parameters (MNBN, MNMONO, and CBPI) in addition to the variables evaluated previously (data not shown). As for previous models, we found no significant predictors of MNBN, including maternal genotoxicity parameters. In addition, maternal genotoxicity parameters did not predict child's MNMONO frequency. However, we found a significant positive association between maternal CBPI and child's CBPI (data not shown). We also evaluated predictors for maternal genotoxicity parameters ([Table 3](#t3){ref-type="table"}). In the total population (*n* = 163), mothers delivering a girl had significantly fewer MNBN than did those delivering a boy. MNMONO frequencies increased significantly with maternal age. Mothers who delivered by cesarean section had significantly higher CBPI values compared with those who delivered naturally, and higher prepregnancy BMI also was associated with higher CBPI values. Discussion ========== The present study investigated factors that predict the frequency of MN in blood samples from newborns (full term and preterm) and mothers in the Rhea cohort (Crete), conducted within the European Union project NewGeneris. Detailed questionnaires provided information concerning gestational factors, demographic data, medical status, smoking, alcohol consumption, and other factors. The CBMN assay is a well-validated reference biomarker for early genetic effects, and the present study is, to the best of our knowledge, the first to analyze MN frequencies in peripheral and umbilical cord blood from mothers and newborns using a semiautomated image analysis system. This system has several advantages, including a well-defined detection algorithm; applicability to human T lymphocytes; discrimination among MONO, BN, and polynucleated cells; scoring MN according to HUMN scoring criteria; thorough validation with false-positive and false-negative rates as low as possible; and determination of cell proliferation (for review, see [@r9]). Unique to this study is the evaluation of MNBN, MNMONO, CBPI, and GA of the newborns, whereas a weakness of the study was the limited data on folate concentration in erythrocytes and plasma vitamin B~12~ concentration. *MN in newborns.* In literature, all studies focusing on MN levels in umbilical cord blood reported low frequencies in BN T lymphocytes from newborns, consistent with our results of 1.53 per 1,000 cells (median) in the total population ([@r6]; [@r7]; [@r25]; [@r33]; [@r34], [@r38], [@r37]; [@r40]; [@r41]; [@r45], [@r44]; [@r52]; [@r58]; [@r59]). We observed a median MNMONO frequency of 0.44 per 1,000 cells and a median CBPI value of 1.58. The few previously reported data on MNMONO and CBPI values in cord blood lymphocytes are consistent with our findings ([@r7]; [@r34]; [@r45]). *MN in newborns versus mothers.* The median MNBN level in mothers was 2.36 per 1,000 cells, which was slightly lower than previously reported levels in literature ([@r24]; [@r39]), as expected from our semiautomated system because of the strict scoring criteria applied for automated image analysis ([@r9]). MN frequencies increased with age and were higher in maternal samples compared with cord blood, in agreement with the literature ([@r7]; [@r33]; [@r34]; [@r45], [@r44]; [@r52]). Although we detected lower levels than previously reported, relative differences in frequencies between newborns and their mothers were comparable with previous reports. *MNMONO, GA, and smoking in newborns.* Multivariable linear regression analysis revealed a significant inverse association between GA and MNMONO in the newborn population, with significantly lower MNMONO frequencies in preterm newborns (GA \< 37 weeks) compared with term births at ≥ 37 weeks. To reduce bias due to pregnancy complications, we performed subsequent multivariate regression analysis using data from full-term infants only. We identified no significant predictors of MNBN levels in full-term newborns, but MNMONO frequency was inversely correlated with GA. [@r59] and [@r45] observed elevated MNBN levels in newborns of smoking versus nonsmoking mothers, whereas [@r37] reported nonsignificantly higher mean MNBN frequencies in infants of nonsmoking mothers. We know of only one report on MNMONO frequencies and exposure to smoking, not in newborns but in 4- to 15-year-old children ([@r26]), and smoking in the household was associated with higher MNMONO values in children. As proposed by [@r3], tobacco smoking may induce damage to the lymphocytes, and damaged cells may not survive the culture period in the CBMN assay or may not divide. If they do not divide, they will not form a BN and will thus be scored as a MONO T lymphocyte. Moreover, some compounds of tobacco smoke can cross the placenta ([@r27]; [@r32]) and can induce DNA damage such as DNA adducts, chromosomal instability, and oxidative damage ([@r10]; [@r19]; [@r46]; [@r47]; [@r49]). Full-term newborns are well equipped to cope with the stress of oxidative damage ([@r7]). Presence of MN within a MONO lymphocyte indicates chromosome breakage/loss before the blood was sampled and thus reflects damage accumulated during pregnancy (*in utero* exposure only). In contrast, MN in BN cells may originate from preexisting MN plus DNA or protein lesions that are expressed as chromosome breaks/loss during *in vitro* replication. MNBN levels may therefore reflect genetic damage induced by stress during delivery. The different (although not significant) response we detected in MNBN levels between preterm and full-term children can result from differences in DNA repair or apoptotic capacity. A possible explanation for the higher MNBN frequencies in preterm children might be the maturation of DNA repair enzymes during the last weeks of gestation and increased transfer of antioxidants such as vitamin B, vitamin C, and beta-carotene during the final days of gestation ([@r20]; [@r21]; [@r50]; [@r51]). *Predictivity of MNMONO in newborns.* Up to now, MNMONO frequencies have been rarely taken into account in biomonitoring studies ([@r1]; [@r12]; [@r23], [@r22]; [@r54]). The value of including both MNBN and MNMONO in biomonitoring studies has been emphasized by [@r30], because of their different but complementary properties. MNMONO cells represent chromosomal damage induced and expressed *in vivo* before the start of the CBMN assay culture. MNBN cells, in addition to *in vivo* accumulated MN, reflect damage present on DNA or key proteins and expressed as MN during *in vitro* cell division. Whether MN frequencies in newborns can be interpreted in the same way as in adults, and whether they are predictive for cancer, is not known at this time. Three major facts need to be considered regarding their biological significance in newborns compared with adults. In adults, circulating T lymphocytes may accumulate MN over several years, in contrast to newborns, where damage accumulates during a short *in utero* exposure time of up to 6 months ([@r48]). Second, the response of T lymphocytes to phytohemagglutinin stimulation in umbilical cord blood is less efficient than the response of T lymphocytes in peripheral blood from adults ([@r11]). Therefore, the significantly higher CBPI in mothers than in newborns was consistent with expectations. Third, despite the fact that MNMONO cells in mothers represent damage accumulated over several years, dependent on the life span of the lymphocytes ([@r53]), and in newborns only during *in utero* exposure, we observed a significantly positive correlation between MNMONO frequencies in mothers and newborns. This suggests that maternal exposure is reflected in their newborns, and one has to consider inherited similarities in DNA repair capacity. Conclusion ========== Although confirmation in a larger study population is needed, multivariable analysis revealed the importance of taking into account GA when studying MN frequencies in newborns. In addition, our results indicate the importance of assessing both MNMONO and MNBN for biomonitoring of newborns, because the first reflects damage expressed during *in vivo* cell division and accumulated *in utero*, and the latter includes additional damage expressed as MN during the *in vitro* culture step. Because of physiological differences and the age of circulating T lymphocytes, it is not yet clear whether MN frequencies in newborns can be interpreted in the same way as in adults---that is, whether they are predictive for cancer and childhood cancer in particular. This work was financed by the European Union Integrated Project NewGeneris, 6th Framework Programme, Priority 5: Food Quality and Safety (contract FOOD-CT-2005-016320). NewGeneris is the acronym of the project "Newborns and Genotoxic Exposure Risks" (<http://www.newgeneris.org>). The authors declare they have no actual or potential competing financial interests.

Introduction {#s1} ============ Prostate cancer (PCa) is a heterogeneous trait, and it is the most common malignancy among men in western countries, including Finland. It is a multifactorial disease, and definitive risk factors include age, ethnic origin and family history. In Finland, the incidence of PCa is 89.4/100,000, and in 2010, 4697 new prostate cancer cases were diagnosed (<http://www.cancer.fi/syoparekisteri/en/>). Despite extensive research over the last decade, the etiological risk factors and genes that cause genetic susceptibility remain largely unknown. This lack of knowledge has hampered effective cancer prevention and development of better treatments. Mutations in the known high-penetrance PCa predisposition genes explain only a small fraction of PCa cases. A polygenic model for familial aggregation of cancer has been proposed where several low-penetrance alleles may have a multiplicative and/or modifying effect. One low-penetrant candidate gene is *ARLTS1* (*ARL11*), ADP-ribosylation factor like tumour suppressor protein 1, a putative tumour-suppressor gene on chromosome 13q14, which has been shown to function in many human cancers [@pone.0072040-Calin1]--[@pone.0072040-Sellick1]. *ARLTS1* is a member of the ADP-ribosylation factor family that plays a role in apoptotic signalling. The same chromosomal area on 13q has been indicated in a multi-centre genome-wide linkage study in families with at least five affected members [@pone.0072040-Xu1]. In another recent study, the immediate adjacent region 13q13 showed a suggestive linkage to PCa, with a HLOD\>1.9 [@pone.0072040-Cropp1]. In addition, the locus 13q14 is among the most frequently deleted chromosomal regions in somatic tumour tissues in both unselected and hereditary prostate cancers [@pone.0072040-Visakorpi1], [@pone.0072040-Rokman1], suggesting that *ARLTS1* could be a target for both germline and somatic mutations. We previously reported a significant association of *ARLTS1* T442C (rs3803185) homozygote carriers with PCa [@pone.0072040-Siltanen1]. This risk genotype was also associated with decreased *ARLTS1* expression in the lymphoblastoid cell line samples of PCa patients, and *ARLTS1* co-expression signatures from data mining revealed that *ARLTS1* expression was strongly associated with immune system processes. These processes are of interest because a link between chronic inflammation and PCa progression has been repeatedly proposed, and multiple genes acting in inflammatory pathways have been linked to PCa susceptibility [@pone.0072040-Sfanos1]--[@pone.0072040-DeMarzo1]. Complex diseases, such as cancer, are caused by a combination of multiple genetic interactions and environmental factors, which are hard to detect and link to each other. To determine true causal associations using statistical methods and phenotype information, it is advisable to organise individual markers into groups according to meaningful biological criteria, such as inflammation. By composing variant sets, it is possible to reduce the number of hypotheses being tested, which allows the association between a genomic feature and a phenotype to be more easily detected. Epistasis in genotype level is defined as the interaction among multiple genes or loci, and this joint genetic effect may be the factor behind "missing heritability", a phenomenon linked to the unexplained portion of hereditary cancer susceptibility, which is observed in PCa. The genome-wide expression quantitative trait loci, eQTL, analysis is a method for studying epistasis in complex traits that is able to detect associations between genotypic and expression data. The eQTL analysis is a widely used approach to gain insight into the role of single nucleotide polymorphisms (SNPs) affecting transcript levels. In this study, we investigated the mechanisms behind the previously observed association of *ARLTS1* and PCa by focusing on finding gene/expression interactions that dispose patients to PCa, including interactions involving the *ARLTS1* gene. To further investigate the *ARLTS1* expression differences seen previously in tumor samples, prostate cancer and lymphoblastoid cell lines [@pone.0072040-Siltanen1], we performed functional eQTL analysis from whole blood derived total RNA from PCa patients. The previously reported *ARLTS1* co-expression with immune system processes [@pone.0072040-Siltanen1] was tested by MDR analysis. To our knowledge, this is the first study reporting the findings of *ARLTS1* interacting variants and prostate cancer eQTLs at the 13q14 region. Materials and Methods {#s2} ===================== Study population {#s2a} ---------------- All the samples were of Finnish origin. The identification and collection of the Finnish HPC families has been described elsewhere [@pone.0072040-Schleutker1]. The familial samples analysed in this study had at least two affected first or second degree relatives. Altogether, 102 prostate cancer cases and 33 healthy male family members belonging to 31 families were initially taken into the study population. The clinical characteristics of the familial patients used in RNA sequencing (n = 84) are referred to in [Table 1](#pone-0072040-t001){ref-type="table"}. Average age at diagnosis was 63.0 y. 10.1371/journal.pone.0072040.t001 ###### Clinicopathologic findings at diagnosis of the PCa patients used in RNA sequencing (n = 84). {#pone-0072040-t001-1} *n (%)* ------------------------------------------ ----------- Age at diagnosis  \<65 years 48 (57.1)  ≥65 years 35 (41.7) Stage T stage  T1 (clinically undetectable) 31 (36.9)  T2--T4 (clinically detectable) 48 (57.1) M stage  M0 (no evidence of metastasis) 53 (63.1)  M1 (bone metastasis) 0 (0)  MX (bone metastasis cannot be assessed) 27 (32.1) PSA value  \<20 ng/ml 60 (71.4)  ≥20 ng/ml 14 (16.7) Grade Gleason score  \<7 44 (52.4)  7 10 (11.9)  \>7 0 (0) Patient information and samples were obtained with full written informed consent. The study was performed under appropriate research permissions from the Ethics Committees of the Tampere University Hospital, Finland, as well as the Ministry of Social Affairs and Health in Finland. Genome-wide SNP Genotyping {#s2b} -------------------------- Genome-wide SNP genotyping was performed using the HumanOmniExpress BeadChip microarray (Illumina, Inc., San Diego, CA, USA) by the Technology Centre, Institute for Molecular Medicine Finland (FIMM), University of Helsinki. This array covers more than 700,000 markers with an average spacing of 4 kb across the entire genome. DNA Sequencing {#s2c} -------------- The *ARLTS1* variant status of the PCa patients used in Illumina genotyping was examined by direct sequencing. Sequencing was performed in an Applied Biosystems 3130xl Genetic Analyzer (Life Technologies Corporation, Carlsbad, CA, USA) according to the manufacturer\'s instructions. Primers and PCR conditions used in the mutation screening are available upon request. RNA Extraction and sequencing {#s2d} ----------------------------- Total RNA was extracted from 84 PCa cases and 15 healthy male relatives. All subjects belonged to the 31 Finnish HPC families mentioned above. Total RNA was purified from whole blood collected in PAXgene® Blood RNA Tubes (PreAnalytiX GmbH, Switzerland/Qiagen/BD) using the MagMAX™ for Stabilized Blood Tubes RNA Isolation Kit (Ambion®/Life Technologies, Carlsbad, CA, USA) and the PAXgene Blood miRNA Kit (PreAnalytiX GmbH, Switzerland/Qiagen/BD). The RNA quality was assessed using the Agilent 2100 Bioanalyzer and the Agilent RNA 6000 Nano Kit (Agilent Technologies, Santa Clara, CA, USA). Library preparation, target enrichment and massively parallel paired-end sequencing of expressed transcripts was performed by Beijing Genomics Institute (BGI Hong Kong Co., Ltd., Tai Po, Hong Kong) using Illumina HiSeq2000 technology (Illumina Inc., San Diego, CA, USA). eQTL analysis {#s2e} ------------- To obtain *ARLTS1* expression values, first, the reads from RNA sequencing were aligned with tophat2 using hg19 as the reference genome [@pone.0072040-Trapnell1]. The raw read count for *ARLTS1* was calculated using HTseq, and read counts were transformed to normalised expression values using DESeq ([www-huber.embl.de/users/anders/HTSeq](http://www-huber.embl.de/users/anders/HTSeq)/, [@pone.0072040-Anders1]). The linear regression model implemented in PLINK was used to detect transcript specific variants. Associations in *cis* were delineated by a 1 Mb window upstream or downstream of the *ARLTS1* SNP rs9526582, as most of the *cis*-eQTLs are located within or close to the gene of interest [@pone.0072040-Veyrieras1]. In addition, we applied a new method based on probabilistic indices to test for eQTL. The new method is a non-parametric directional test (similar to the well-known Jonckheere-Terpstra test), implemented in our R-Package GeneticTools that is available on the Comprehensive R Archive Network (<http://cran.r-project.org/package=GeneticTools>), and a package description is under development (unpublished data). P-values were calculated using permutation tests and were adjusted for multiple testing by applying the Benjamini-Hochberg correction. We also calculated for each test size α in \[0,0.1\] the ratio of the amount of expected test rejections and the amount of observed rejections. The α for which the ratio of these two values was maximal, we chose also as an optimal test size. Gene expression dataset {#s2f} ----------------------- All microarray gene expression data on cell lines (n = 1445) included in these analyses are publicly available via the Gene Expression Omnibus (GEO) (<http://www.ncbi.nlm.nih.gov/geo/>, accession numbers; GSE36133, GSE7127, GSE8332, GSE10843, GSE10890, GSE12777, GSE15455, GSE18773, GSE20126, GSE21654 and GSE24795) and GSK Cancer Cell Line Genomic Profiling Data (<https://cabig.nci.nih.gov/tools/caArray_GSKdata>) (2008) [@pone.0072040-George1] (GlaxoSmithKline). The bulk of the gene expression data was acquired from the Cancer Cell Line Encyclopedia (CCLE) (GSE36133) [@pone.0072040-Barretina1]. We used samples from the most recent and widely cited Affymetrix microarray platform, HGU133_plus 2.0, to perform these analyses. Gene expression data Normalisation {#s2g} ---------------------------------- Gene expression data normalisation was performed from the raw CEL files using the Aroma Affymetrix (Version 1.3.0) R package (<http://www.aroma-project.org>) based on custom CDF files (version 16) found at <http://brainarray.mbni.med.umich.edu> [@pone.0072040-Hu1]. We processed expression for 19,003 distinct genes. All computations were performed in the R statistical environment, employing the BioConductor suite of packages. ### Co-expression analysis of *ARLTS1* {#s2g1} The co-expression analysis method was described previously [@pone.0072040-Siltanen1]. The meta cell line data (n = 1445) originated from over 48 human anatomical parts. A correlation value \>0.30 and a p-value \<0.05 were used as determinants for a statistically significant association. We performed multiple test corrections using Benjamini Hochberg and Bonferroni methods. We decided to use the Benjamini Hochberg (BH) method for selecting significantly co-expressed genes because it was moderately strict at calling a gene pair correlation a false positive. *In silico* functionality prediction {#s2h} ------------------------------------ RegulomeDB (<http://regulome.stanford.edu/>) and HaploReg (<http://www.broadinstitute.org/mammals/haploreg/haploreg.php>) [@pone.0072040-Ward1], [@pone.0072040-ENCODE1] databases were used to further elucidate the role of eQTLs in gene regulation. RegulomeDB allows he features of DNA and regulatory elements of non-coding regions to be assessed, and HaploReg is a tool for developing mechanistic hypotheses of the impact of candidate regulatory non-coding variants on clinical phenotypes and normal variation. MDR analysis {#s2i} ------------ The multifactor dimensionality reduction (MDR) program is publicly available from the internet ([www.epistasis.org](http://www.epistasis.org)). We used version 2.0_beta_8.4 to examine gene-gene interactions. MDR detects interactions in relatively small sample sizes. MDR is a non-parametric, model-free data mining approach constructive induction algorithm that transforms the high-dimensional data into one-dimensional variables by pooling genotypes into high and low risk groups based on the ratio of cases to controls that have the genotype in question [@pone.0072040-Hahn1]. MDR selects one genetic model (a one, two, three or four stage locus) that most successfully predicts the phenotype or disease status, e.g., cancer. Data are then divided into ten equal parts to perform a 10-fold cross-validation. The model creates a training set (9/10 of the data) and testing set (1/10 data) to evaluate the prediction ability. The procedure repeats this protocol ten times and calculates the cross-validation consistency (CVC). CVC depicts the number of times that particular model is chosen as the best one of those ten intervals. From the MDR results section, testing balanced accuracy (TBA) shows how many instances are correctly classified. The *ARLTS1* genotypes from 102 PCa cases and 33 controls were combined with the GWAS data of 700,000 SNPs. Selection of genes and SNPs for MDR {#s2j} ----------------------------------- The genes functioning in immune system processes and inflammation pathways were selected from a previously published comprehensive collection made by Loza MJ et al. [@pone.0072040-Loza1] because MDR is not able to run data sets of thousands of SNPs within reasonable time limits. In short, the SNPs selected included those involved in apoptosis, cytokine signalling, Toll-like receptor signalling, leukocyte signalling, complement, adhesion and natural killer cell signalling. We also ran MDR with a subset of genes gathered from our previous *ARLTS1* co-expression studies (e.g., twelve genes within B-cell receptor \[BCR\] signalling pathway). The total amount of SNPs was 12,011 of which 4,764 were found in our Illumina Human OmniExpress GWAS data. Results {#s3} ======= RNA expression {#s3a} -------------- The *ARLTS1* RNA expression levels were analysed from total RNA of 84 PCa cases and 15 controls. A significant decrease of the *ARLTS1* expression level in PCa cases was detected (p = 0.0037, [Fig. 1](#pone-0072040-g001){ref-type="fig"}). This is in concordance with our previous results from cell line, benign prostatic hyperplasia (BPH) and tumour specimen RNA expression data [@pone.0072040-Siltanen1]. {#pone-0072040-g001} eQTL analysis {#s3b} ------------- To identify possible *cis*-acting genetic variants associated with *ARLTS1* transcript levels, we performed an eQTL analysis within a special area of the 13q14 region. By a linear regression model (PLINK), we were able to detect 5 eSNPs affecting *ARLTS1* expression ([Table 2](#pone-0072040-t002){ref-type="table"}). When the calculation window was diminished from 1 Mb to 200 kb, only one SNP, rs7997377, remained. With the directional test performed by the R-Package analysis tool, 11 statistically significant eSNPs were found ([Table 2](#pone-0072040-t002){ref-type="table"}), and two were in concordance with the PLINK linear regression model results. For both test procedures a test size of 0.01 was chosen. The eSNPs found by the directional test were located mainly in non-coding regions ([Table 2](#pone-0072040-t002){ref-type="table"}). The genomic locations of the eSNPs found in the 13q14 region are visualised in [Figure 2](#pone-0072040-g002){ref-type="fig"}. Altogether, 468 genomic variants within the 1 Mb region originating from *ARLTS1* SNP rs9526582 were tested by a linear regression model and directional test. ![Schematic diagram showing the genomic locations of eSNPs gathered by eQTL analysis in PCa patients.\ The gene symbols are as follows: *CYSLTR2* (cysteinyl leukotriene receptor 2), *FNCD3A* (fibronectin type III domain containing 3A), *MLNR* (motilin receptor), *CDADC1* (cytidine and dCMP deaminase domain containing 1), *CAB39L* (calcium binding protein 39-like), *SETDB2* (SET domain, bifurcated 2/*CLLD8*), *PHF11* (PHD finger protein 11/NY-REN-34 antigen), *RCBTB1* (regulator of chromosome condensation \[RCC1\] and BTB \[POZ\] domain containing protein 1/*CLLD7*), *ARLTS1* (/*ARL11*, ADP-ribosylation factor-like 11), *EBPL* (emopamil binding protein-like), *KPNA3* (karyopherin alpha 3, importin alpha 4), *SPRYD7* (SPRY domain containing 7*/C13orf1*, chromosome open reading frame 1), *MIR3613* (microRNA 3613), *TRIM13* (tripartite motif containing 13), *KCNRG* (potassium channel regulator), *MIR-15A* and *MIR16-1* (microRNA genes 15a and 16-1) and *DLEU2* (deleted in lymphocytic leukemia 2).](pone.0072040.g002){#pone-0072040-g002} 10.1371/journal.pone.0072040.t002 ###### Chromosomal region 13q14 risk variants (eSNPs) associated with differential *ARLTS1* expression. {#pone-0072040-t002-2} SNP Gene Position Allele1 Allele2 P-value Adjusted P-value --------------------------- ---------- ---------- --------- --------- --------- ------------------ *Linear regression model* RS1886014 N/A 49321044 A G 0,007 0,369 RS7997737 *SETDB2* 50033188 G A 0,006 0,322 RS7337547 N/A 50443527 C A 0,008 0,384 RS7995192 N/A 50782599 G A 0,008 0,331 RS2532975 N/A 50945011 G A 0,005 0,322 *Directional test* RS2075610 *MLNR* 49795705 G A 0,010 0,322 RS7997737 *SETDB2* 50033188 G A 0,008 0,322 RS1543513 *SETDB2* 50034684 A C 0,007 0,322 RS9568232 *PHF11* 50089844 A G 0,000 0,322 RS9562905 N/A 50210212 A C 0,008 0,322 RS9568354 *SPRYD7* 50487993 A G 0,002 0,000 RS2580189 N/A 50806640 A G 0,009 0,322 RS2532975 N/A 50945011 G A 0,001 0,322 RS1262781 N/A 51066171 A G 0,010 0,322 RS1262774 N/A 51068896 A G 0,006 0,322 RS17074618 N/A 51153475 A G 0,006 0,322 After adjusting for multiple testing, a FDR of 39% in the linear model and approximately 32% in the directional test has to be accepted to keep the significant test results from the marginal p-values. For a more common FDR level of 10% one significant eSNP from the directional test remained significant (rs9568354). When we considered the ratio of expected and observed significant tests, a maximum ratio for α  =  0.011 in directional test was identified. For that α approximately 2.5 times more significant test results appeared than expected. Hence, we report also the above mentioned non-adjusted p-values for a significance level 0.01. With linear regression model, the amount of observed significant tests matched the amount of expected test rejections under the null hypothesis. The association of the eSNP genotypes with the *ARLTS1* expression level was calculated. A statistically significant correlation was naturally observed between all the 14 SNPs and *ARLTS1* transcript levels. The eSNP genotype - *ARLTS1* expression association box plots are depicted in [Figure 3](#pone-0072040-g003){ref-type="fig"}. {#pone-0072040-g003} The functionality and possible transcriptional regulatory effect of the 14 eSNPs within genes found by eQTL was evaluated using ENCODE-data in the RegulomeDB and HaploReg databases. Altogether nine of the fourteen eQTLs are reported in RegulomeDB. The findings in the GM12878 lymphoblastoid cell line are emphasized below because this cell line resembles the tissue type from which the RNA sequencing data was retrieved. The most substantial evidence for the regulation of *ARLTS1* was found for SNP rs2532975. Its regulatory role is supported by its location in a regulatory active region. According to the Chip-Seq data from the GM12878 cell line, rs2532975 resides in a BATF (basic leucine zipper transcription factor) binding site. In addition, using position weight matrix (PWM) matching, a CDC5 (cell cycle serine/threonine-protein kinase) binding motif has been identified that spans the genomic position of this variant. Furthermore, a promyelocytic leukemia zinc finger (PLZF) motif is reported in HaploReg. As reported by HaploReg, CDC5 binding efficiency decreases while PLZF binding increases. Additionally, the chromatin state in the region surrounding rs2532975 might adopt weak enhancer characteristics. This prediction is further strengthened by the presence of two histone marks in this region, H3k4me1 and H3k4me2, identified in the GM12878 cells. Another possible candidate for *ARLTS1* regulation is rs9562905. In a Chip-Seq study, a POLA2 binding site was identified in a human embryonic stem cell line, H1-hESC, in the region surrounding rs9562905. HaploReg predicts active enhancer characteristics in the chromatin surrounding rs9562905 in the lymphoblastoid GM12878 cells, similar to rs2532975. This interpretation is supported by the presence of two enhancer associated histone marks, H3k4me1 and H3k4me2. In addition, a FAIRE-sequencing study conducted for the GM12878 cells also implies that this region is in an open chromatin state and is therefore likely to have regulatory activity. The variants rs1543513, rs7997737 and rs9568354 share similar chromatin structural features in the GM12878 cells. According to HaploReg, the chromatin state associates with the weakly transcribed region. This prediction is confirmed by the elongation of histone mark H3k36me3 in the surrounding regions of the variants rs1543513, rs7997737 and rs9568354. According to RegulomeDB, rs1543513 is located within the Irx3 and Irx6 motifs. However, in HaploReg, the presence of these motifs is not reported. Similarly, the transcription factor binding motif of AIRE (autoimmune regulator) is reported for rs1262781 in RegulomeDB but not in HaploReg. A MZF1 (myeloid zinc finger 1) motif surrounding rs7997737 is reported in both databases. HaploReg reports a negative LOD score difference for rs7997737, which can be interpreted as lowered binding efficiency of MZF. Compared to the three variants mentioned above, rs9568232 shares similar characteristics regarding its chromatin state. According to HaploReg, this chromatin state is related to elongated transcription. The variants rs2075610 and rs1262774 are located within regions most likely under the control of epigenetic regulation by the polycomb-group proteins in the GM12878 cells. In addition, DNase-Seq studies performed for several cell lines indicate an open chromatin state surrounding rs2075610. However, two repressed state chromatin histone marks, H3k27me3 and H3k9me3, have also been identified. Variant rs1262774 is not reported in the RegulomeDB. The remaining variants are located in heterochromatin regions, according to HaploReg. For rs7995192, rs2580189 and rs1262781, this prediction is further confirmed by the presence of H3k27me3. In addition, another repressive state associated histone mark, namely H3k9me3, is present in the rs7995192 and rs2580189 regions. Although there is evidence of repressed state chromatin, RegulomeDB reports that transcription factor binding motifs do in fact span the genomic positions of rs2580189 and rs1262781. Two motifs for XBP-1 (X-box binding protein 1) and ATF6 (activating transcription factor 6) span the region of rs1262781, according to RegulomeDB. HaploReg confirms the presence of XBP-1 and ATF6 motifs and an additional motif for SOX-17 (SRY \[sex determining region Y\] box 17). HaploReg reports lower predicted binding efficiencies for XBP-1 and ATF6 and a slightly increased binding efficiency for SOX-17. RegulomeDB reports a ZNF143 (zinc finger protein 143) motif in the region surrounding SNP rs2580189. However, HaploReg does not report the presence of this motif. Taken together, the results gathered from RegulomeDB and HaploReg indicate that the *ARLTS1* eQTLs are located within regulatory areas of the 13q14 region. Co-expression analysis {#s3c} ---------------------- The GeneSapiens mRNA expression database data, including *ARLTS1* expression, from 1445 cell lines (48 cancer subtypes) and prostate cancer tumours was available for interaction studies. Altogether 1381 genes with correlation value \>0.30 and p-value \<0.05 was found to be positively correlating with the *ARLTS1* gene. Using DAVID GO functional clustering, (<http://david.abcc.ncifcrf.gov/tools.jsp>) [@pone.0072040-Huangda1], [@pone.0072040-Huangda2] a strong association with nucleus and zinc-finger protein processes was illustrated when all the genes positively correlating with *ARLTS1* expression (n = 36) in the PCa cell line cohort were taken into account (with a more stringent correlation value \>0.50). The group of nuclear processes (nucleus, intracellular organelles, transcription and DNA-binding) was enriched when data of PCa cell lines was studied. The enrichment score was 13.49 with a p-value 1.1E-32. The adjusted Benjamin score was 5.1E-30, and the cluster of zinc-finger binding proteins revealed a p-value of 9.0E-22. The same phenomenon was observed within the whole data of cell lines (meta cohort) with genes showing a correlation value \>0.30. *ARLTS1* co-expression genes (with correlation value \>0.50) from the meta cell line data revealed a category of immune system processes (B-/T-cell activation, leukocyte/lymphocyte differentiation and activation), with an enrichment score of 2.94 (p-value 5.57E-7, adjusted Benjamin score 2.9E-5). *ARLTS1* co-expression data of genes negatively correlated to *ARLTS1* identified a strong gene ontology of glycoprotein and plasma membrane protein genes in PCa cell lines (n = 2722, correlation value\<−0.50, enrichment score 52.13, p-value 3.4E-72 and 38.35, p-value 7.9E-32, respectively). Within the negatively correlating genes, a cluster of immunoglobulin domain containing proteins harboured an enrichment score of 12.23 with a p-value of 5.4E-24. The GO term "cytokine activity" revealed an enrichment score of 10.43 with a p-value of 6.5E-10 within negatively correlating genes. In addition to the result of *ARLTS1* negatively correlating genes, we also identified clusters of G-protein coupled receptors and cell-cell signalling. Top five positively and negatively *ARLTS1* correlating genes in cell line data are presented in [Table 3](#pone-0072040-t003){ref-type="table"} (upper panel). Additionally, results of the *ARLTS1* co-expression with genes (*SETDB2, PHF11, SPRYD7, MLNR*) that harbored eSNPs are presented in the lower part of the [Table 3](#pone-0072040-t003){ref-type="table"}, from the co-expression analysis performed in the meta cell line, prostate cell line data and prostate tumor data. 10.1371/journal.pone.0072040.t003 ###### *ARLTS1* Co-expression signatures from tumor specimens and cell lines. {#pone-0072040-t003-3} Gene Correlation value P-value Samples (n) pval_corrected[\*](#nt101){ref-type="table-fn"} pval_corrected[\#](#nt102){ref-type="table-fn"} ------------------------------------------- ------------------- ------------------ ------------- ------------------------------------------------- ------------------------------------------------- ***Top five genes*** *Co-expression in meta cell line data* *Positive correlation*   *BTK* 0,69 0 2818 0 0   *GPR18* 0,66 0 2818 0 0   *CXorf21* 0,66 0 2818 0 0   *P2RY8* 0,66 0 2818 0 0   *PIK3CG* 0,65 0 2818 0 0 *Negative correlation*   *NCKAP1* −0,62 1,69E-295 2818 3,22E-291 7,70E-295   *CDC42BPB* −0,59 1,64E-263 2818 3,11E-259 7,43E-263   *GIPC1* −0,57 1,10E-239 2818 2,10E-235 5,01E-239   *PTMS* −0,53 2,68E-208 2818 5,09E-204 1,22E-207   *KIAA0284* −0,53 1,04E-203 2818 1,97E-199 4,71E-203 ***Genes harboring eSNPs*** *Co-expression in meta cell line data*   *MLNR* −0,15 1,35E-16 2818 2,57E-12 2,96E-16   ***PHF11*** **0,44** **0,000** **2818** **0,000** **0,000**   ***SETDB2*** **0,64** **0,000** **2818** **0,000** **0,000**   *SPRYD7* 0,28 0,000 2818 0,000 0,000 *Co-expression in PCa cell lines*   ***MLNR*** **−0,35** **0,0352** **36** **1** **0,081**   *PHF11* 0,09 0,608 36 1 0,704   ***SETDB2*** **0,61** **9,01E**-**05** **36** **1** **0,001**   *SPRYD7* 0,17 0,336 36 1 0,451 *Co-expression in prostate tumor samples*   *MLNR* 0,03 0,821 75 1 0,934   *PHF11* −0,07 0,576 75 1 0,806   *SETDB2* −0,07 0,547 75 1 0,788   ***SPRYD7*** **−0,34** **0,003** **75** **1** **0,045** Bonferroni correction. Benjamini Hohchberg multiple testing correction. The expression of *ARLTS1* was positively correlated with the expression of *SETBD2*, in both the whole data of cell lines (meta cell line cohort) (correlation value 0.64, p-value 0.000) and in the specific PCa cell lines (correlation value 0.61, p-value 0.00009) ([Table 3](#pone-0072040-t003){ref-type="table"}). Expression of the *PHF11* gene was positively correlated with *ARLTS1* expression in the meta cell line data (correlation value 0.44, p-value 0.000). Expression of *MLNR* and *SPRYD7* was negatively correlated with *ARLTS1* expression. *ARLTS1* and *MLNR* had a correlation value of −0.35 (p-value 0.04) in PCa cell lines, and *ARLTS1* and *SPRYD7* had a correlation value of −0.34 (p-value 0.003) in prostate tumour specimens ([Table 3](#pone-0072040-t003){ref-type="table"}). The strong negative correlation of *ARLTS1* and *SPRYD7* expression levels was also validated in our transcriptome data of 84 PCa cases and 15 controls. MDR analysis {#s3d} ------------ By direct sequencing, we were able to detect six *ARLTS1* variants at the same amplicon from PCa patients included in the Illumina genotyping (n = 135). All the variants were previously known (rs117251022, rs3803186, rs147120792, rs3803185, rs138452698 and G446A \[Trp149Stop\]). To elucidate the genotypic *ARLTS1* interactions we used multifactor dimensionality reduction (MDR). Gene-gene interaction status between *ARLTS1* and genes functioning in immune system processes within 102 PCa cases and 33 controls was calculated, but we were not able to find any statistically significant *ARLTS1* interactions in this study cohort (data not shown). Discussion {#s4} ========== *ARLTS1* is a cancer-predisposing gene with proven tumour suppressor properties. However, very little evidence on function, especially on pathways, is currently available. In this study, we were able to verify downregulated *ARLTS1* expression in the blood-derived RNA of PCa patient samples, demonstrating for the first time the effect of germline alteration on *ARLTS1* expression levels. Tumour suppressor function of the *ARLTS1* gene has been previously proven by Calin GA et al., who found that transduction of full-length *ARLTS1* to A549 cells in Nu/Nu mice decreased tumour growth when compared to empty vector [@pone.0072040-Calin1]. The ability of *ARLTS1* to suppress tumour formation in preclinical models has also been observed with ovarian [@pone.0072040-Petrocca1] and lung cancer cells [@pone.0072040-Yendamuri1]. However, these results are based on somatic mutations in cancerous cell lines, whereas our result reveals a novel expression difference at the germline level, supporting the role of *ARLTS1* as a tumour suppressor gene. In the future, these types of findings could be used to enable screening and detection of at-risk patients even before clinical diagnoses. We have previously genotyped *ARLTS1* variants in prostate, breast and colorectal cancer [@pone.0072040-Siltanen2] and produced a prostate cancer follow-up study [@pone.0072040-Siltanen1]. In the first study [@pone.0072040-Siltanen2], we reported a statistically significant association with *ARLTS1* variants T442C, G194T and prostate cancer risk. However, after adjusting for multiple testing, none of the results were significant. In the follow-up study with larger sample size, we reported a statistically significant association with T442C variant and prostate cancer risk [@pone.0072040-Siltanen1]. We reported also a decreased or lost *ARLTS1* RNA or protein expression in clinical prostate tumors, prostate cancer cell lines and xenografts, supporting the role of *ARLTS*1 as a tumor suppressor gene. Thus, there is no conflict between the previous publications and this study that is confirming the tumor suppressor role of *ARLTS1*. Here, 14 eSNPs located at the 13q14 region were shown to significantly influence *ARLTS1* transcript levels in PCa patients. The eQTL analysis was performed with two methods, a linear model approach and a directional test. One advantage of our directional test method over the commonly used linear model approach is its robustness against outliers and the weaker model assumptions. In the linear model approach, the expression values of the different genotype groups are considered to follow the same normal distribution up to a location shift parameter. If the location shift parameter is not equal to zero, it generates a testing problem. The directional test only assumes that the different expression values follow certain distribution functions, so a testing problem only occurs if the distributions of the genotype groups are stochastically ordered. Interestingly, 5 of the eSNPs are located within the protein-coding genes *SETDB2*, *PHF11*, *SPRYD7* and *MLNR*. Two eQTLs were positioned in the *SETDB2* (SET domain, bifurcated 2) gene, also known as the *CLLD8* (chronic lymphocytic leukemia deletion region gene 8 protein) gene, which functions mainly in epigenetic regulation [@pone.0072040-Zhang1]. The *PHF11* gene, another hit for the eSNPs, is a positive regulator of Th1-type cytokine gene expression through nuclear factor kappa B, (NF-kB) [@pone.0072040-Zhang1], [@pone.0072040-Clarke1]. The third eSNP gene, *SPRYD7*, (SPRY domain containing 7/chromosome 13 open reading frame 1 *\[C13or1\]*) also known as chronic lymphocytic deletion region gene 6 protein, (*CLLD6*) [@pone.0072040-Mabuchi1], is proposed to have a tumour suppressor function because it has been detected to be downregulated in B-CLL patients [@pone.0072040-vanEverdink1]. The SPRY/B30.2 protein domain occurs in a variety of cellular proteins, mediates protein-protein interactions and negatively regulates cytokine activities [@pone.0072040-Nicholson1], [@pone.0072040-Woo1]. The fourth eSNP gene, *MLNR/MTLR1*, (G-protein-coupled receptor 38, GPR38) is a member of the G-protein coupled receptor 1 family and is identified as the motilin receptor [@pone.0072040-Feighner1]. The found eSNPs were not affecting the expression of the gene they resided in, suggesting the role of these eSNPs as specific regulators targeting *ARLTS1* expression. However, no explicit conclusions about the actual causal interaction between these variants and *ARLTS1* can be made without further functional validation. The eQTL result gathered in this study was from a relatively small sample set in which analyses encompassed 1 Mb up- and downstream of the *ARLTS1* gene. Thus, further validation in larger sample sets and genomic areas are warranted. However, in our results, one eSNP remained significant even after adjustments for multiple testing. Those adjustments are intricate in the case of eQTL analysis and studies with small sample sizes often suffer from a low detection rate after multiple testing adjustment [@pone.0072040-Loo1]. One possible strategy to deal with this is to accept a relatively high FDR for the sake of later validation in a larger population. Here, no potential candidate eSNPs were excluded from further studies because the α with largest ratio between observed and rejected tests was considered as optimal test size. Lymphoblastoid cell lines (LCLs) are widely used in eQTL studies because they are an easily accessible source of patient samples of a single cell type. Particularly with prostate cancer, the tumour specimens of multiple foci are hard to collect in large amounts making it difficult to reliably detect disease-associated traits. By using prostate cancer specimens, it is possible to find tissue-specific genetic effects, but the use of lymphoblastoid cell lines or whole blood of familial prostate cancer patients enables one to detect the possible heritable germline differences which may contribute to prostate cancer susceptibility. It has been argued that SNP-transcript approaches using LCLs of small sample sets are underpowered [@pone.0072040-Min1], but many studies have been able to find overlapping eQTLs in cell lines and primary tissues [@pone.0072040-Bullaughey1]--[@pone.0072040-Nica1]. Our finding of the eSNP in the *SPRYD7* gene was endorsed in the co-expression data, in which the *ARLTS1* expression levels were significantly correlated with *SPRYD7* expression, and in the prostate tumour specimens ([Table 3](#pone-0072040-t003){ref-type="table"}). Additionally, in the data from the ENCODE consortium, the significance and usage of LCLs are justified because the lymphoblastoid cell line (GM12878) is one of the three cell lines in the higher priority Tier 1 cohort [@pone.0072040-ENCODE2]. There may be a joint effect between the 13q14 genes. For example, one larger transcript variant that consists of more than one gene has been proposed with *SETDB2* and *PHF11* [@pone.0072040-Zhang1]. Very little is known about the interactions in this area, particularly the genomic collaborators of the *ARLTS1* gene. A recent study identified cellular retinoic acid binding protein 2 (CRABP2) and phosphoglycerate mutase 1 (PGAM1) as novel ARLTS1-binding proteins using the in-frame cDNA library technique [@pone.0072040-Lee1]. These proteins were not observed in the expression or genotype level of our data, so further studies are needed to elucidate the actual *ARLTS1* interacting genes and proteins. Considering the interactions between *ARLTS1* and inflammation pathway genes, it was our hypothesis that *ARLTS1* would have interactions with inflammatory genes, as chronic inflammation has been proposed as a prostate cancer risk factor. With MDR analysis, we aimed to investigate the previously observed *ARLTS1* connection with immune system processes [@pone.0072040-Siltanen1]. Another aim was to test the hypothesis of genes functioning in inflammation processes affecting prostate cancer risk. However, we failed to substantiate our hypothesis, and statistically significant association between *ARLTS1* T442C SNP and the inflammatory/immune system SNPs was not found within the 700,000 SNPs analysed. For the analysis, we extracted distinct SNP groups from 4,764 markers found from our data of 700,000 variants in total. One possible reason for the negative results may be the relatively limited sample set of 135 cases and controls because we were not able to generalise the results to the test data. We were also concerned about overfitting the data with MDR. Our previously found association between *ARLTS1* T442C and prostate cancer risk may be caused by the interaction network of different SNPs in the 13q14 region. Variant T442C may be included in a very rare haploblock, or there may still be a missing variant because the region has been connected to other cancers in addition to PCa. In conclusion, in this study we have shown that the *ARLTS1* expression level is influenced by changes in germline expression levels and that the *ARLTS1* gene is a quantitative trait locus for 14 eSNPs in the 13q14 region. Thus, our results indicate a more complicated network of changes that increase PCa risk than merely one genetic variant in *ARLTS1*. We thank all of the patients who participated in our study and acknowledge the contribution of Riina Liikala and Riitta Vaalavuo for technical assistance. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SS TW DF JPM JS. Performed the experiments: SS VL DF JPM TR. Analyzed the data: SS DF TR JPM. Contributed reagents/materials/analysis tools: JS OK. Wrote the paper: SS TW JS.

Introduction {#S0001} ============ A number of persistent toxic substances (PTS), such as persistent organic pollutants (POPs) and certain toxic metals (especially cadmium, lead and mercury), are recognised as being responsible for adverse development and health effects in children \[[1](#CIT0001)--[5](#CIT0005)\]. Foetuses and newborns are particularly sensitive to the toxic effects of certain metals and organochlorine environmental pollutants that are resistant to degradation -- they have long environmental life times (often in years). Due to their bioaccumulation along the food chain, the diet has become the primary source for humans. Foetal exposure occurs via the umbilical cord, while for breast-fed infants mother's milk can constitute a primary source. Clearly, maternal exposure is of special relevance \[[6](#CIT0006)--[8](#CIT0008)\]. The levels of PTS in maternal blood during pregnancy usually constitute a measure of the potential foetal risk. Of particular concern are long-term negative impacts on a child's mental development, immunity to disease, and future reproductive health and cancer risk. Several ongoing multidisciplinary study projects assess human exposure to PTS in different geographical regions of the world to establish linkages between environmental chemicals and health. In general, the coastal populations of the Arctic are known to have high body burdens, especially those consuming marine fish and mammals \[[9](#CIT0009)\]. The Arctic Monitoring and Assessment Programme (AMAP) started in 1991 and currently all Arctic countries are members, namely: Canada, Denmark, Finland, Iceland, Norway, Russia, Sweden and USA \[[9](#CIT0009)--[12](#CIT0012)\]. Countries located in tropical and southern areas have joined more recently, including South Africa \[[8](#CIT0008),[13](#CIT0013)\], Brazil \[[6](#CIT0006),[14](#CIT0014)\], Vietnam \[[15](#CIT0015)\] and Australia \[[16](#CIT0016),[17](#CIT0017)\]. The extent of human exposure, as measured by biological contaminant concentrations, has not been adequately investigated in South America. As Argentina features the most southern towns and settlements in the world, it is relevant to compare exposures in the Southern-Hemisphere Antarctic with those known for the northern Arctic. Furthermore, Argentina also allows a comparison between high altitude populations with coastal inhabitants. Due to the agricultural profile of South American countries and the inherent practice of pesticide utilisation, mismanagement of the latter constitutes an important environmental problem for this continent \[[18](#CIT0018)\]. Furthermore, widespread land degradation and deforestation have taken place in productive areas as a consequence of poor governance and lack of plans for using natural resources \[[19](#CIT0019)\]. Such activities inevitably accelerates climate change and, thus, land degradation accompanied by increased water pollution have negative consequences for biodiversity, the natural environment and human health. These changes have high economic costs that lead to increased poverty. The rising temperature related to the climate change will also alter human exposure to environmental contaminants significantly, especially for metals released from icebergs, oceans and rivers (including mercury) \[[20](#CIT0020)\]. Furthermore, extensive distribution by long-range transport across oceans implies that remote areas (regarded initially as environmentally clean) receive contaminants that are transferred to the food web and humans \[[9](#CIT0009)\]. Due to the volatility and particulate properties of contaminants and the dominant atmospheric and oceanic south--north circulations, contaminants have been shown to be transported towards the circumpolar Arctic regions of the Northern Hemisphere \[[9](#CIT0009)\]. In comparison, there is less knowledge about comparable long-distance transfer in the Southern Hemisphere. In the current paper we describe the framework for an AMAP-compatible study that was given the acronym EMASAR (*Estudio del Medio Ambiente y la Salud Reproductiva* -- Study on Environment and Reproductive Health). It was designed to investigate maternal/foetal health risks related to food security and exposure to PTS in two regions of Argentina and food. The EMASAR assessed concentrations of environmental toxins in the blood of delivering women and collected pertinent personal, dietary and medical information. More specifically, we describe the overall study design and report and compare basic demographic features of the two study populations and clinical chemistry findings for selected residents of the Argentinian cities of Salta in the northwest and Ushuaia in the south. The current article constitutes the background and context for a number of biomonitoring publications concerning PTS exposures. Materials and methods {#S0002} ===================== Geographical information {#S0002-S2001} ------------------------ The geographical locations of the EMASAR study sites are shown in [Figure 1](#F0001). The City of Salta is the capital of Salta Province, which is located in the Lerma Valley of the Andes Mountains foothills and is at 24.5° south of the Equator. The metropolitan area has a population of around 620,000 inhabitants, which makes it the second most populated city in northwestern Argentina, while it is 1,210,000 for Salta Province. Its climate is highland subtropical, with average monthly temperatures ranging from 12 to 22°C. Based on its inland location and its livestock economy and tradition, beef is the main component of the diet. Salta's economy is diverse, but relatively under-developed; poverty is a general feature and there are large socioeconomic inequalities.Figure 1.Map of South America with the study areas Salta and Ushuaia, Argentina. Ushuaia is the capital of the Argentinian Province of Tierra del Fuego, Antártida e Islas del Atlántico Sur. It is the southernmost city in the world, located at sea level and 54.5° south of the Equator and has a sub-polar oceanic climate. The average monthly temperature varies from 1 to 10°C. The city has a population of some 60,000 people, while the province has around 130,000 inhabitants. The main economic activities in Ushuaia are fishing, natural gas and oil extraction, sheep farming and ecotourism. Having a free port status, the city also attracts advanced industry with salaries above the national standards. Qualified personnel come from other parts of Argentina to work for both short and long periods. Consequently, the population of Ushuaia is neither homogenous nor stable over time, while the socioeconomic conditions are among the most prosperous in Argentina. Study design and study populations {#S0002-S2002} ---------------------------------- EMASAR is an observational study, with a cross-sectional design that was compatible with the circumpolar programme in the context of the AMAP \[[9](#CIT0009)\]. The field work was conducted at the Hospital Público Materno Infantil de Salta and at the Clínica San Jorge in Ushuaia. The hospital in the City of Salta is a public institution, which is responsible for all in-hospital deliveries in the city and is the referral hospital for the province. Clínica San Jorge in the City of Ushuaia is a private institution co-responsible with a public hospital for the in-hospital deliveries in the city and the surrounding provincial areas. EMASAR is a collaborative project between the UiT The Arctic University of Norway, Tromsø, the Stavanger University Hospital, Stavanger (Norway) and the two Argentinian partner hospitals mentioned. During the preparation period and throughout the study period, the project group carried out several organisational meetings at the two study sites. The local research co-workers were thoroughly informed about the protocol and the sampling procedures. All the questionnaires were prepared in English and subsequently translated into Spanish. Individual pre-numbered sampling kits were assembled in Norway and included printed questionnaires, blood sampling equipment, sample transfer vials and comprehensive written instructions for the local staff. Recruiting of the study cohort took place over the period April 2011--March 2012 and included women who either were about to deliver or had given birth within the last 48 hours at one of the two hospitals. The participants were recruited shortly before or at admission to the delivery unit. The mothers had to be above 18 years of age, be willing to participate and written informed consent was required. We estimate that more than 90% of invited women consented to participate, although refusals were not systematically registered. A total of 717 women were recruited; of these, 19 from Salta were excluded due to a lack of biological samples, yielding a final study population of 698 (200 from Ushuaia and 498 from Salta). The study (\#2010/7317) was approved by the Ethics Committee of the Salta Medical Association and the Ministries of Health in both provinces. As required by law, the Norwegian Regional Committee for Medical and Health Research Ethics (REC North) approved the study (\#2011/706). The study was conducted in accordance with the Helsinki declaration. Data collection and analyses {#S0002-S2003} ---------------------------- ### Questionnaires {#S0002-S2003-S3001} Participants were examined and interviewed by the presiding midwife or obstetrician. The questionnaire was based on that used in comparable studies \[[13](#CIT0013),[15](#CIT0015)\], with additional questions and adjustments pertaining to the Argentinian context. The information collected included: maternal age, previous children and breastfeeding experiences, medical history (e.g. chronic diseases, number of amalgam fillings and silicone implants), socioeconomic factors (place of birth, family conditions, housing, education, employment), environmental factors (use of fuel for warming or cooking, potential exposures to chemicals/pollution/pesticides at work or at home), lifestyle (use of coffee, tea/maté, tobacco, coca leaves and alcohol) and information on diet before and during pregnancy. The questions about diet pertained to the frequency of intake (never or seldom, at least once a week or almost every day) of various basic food categories: eggs, meat (red meat, poultry, processed or tinned meat); fish (tinned, smoked, processed, freshwater or seafood); fruits and vegetables (root, leafy/ground and other); dairy products (milk, butter, cheese); fats (oils, margarine); carbohydrates (cereals, noodles, pasta, bread, sugar); and fluids (juices, soft drinks, bottled water). We also inquired about the use of vitamin supplements and personal care products (e.g. creams, lotions, sun protection products, deodorant, soap, hair products/hair dye and cosmetics). ### Clinical information {#S0002-S2003-S3002} Clinical obstetrical data are based on hospital records and a medical doctor completed a standardised form. Information on the history of earlier pregnancies was also sought, along with obstetrical and neonatal data for the current delivery. The latter included date of delivery, gestational age (as clinically estimated; Naegele term), as well as the weight, length, head circumference, sex of the neonate and any notable malformation. ### Sampling and analyses {#S0002-S2003-S3003} Maternal blood, urine and hair samples were obtained at 36±12 hours after the delivery (optimally in the morning). Use of personal products prior to blood sampling, time since last meal and last cup of coffee, as well as time of blood sampling and of freezing were recorded. Measurement of maternal height (in cm) and weight (in kg) were obtained after the delivery and, when possible, coincided with the blood sampling. Non-fasting venous blood samples were collected from the antecubital vein from the mothers. For the chemical analyses of toxic and essential elements, blood was drawn into a BD Vacutainer® for trace elements (BD Hemogard™ Royal Blue, Ref\# 368381; plastic, 6-ml, with 10.8 mg K~2~ EDTA) and transferred into pre-rinsed cryo-vials (Sarstedt CryoPure 5.0 ml and 2.0 ml tube). Additionally, five BD Vacutainers® (BD SST II Plus Advance 10/8.5 ml) were sampled after centrifugation at 2000 relative centrifugal force (RCF) for 10 minutes. The maternal serum was then transferred into four cryo-vials (Sarstedt CryoPure 2.0 ml tubes) for general analyses, was subsequently apportioned into three glass vials (4.0 ml, pre-rinsed with *n*-hexane/acetone) and retained for chemical analyses of POPs and selected haematological/hormonal parameters. The latter were analysed immediately at the local hospital laboratories in Salta and Ushuaia and included the routine analyses of reactive protein C (CRP), ferritin, follicle-stimulating hormone (FSH), luteinizing hormone (LH), prolactin, oestrogen, progesterone and lipid profile (cholesterol, triglycerides, phospholipids, HDL, LDL). Blood samples from the newborns were collected as part of routine sampling at the hospital. Venous samples were drawn from the heel capillary with standard equipment into vials (BD Microtainer® Brand Tubes, 2.0 ml). Samples were centrifuged at 1200 RCF for 10 minutes and the plasma was transferred into cryovials (Sarstedt CryoPure 2.0 ml tubes). A maternal morning spot urine was also collected (Sarstedt inc., Newton, NC), which was transferred into two 10 ml vials (BD Falcon; Becton Dickinson (BD), Plymouth, UK). Maternal hair strands (\~5 cm length) were taken close to the scalp from the occipital lobe and tied with a cotton thread and subsequently stored in an envelope at room temperature until shipping to Norway. Both the urine and blood samples collected for contaminant analyses were frozen immediately after preparation and stored at −20°C at the local hospital until they were shipped frozen to Norway and stored in the EMASAR biobank at the the UiT The Arctic University of Norway at −35°C until analysis. For the chemical analyses, blood samples were later transferred in a frozen state to the Department of Environmental Chemistry, Institute of Environmental Assessment and Water Research (IDAEA), Spanish Council for Scientific Research, Barcelona, Spain for chemical analyses of persistent organic pollutants and to the Department of Environmental Sciences, "Jožef Stefan" Institute (JSI), Ljubljana, Slovenia for determination of toxic and essential elements. More detailed descriptions of the analytical procedures and the quality control measures employed will be reported in upcoming articles. Follow-up of the children {#S0002-S2004} ------------------------- The participating mother was invited to bring her child to a 1-year follow-up examination during which a clinical examination was performed and a questionnaire on the health and development of the child during the first year of life was completed. Anthropometric measurements were also made (specifically weight, height, head and abdominal circumference). Statistical analyses {#S0002-S2005} -------------------- Statistical analyses were carried out using the IBM SPSS-Statistics for Windows statistical package version 24 (SPSS Inc. Chicago, IL). For comparisions of information between the two study sites, non-parametric tests were performed for interval data with non-normal distributions; Kruskal Wallis or Mann--Whitney U-test when unequal or equal variance, respectively. For categorical data, the Mann--Whitney U-test was used. The level of significance was set at p \< 0.05. Results {#S0003} ======= Some personal and obstetrical characteristics of the women and the pregnancy outcomes in the study are presented in [Tables 1](#T0001) and [2](#T0002), respectively. The women in Ushuaia were, on average, 4 years older than the women in Salta (28.8 vs. 24.7 years). In Ushuaia, most women were married/cohabitants, while in Salta one third were single; the latter were less educated, had more amalgam fillings and the proportion of current smokers was higher (9.6 vs. 4.5%). Nevertheless, the overall smoking rate prior to pregnancy was comparable, but significant (≥ 25% did so). More Saltanean women were exposed to passive smoking at home (42 vs. 28%). The mean pre-pregnancy body mass indeces (BMIs) were the same, namely 24. During pregnancy, Ushuaian women gained 5 kg more body weight than those in Salta. The women in Ushuaia had in general more permanent work compared to the Salta women (66% vs. 17%). Selected maternal and newborn characteristics are summarised in [Table 2](#T0002). The proportion of first-time mothers was 44% in Salta and 41% in Ushuaia and the former had a higher birth rate, with 16% being para four or more. Saltanean women reported longer breastfeeding periods and fewer used vitamin and folic acids supplements both before and during pregnancy. According to the medical reports, gestational diabetes, hypertensive disorders and anaemia (Hgb \< 9.0 g/dL) were almost non-existent among the women from Salta, with only one case of each reported; by contrast in Ushuaia there were 12, 11 and 2 cases, respectively (data not shown). The births took place at a gestational age close to 39 weeks at both sites, with a range of 32--41 for Ushuaia and 33--42 for Salta. Caesarean sections were more frequent in Ushuaia, specifically 43% of the deliveries compared to only 6% in Salta. The newborns in Ushuaia were slightly heavier, as shown in [Table 2](#T0002).Table 1.Personal characteristics of the study population from Ushuaia and Salta. Ushuaia (n = 200)Salta (n = 98)  nMean (SD) or n (%)50thmin--maxnMean (SD) or n (%)50thmin--maxp-value^a^Age20028.8 (6.5)2816--4549824.7 (6.2)2314--44\*\*\* ≤20 28 (14.0)   177 (35.5)    21--25 45 (22.5)   133 (26.7)    26--30 55 (27.5)   102 (20.5)    31--35 38 (19.0)   57 (11.4)    \>35 34 (17.0)   29 (5.8)   Marital status200   498   \*\*\* Married 72 (36.0)   56 (11.2)    Cohabiting 106 (53.0)   283 (56.8)    Divorced 1 (0,5)   3 (0.6)    Single 21 (10.5)   156 (31.3)   Education200   498   \*\*\* Primary 7 (3.5)   161 (32.3)    Secondary 97 (48.5)   284 (57.0)    Tertiary 56 (28.0)   40 (8.0)    University 40 (20.0)   12 (2.4)   Permanent job200132 (66.0)  49484 (17.0)   Height, cm200162 (6.0)162147--181496158 (5.8)158140--176\*\*\*Weight, pre-pregnancy, kilo^b^19262 (11.0)6040--11144959 (11.0)5735--109\*\*\*BMI, pre-pregnancy, kg/m^2^19224 (4.0)2316--4144724 (4.2)2315--40 Weight, post-partum, kilo20074 (11.0)7250--12049065 (11.2)6440--111\*\*\*Smoking, current2009 (4.5)  49848 (9.6)  \*Smoking last year20057 (28.5)  498127 (25.5)   Home indoor smoking20056 (28.0)  495208 (42.0)  \*\*Amalgam fillings ever19989 (44.7)  481338 (70.3)  \*\*\*Year of living current home2006.7 (7.6)31--3249711.1 (9.7)91--42\*\*\*Nationality200   497     Argentina 197 (98.5)   482 (96.8)    Other^c^ 3 (1.5)   15 (3)   Year of examination200   498   \*\*\* 2011 200 (100)   366 (73.5)    2012     132 (26.5)   [^1] Table 2.Selected maternal and newborn characeristics for the two study groups. Ushuaia (n = 200)Salta (n = 498)p-value^a^ nMean (SD) or n (%)50thmin-maxnMean (SD) or n (%)50thmin-maxParity^b^2001.9 (0.96)21--74982.2 (1.5)21--8\*\* 1 82 (41.0)   221 (44.4)    2 75 (37.5)   121 (24.3)    3 33 (16.5)   78 (15.7)    ≥4 10 (5.0)   78 (15.6)   Previous breastfeeding, month11620.4 (19.6)160--15626834.3 (29.0)240--217\*\*\*Vitamin supplements before pregnancy19830 (15.2)  49332 (6.5)  \*\*\*Vitamin supplements during pregnancy19877 (38.9)  490137 (28.0)  \*\*Folic acid before pregnancy19646 (23.5)  48242 (8.7)  \*\*\*Folic acid during pregnancy199168 (84.4)  494289 (58.5)  \*\*\*Caesarean Section, elective19946 (23.0)  4989 (1.8)  \*\*\*Caesarean Section, acute19941 (20.5)  49819 (3.8)  \*\*\*Gender newborn, girl*200*104 (52)  479257 (53.7)   Length, cm19849.6 (2.1)5042--5449248.5 (2.2)4941--55\*\*\*Weight, kilo1983.38 (0.44)3.382.12--4.504913.29 (0.48)3.31.65--5.20\*Head, cm19734.9 (1.5)3531--4049134.3 (1.4)3428--38\*\*\*Gestational age, weeks19938.8 (1.4)3932--4146138.8 (1.3)38.833--42 [^2] Various characteristics concerning housing, family situation and social environment are summarised in [Table 3](#T0003). In both communities the participants lived in homes, although almost twice as many people lived together in the same house in Salta (seven vs. four in Ushuaia). In Salta, most (\~80%) owned their home, while in Ushuaia more dwellings were rental units and there was more dwelling place mobility. Although partners in Salta had lower education, employment was high in both communities and most study participants resided in urban areas.Table 3.Environmental characteristics of the study population from Ushuaia and Salta. Ushuaia (n = 200)Salta (n = 498)p-value^a^ nMean (SD) or n (%)50thmin--maxnMean (SD) or n (%)50thmin--maxArea home          Urban residence 183 (91.5)   432 (86.7)    Semi-urban 15 (7.5)   35 (7)    Rural 2 (1)   31 (6.2)   Current home          Years of living 6.7 (7.6)31--32 11.1 (9.7)91--42\*\*\* Owned 110 (55.0)   387 (77.7)  \*\*\* Rented 86 (43.0)   56 (11.2)  \*\*\* Other 9 (4.5)   72 (14.5)   Type of home          House 138 (69.0)   393 (78.9)  \*\* Flat 62 (31.0)   23 (4.6)  \*\*\* Shared     108 (21.7)    Tenement     12 (2.4)    Informal     43 (8.6)    Other     125 (25.1)   Fuel cooking          Gas 200 (100)   492 (98.8)    Wood 1 (0.5)   34 (6.8)  \*\* Coal     24 (4.8)  \*\* Electricity 1 (0.5)   6 (1.2)    Other 4 (2.0)   2 (0.4)  \*Fuel heating          Gas 195 (97.5)   73 (14.7)  \*\*\* Wood 5 (2.5)   22 (4.4)    Electricity 11 (5.5)   242 (48.6)  \*\*\* Coal     27 (5.4)    None     152 (30.5)    Other 13 (6.5)   4 (0.8)  \*\*\*Drinking water, sources          Running/community tap 163 (81.5)   491 (98.6)  \*\*\* Well, rain, river 1 (0.5)   1 (1.4)    Other sources (bottled) 56 (28.0)   5 (1.0)  \*\*\*People in home1994.3 (1.6)42--154976.6 (3.5)61--23\*\*\* Partner, permanent job187177 (94.7)  338325 (96.2)    Partner, education186   340   \*\*\*Primary education 19 (10.2)   144 (42.4)   Secondary education 106 (57.0)   169 (49.7)   Tertiary education 25 (13.4)   17 (5.0)   University education 35 (18.8)   8 (2.4)   No education 1 (0.5)   2 (0.6)   Harvesting own food          Growing own food2004 (2.0)  49152 (10.6)  \*\*\* Go fishing (family) 67 (33.5)   139 (28.1)    Eat their own fish 41 (61.2)   101 (71.6)   Pollutants          Lead use repairing home20053 (26.5)  497231 (46.5)  \*\* do not know 57 (28.5)   28 (5.6)    Pollution around home20018 (9.0)  494277 (56.1)  \*\*\* Pesticide insect control home1959 (4.6)  497398 (80.1)  \*\*\* Pesticide use garden1972 (1.0)  49053 (10.8)  \*\*\* Control programme insects^b^200   467157 (33.6)   [^3] Gas was the primary home cooking fuel in both communities; while it dominated as the heating fuel in Salta, electricity did so in Ushuaia (see [Table 3](#T0003)). Running/community tap water were the dominating sources of drinking water in both Salta (96%) and Ushuaia (81%); for the latter, bottled water was the secondary source. Growing/harvesting of food was seldom done (11% in Salta and 2% in Ushuaia) and the consumption of caught fish was relatively common (60--70%) in both places. Almost half of the Salta women and very few from Ushuaia reported use of lead-containing materials during home repair. Environmental pollution sources in the vicinity of dwellings was more widespread in Salta (56%) than in in Ushuaia (9%) and the use of pesticides for insect control in homes was most common in Salta (80%), where a control programme for chagas, yellow fever or malaria existed. Details about the food frequency intake during pregnancy are reported in Supplementary Table S1 and illustrated in [Figure 2(a](#F0002),[b](#F0002)). There were almost no differences in frequency intake before (not shown) or during during pregnancy, nor between the communities for items like vegetables (root and leafy), butter and cheese and noodles, rice and pasta. Women in Salta reported a more frequent intake of meat, eggs, vegetables, cereals, fats and sugar. Fish intake was generally low for both sites, with women from Ushuaia eating slightly more, although there was somewhat higher intake of freshwater and tinned fish in Salta.Figure 2.Maternal intake (%) of dietary items in (a) Ushia and (b) Salta. The intake frequencies depicted were significantly different between the two communities at p \< 0.05 or better, with the exception of vegetables roots, fruits and butter/cheese (p ≥ 0.05) (see Supplementary Table S1). Haematological data are presented in [Table 4](#T0004) and significant differences (p \< 0.001) are evident for all. The clinical chemistry data for CRP, ferritin, total cholesterol, triglyceride and LDL were higher in Ushuaia, while those of estradiol were lower (p\<0.001). The serum estradiol data depicted in [Figure 3](#F0003) indicate that significant changes occurred during the first days post-partum.Table 4.Concentrations of maternal serum CRP, ferritin, lipids and estradiol sampled at a median of 1-day post partum. Ushuaia (n = 200)Salta (n = 498)p-value nMeanSDMedianMinMaxnMeanSDMedianMinMaxC-reactive protein (CRP), mg/L19956424532804714133315331\<0.001Ferritin, ng/mL19936322662454641816132181\<0.001Cholesterol, total, mg/dL199225652198860247121045203119394\<0.001Triglycerides, mg/dL19922076207724884711836817353496\<0.001Phospholipids, mg/dl      47120845201118357 HDL Cholesterol, mg/dl      47147124722137 LDL Cholesterol, mg/dl19613453131333864711273512812297\<0.001Estradiol, pg/mL^a^19918112414829883470731672499743813\<0.001[^4] Figure 3.Distribution of maternal estradiol (pg/mL) by day of sampling post-partum. Discussion {#S0004} ========== Differences in socio-economic status and in regional development between the southern area (Ushuaia) and the poorer region of Salta are evident. The latter has more unemployment and a different social structure. Salta had a higher birth rate (16% being para four or more), longer breastfeeding periods and less use of vitamin and folic acids supplements, both before and during pregnancy. The single case of anaemia in Salta, compared to two in Ushuaia, seems inconsistent with the lower serum ferritin concentrations observed there. Perhaps this and the absence of reported cases of gestational diabetes and hypertensive disorders in Salta may well reflect under-reporting or under-diagnosis. In Ushuaia, 43% Caesarean deliveries were registered compared to only 6% in Salta and this reflects its globally established association with a higher socio-economic status \[[21](#CIT0021)\]. The elevated CRP concentrations observed in Ushuaia presumably reflect the higher rate of Caesarean deliveries \[[22](#CIT0022)\]. In a preliminary analysis of the EMSAR data \[[23](#CIT0023)\], a limited set of organochlorine compounds (OCs) in blood serum indicated dependencies of the concentrations on BMI, parity and region of residence. It is generally recognised that lactation transfers OCs to a newly born baby \[[24](#CIT0024)\]. There is also enough evidence that diet constitutes an important source of such environmental contaminants \[[9](#CIT0009)\]. More detailed analyses of the full EMASAR study data set on concentrations, sources and pathways of both OCs and inorganic substances (including essential elements) will be presented in forthcoming papers. Regional differences in exposure to POPs and inorganic elements will be evaluated, respectively, by comparisons of the more historic *p,p′*-DDT/*p,p′*-DDE ratio and selected PCB isomer ratios in sera, as well as of pertinent toxic-to-essential element ratios in whole blood \[[25](#CIT0025)\]. Although the personal and lifestyle information obtained and reported is somewhat limited, we feel that the described data set is adequate to examine in some detail the influence/modulation of maternal body burdens of PTS and foetal exposure in relation to maternal characteristics, parity, breast feeding history, living and environmental conditions/issues \[[26](#CIT0026),[27](#CIT0027)\]. Potential relationships of PTS to the clinical chemistry parameters measured are anticipated, such as the association of OCs with circulating hormones such as estradiol \[[28](#CIT0028),[29](#CIT0029)\] and the influence of reduced maternal lipid concentrations post-partum \[[30](#CIT0030),[31](#CIT0031)\], as well as factors that can enhance the uptake of inorganic elements, such as low iron status \[[25](#CIT0025)\]. Associations of the maternal body burdens of organochlorine and inorganic toxicants with birth weight, birth length and gestational age of the newborns will also be examined. Concluding remarks {#S0005} ================== The documented misuse of pesticides in agricultural production and food production is a real threat to the food security, especially for those of reproductive age and pregnant women, since the most vulnerable period in human life is during pregnancy and early childhood \[[18](#CIT0018)\]. There is an urgent need for a comprehensive assessment of exposures in areas of the Southern Hemisphere. We conclude that our comprehensive data set and the planned publications of observed concentrations of inorganic and organic environmental contaminants in both mothers and their newborns, as well as their interpretation and the anticipated follow-up of the children, will contribute to this objective. Supplementary Material ====================== ###### Supplemental_Materials.docx ###### Click here for additional data file. The authors acknowledge the Health Ministries of the provinces Salta and Tierra del Fuego, Argentina for their administrative support and facilitation. We acknowledge Dr. Martin de la Arena for his assistance in obtaining the ethics approval and initiating the study in Argentina, Bente A. Augdal for designing the biological sampling procedures and managing the study biobank in Norway, Dra. Silvia dib Ashur for implementation of the routines for biological sampling and Lic. Maria José Aleman (Salta) and Maria Florence Bressan (Ushuaia) for their key roles in the local administration of the study. The authors also wish to thank The Norwegian Ministry of Foreign Affairs and The Arctic Monitoring and Assessment Programme (AMAP) for generous funding of the study. Disclosure statement {#S0006} ==================== No potential conflict of interest was reported by the authors. Supplemental data {#S0007} ================= Supplemental data for this article can be accessed [here](https://doi.org/10.1080/22423982.2017.1364598). [^1]: ^a^\*, \*\*, \*\*\* represent p-value \<0.05, 0.010, 0.001, respectively; ^b^ based on self-reported pre-pregnancy weight; ^c^ Chile, Peru, Korean (Ushuaia); Bolivia (n = 13) and Paraguay (n = 2) (Salta). [^2]: ^a^\*, \*\*, \*\*\* represent *p*-value \<0.05, 0.010, 0.001, respectively; ^b^ Total number of childbirths including stillborns after week 23. Twins count as one birth and one pair of twins is included. [^3]: ^a^\*, \*\*, \*\*\* represent p-value \<0.05, \<0.010, \<0.001, respectively; ^b^Changas, Dengue, Yellow Fever or Malaria Control Programme. [^4]: ^a^ The concentrations mU/mL of follicle-stimulating hormone (FSH) and luteinising hormone (LH) were mostly at around the detection limit.

Background ========== The health and social costs associated with alcohol and other drug (AOD) use are considerable and fall most heavily on young people \[[@B1]-[@B3]\]. This in itself is an argument for early AOD prevention programs. School drug education offers the potential to prevent problems by equipping young people with the knowledge and skills to make responsible decisions about AOD use, and it has near universal reach in developed countries where the great majority of young people attend secondary school \[[@B4]\]. However, historical approaches to school drug education have not been particularly successful at reducing AOD use \[[@B5]-[@B7]\]. This then poses the question as to whether effectiveness should be measured by abstinence or reduced use, or whether harm reduction is a more realistic and useful measure. Harm reduction programs offer greater promise of achieving worthwhile benefit because they have the flexibility to select strategies on the basis of evidence of effect. Within this model abstinence or reduced use strategies may be chosen if there is evidence that they reduce harm, but they are not goals in their own right \[[@B8]\]. The School Health and Alcohol Harm Reduction Project (SHAHRP) demonstrated the effectiveness of a skills based, harm reduction intervention. Students who received the SHAHRP program experienced 22.9% less alcohol-related harm than their control peers \[[@B9]\]. A more recent study of computerised harm reduction prevention similarly reported that alcohol consumption, risky drinking and alcohol-related harms increased to a lesser extent among female intervention students, although there was no program effect among male students \[[@B10]\]. A related study that targeted cannabis, as well as alcohol, found that weekly alcohol consumption of intervention students decreased, while that of their control peers increased. The frequency of drinking to excess also increased to a lesser extent among the intervention students \[[@B11]\]. Drug education has developed considerably in Australia over the past decade. However, evidence-based practice tends to be best reflected in demonstration programs that focus on a single drug type, such as alcohol \[[@B9],[@B12]\]. The most recent findings have not been brought together in a mass drug education program that targets all forms of drug use and can be readily accommodated within a secondary school health curriculum. Past classroom-based programs have not well exploited the important influence that parents have on their children in terms of AOD choices and the likely benefits of increased communication on this issue \[[@B13]\]. Finally, while most Australian school drug education programs are grounded in the research literature and principles of effective practice there has been insufficient attention to the pedagogy required to deliver effective classroom lessons \[[@B14]\]. Drug education is most effective when inclusive, interactive teaching strategies actively engage students in the learning process \[[@B14],[@B15]\]. However, reviews of drug education program implementation consistently identify a breakdown in fidelity when learning tasks are interactive \[[@B16]-[@B18]\]. This research study seeks to extend earlier Australian harm reduction education research by investigating the prevention effects of a single program for both licit and illicit drugs. The size and nature of the intervention have been tailored to facilitate incorporation within an already crowded school curriculum. It draws on the recent literature as to the common elements in effective school drug education programs \[[@B19]-[@B22]\]. It seeks to incorporate the social influence of parents through structured home activities that complement the classroom lessons. It emphasises the use of appropriate pedagogy through two days of training, where participatory delivery of each lesson is modelled to ensure teachers are equipped to teach the program as intended. All these elements have been trialled and refined during the course of a two year pilot program that was undertaken in four Victorian government secondary schools (three intervention and one control) \[[@B23]\]. Conceptual underpinnings ------------------------ The intervention at the heart of this study is grounded in social learning theory, which posits that human learning occurs in a social context \[[@B24]\]. In this model, drug use is socially learned through modelling, imitation and reinforcement, and influenced by an individual\'s cognitions, attitudes, and beliefs. The corollary is that drug education can use the same learning processes to equip students with the skills to recognise these influences and develop a repertoire of counter behaviours. Social learning theory underpins all of the most effective school drug education programs, and a number of researchers have indicated that it should be the model of choice in any prevention program \[[@B7],[@B19],[@B25]\]. The intervention also draws on two other theoretical models, poststructuralist subjectivity and cognitive dissonance. Poststructuralist subjectivity theory provides a way of understanding how students\' self concepts, and hence their approach to drug use, can change. Cognitive dissonance theory provides a way of understanding how students resolve conflicting ideas on drug use. Poststructuralist subjectivity theory refers to the ways in which the social world is involved in shaping an individuals\' sense of self \[[@B26],[@B27]\]. It posits that cultural norms and expectations shape the individual\'s sense of who they can or should be, and identifies the importance of a \'sense of belonging\' to identity formation \[[@B28]\]. In this process of subjectification, identity is continuously under construction, maintenance and re-construction, and hence open to change, particularly change that is social and collective in nature. Thus participatory and collective learning methods become the vehicle for the creation of new norms and possibilities in the peer group \[[@B14],[@B15],[@B29]\]. Festinger\'s theory of cognitive dissonance posits that contradictory cognitions, or cognitive dissonance, cause mental discomfort \[[@B30]\]. This serves as a driving force that compels a person to adopt new thoughts or beliefs, or to modify existing beliefs, so as to reduce the amount of dissonance between cognitions. The education intervention used in this study does not tell students what they should think or how they should behave in terms of drug use. Rather, it provides a social learning process whereby the students reach their own conclusions through exploration of drug use issues. By guiding students through an interactive discovery process that involves articulation of responsible drug use behaviour their ownership of that behaviour is reinforced and they are less likely to behave in a contrary manner. Aims and hypothesis ------------------- The aim of this study is to develop and assess the effectiveness of a comprehensive, evidence-based, harm reduction focused school drug education program for year eight and nine students by undertaking a cluster randomised trial with 21 secondary schools in the Victorian state of Australia. The specific hypothesis is that while the intervention students may not be any less likely to take up use of alcohol, tobacco and illicit drugs they will be more likely to consume in a less risky manner and experience less harms associated with use. Methods/design ============== Study design ------------ Twenty one Victorian government high schools have been recruited to the study. The schools were then allocated to four strata: six schools to metro location, high SES; three schools to rural location, high SES; six schools to metro location, low SES; six schools to rural location, low SES. This approximated the proportion of Victorian secondary schools in each category. Within each strata the schools were then randomly allocated to intervention or control conditions on a two to one proportion to allow more precise statements about the effects of the intervention \[[@B31]\]. Random allocation and allocation concealment were achieved by one member of the research team writing the names of each school on identical paper chits which were then folded to conceal the name. The same researcher also created seven chits with the word control and 14 with the word intervention. These were similarly folded to conceal the words. The chits with the school name and those designating allocation were placed in separate containers. A different member of the team then simultaneously drew out a chit with the name of a school and a chit designating either control or intervention allocation. The study will follow a cohort of students from the start of year eight, when most are 13 years old, to the end of year ten, when most would have turned 16. Pre intervention, baseline testing will occur at the beginning of year eight (Mar/Apr). The first phase of the drug education intervention will occur in the middle of that year (Jul/Sep). Post 1 testing will occur at the end of the same year (Nov/Dec), approximately 8 months subsequent to baseline. The second phase of the drug education intervention for the students, now in year nine, will occur in the middle of the second year of the study. Post 2 testing will occur at the end of the second year. In the third year, when the students are in year ten, no further drug education lessons will be provided as part of this study, but schools will continue with their usual year ten drug education curriculum. Post 3 testing will occur at the end of the year. Throughout the study control students will receive the drug education normally provided by their school. The research sequence is illustrated in Figure [1](#F1){ref-type="fig"}. Fidelity data will be collected from all participating teachers on a lesson by lesson basis via an on-line survey. In addition, focus groups/interviews will be used to collect qualitative information about the relevance and suitability of lesson materials from students and teachers in a sample of five of the 14 intervention schools following delivery of both the year eight and the year nine program. The student focus groups will be one hour in duration with a roughly even mix of six to eight males and females. Teacher interviews will also be one hour long with either individuals, or small groups of two to three respondents. {#F1} Sample size calculation ----------------------- The ability to detect small changes in substance use, knowledge, attitudes and harms experienced is central to this study. The primary outcome is a change in alcohol consumption patterns and associated harm, and accordingly sample size estimations are based on detecting a small effect size of .15 in relation to these measures. This effect size was chosen on the basis of previous school drug education studies \[[@B9],[@B32]\]. The target sample size has been estimated using G\*Power v.3.1.3 software where α = 0.05 and 1-β error probability = 0.95 \[[@B33]\]. Assuming simple random sampling, a total sample size of 364 is required at the end of the study. However, there is a design effect due to the loss of effectiveness created by cluster sampling. The design effect for the SHAHRP study, which took into consideration the effect of clustering by school and a 15% annual rate of student attrition, was calculated to be 1.48. Using this correction, a total sample size of at least 539 is required to test the effectiveness of the intervention. Sample ------ The participating students attended the 21 Victorian government secondary schools that had agreed to participate in this project. Four schools are located in metropolitan Melbourne; ten are located in metropolitan fringe and major regional areas; and seven are in regional and rural areas. The student populations of six schools are in the medium high/high range in term of socioeconomic status, as measured by the Department of Education and Early Childhood Development\'s (DEECD) Student Family Occupation (SFO) index. The student populations of ten schools are in the medium range. The student populations of five schools are in the low range. Written active consent was sought from the 2700 year eight students in the 21 participating schools and their parents. Of this total population 1746 or 64.7% agreed to participate in the research. Intervention students numbered 1230. Control students numbered 516. The drug education intervention ------------------------------- The student intervention material and an accompanying teacher implementation manual and teacher training program have been developed from material trialled in the pilot program \[[@B23]\], which in turn drew on a range of earlier Australian research and development projects in drug education and resilience education. These included the School Health and Alcohol Harm Reduction Project (SHAHRP) and GET WISE: Working on Illicits in School Drug Education \[[@B9],[@B34]\]. The education program comprises 10 lesson plans in year eight and 8 in year nine that address issues around the use of alcohol, tobacco, cannabis and other illicit drugs (see Table [1](#T1){ref-type="table"}). Experience from the SHAHRP study indicated that this number of lessons can provide adequate coverage and be accommodated within a secondary school curriculum. A number of tasks are designed to be undertaken at home in collaboration with a parent so as to draw out their influence on decision-making about drug use \[[@B35]\]. As alcohol is the most commonly used drug by Australian young people, and the drug that causes them the greatest harm, it will receive the greatest coverage, followed by tobacco and cannabis. ###### Year eight and year nine lesson plans Lesson Year 8 Year 9 -------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 1 WHAT IS A DRUG? - Introduction, agreements, definitions and drug categories PRIORITIES AND CONCERNS - Identifying what young people value and worry about and what worries they have around drugs 2 ALCOHOL AND EFFECTS AND STANDARD DRINKS - How alcohol effects the body, assessing harms associated with use, pouring standard drinks, understanding blood alcohol content and safer levels of use FACING FACTS AND FINDING SOLUTIONS - Alcohol and Cannabis- guidelines on use and the research that informs them 3 PARTY BEHAVIOURS AND ALCOHOL - The relationship between levels of alcohol use and the risk of harm to self and others USING YOUR RESOURCES - Pouring standard drinks, matching harms to levels of alcohol use, identifying strategies to reduce harm 4 PREVALENCE AND NORMS - Dispelling myths about levels of drug use amongst young people, identifying reasons for use/non -use WINDING UP, WINDING DOWN - Learning about the effects/risks of Amphetamine type stimulants, identifying drug-free ways of achieving \'high\' and \'serene\' states of mind 5 TOBACCO - Considering gender differences in relation to smoking; the impact of media messages DRUGS, DISINHIBITION, SEXUAL VULNERABILITY AND VIOLENCE - Discussing sexual vulnerability in relation to drug use, identifying strategies for avoiding or reducing harm 6 CANNABIS - Information about cannabis and its effects, identifying risks associated with Cannabis use INVISIBLE RISKS - Information about injecting drug use, blood-borne viruses and methods of protection 7 RISK REDUCTION - Assessing risk and developing strategies to avoid or minimise harm PERSONAL CONFIDENCE - AND DRUG USE - Developing and rehearsing positive self talk, refusal skills and tactics for peer negotiation 8 INFLUENCES - Identifying social and media influences to use alcohol GETTING HELP AND TALKING WITH ADULTS - Information about heroin, rehearsing steps for practical first aid in situations involving overdose, rehearsing help seeking with adults 9 OPTIONS AND DECISIONS - Generating and rehearsing strategies to reduce harms associated with drug use 10 STANDING UP FOR YOURSELF - Providing peer support, using assertion skills in situations involving alcohol The education program incorporates the elements of effective practice identified by reviews of school drug education \[[@B19]-[@B22],[@B25],[@B36]\]. This literature identifies the importance of basing programs on evidence of proven effect; the needs of students; provision of essential knowledge; adequate coverage of salient issues; and use of interactive learning strategies that enhance negotiation skills, involve participants in problem-solving and engage them in deconstructing the social pressures and perceived norms around drug use. The curriculum is also informed by research in the field of resilience education that identifies social competence, problem-solving, autonomy and a sense of purpose as key attributes of resilient young people \[[@B37]\], and highlights the importance of participatory and developmentally appropriate learning strategies in enhancing social and emotional learning \[[@B38]\]. Each year all teachers delivering the classroom program will participate in an intensive two-day professional learning program that provides a grounding in the evidence-base informing the study and active sampling of each of the lesson activities they will be teaching their students. Emphasis will be given to modelling and explicit leadership coaching in use of the participatory methods. Student survey instrument and measurement of change --------------------------------------------------- The survey instrument to measure change is a development of the self-completion questionnaire used in SHAHRP and was trialled in the pilot research that preceded this study \[[@B9],[@B23]\]. Self-report is well accepted practice in studies of this type and research indicates little inconsistency between self report and other measures of drug use \[[@B39],[@B40]\] The instrument will collect information on knowledge, patterns and context of use, attitudes and harms experienced in relation to alcohol, tobacco, cannabis and other illicit drug use. As was done in the pilot study, scales will be constructed to measure overall change in AOD knowledge (38 items), attitudes (4/5 items per drug type), and harms (5/10 items per drug type) A student generated code, based on easily remembered fragments of personal information, is used to maintain confidentiality, while allowing individual matching over the course of the study. As part of the pilot research, feedback was obtained from experts in school drug education and students in the target group. This indicated the instrument had both content and face validity. The Cronbach\'s alpha test was used to measure the internal consistency of the AOD knowledge, attitude and harm scales generated from the pilot survey responses. The results for the pilot AOD knowledge and alcohol, tobacco, cannabis and other drugs attitude and harm scales are presented in Table [2](#T2){ref-type="table"}. The internal consistency co-efficient of all indices was significant. In the case of the knowledge scale and all the harm scales the respective co-efficients were high (α = .609-.949), indicating each scale was a good measure of the intended single latent construct \[[@B41]\]. The attitude scale co-efficients were lower, although still significant. Factor analysis of the attitudes towards alcohol, tobacco and other drugs indicated there were two main factors in the data accounting for between 53.5% and 57.9% of the total variation in each case. One factor pertained to attitudes about knowledge and communication; the other pertained to attitudes about harm. ###### Internal consistency of knowledge, attitude and harm scales Scale alpha p Component factors \% of variance ------------------------------- ------- ---------- ----------------------------- ---------------- Knowledge scale .859 \< 0.001 Attitudes towards alcohol .387 \< 0.001 Knowledge and communication 30.6 Harm 23.5 Alcohol harm scale .949 \< 0.001 Attitudes towards tobacco .445 \< 0.001 Harm 32.4 Knowledge and communication 21.1 Tobacco harm scale .802 \< 0.001 Attitudes towards cannabis .548 \< 0.001 Cannabis harm scale .891 \< 0.001 Attitudes towards other drugs .284 \< 0.001 Knowledge and communication 32.7 Harm 25.2 Other drugs harm scale .609 \< 0.001 Blinding -------- Participants cannot be blinded to intervention in this sort of psycho-social educational intervention, nor can program deliverers or outcome assessors. Research ethics --------------- The study was approved by Edith Cowan University\'s and the University of Melbourne\'s human research ethics committees. It was also approved by the Research Branch, Education Policy and Research Division of the Victorian Department of Education and Early Childhood Development. Statistical analysis -------------------- Multi-level modelling can best accommodate hierarchically structured data and will be used in this analysis, adjusting for any baseline differences between the intervention and control groups \[[@B42],[@B43]\]. The hierarchy in this longitudinal multi-level dataset comprises level 1 units (occasions of repeated measurement), nested within the level 2 units (the individual student), nested within the level 3 units (the school). Stata version 10.0 will be used for model fitting, with a three-level mixed regression model fitted to the data to account for the repeated observations \[[@B44]\]. If necessary, the knowledge, consumption and harm scales will be log-transformed to satisfy the assumption of normality. If normality cannot be achieved by log-transformation, non-parametric procedures will be used for analysis. All analyses will be conducted on an Intent-to-Treat basis. Complete-case analysis (CCA) will be complemented with multiple imputation analysis (MIA) to account for missing data. Discussion ========== The benefits of undertaking this harm reduction focused drug education demonstration program derive both from the knowledge gained by trialling an optimum combination of evidence derived approaches with a large, representative student sample, and the resultant product. The program targets alcohol, tobacco and illicit drug use; draws on research that consistently identifies a core set of effective practice elements and is rigorously evaluated over an extended period. The program also does not simply rely on the traditional measures of effectiveness, namely abstinence or reduced use. Harms associated with use are also measured as valid indicators of program effect. This will provide a broader understanding of what benefits can be achieved by school drug education. By the end of the study a well documented, comprehensive drug education package will have been produced. It will be of a size able to be accommodated within a secondary school\'s health curriculum and the evaluation findings will provide evidence as to its potency. This makes available to schools a readily useable drug education program with well understood prevention characteristics. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= RM obtained funding for the study, co-ordinated the design of the study, contributed to the design of the intervention materials and was the lead author of this study protocol. HC co-ordinated the design of the intervention materials and contributed to the study design and the drafting of this study protocol. DF contributed to the study design and the drafting of this study protocol. LL contributed to the study design and the drafting of this study protocol. LV recruited the study schools and contributed to the study design, the design of the intervention materials and the drafting of this study protocol. RR contributed to the study design, the design of the intervention materials and the drafting of this study protocol. MP contributed to the design of the intervention materials and the drafting of this study protocol. All authors read and approved the final manuscript Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/12/112/prepub> Acknowledgements ================ This study is jointly funded by the Australian Research Council and the Victorian Department of Education and Early Childhood Development through Australian Research Council linkage grant LP100100798

Learning objectives =================== Describe the use of current imaging techniques to reach a correct diagnosis of hepatocellular carcinoma. Emphasise the knowledge of complementary tools for the evaluation of early response to treatment tumour. Content organisation ==================== Our cases will be presented in a pictorial essay mode. It is important to consider: The radiologic diagnosis of HCC can be made by either CT or MRI methods. Typically, HCC enhances during the arterial phase due to the presence of an intense arterial blood supply of the hepatic arteries. TC dual Energy: Dual-energy CT (DECT) is an innovative imaging technique that operates on the basic principle of application of two distinct energy settings that make the transition from CT attenuation--based imaging to material-specific or spectral imaging. DECT can also aid in evaluation of response to therapy and detection of oncology-related disorders. MDCT perfusion: non-invasive method of quantification of tumour blood supply, related to the formation of new arterial structures (neoangiogenesis), which are essential for tumour growth. Conclusion ========== Dual energy and perfusion CT are innovative techniques to detect suspicious lesions of hepatocarcinoma. Dual Energy CT also helps to reduce radiation exposure by using low dose imaging protocols without affecting diagnostic purpose. The Perfusion provides information on the neovascularization of an injury (predicting malignancy) and assesses early therapeutic response (before displaying morphological changes).

Introduction {#Sec1} ============ Lung cancer is one of the deadliest types of cancer in the world^[@CR1]^ and approximately 80--85% of lung cancers are non-small-cell lung cancers (NSCLCs). Despite many advances in the treatment of NSCLC in recent years, the median survival is still \<12 months and \<15% of NSCLC patients survive more than 5 years from the initial diagnosis.^[@CR2]^ Even if NSCLC patients are treated with chemotherapy or radiotherapy, the incidence of recurrence is still high.^[@CR3],[@CR4]^ Compared with traditional chemotherapy, precise tumor treatment is helpful to improve the treatment effect and quality of life. Targeted therapy has become an important means of disease treatment for NSCLC patients.^[@CR5]^ The growth, invasion, and metastasis of tumors cannot be separated from the new blood vessel supply of additional oxygen and nutrients.^[@CR6]^ Tumor angiogenesis is the key condition for tumor growth and metastasis, so antiangiogenic treatment is currently one of the main treatment strategies for lung cancer. Angiogenesis inhibitors, such as epidermal growth factor receptor (EGFR) inhibitors, have been used to limit tumor growth;^[@CR7]^ however, traditional angiogenesis inhibitors are not ideal in solid tumor treatment, which suggests that there may be another relatively independent microvascular system involved in tumor blood supply. In recent years, tyrosine kinase inhibitors (EGFR-TKIs) have been used in NSCLC patients.^[@CR8]^ Gefitinib, an EGFR-TKI, is used to treat advanced NSCLC patients. Despite the significant response to gefitinib, these tumors always develop resistance within an average of 9--12 months.^[@CR9]^ In 1999, Maniotis et al.^[@CR10]^ found that cancer cells cover nonendothelial vascular channels containing red blood cells, which is called vasculogenic mimicry (VM). The formation of VM not only accelerates tumor metastasis but also increases the risk of resistance to antiangiogenic therapy in NSCLC.^[@CR11]^ In addition, it is an interesting possibility that antiangiogenic therapy may lead to the formation of VM, thus allowing drug-induced resistance to occur.^[@CR12]^ However, current antiangiogenic drugs are mainly based on the inhibition of endothelial cell-dependent angiogenesis, ignoring the existence of VM. Therefore, targeting the formation of VM may bring breakthroughs in cancer treatment. The transcriptional signature of VM^[@CR13]^ shares components with signatures of "stemness" and epithelial-to-mesenchymal transition (EMT), key attributes related to tumor plasticity during metastasis and drug-induced resistance.^[@CR14]--[@CR16]^ A series of studies have confirmed that EMT is involved in the formation of VM.^[@CR16]^ Epithelial endothelial transition, a subtype of EMT, produces endothelial-like phenotype of tumor cells, whereas endothelial-like phenotype can form VM to allow serum into tumor tissues.^[@CR17]^ Serum contains prothrombin, which is converted into thrombin via proteolytic cleavage in the tumor microenvironment.^[@CR18]^ However, the level of thrombin in NSCLC tumor tissues and the relationship between thrombin and VM are still unclear. It has long been presumed that tumors may take advantage of the hemostatic system. A relationship between increased clotting and malignancy was recognized more than a century ago. Thrombin exerts a variety of biological effects by interacting with various receptors on the surface of vascular and nonvascular cells.^[@CR19]^ Studies have shown that thrombin plays important roles in tumorigenesis, inflammation, and metastatic dissemination of tumor cells through its PAR-1 receptor.^[@CR20]--[@CR22]^ Thrombin binds to the hydrophobic hirudin-like sequence DK^51^YEPF^55^ on the PAR-1 extracellular face via exosite I, facilitating the interaction of the PAR-1 cleavage site with the active site of thrombin. Thrombin cleaves the N-terminal extracellular domain of PAR-1 and the newly exposed N terminus interacts with the second extracellular domain of the cleaved receptor to initiate signaling. The cleavage of PAR-1 by thrombin initiates potent inflammatory responses, including the upregulation of cell surface adhesion molecules and the induction of hyperpermeability.^[@CR23]^ It was reported that thrombin, acting through PAR-1, promotes EMT in embryo development.^[@CR24],[@CR25]^ However, the relationship between thrombin, PAR-1, and VM in human NSCLC has not been reported thus far; we suppose that thrombin may induce VM through PAR-1 activation. Direct thrombin inhibitor peptide (DTIP) and recombinant hirudin (r-hirudin), which are derivatives of wild-type hirudin variant 2, were developed by our group.^[@CR26],[@CR27]^ DTIP and r-hirudin bind to exosite I and to the apolar region of thrombin, whereas the N-terminal moiety of r-hirudin and DTIP blocks access to the thrombin active site, to inhibit the activity of thrombin. r-hirudin has entered phase I clinical trials and DTIP is a novel antithrombotic agent that can be used to prevent thrombosis without conferring an increased bleeding risk for subcutaneous (s.c.) injection. In our recent studies, we found that r-hirudin and DTIP could suppress progression, dissemination, and spontaneous metastasis in NSCLC. Therefore, in this study, we first investigated the levels of thrombin in clinical NSCLC samples and explored the relationship between thrombin level and VM formation. We evaluated the effects of r-hirudin and DTIP on VM formation in vitro and in vivo. We first examined the potential roles of thrombin and PAR-1 in regulating the formation of VM in NSCLC. We evaluated the effects of combination therapy with DTIP and gefitinib in NSCLC. Thrombin may be a potential target for future anticancer therapies and combination therapy with direct thrombin inhibitors and EGFR inhibitors might achieve a superior therapeutic effect. Results {#Sec2} ======= The formation of VM in NSCLC is closely related to the level of thrombin and the prognosis of patients {#Sec3} ------------------------------------------------------------------------------------------------------ VM is detected in clinically by CD31/periodic acid schiff (PAS) double-staining of PAS positive, CD31 negative vessels.^[@CR11],[@CR14],[@CR28],[@CR29]^ We analyzed the expression levels of thrombin and the formation of VM in 152 NSCLC cases against detailed clinical and pathologic information. We found that the thrombin score was higher in NSCLC tissues (3.12 ± 2.82) than in adjacent nontumor lung tissues (0.77 ± 1.13, Fig. [1a](#Fig1){ref-type="fig"}) and upregulated thrombin expression was observed in 126 of 152 human NSCLC tumor samples (82.9%). Furthermore, thrombin expression in tumor tissue was correlated with clinicopathological features, such as tumor-node-metastasis (TNM) stage (Fig. [1b](#Fig1){ref-type="fig"}) and tumor size (Fig. [1c](#Fig1){ref-type="fig"}). Thrombin expression in patients in phase III (3.69 ± 3.29) or phase IV (3.91 ± 3.75) was significantly increased compared with that in patients in phase I (1.96 ± 2.30) and thrombin expression was higher in patients with a maximum tumor diameter \> 5 cm (4.28 ± 3.29) than in patients with a maximum tumor diameter ≤ 2 cm (2.05 ± 1.90). Tissues were stained with PAS and anti-CD31 antibodies to investigate VM. Typical VM structures were found in 68 (68/152) lung cancer specimens. Univariate survival analysis demonstrated that patients with VM (35.4 ± 4.8 months) had a reduced 16-month survival compared with patients without tumor VM (51.2 ± 5.2 months, Fig. [1f](#Fig1){ref-type="fig"}). Patients with VM (20.0 ± 3.5 months) had a worse prognosis and were more likely to relapse after surgery than patients without VM (38.0 ± 5.3 months, Supplementary Fig. [S2a](#MOESM1){ref-type="media"}). Importantly, compared with patients without VM (1.93 ± 1.86), patients with VM (4.4 ± 3.09) had a high expression of thrombin in lung tumors (Fig. [1d, e](#Fig1){ref-type="fig"} and Supplementary Fig. [S1](#MOESM1){ref-type="media"}). Moreover, compared with that in a normal lung cell line (BEAS-2B), thrombin expression in four NSCLC cell lines was significantly increased (Fig. [1g](#Fig1){ref-type="fig"} and Supplementary Fig. [S2b](#MOESM1){ref-type="media"}). The expression of thrombin in tumor tissues of mice was also increased compared with that in paired normal lung tissues (Fig. [1h](#Fig1){ref-type="fig"} and Supplementary Fig. [S2c](#MOESM1){ref-type="media"}).Fig. 1Thrombin is overexpressed in NSCLC and the formation of VM in NSCLC is closely related to the level of thrombin and the prognosis of patients.**a** Score of thrombin expression in adjacent nontumor lung tissue and in NSCLC tissue. **b** Thrombin expression in different stages (AJCC Tumor-node-metastasis (TNM) stages) of NSCLC. **c** The association of thrombin expression with tumor size in NSCLC samples, where D is the maximum tumor diameter. **d** Region showing typical malignant morphology of NSCLC and thrombin expression in corresponding tissues. PAS+/CD31− VM tubes (black arrow) are also shown. **e** The relationship between the expression level of thrombin and the formation of VM in tumor tissues. **f** Univariate survival analysis according to VM in NSCLC patients and Kaplan--Meier survival analysis for patients dichotomized by the presence VM. **g** The expression of thrombin in BEAS-2B, PC9, 95D, A549, and Lewis cells was determined by western blotting. **h** The expression of thrombin in tumor tissues of mice and normal lung tissues of mice was determined by western blotting. All the results are expressed as the mean ± SD. The error bars indicate the SD. ANOVA followed by Dunnett's test was applied for multiple comparisons. \**p* \< 0.05 r-hirudin and DTIP inhibit thrombin-promoted VM formation in vitro by inhibiting EMT {#Sec4} ------------------------------------------------------------------------------------ To further explore the role of thrombin in VM formation, NSCLC cell lines were treated with thrombin in vitro. DTIP and r-hirudin, which are direct thrombin inhibitors, were developed by our group. Tubular network formation by tumor cells in Matrigel was used to detect the ability for VM formation by tumor cells in vitro.^[@CR10],[@CR11],[@CR13],[@CR29]^ When tumor cells are cultured, the cells elongate and present cellular protrusions. Through the interaction of filaments, cells are arranged in a ring.^[@CR30]^ With thrombin treatment, A549 and Lewis cells elongated and proliferated in a decentralized manner. Cellular protrusions were more obvious and the cells formed more networks compared with the NS cells. However, with exposure to r-hirudin or DTIP, intercellular connections were damaged and tube structures were reduced or even disappeared (Fig. [2a, b](#Fig2){ref-type="fig"}). EMT is the key event underlying VM formation.^[@CR31]^ E-cadherin, vimentin, N-cadherin, and snail were used as markers to reflect EMT during VM formation.^[@CR32]^ We detected E-cadherin, N-cadherin, and snail expression in NSCLC cells pretreated with thrombin, r-hirudin, or DTIP. The results indicated that the expression of snail and N-cadherin was reduced, whereas the expression of E-cadherin was increased in both r-hirudin- and DTIP-treated A549 and Lewis cells compared with thrombin-treated cells (Fig. [2c](#Fig2){ref-type="fig"}). These results suggested that r-hirudin and DTIP could inhibit the VM formation induced by thrombin in NSCLC cells by regulating EMT.Fig. 2r-hirudin and DTIP inhibit VM formation of A549 and Lewis cells in vitro.**a**, **b** A549 and Lewis cells were pretreated with PBS, 10 nmol/L thrombin, 25 nmol/L r-hirudin, 50 nmol/L DTIP, 10 nmol/L thrombin+25 nmol/L r-hirudin, or 10 nmol/L thrombin+50 nmol/L DTIP, and VM-like network formation on Matrigel in different groups was examined. Left, representative photographs of experiments. Right, quantification of the inhibitory activity of r-hirudin and DTIP on VM tube formation. **c** The effect of thrombin, r-hirudin and DTIP on the expression of snail and other EMT markers was determined by western blotting analysis. GAPDH was used as the loading control. The summarized western blot data for EMT markers are given (bottom). **d** Different groups were probed for phosphorylated p65, phosphorylated IκBα and total p65 and IκBα. GAPDH was used as the loading control. Right, summarized western blot data. **e** The expression of snail and other EMT markers in different groups was determined by western blotting analysis. Right, the summarized western blot data for EMT markers. All the results are expressed as the mean ± SD. The error bars indicate the SD. ANOVA followed by Dunnett's test was applied for multiple comparisons. \**p* \< 0.05 Thrombin has been reported to activate nuclear factor-κB (NF-κB) signaling in human pleural mesothelial^[@CR33]^ and SH-SY5Y cells^[@CR34]^ via PAR-1. The effect of r-hirudin and DTIP on the NF-κB pathway in thrombin-stimulated NSCLC cells was analyzed. The results indicated that thrombin could activate NF-κB signaling in NSCLC cells. Compared with the thrombin-treated group, r-hirudin and DTIP exhibited diminished IκBα and p65 phosphorylation (Fig. [2d](#Fig2){ref-type="fig"}), suggesting that r-hirudin and DTIP can inhibit thrombin-induced NF-κB activation. Studies have shown that NF-κB plays an important role in the EMT process during liver fibrogenesis^[@CR35]^ and EMT of NSCLC cells occurs via NF-κB activation.^[@CR36]^ Tian et al.^[@CR37]^ reported that NF-κB is a master regulator of EMT autocrine loops. We also found that pyrrolidinedithiocarbamate ammonium (PDTC), an inhibitor of NF-κB, could inhibit thrombin-induced increased expression of snail and N-cadherin and inhibit thrombin-induced decreased expression of E-cadherin. lipopolysaccharide (LPS), an activator of NF-κB, could prevent the inhibition induced by r-hirudin and DTIP (Fig. [2e](#Fig2){ref-type="fig"}), suggesting that thrombin can regulate EMT via the NF-κB pathway. r-hirudin and DTIP inhibit VM formation in a mouse lung cancer model {#Sec5} -------------------------------------------------------------------- The aforementioned results suggested that thrombin could regulate EMT and promote VM formation in NSCLC cells, and that the novel thrombin inhibitors r-hirudin and DTIP could inhibit thrombin-promoted VM formation. We further confirmed these effects in vivo. Spontaneous metastasis generated through s.c. inoculation of tumors in mice, which involves a comprehensive process, was examined. Daily treatment with 1.0 mg/kg DTIP or 0.5 mg/kg r-hirudin for 21 consecutive days, one week after the injection of LLC cells, was applied. Mice were humanely killed and the tumors were resected 45 days after injection. r-hirudin and DTIP inhibited tumor growth (Fig. [3a, b](#Fig3){ref-type="fig"}). Consequently, we counted the number of mice with panniculus invasion and metastases after mice were sacrificed. All mice from the normal saline-treated group showed signs of panniculus invasion, whereas only five of nine mice from the r-hirudin-treated group and five of ten mice from the DTIP-treated group had any noticeable signs of panniculus invasion. Furthermore, the number of mice with lung, liver, and colon metastases was largely reduced in the r-hirudin- or DTIP-treated groups compared with that in the normal saline-treated group (Fig. [3c](#Fig3){ref-type="fig"}).Fig. 3r-hirudin and DTIP inhibited tumor development and VM formation.LLC cells at a density of 1 × 10^6^ in 0.1 mL serum-free media were injected subcutaneously into the right flank of mice. One week after the injection of LLC cells, mice were administered normal saline, 1.0 mg/kg DTIP, or 0.5 mg/kg r-hirudin for 21 consecutive days. **a** Volume of resected tumors derived from the normal saline-, r-hirudin-, and DTIP-treated groups at 45 days after cell injection. **b** Mice were humanely killed and the tumors were resected at 45 days after cell injection. The weights of tumors resected from the normal saline-, r-hirudin-, and DTIP-treated groups. **c** Number of NS-, r-hirudin-, and DTIP-treated mice with panniculus invasion, lung metastasis, liver metastasis, and colon metastasis. **d** Number of NS-, r-hirudin-, and DTIP-treated mice with VM. **e** Representative images of anti-CD31/PAS staining in tumor tissues. Right, the number of VM tubes per tumor tissue sample in different groups. **f** DAPI and FITC-dextran (dextran inverted)-stained tumor sections from resected tumors. **g** Leakiness index of resected tumors. **h** The expression of snail and other EMT markers in tumors was determined by western blotting analysis. **i** Summarized western blot data. **j** Immunohistochemical analysis was performed on tumor samples to determine the expression of E-cadherin, N-cadherin, vimentin, and Snail in tumors from NS-, r-hirudin-, and DTIP-treated mice. Right, scoring of immunostained specimens. All the results are expressed as the mean ± SD. The error bars indicate the SD. Compared with the NS group by one-way ANOVA. \**p* \< 0.05, \*\**p* \< 0.01, \*\*\**p* \< 0.001 CD31/PAS double staining was performed after tumor tissue was stripped from the mouse model. The results showed that six of ten mice from the normal saline-treated group formed obvious VM structures, whereas in the r-hirudin- and DTIP-treated groups, only two (two of nine) and three (two of ten) mice, respectively, formed VM channels (Fig. [3d](#Fig3){ref-type="fig"}); r-hirudin and DTIP reduced the number of VM channels (Fig. [3e](#Fig3){ref-type="fig"}). Thrombin and VM have been associated with vascular leakiness.^[@CR10],[@CR29]^ The results revealed that the vessels within normal saline-treated tumors were leakier than those in r-hirudin- or DTIP-treated tumors (Fig. [3f, g](#Fig3){ref-type="fig"}). EMT has also been confirmed to be associated with cancer cell invasion and is a key step in VM formation.^[@CR38],[@CR39]^ We further examined the expression of E-cadherin, N-cadherin, vimentin, and snail by western blotting and immunohistochemistry (IHC). r-hirudin and DTIP reduced the expression of N-cadherin, vimentin, and snail, and increased the expression of E-cadherin in primary tumors (Fig. [3h--j](#Fig3){ref-type="fig"}). Taken together, the results of vascular leakage analysis, the characteristics of VM, and the results of PAS/CD31 staining suggest that r-hirudin and DTIP could reduce vascular leakage and inhibit EMT, which strengthens the hypothesis that VM is one of the targets of r-hirudin and DTIP in inhibiting tumor development. Furthermore, we did not find increased bleeding after administration of DTIP. Slight s.c. hemorrhage was observed after r-hirudin administration for 3 weeks continuously (Supplementary Fig. [S3a](#MOESM1){ref-type="media"}) and no obvious bleeding was found with either r-hirudin or DTIP after 1 week of discontinuation (Supplementary Fig. [S3b](#MOESM1){ref-type="media"}). These results show that DTIP, a direct thrombin inhibitor, could be applied in anticancer therapy. Thrombin expression level with different PAR-1 backgrounds is associated with prognosis of NSCLC patients and formation of VM {#Sec6} ----------------------------------------------------------------------------------------------------------------------------- Thrombin is the main activator of PAR-1.^[@CR40]^ Our previous experimental results also showed that PAR-1 was highly expressed in NSCLC patients and mice (Supplementary Fig. [S4a, f, g](#MOESM1){ref-type="media"}), and NSCLC cells (Supplementary Fig. [S4h, i](#MOESM1){ref-type="media"}). However, we did not find an obvious relationship between the PAR-1 expression levels of tumors and clinical variables, such as the stage of NSCLC differentiation, disease progression, and overall survival (Supplementary Fig. [S4a--c](#MOESM1){ref-type="media"}). Compared with patients without VM, patients with VM also had a high expression of PAR-1 in lung tumors (Supplementary Fig. [S4d](#MOESM1){ref-type="media"}), whereas we did not find a consistent relationship between the expression of thrombin and PAR-1 (Supplementary Fig. [S4e](#MOESM1){ref-type="media"}). Interestingly, we found that in patients with high PAR-1 expression, positive thrombin (thrombin+) expression was associated with a much shorter survival time than negative thrombin (thrombin−) expression (Fig. [4a, b](#Fig4){ref-type="fig"}) and patients with thrombin+ expression were more likely to form VM than those with thrombin− expression (Fig. [4a, d](#Fig4){ref-type="fig"}). However, no significant difference in the survival and the formation of VM was observed between patients with thrombin+ and thrombin− expression with low expression of PAR-1 (Fig. [4c, e](#Fig4){ref-type="fig"}).Fig. 4Thrombin with PAR-1 expression is associated with the prognosis of NSCLC patients and the formation of VM.**a** Different expression statuses of thrombin and PAR-1 detected by immunohistochemical staining in consecutive sections of NSCLC tissues from the same patient. The NSCLC tissues of case 1 were positive for thrombin and had high expression of PAR-1; case 2 had tissues positive for thrombin but with low expression of PAR-1; case 3 had tissues negative for thrombin but with high expression of PAR-1; and case 4 had tissues negative for thrombin with low expression of PAR-1. Based on the score of thrombin or PAR-1 in the central positive staining area in the tumor section as a cutoff value, the scores for thrombin and PAR-1 were classified as positive (thrombin+ ≥ 2 and PAR-1 high expression≥6) or negative (thrombin− \< 2 and PAR-1 low expression \< 6). **b** In Kaplan--Meier analyses, an association of thrombin expression with overall survival (OS), were demonstrated in the PAR-1 high-expression subgroup of NSCLC patients. **c** An association of thrombin expression with overall survival was demonstrated in the PAR-1 low-expression subgroup. **d** An association of thrombin expression with VM formation was demonstrated in the PAR-1 high-expression subgroup. **e** An association of thrombin expression with VM formation was demonstrated in the PAR-1 low-expression subgroup. All the results are expressed as the mean ± SD. The error bars indicate the SD PAR-1 is a major determinant in thrombin-promoted formation of VM in NSCLC {#Sec7} -------------------------------------------------------------------------- To observe the role of PAR-1 in thrombin-mediated VM formation more clearly, A549^*PAR-1−/*−^ and LLC^*Par-1−/*−^ cells were constructed. Our previous studies showed that A549, 95D, PC9, and LLC cells express high levels of PAR-1 compared with normal lung epithelial cells. After PAR-1 knockout, the proliferation of lung cancer cells decreased in vitro. PAR-1 depletion almost completely abrogated thrombin-promoted cell invasion (Supplementary Fig. [S5a, b](#MOESM1){ref-type="media"}). The VM formation ability of A549^NC^ and A549^*PAR-1−/−*^ cells was also detected in vitro. A549^*PAR-1−/−*^ cells could not form tube-like structures or formed only a few VM-like tubes, and thrombin had no effect on VM-like channels in A549^*PAR-1−/*−^ cells (Fig. [5a](#Fig5){ref-type="fig"}). Similarly, PAR-1 knockout also reduced the formation of tube-like structures in LLC^*Par-1−/−*^ cells (Fig. [5b](#Fig5){ref-type="fig"}). PAR-1 deficiency diminished IκBα and p65 phosphorylation. The ability of thrombin to activate NF-κB was also inhibited (Fig. [5c](#Fig5){ref-type="fig"} and Supplementary Fig. [S5c, d](#MOESM1){ref-type="media"}). After knocking out PAR-1 in A549 and LLC cells, the expression of N-cadherin and snail decreased, whereas that of E-cadherin increased, and thrombin could not rescue the expression level of EMT markers (Fig. [5d](#Fig5){ref-type="fig"} and Supplementary Figs. [S5e, f](#MOESM1){ref-type="media"} and [S6](#MOESM1){ref-type="media"}). Incubation with LPS rescued the expression level of EMT markers (Fig. [5d](#Fig5){ref-type="fig"} and Supplementary Fig. [S6e, f](#MOESM1){ref-type="media"}). Importantly, thrombin-regulated EMT expression was inhibited by the specific PAR-1 inhibitor ML161 (Supplementary Fig. [S6](#MOESM1){ref-type="media"}). These results suggested that thrombin could promote EMT and VM formation by PAR-1-mediated NF-κB signaling cascades.Fig. 5PAR-1 is a major determinant in thrombin-promoted metastasis of lung cancer and formation of VM.**a**, **b** PAR-1 gRNA constructs targeting the PAR-1 gene via lentiviral transduction were used to construct stable A549^*PAR-1−/*−^ and LLC^*Par-1−/−*^ cells. An empty construct was also used as a negative control. VM channel formation in PAR-1-deficient A549 and LLC cells. Left, representative photographs of experiments. Right, quantification of the number of VM tubes. **c** Cell lysates were probed for phosphorylated p65, phosphorylated IκBα, and total p65 and IκBα. **d** The expression of snail and other EMT markers in PAR-1-deficient A549 and LLC cells was determined by western blotting analysis. **e**--**j** LLC cells infected with gRNA-PAR-1 lentivirus (*Par-1*^*−/−*^ group) or LV-negative control (NC, vehicle group) were injected subcutaneously into the right flank of mice. **e** Volumes of resected tumors derived from injection of LLC^*Par-1−/−*^ and LLC^NC^ cells into mice at 5 weeks after cell injection. **f** Photograph of tumors resected from each group at 5 weeks after cell injection. **g** Number of mice from different groups with panniculus invasion, lung metastasis, liver metastasis, and colon metastasis. **h** Representative images of anti-CD31/PAS staining in tumor tissues. Right, the number of VM tubes per tumor tissue in the vehicle and *Par-1*^*−/−*^ groups. **i** E-cadherin, N-cadherin, and Snail in resected tumors. The summarized western blot data for EMT markers are given (bottom). **j** DAPI and FITC-dextran (dextran inverted)-stained tumor sections from resected tumors. Right, leakiness index of resected tumors. All the results are expressed as the mean ± SD. The error bars indicate the SD. ANOVA followed by Dunnett's test was applied for multiple comparisons. \**p* \< 0.05, \*\**p* \< 0.01, \*\*\**p* \< 0.001, NS, not significant To further analyze the effects of PAR-1 on lung cancer growth, metastasis, and VM formation, we established a lung cancer model in mice using LLC cells infected with gRNA-PAR-1 lentivirus (*Par-1*^*−/−*^ group) or LV-negative control (NC, vehicle group). Mice with s.c. inoculated tumors were injected s.c. with LLC^*Par-1−/−*^ and LLC^NC^ cells and monitored for tumor growth every 3 days. Five weeks after injection of tumor cells, mice injected with control cells had large tumors with an average weight of 1.4 ± 0.4 g and a volume of 884.9 ± 181.4 mm^3^. Interestingly, tumors from the PAR-1-deficient group were smaller than those from the NC group, with an average weight of 0.41 ± 0.10 g and a volume of 178.4 ± 86.7 mm^3^ (Fig. [5e, f](#Fig5){ref-type="fig"}). Panniculus invasion, lung metastases, liver metastases, and colon metastases were all largely reduced in the PAR-1-deficient group (Fig. [5g](#Fig5){ref-type="fig"}). In addition, VM channel formation was decreased in the PAR-1-deficient tumors (Fig. [5h](#Fig5){ref-type="fig"}) and the tumors from the control group were leakier than the PAR-1-deficient tumors (Fig. [5j](#Fig5){ref-type="fig"}). The expression of snail and N-cadherin was reduced in tumors with PAR-1 deficiency and E-cadherin expression was increased (Fig. [5i](#Fig5){ref-type="fig"}). Together with the in-vitro experiments, these results allowed us to conclude that PAR-1 plays an important role in thrombin-induced VM formation in NSCLC. Combination therapy with DTIP and an EGFR inhibitor resulted in improved antitumor efficacy {#Sec8} ------------------------------------------------------------------------------------------- EGFR-TKIs, including gefitinib, erlotinib, and so on, as angiogenesis inhibitors have significantly improved clinical outcomes. However, most NSCLC patients do not respond (primary resistance) and those who initially respond eventually progress (acquired resistance). There is an urgent need to develop a combination therapy of TKIs with other reagents to enhance antitumor activity and overcome drug resistance. Van der Schaft et al.^[@CR41]^ reported that angiogenesis inhibitors had no effect on VM and Naumov et al.^[@CR42]^ reported that EGFR inhibitors could induce a rise in both host and tumor-derived vascular endothelial growth factor (VEGF), which is associated with resistance. One week after LLC inoculation, mice were randomly divided into four groups: normal saline, DTIP (1 mg/kg per day, s.c.), gefitinib (100 mg/kg per day, intragastric), and combination DTIP with gefitinib treatment groups. Next, the mice were administered DTIP, gefitinib, or normal saline for 21 consecutive days. The tumor volume and tumor weight in the combination treatment group were significantly lower than those in the groups administered DTIP alone or gefitinib alone (Fig. [6a--c](#Fig6){ref-type="fig"}). The number of mice with lung metastases was largely reduced in the combination treatment group compared with that in the single-treatment groups (Fig. [6d, e](#Fig6){ref-type="fig"}). Consistent with previous reports,^[@CR41],[@CR42]^ we found that gefitinib had no effect on VM formation (Fig. [6f, g](#Fig6){ref-type="fig"}), and after gefitinib treatment the expression of VEGF was increased (Fig. [6g, h](#Fig6){ref-type="fig"}). However, combination therapy with DTIP and gefitinib could inhibit VM formation (Fig. [6f, g](#Fig6){ref-type="fig"}), VEGF expression (Fig. [6g, h](#Fig6){ref-type="fig"}), and vascular leakage (Supplementary Fig. [S7](#MOESM1){ref-type="media"}) compared with administration of gefitinib alone. These results indicated that DTIP potentiated gefitinib-induced inhibition of lung cancer and inhibited gefitinib resistance in mice.Fig. 6Combination therapy with DTIP and gefitinib results in improved antitumor efficacy.LLC cells (1 × 10^6^) were subcutaneously injected into the right dorsal region of mice. One week after LLC inoculation, the mice were randomly divided into four groups: normal saline, DTIP (1 mg/kg per day, s.c.), gefitinib (100 mg/kg per day, i.g.), and combination DTIP with gefitinib treatment groups, and then the mice were administered DTIP, gefitinib, or normal saline for 21 consecutive days. **a** Tumor growth curve after injection of LLC cells. **b** Mice were humanely killed and the tumors were resected 45 days after cell injection. **c** The weight of resected tumors was determined in different groups. **d** Representative photograph of metastatic nodules in the lungs (upper). H&E-stained sections of representative single lung lobes (bottom). **e** Number of mice with lung metastasis. **f** The number of VM tubes per tumor tissue sample was determined in different groups. **g** Representative images of anti-mouse anti-CD31/PAS staining in tumor tissues (upper). Immunohistochemical analysis was performed to determine the expression of VEGF (bottom). **h** The expression score for VEGF was determined. All the results are expressed as the mean ± SD. The error bars indicate the SD. ANOVA followed by Dunnett's test was applied for multiple comparisons. \**p* \< 0.05, \*\**p* \< 0.01, \*\*\**p* \< 0.001, NS, not significant Discussion {#Sec9} ========== In certain circumstances, tumors might have the potential to promote their own growth via alternative metastasis pathways, and metastasis is closely linked to the capacity for VM. VM is related to the expression of markers of pluripotent stem cells, invasive disease, and poor prognosis of patients, and VM is also related to tumor spread and metastasis.^[@CR11]^ We found that there were VM tubes in tumor samples of NSCLC patients and a high VM ratio was associated with poor prognosis. This result is consistent with the hypothesis that VM occurs through aggressive tumor "stem-like" phenotypes,^[@CR43]^ leading to worse clinical outcomes. Patients who survived \>5 years had low VM ratios. VM formation may be a biomarker for tumor prognosis. VM may be a potential therapeutic target. Prospective studies will be helpful to explore the molecular regulatory mechanism of VM. It has long been presumed that tumors may take advantage of the hemostatic system. There are many significant hemostatic abnormalities in cancer patients. In fact, hemostasis complications are a common cause of death in cancer patients. Studies have shown that exogenous thrombin is capable of enhancing tumor adhesion to platelets,^[@CR44]^ endothelial cells,^[@CR45]^ and fibronectin^[@CR46]^ in vitro and revealed that exogenous thrombin promotes tumor growth.^[@CR47]^ In our study, we found that thrombin was overexpressed in NSCLC, and that the level of thrombin was correlated with poor prognosis of NSCLC patients, which might be a reason why malignant tumors often exhibit hypercoagulability. The mechanism by which thrombin expression is increased in NSCLC needs to be further studied in the future. Interestingly, we also found that the expression of thrombin in tumors is closely related to the formation of VM. The expression of thrombin in tumors was increased in patients with VM. Hence, we hypothesize that VM might be driven by increased expression of thrombin. To further study the effect of thrombin on VM formation, NSCLC cell lines were treated with thrombin in vitro. The results showed that 10 nmol/L thrombin could promote NSCLC cell elongation and proliferation to varying degrees. It also promoted VM tube formation. r-hirudin and DTIP are new bivalent direct thrombin inhibitors that could effectively prevent the formation of thrombosis and embolism.^[@CR26],[@CR27]^ However, in vitro, r-hirudin and DTIP were effective against thrombin-induced tubular network formation in lung cancer cells. In lung cancer models, r-hirudin and DTIP inhibited tumor progression and spontaneous metastasis, and in r-hirudin- or DTIP-treated mice, VM formation was likely inhibited compared with that in normal saline-treated mice. This suggests that VM might be a target of r-hirudin and DTIP in suppressing tumor progression. A high degree of tumor leakiness has been associated with VM, in which tumor cells differentiate into endothelial-like cells and form extracellular matrix-rich tubular structures to carry blood from the vasculature to hypoxic regions of the tumor.^[@CR10],[@CR29]^ Vascular leakage^[@CR10],[@CR29]^ and EMT^[@CR11],[@CR38],[@CR39]^ are two important characteristics of VM formation. Furthermore, r-hirudin and DTIP inhibited EMT in vitro and in vivo, and reduced vascular leakage in mouse tumor models. Taken together, the results of vascular leakage analysis, the characteristics of VM, and the results of PAS/CD31 staining suggest that r-hirudin and DTIP could reduce vascular leakage and inhibit EMT, which strengthens the hypothesis that VM is one of the targets of r-hirudin and DTIP in inhibiting tumor development. It was reported that thrombin, acting through PAR-1, promotes EMT in embryo development.^[@CR25]^ Our studies have found that PAR-1 is a major determinant in thrombin-promoted metastasis of lung cancer; however, the contribution of thrombin/PAR-1 to EMT and VM formation is still unknown. Our previous studies suggested that PAR-1 was highly expressed in human NSCLC tissues but showed no difference based on subtype and clinical stage. Interestingly, we found that the relationships between thrombin and NSCLC prognosis, and VM formation were different in patients with various PAR-1 expression levels. Thrombin expression was closely associated with VM formation and survival in NSCLC patients with higher PAR-1 levels; however, it was not significantly related with VM formation and survival in NSCLC patients with low PAR-1 expression. NSCLC patients with thrombin-positive/PAR-1-high expression had the poorest prognosis and were the most likely to form VM. These findings suggest that thrombin, together with PAR-1, substantially contributes to VM formation and NSCLC malignancy. To further investigate the effect of thrombin on PAR-1-mediated VM formation, PAR-1-deficient NSCLC cells were treated with thrombin in vitro. We found that the ability of PAR-1-deficient NSCLC cells to form VM tubes was lost; thrombin had no effect on their VM formation capacity. The cleavage of PAR-1 by thrombin leads to the activation of NF-κB signaling cascades.^[@CR33]^ Studies have shown that NF-κB plays an important role in the EMT process. In this study, we found that thrombin could activate NF-κB signaling and induce EMT, and that r-hirudin and DTIP could inhibit thrombin-induced NF-κB activation and EMT. PDTC, an inhibitor of NF-κB, could inhibit thrombin-induced EMT. Thrombin is the main activator of PAR-1. In PAR-1-deficient NSCLC cells, thrombin had no effect on NF-κB activation, EMT, and VM formation, whereas LPS could rescue NF-κB activation and EMT. These results suggested that thrombin promoted the activation of NF-κB through the activation of PAR-1, thereby inducing EMT and VM formation. r-hirudin and DTIP could bind to thrombin specifically and block thrombin binding to PAR-1, thereby inhibiting NF-κB activation and reducing the expression of N-cadherin and snail, while increasing the expression of E-cadherin in NSCLC cells, and this mechanism is very beneficial for suppressing VM formation. This study is the first to illustrate the effect of thrombin inhibitors on the formation of VM. A series of studies in the early 2000s observed that distinct epidemiological subgroups of patients with NSCLC had dramatically enhanced responses to treatment with the ATP-competitive, reversible EGFR inhibitors gefitinib and erlotinib.^[@CR48]^ Meanwhile, EGFR inhibitors as angiogenesis inhibitors are significantly associated with an increased risk of arterial thromboembolism.^[@CR49]^ Less than 10% of NSCLC patients have an objective tumor response. In addition, it is an interesting possibility that antiangiogenic therapy may lead to a selective growth advantage favoring VM, thus leading to drug-induced resistance.^[@CR12]^ Based on the findings of this study, we conclude that the highly specific, potent thrombin inhibitor DTIP inhibits VM formation in NSCLC. In our studies, we found that gefitinib, an EGFR inhibitor, did not inhibit VM formation. Similar results have been reported by Rybak et al.^[@CR50]^ and Van der Schaft et al.^[@CR41]^ Gefitinib could increase the expression of VEGF, which is consistent with the results reported by Naumov et al.^[@CR42]^ These results might be associated with gefitinib resistance. Interestingly, combination therapy with DTIP with gefitinib could inhibit VM formation and VEGF expression. DTIP potentiated gefitinib-induced inhibition of lung cancer and inhibited gefitinib resistance in mice. Combination therapy with DTIP with gefitinib resulted in improved antitumor efficacy. This approach merits further evaluation in more extensive studies and in NSCLC patients. In this study, we not only showed that thrombin plays a crucial role in VM but also explained why hypercoagulability promotes cancer progression. VM formation may be a biomarker for tumor prognosis. We conclude that thrombin plays an important role in PAR-1-mediated VM formation via NF-κB signaling. We suggest that treatment with DTIP should start immediately after diagnosis (before extensive tumor development) and should be administered in conjunction with EGFR inhibitors. Our future experiments will focus on the mechanisms, the reason why the expression of thrombin is increased in NSCLC, and the ability of combination therapy with thrombin inhibitors and angiogenesis inhibitors to achieve a better therapeutic effect. Materials and methods {#Sec10} ===================== Materials {#Sec11} --------- The expression and purification of r-hirudin and DTIP have been described previously.^[@CR26],[@CR27]^ Detailed materials are described in the [Supplementary Materials](#MOESM1){ref-type="media"}. Animals {#Sec12} ------- C57BL/6 mice (5--6 weeks, male) were purchased from Shanghai SLAC Laboratory Animal Limited Company. All animal procedures were carried out in accordance with institutional guidelines at Fudan University. Patients and tissue slides {#Sec13} -------------------------- All experiments using human tissue slides were performed in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Fudan University. The clinical samples of lung cancer patients were obtained from Fudan University Shanghai Cancer Center (Shanghai, China). The tissue slides consisted of samples from 152 patients with NSCLC, including 118 cases of adenocarcinoma, 25 cases of squamous cell carcinoma, and 9 cases of other types of NSCLC at different stages. Healthy lung tissues from each sample were also included. The clinicopathological features of 152 NSCLC tissues are shown in Supplementary Table [1](#MOESM1){ref-type="media"}. Vasculogenic mimicry formation assay {#Sec14} ------------------------------------ Ninety-six-well plates were precoated with 50 μL liquid Matrigel per well. After incubation at 37 °C for 1 h, A549 cells (1.5 × 10^4^ cells/well) or Lewis cells (0.5 × 10^4^ cells/well) suspended in RPMI-1640 medium with phosphate-buffered saline, 10 nmol/L thrombin, 25 nmol/L r-hirudin, 50 nmol/L DTIP, 10 nmol/L thrombin + 25 nmol/L r-hirudin, or 10 nmol/L thrombin + 50 nmol/L DTIP were seeded and cultured for 4 h. The cells in five randomly chosen fields were photographed and counted. VM formation as measured by the capacity to form loops was quantified by AngioSys 2.0 (TCS Cellworks, Buckingham, England) and the number of loops was regarded as the number of tubes. Immunohistochemistry and scoring of immunostained specimens {#Sec15} ----------------------------------------------------------- Immunostained specimens were reviewed by two pathologists blinded to the patients' clinical status. To evaluate the expression of proteins using IHC, the intensity of protein staining was graded by consensus on a scale from 0 to 3 (0 was negative staining; 1 was weakly positive staining; 2 was moderately positive staining; and 3 was strongly positive staining). The frequency of positive cells was graded by consensus on a scale from 0 to 4 (0 was positivity in \<1% of cells; 1 was positivity in 1--10% of cells; 2 was positivity in 10--50% of cells; 3 was positivity in 50--80% of cells, and 4 was positivity in \>80% of cells). The immunohistochemical scores were determined by the staining intensity and the frequency of positive cells (staining intensity score × positive cell score). Based on the score of thrombin or PAR-1 staining in the central positive staining area in the tumor section as a cutoff value, the score (staining intensity score × positive cell score) of thrombin or PAR-1 was classified as positive (thrombin+ ≥ 2 and PAR-1 high expression ≥ 6) or negative (thrombin− \< 2 and PAR-1 low expression \< 6). VM detection {#Sec16} ------------ PAS staining and CD31 IHC were used to evaluate the presence and extent of mimicry as previously described.^[@CR11],[@CR14],[@CR29]^ Each sample was scored for the number and size of areas with morphology consistent with mimicry. The criteria used were as follows: (1) the presence of PAS-positive channels that contained red cells and fluid, (2) the absence of CD31 staining in these channels, and (3) the polarization of tumor cells on an indistinct or imperceptible matrix lining vascular channels with red cells and/or fluid and no evidence of endothelization or tumor cells lining vascular spaces with no evidence of a matrix. After PAS and CD31 staining, the whole sections were observed via slice digital scanning (3DHISTECH, Ltd). If typical VM structures could not be found in the whole slice, the sample was considered to be from a "patient without VM." Subcutaneous inoculation of tumors in mice {#Sec17} ------------------------------------------ This animal study was conducted in accordance with the rules and regulations of the Institutional Animal Care and Use Committee at the Department of Laboratory Animal Science, Fudan University (Shanghai, China). Tumor cells at a density of 1 × 10^6^ in 0.1 mL serum-free media were injected s.c. into the right flank of mice. The mice were monitored daily for the development of visible tumors. Once a tumor was clearly visible, it was calipered three times each week, and the volume was estimated using the formula *V* = (*LW*^2^)/2, where *V* is the volume, *L* is the longest diameter, and *W* is the shortest diameter. After mice were killed, the tumors were removed; some were fixed in 4% paraformaldehyde (PFA) and the other tumors were lysed for western blotting. Paraffin sections were made from the largest section of mouse tumor tissues and were used for PAS and CD31 staining. After PAS and CD31 staining, we observed the whole section via slice digital scanning (3DHISTECH, Ltd). The number of typical VM channels in the whole tumor tissue section of each mouse was counted. If there were no typical VM channels in the whole tumor tissue section, the number of VM components was regarded as 0. Vascular leakage {#Sec18} ---------------- Mice were injected with 0.1 mL fluorescein isothiocyanate dextran (10 mg/mL, 20 kDa, Sigma Aldrich) by tail vein injection. After 30 min, the mice were anesthetized and perfused with 4% PFA. After fixation, tumors were collected and placed in 4% PFA overnight at 4 °C. After this, samples were infiltrated with 20% sucrose overnight at 4 °C. Tumors were frozen in OCT compound (Sakura Finetek) and 15 μm-thick sections were cut, washed, incubated with 4′,6-diamidino-2-phenylindole, and mounted in ProLong Gold antifade reagent (Life Technologies). Sections were examined under a fluorescence microscope. An average of five fields were analyzed from each sample. Images were quantified using ImageJ software. Statistical analysis {#Sec19} -------------------- All data are expressed as the mean ± SD. Unless otherwise stated, differences between the two groups were analyzed by unpaired *t*-test when variances were equal and one-way analysis of variance followed by the Newman--Keuls test was used for multiple comparisons with Prism 6 (GraphPad, Inc.). *P* \< 0.05 was considered to be statistically significant. Supplementary information ========================= {#Sec20} ###### Supplementary information Supplementary information ========================= The online version of this article (10.1038/s41392-020-0167-1) contains supplementary material, which is available to authorized users. This study was supported by the National Natural Science Foundation of China (NSFC 81902995 and NSFC 81673498), the Science and Technology Commission of Shanghai Municipality (STCSM 16431904600), and a project funded by China Postdoctoral Science Foundation (2018M641936). We thank the Duoease Scientific Service Center for excellent language editing service and suggestions for figure revision. Z.B., W.M., and Z.Y. performed all of the experiments. M.Y., W.Q., and L.Y. participated in the research. Z.B. and M.W. designed experiments, analyzed data, and wrote the paper. Y.M. is the supervisor of Z.B. H.Z. and W.H. collected the clinical samples. All authors read and approved the final manuscript. The authors declare no competing interests.

Background {#Sec1} ========== Oxidative tress occurs when the production of reactive oxygen species (ROS) or their products are in excess of antioxidant defense mechanism. The brain is especially vulnerable to oxidative stress compared to other organs because it exhibits lower antioxidant enzyme activity and contains an abundance of unsaturated fatty acids, which are the targets of lipid peroxidation. Thereafter oxidative stress contributes to the neurodegenerative disorders, and is one of the earliest pathological changes in Alzheimer's disease \[[@CR1]\]. Apart from quenching free radicals, antioxidants may combat the ROS generation and block ROS-mediated deleterious effects by gene regulation. Supplementations with exogenous antioxidants from diets as well as medicinal herbs catch attention on avoid unwanted neuropathophysiology. Tertiary butyl hydroperoxide (t-BHP), a simple lipophilic alkyl hydroperoxide, has been reported to induce dose-dependent oxidative stress and damage in brain cells \[[@CR2]\]. The alkoxyl and alkyl radicals derived from t-BHP are a species that produce cell damage \[[@CR3], [@CR4]\]. Furthermore, t-BHP-induces a cellular redox imbalance may induce cell apoptosis \[[@CR5]\]. Therefore, t-BHP is a useful inducer for investigating the effectiveness of various neuroprotection products against oxidative stress. In addition, D-galactose at normal concentrations can be metabolized by D-galactokinase and galactose-1-phosphate uridyltransferase. However, high concentrations are converted into aldose and hydroperoxide by galactose oxidase, resulting in the generation of ROS in the brain \[[@CR6], [@CR7]\], which can affect bimolecules including protein, DNA and lipids and which may lead to consecutive functional disturbance and cell death. Thus, treatment of animals with D-galactose may serve as a useful model for studying the effect of neuroprotective agents. *Wedelia chinensis,* a *Wedelia* genus plant belonging to the *Compositae*, has been traditionally used as a hepatoprotective tea material in Taiwan \[[@CR8]\]. *W. chinensis* possesses multiple activities such as anti-microbial, anti-inflammatory, anti-cancer, and CNS-depressant activity \[[@CR9]--[@CR11]\]. Previously, Lin reported that *W. chinensis* contained four compounds capable of suppressing androgen activity: luteolin, apigenin, indole-3-carboxyaldehyde and wedelolactone \[[@CR12]\]. Luteolin and apigenin, belonging to flavonoid structure, have been reported to reveal anti-oxidant, anti-inflammatory, and anti-cancer effects \[[@CR13]--[@CR15]\]. Wedelolactone, belonging to coumarin structure, has been demonstrated to exhibit a wide range of biological effects including anti-inflammation, immunomodulatory, anti-myotoxic, anti-oxidant, anti-phlogistic, anti-haemorrhagic, anti-hepatotoxic and anti-cancer activity \[[@CR16]--[@CR20]\]. Since the neuroprotection potential of *W. chinensis* is not well understood, in the present study we used two models, *in vitro* and *in vivo*, to evaluate its neuroprotective potential. Methods {#Sec2} ======= Reagents and materials {#Sec3} ---------------------- Dulbecco's modified Eagle's medium (DMEM), fetal bovine serum, and penicillin/streptomycin were purchased from GIBCO Ltd. (Grand Island, NY, USA). Antibodies, such as anti-Bcl-2, -Bcl-X~L,~ -Bax, -Nrf2 and -β-actin antibodies were obtained from Santa Cruz Biotechnology (Santa Cruz, CA). The herb, *W. chinensis*, was purchased from a reputable folk medicine store in Taiwan and identified by associate professor Ko-Kaung Liao, who majors in Biology and is familiar with medicinal plants. Wedelolactone, 7-methoxy-5, 11, 12-trihydroxy-coumestan, was synthesized from commercially available phloroglucinol \[[@CR21]\]. All chemicals were analytical grade purchased from Sigma-Aldrich (USA) and Merck (Germany). Preparation of the crude extract {#Sec4} -------------------------------- The air-dried whole plant of *W. chinensis* was shredded in a blender and extracted with 5 volumes of 95% ethanol at room temperature for 2 days. The mixture was filtered through filter paper (5 μm pore size), and the filtrate was dried using rotary evaporation under vacuum at 40°C. The percentage yield was 9.3% (w/w). The crude ethanolic extract was suspended in distilled water and partitioned successively with n-hexane and ethyl acetate, respectively to obtained semi-crudes. All dried extracts, including ethanol extract, n-hexane extract (HEW), and ethyl acetate extract (EAW) were stored at -20°C prior to use in the following studies. DPPH radical scavenging assay {#Sec5} ----------------------------- The 2,2-diphenyl-1-picrylhydrazyl (DPPH) method measures the reaction of the antioxidants with the stable DPPH radical in a methanol solution. Briefly, a 60 μM DPPH radical solution was freshly made in methanol. Various concentrations of the sample extracts were reacted with the DPPH radical solution (3 mL) for 45 min at room temperature, and the absorbance was measured at 517 nm. The affinity of the test material to quench the DPPH free radical was evaluated according to the equation scavenging % = (absorbance of control group--absorbance of the extract added group)/absorbance of control group × 100%. HPLC analysis {#Sec6} ------------- EAW was analyzed using a Hitachi L7100 HPLC system with a 5-μm ODS-Hypersil column (250 × 4.6 mm). The mobile phase was generated from solvent A \[acetonitrile: H~2~O: acetic acid (90:10:3)\] and solvent B \[acetonitrile: H~2~O: acetic acid (10:90:3)\] using the following gradient program: 0--3 min 100% solvent B, 3--8 min 85% solvent B, 8--15 min 80% solvent B, 15--40 min 60% solvent B. The detection wavelength was 360 nm, and the flow rate 0.8 mL/min. Quantization was carried out by the external standard method on the basis of the area at 360 nm using calibration curve of wedelolactone and luteolin. Cell culture {#Sec7} ------------ Adrenal pheochromocytoma (PC12) cells were maintained in DMEM medium containing 5% fetal bovine serum, 10% horse serum, and 100 U/mL penicillin and streptomycin in a 5% CO~2~ incubator at 37°C. Preparation of nuclear protein and analysis of antioxidant-response element (ARE) binding activity of nuclear factor E2-related factor 2(Nrf2) {#Sec8} ---------------------------------------------------------------------------------------------------------------------------------------------- PC12 cells (1 × 10^6^ cells/ml) were exposed to the indicated reagent or vehicle (0.1% DMSO) for 6 h. Nuclear extracts were harvested using NE-PER nuclear extraction reagent (Thermo Fisher Scientific, Rockford, IL, USA) according to the manufacture's instructions. Nuclear protein concentrations were determined using the Bio-Rad protein assay reagent. The amount of Nrf2 available in the nucleus to bind AREs was determined using a TransAM Nrf2 Transcription Factor ELISA Kit (Active Motif Inc., Carlsbad, CA, USA) according to the manufacturer's instructions. Briefly, nuclear extracts (2.5 μg) were added to wells that contained the immobilized consensus ARE oligonucleotide. A primary anti-Nrf2 antibody was added to each well, followed by a horseradish peroxidase-conjugated secondary antibody. The signal was detected at 450 nm, and Nrf/ARE binding activity is reported as A~450~. Intracellular ROS analysis {#Sec9} -------------------------- PC12 cells were pretreated with EAW for 24 h and then treated with or without 100 μM of t-BHP for 3 h. After the indicated treatments, PC12 cells were incubated with the fluorescent probe 2′,7′-dichlorofouorescein diacetate (DCFH-DA; 10 μM) for 30 min. DCFH-DA becomes highly fluorescent 2′,7′-dichlorofouorescein (DCF) upon oxidation by ROS, and then washed twice with PBS. Finally the fluorescence intensity of DCF was measured in a microplate reader with an excitation wavelength 485 nm and an emission wavelength of 535 nm. Determination of cytotoxicity by LDH release assays {#Sec10} --------------------------------------------------- Cells were seeded in 24-well plates at a density of 2 × 10^5^ cells/ml. PC12 cells were pretreated with EAW (0, 10, 25 and 50 μg/mL) for 24 h and then treated with or without 100 μM of t-BHP for 3 h. The cytotoxicity was quantitatively assessed by measuring the activity of the lactate dehydrogenase (LDH) released from the damaged cells into the culture medium. At the end of the treatment, the media was used for the LDH activity assay. The enzyme was determined by using an assay kit according to the manufacturer's protocol. The absorbance of the samples was read at 490 nm using a microplate reader. The LDH release was proportional to the number of damaged PC12 cells. Sub-G~0~/G~1~ hypodiploid FACS analysis {#Sec11} --------------------------------------- After treatment, the cells were fixed with ice-cold 70% ethanol overnight. The cells were then centrifuged at 1500 g for 10 min and resuspended in PBS containing 0.5% Triton X-100, 0.1 mg/mL RNase, and 40 μg/mL propidium iodide (PI) at room temperature for 30 min. Cellular DNA content (10,000 cells per analysis) was determined by flow cytometric analysis of PI-labeled cells using a FACSCalibur flow cytometer (Becton Dickinson, San Jose, CA) equipped with a single argon-ion laser (488 nm). Cell cycle was analyzed using Modfit software (Verity Software House, Topsham, ME) and those cells in apoptosis were indicated by the substantially lower DNA content compared to the G~0~/G~1~ peak in the DNA histogram. Western blotting analysis {#Sec12} ------------------------- The cells were washed in PBS with 1 mM zinc ion and lysed in radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris-buffer, 5 mM EDTA, 150 mM NaCl, 1% NP 40, 0.5% deoxycholic acid, 1 mM sodium orthovanadate, 81 μg/mL aprotinin, 170 μg/mL leupeptin, and 100 μg/mL PMSF; pH 7.5). The mixtures were mixed for 30 min at 4°C and centrifuged (10,000 g) for 10 min. The supernatants were collected as whole-cell extracts. The protein content was determined with the Bio-Rad protein assay reagent using bovine serum albumin as a standard. Protein samples (50 to 100 μg protein) were boiled, separated on an SDS polyacrylamide gel, electrophoretically transferred to nitrocellulose membranes (Amersham, Arlington Heights, IL) and blotted with the indicated primary antibodies. The proteins were visualized with horseradish peroxidase (HRP)-conjugated secondary antibodies (Zymed Laboratory, Inc., San Francisco, CA) followed by chemiluminescence (ECL-Plus; Santa Cruz Biotechnology). The relative photographic density was quantified using densitometry. Assessment of cytochrome C release {#Sec13} ---------------------------------- Cells were harvested and washed with ice-cold PBS at the end of treatment to prepare the cytosolic fraction. The cell pellet was resuspended in 500 μL buffer A (20 mM HEPES-KOH, pH 7.5, 10 mM KCl, 1.5 mM MgCl~2~, 1 mM sodium EDTA, 1 mM leupeptin and 1 μg/mL chymostatin) and homogenized using a Pyrex glass homogenizer and a type B pestle (40 strokes). The homogenate was centrifuged at 100,000 g (4°C for 1 h) to generate the cytosolic fraction. A Bio-Rad protein assay kit determined protein concentrations, and 25 μg of each fraction was separated using 15% SDS-PAGE. The proteins were transferred to an NC membrane, and then reacted sequentially with a primary antibody (anti-cytochrome C and anti-β-actin as internal controls) and a secondary peroxidase-conjugated antibody. Protein bands were revealed using enhanced chemiluminescence (ECL commercial kit). Animals and treatments {#Sec14} ---------------------- Five-week-old male ICR mice were purchased from GlycoNex Inc. (Taiwan) and kept in groups of three to four per cage in an animal room at a temperature of 22 ± 2°C, 12 h light--dark cycle, and relative humidity 50-70%. The mice were provided with rodent chow and water *ad libitum*. All experiments were approved by the Animal Care Committee of Chung Shan Medical University (IACUC approval No. 896). After adaptation for a week, the mice were randomly divided into 4 groups (n = 6). Group 1 served as the vehicle control and received daily subcutaneous (s.c.) injections of 0.1 mL of distilled water and three intraperitoneal (i.p.) injections/week of 1% DMSO in distilled water for 9 weeks. Group 2 received daily s.c. injections of 50 mg/kg of D-galactose in distilled water and three i.p. injections/week of 0.1 mL of 1% DMSO in distilled water for 9 weeks. Groups 3 and group 4 received the same s.c. injections of D-galactose plus three i.p. injections/week of 0.1 mL of 1% DMSO containing, respectively, 10 mg/kg or 25 mg/kg of EAW. Preparation of brain samples and measurement of lipid peroxidation {#Sec15} ------------------------------------------------------------------ Mice were euthanized and the brain removed and dissected on an ice-bath plate. The right cerebral cortex was fixed in 10% paraformaldehyde for histological studies and the left cerebral cortex was homogenized in 10 volumes (v/w) of ice-cold saline containing a protease inhibitor cocktail (Sigma-Aldrich, MO, USA). The resultant homogenate was centrifuged and the supernatant was removed for lipid peroxidation determination. The amount of malondialdehyde (MDA) was measured by reaction with thiobarbituric acid and the absorbance read at 532 nm on a spectrophotometer. TUNEL staining and histopathological examination {#Sec16} ------------------------------------------------ The cerebral cortex samples were dehydrated, embedded in paraffin blocks after post-fixation, then sliced and the sections stained with terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL). The TUNEL positive cells were visualized using the peroxidase-DAB reaction and counterstained with hematoxylin. Detailed procedures followed the manufacturer's protocol for the in situ Apoptosis Detection Kit (In Situ Cell Death Detection Kit; Roche Diagnostics, Mannheim, Germany) according to the manufacturer's instructions. The digital images were captured using a digital camera (Canon A640, Tokyo, Japan). Quantification of apoptotic cells was by measuring three randomly selected microscopic fields (100× magnification) for each slide. The percentage of TUNEL-positive cell (%) = 100 × (TUNEL-positive cells/total cells). In addition, the sections were stained with hematoxylin-eosin for light microscopic analysis. Large nuclei and dark staining neurons in the cerebral cortex were counted in five random fields per slide at 400× magnification and the density of neurons expressed as no./mm^2^. Statistical analysis {#Sec17} -------------------- Statistical significance was determined using one-way analysis of variance (ANOVA) followed by Dunnett's or Tukey's post hoc test. *P* values less than 0.05 were considered statistically significant. Results and discussion {#Sec18} ====================== Free radical scavenging activity and main constituent of EAW {#Sec19} ------------------------------------------------------------ Bleaching of DPPH was performed to determine the free radical scavenging capacity of the crude extracts. In our preliminary study, ethanol extract of *W. chinensis* exhibited free radical scavenging activity, and after solvent partition, it showed EAW increased the potency of free radical scavenging activity, with 10 μg/mL of EAW quenching \~50% of free radicals (Figure [1](#Fig1){ref-type="fig"}A). Then HPLC analysis examined the composition of EAW as compared with the authentic samples. This demonstrated that wedelolactone and luteolin were the main constituents (retention time 19.02 min and 28.15 min) (Figure [1](#Fig1){ref-type="fig"}B). The content of wedelolatone in EAW was 12.8% and luteolin was 11.4% respectively.Figure 1**DPPH quenching effect and HPLC analysis of** ***Wedelia chinesis*** **extract. (A)** The indicated concentrations of the ethanol extract, hexane extract (HEW), and ethyl acetate extract (EAW) of *Wedelia chinensis* were evaluated for DPPH quenching capacity as described in the text. % of DPPH scavengering activity = (absorbance of control group -- absorbance of the extract added group)/absorbance of control group × 100%. Means ± SD (n = 3). **(B)** HPLC analysis of EAW*.* For the conditions, see the Materials and Methods. Peak 1 is wedelolactone, and peak 2 is luteolin. EAW activates nuclear Nrf2-ARE binding activity {#Sec20} ----------------------------------------------- Figure [1](#Fig1){ref-type="fig"} shows that EAW exhibited radical scavenging activity. In addition to direct scavenging radicals, whether noncytotoxic concentration of EAW has antioxidant enzyme induction activity in PC12 cells was investigated. The antioxidant response element (ARE) is a cis-acting enhancer sequence found in the promoter region of many genes encoding antioxidant enzymes/proteins. Nrf2 is a major ARE-binding transcription factor and induces the expression of antioxidant enzymes. Thereafter, whether EAW can enhance binding activity of nuclear Nrf2 to the ARE was evaluated. It was found that treatment with noncytotoxic concentration of EAW (25 and 50 μg/mL) significantly activated Nrf2--binding activity (Figure [2](#Fig2){ref-type="fig"}).Figure 2**Effects of EAW on Nrf2 activation.** After pretreatment with EAW, PC12 cells were harvested and nuclear extracts were prepared and analyzed using the TransAM Nrf2 kit, as described in the Materials and Methods. Data represent the means ± SD (n = 3). \**P \<* 0.05, versus solvent control (0.1% DMSO) treated alone. Effects of EAW on t-BHP-induced intracellular ROS, cytotoxicity and apoptosis {#Sec21} ----------------------------------------------------------------------------- The intracellular level of ROS with or without EAW treatment prior to t-BHP exposure was evaluated. The treatment of PC12 cells with 100 μM t-BHP for 3 h increased DCF fluorescence. It implicated t-BHP induced intracellular ROS. In our preliminary study, pretreatment of EAW for 6 h didn't exhibited apparent reduction effect on t-BHP-induced intracellular ROS (data not shown). However, pretreatment of EAW for 24 h showed reducing effect significantly (Figure [3](#Fig3){ref-type="fig"}). Thereafter, pretreatment of EAW for 24 h was used in the following cell culture study. LDH is a stable cytoplasmic enzyme that is present in all cells, including neurons. LDH is rapidly released into the cell culture medium when the cell plasma membrane is damaged. Therefore, LDH is a reliable biochemical index for neuronal cytotoxicity. Treatment with t-BHP for 3 h increased LDH release into the medium to 123.0 ± 3.3% of control (Figure [4](#Fig4){ref-type="fig"}A). Pretreatment with different EAW concentrations prior to t-BHP exposure significantly inhibited LDH leakage in the PC12 cell system compared to t-BHP treated alone group (Figure [4](#Fig4){ref-type="fig"}A). The flow cytometry analysis was used to evaluate the t-BHP-induced apoptosis of PC12 cells (Figure [4](#Fig4){ref-type="fig"}B). It showed t-BHP-induced 23.21% sub-G~0~/G~1~ phase, which indicated DNA degradation and the cells undergoing apoptosis. With the pretreatment of 25 and 50 μg/mL EAW significantly declined to 16.83% and 11.30% respectively compared to t-BHP treated alone group (Figure [4](#Fig4){ref-type="fig"}B).Figure 3**Effects of EAW on t-BHP-induced intracellular ROS level.** After treatment, the intracellular ROS were determined using labeling with DCFH-DA as described in the text. \#\# *P* \< 0.01, compared to the solvent control (0.1% DMSO). \**P* \< 0.05 and \*\**P* \< 0.01, compared to t-BHP treatment alone.Figure 4**Effects of EAW on t-BHP-induced cytotoxicity and apoptosis in PC12 cells.** PC12 cells were pretreated with EAW (0, 10, 25 and 50 μg/mL) for 24 h and then treated with or without 100 μM of t-BHP for 3 h. **(A)** The cytotoxicity was determined by the LDH assay kit. **(B)** The DNA content was analyzed by the flow cytometry as described in the text (upper panel). The percentage of apoptotic cells (sub-G~1~ phase) is expressed as mean ± SD (n = 3). \#\# *P* \< 0.01, compared to the solvent control (0.1% DMSO). \**P* \< 0.05 and \*\**P* \< 0.01, compared to t-BHP treatment alone. Effects of EAW and its major constituents on t-BHP-induced changes in cytochrome C and the Bcl-2 family {#Sec22} ------------------------------------------------------------------------------------------------------- Oxidative stress initiates mitochondrial apoptotic signals \[[@CR22], [@CR23]\]. The loss of mitochondrial cytochrome C and its release into the cytoplasm are essential events in the initiation of mitochondrial caspase signaling. It was found that t-BHP treatment increased cytosolic cytochrome C, which was decreased by pretreatment with EAW or luteolin or wedelolactone (Figure [5](#Fig5){ref-type="fig"}A). The Bcl-2 family members have been found to play important roles in regulating mitochondrial-mediated apoptosis. High ratio of Bcl-2/Bax and Bcl-X~L~/Bax were observed to promote cell survival by preserving the integrity of the external mitochondrial membrane, which prevents the released of cytochrome C from the mitochondria, inducing cell death. It has been reported that overexpression of Bcl-2 in neurons was found to protect from neuronal loss in mouse model of cerebral ischemia \[[@CR24]\]. In addition, Bax inhibition prevented neuronal death in a mouse model of Parkinson's disease \[[@CR25]\]. The ratio of Bcl-2/Bax and Bcl-X~L~/Bax are important in determining sensitivity to apoptotic stimuli. Therefore, we determined the effect of EAW, luteolin and wedelolactone on t-BHP--induced changes in these proteins. The treatment of PC12 cells with 100 μM t-BHP for 3 h decreased the ratio of Bcl-2/Bax and Bcl-X~L~/Bax significantly, which were reversed by the pretreatment of EAW or luteolin or wedelolactone (Figure [5](#Fig5){ref-type="fig"}B). However, PC12 cells cannot totally represent the characters of primary cultured neurons, so, the further work needs to focus directly on primary neurons.Figure 5**Effects of EAW, luteolin and wedelolactone on t-BHP-induced cytochrome C release and changes in Bcl-2 family proteins.** PC12 cells were pretreated with EAW (25 and 50 μg/mL) or luteolin (L10, 10 μM; L20, 20 μM) or wedelolactone (W10, 10 μM; W20, 20 μM) for 24 h, then 100 μM of t-BHP was added for 3 h. **(A)** Cells were harvested after incubation, and the cytosolic protein extracts were prepared as described in the text and subjected to Western immunoblotting analysis against anti- cytochrome C. The average densitometric value of cytochrome C is shown as the mean ± SD (n = 3). \# *P* \< 0.05, compared to the solvent control (0.1% DMSO). \**P* \< 0.05, compared to t-BHP treatment alone. **(B)** After treatment, the total cell lysates were prepared and subjected to Western immunoblotting analysis against against anti-Bcl-2, -Bcl-xL, -Bax and anti-actin. Actin was used as the loading control. The ratio of Bcl-2/Bax and Bcl-xL/Bax are shown as the mean ± SD (n = 3). \# *P* \< 0.05, compared to the solvent control (0.1% DMSO). \**P* \< 0.05, compared to t-BHP treatment alone. Effects of luteolin and wedelolactone on nuclear translocation of Nrf2 and γ-GCS expression {#Sec23} ------------------------------------------------------------------------------------------- It has been demonstrated that some antioxidants are exerting their antioxidative effect by activating Nrf2, which is a master regulator of the antioxidant response \[[@CR26]\]. Activation of Nrf2 requires its translocation into the nucleus. It showed luteolin and wedelolactone increased the nuclear translocation of Nrf2, which implicated activation of Nrf2 (Figure [6](#Fig6){ref-type="fig"}). In addition, luteolin and wedelolactone increased the protein expression of γ-glutamyl-cysteine synthetase (γ-GCS), a product of Nrf2 target gene (Figure [6](#Fig6){ref-type="fig"}). Luteolin and wedelolactone containing phenolic structure have been reported to exhibit free radical quenching properties \[[@CR27], [@CR28]\]. Therefore, luteolin as well as wedelolactone contributed to the neuroprotective potential of EAW by activating Nrf-2 and quenching ROS. In addition to oxidative stress, degenerative lesions are associated with inflammatory reaction \[[@CR29]\]. Since luteolin and wedelolactone also have anti-inflammatory potential \[[@CR13], [@CR16]\], this may also contribute to the neuroprotective potential of *W. chinensis.*Figure 6**Effects of luteolin and wedelolactone on Nrf2 nuclear translocation and γ-GCS expression.** The cells were treated with luteolin or wedelolactone for 24 h prior, then the cells were collected for protein extraction. The nuclear protein extract and total protein extract were applied to western blotting analysis against anti-Nrf2 and anti-γ-GCS respectively. The average densitometric value is shown as the mean ± SD (n = 3). \**P* \< 0.05, compared to the solvent control (0.1% DMSO). Effect of EAW on MDA levels and neuronal cell density in the cerebral cortex of D-galactose-treated mice {#Sec24} -------------------------------------------------------------------------------------------------------- Oxidative stress is proposed to be major contributor of the aging process and is linked to neurodegenerative disease such as Alzheimer's disease \[[@CR30], [@CR31]\]. Neurodegenerative diseases are characterized by progressive neuronal loss that leads to functional declines. There has been a growing interest in preventing various age-related pathologic conditions in brain. One of strategies to slow or delay neuronal damage is reducing brain oxidative stress. MDA is a metabolic product of lipid peroxidation, which is generated under oxidative stress. It has been reported that high concentration of D-galactose may induce oxidative stress in brain \[[@CR7]\]. Figure [7](#Fig7){ref-type="fig"}A shows that D-galactose treatment caused an increase in MDA levels in the cortex compared to the control group (*P* \< 0.05) and that this increase was reduced by co-administration of 10 or 25 mg/kg of EAW. Since oxidative stress might induce cell apoptosis we investigated the effect of D-galactose on cell apoptosis in the cortex by TUNEL assay. This showed TUNEL-positive cells were significantly increased in the D-galactose treated group compared to the vehicle control group (*P* \< 0.05), and that they were reduced by co-administration of EAW (*P* \< 0.05; Figure [7](#Fig7){ref-type="fig"}B and C). In addition, Figure [7](#Fig7){ref-type="fig"}D shows representative photomicrographs of hematoxylin-eosin staining in the cortex of control mice and mice treated with D-galactose with or without co-administration of EAW. Neuron cells (large nucleus and dark staining) in the sections are stained by hematoxylin-eosin. As shown in Figure [7](#Fig7){ref-type="fig"}E, cell counts revealed that mice treated with D-galactose for 9 weeks showed neuronal loss in the cortex compared to the vehicle control group (*P* \< 0.05) and that this D-galactose-induced neuronal cell loss was inhibited by co-administration of EAW significantly (*P* \< 0.05). One might be interested in knowing the dosage of EAW when applying in the human. By using the body surface area normalization method \[[@CR32]\], the effective dose, 10 and 25 mg/kg, of EAW in preventing neuronal loss in the cortex in mice can be translated to the human equivalent dose, 0.81 and 2.03 mg/kg. However, further study is needed to evaluate the effective amount of *W. chinensis* when used in human for improving the health or for a therapeutic purpose.Figure 7**Effects of EAW on lipid peroxidation, TUNEL assay, and neuronal density in the cortext of D-galactose-treated mice. (A)** MDA levels were analyzed as described in the text. **(B)** Representative of TUNEL stained brain section. Arrow indicates TUNEL positive cells. **(C)** TUNEL positive cells were quantified as described in the text. Data are expressed as mean ± SD (n = 6). **(D)** Representative hematoxylin-eosin stained sections of the cortex in each group (200×). Scale bar: 100 μm. **(E)** Neuronal density calculated from the number of neurons in a 0.218 mm^2^ field (400x) and expressed as the mean ± SD for the number of mice indicated in the text. \# *P \<* 0.05, compared to the control group; \**P \<* 0.05 and \*\**P \<* 0.01, compared to the D-galactose-treated group. Conclusion {#Sec25} ========== EAW suppressed t-BHP-induced cytotoxicity and apoptosis by decreasing intracellular ROS through scavenging radicals and activating endogenous antioxidant capacity in PC12 cells. *In vivo* experiment, EAW reduced D-galactose-induced lipid peroxidation and neuron apoptosis in the cerebral cortex of mice. Further, in addition to luteolin, we found wedelolactone, induced the nuclear translocation of Nrf2 and the expression of γ-GCS. These results demonstrate that *W. chinensis* possesses neuroprotective potential by blocking oxidative stress-induced cell damage and that luteolin and wedelolactone contribute to the action. Research on other mechanisms by which EAW exerts its neuroprotective effect is currently in progress. **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** The authors participated in the data analysis and manuscript preparation. All authors approved the final manuscript. This study was supported by grants from the Chung Shan Medical University Hospital (CSH-2012-C-018) and the Ministry of Science and Technology (NSC 99-2632-B-040-001-MY3). Densitometry was performed at the Instrument Center of Chung Shan Medical University, which is supported by the Ministry of Science and Technology, Ministry of Education and Chung Shan Medical University.

Background ========== The in-hospital resuscitation of patients suffering from major traumatic haemorrhage has undergone a revolution over recent years, with the previous mainstay of resuscitation---crystalloid---being replaced by blood component therapy. Observational evidence from both military and civilian practice suggests a survival benefit from resuscitation with high ratios of plasma to packed red blood cells (pRBC) (so-called "haemostatic resuscitation" (HR)) \[[@B1],[@B2]\]. It has been suggested that early use of blood products rather than crystalloids may convey a survival advantage, perhaps due to improved oxygen-carrying capacity, more effective volume expansion and by lessening the coagulopathy of trauma \[[@B3],[@B4]\]. The pre-hospital setting is more complex. A previous randomised controlled trial \[[@B5]\] demonstrated that pre-hospital crystalloid resuscitation increased mortality and morbidity in patients who had suffered penetrating trauma. It is believed that early aggressive volume administration may lead to "clot blow-off", i.e. clots dislodged due to increased arterial pressure, and rebleeding. Consequently, pre-hospital resuscitation moved towards restricted fluid regimes. Military experience has proven that it is feasible and practical to provide blood products for transfusion in the pre-hospital arena \[[@B6]\]. Anecdote and limited observational evidence from British military casualty retrieval missions suggest a potential survival benefit from pre-hospital HR \[[@B7],[@B8]\] but are confounded by factors such as increasingly liberal in-hospital transfusion practices \[[@B9]\]. However, a large observational study of pre-hospital blood products failed to identify any reduction in mortality or coagulopathy \[[@B10]\]. This is consistent with a civilian in-hospital study which suggested that HR's effect is dependent on surgical haemorrhage control \[[@B11]\]. Other confounders of military studies include incomplete data (due to the nature of the combat environment) and inability to follow up patients---particularly those of other nationalities---following discharge. The military population is younger, almost exclusively male and fitter, with minimal comorbidity compared to civilian trauma patients. Mechanisms of injury differ; over 95% of severe battlefield trauma results from explosive events or gunshot from military weapons \[[@B10]\]---mechanisms almost unknown in UK civilian practice and rare in other countries. Consequently, perceived benefits from military pre-hospital cannot be assumed to translate directly to civilian practice. Nonetheless, an increasing number of civilian pre-hospital retrieval services are carrying pRBCs for trauma resuscitation, which has significant cost and logistical implications. Pilot studies \[[@B12]-[@B14]\] have demonstrated the feasibility of pre-hospital pRBC use with complete traceability and minimal wastage (0.0%--1.6%), with no early transfusion reactions recorded. Studies on the effectiveness of pre-hospital pRBC \[[@B3],[@B15],[@B16]\] or plasma \[[@B17]\] transfusion in a civilian population suggest a benefit; however, these findings are limited by factors including retrospective design, non-standardised care and insufficient statistical power. There are at least two relevant ongoing randomised controlled trials \[[@B18],[@B19]\]. Despite the existence of a number of primary studies as outlined above, a scoping search identified no existing systematic reviews. A systematic review \[[@B20]\] on all aspects of the acute management of trauma from 2011 was limited to RCTs and did not find any relevant studies. US guidelines from 2009 \[[@B21]\] had a limited search strategy, and searches were performed in 2007; this report included two potentially relevant primary studies \[[@B22],[@B23]\]. A narrative review from 2013 \[[@B24]\] with a limited search strategy identified only one relevant study \[[@B12]\]. Pre-hospital resuscitation with blood components has already been adopted as routine practice in some emergency services. Urgent clarification of the evidence base is required. The aim is therefore to undertake a systematic review of the evidence on the clinical effectiveness of pre-hospital blood components (red blood cells and/or plasma or whole blood), in both civilian and military settings, compared with other resuscitative fluids in patients with major traumatic haemorrhage. Methods/design ============== Standard systematic review methodology aimed at minimising bias will be employed, and reporting will follow the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines \[[@B11]\]. This protocol is registered with PROSPERO (CRD42014013794). Searches -------- The following sources will be searched for primary studies: ● Bibliographic databases---MEDLINE, MEDLINE In Process and EMBASE via Ovid and The Cochrane Library (CENTRAL databases) ● UK Blood Services Transfusion Evidence Library ● Defence Medical Library Service (Ministry of Defence) ● Science Citation Index (ISI) for citation searching ● Current controlled trials (<http://www.controlled-trials.com/>), WHO International Clinical Trials Registry Platform (ICTRP) and ClinicalTrials.gov for ongoing studies ● Specialist abstract and conference proceeding resources (British Library's ZETOC and ISI Proceedings) ● Checking of citation lists of included studies and relevant reviews ● Contact with study authors and researchers of ongoing trials ● Contact with experts in the field (e.g. THOR---The Trauma Hemostasis and Oxygenation Research Network, the Cochrane Pre-hospital and Emergency Care group) ● Hand searching may be performed for relevant conference abstract books A combination of alternative text and MeSH terms relating to the condition (haemorrhage), intervention (blood components) and setting (pre-hospital) will be utilised. There will be no language restrictions applied to the searches. Given that the number of relevant primary studies is likely to be small, and the study design (terminology) variable, searches will be broad and study design filters will not be used. There will also be no restriction of searches by comparators or outcomes. A sample search strategy for MEDLINE is provided in Additional file [1](#S1){ref-type="supplementary-material"}. Search results will be entered into electronic databases (EndNote version X7.1, Thomson Reuters, New York) to facilitate record keeping, duplicate removal, study selection and document writing. Titles (and abstracts where available) of articles identified by the searches will be screened by two reviewers, independently, for relevance to the review question using pre-specified screening criteria. This process will be aimed at removing non-relevant studies. Full text copies of potentially relevant articles will be acquired and assessed independently against the inclusion criteria by two reviewers. Discrepancy between reviewers will be resolved by discussion or by referring to a third reviewer. Where necessary, translation (full/part) of non-English language articles will be undertaken to facilitate this process and subsequent reviewing; the review team has access to a wide range of translators. The study selection process will be illustrated using a PRISMA flow diagram. Reference management software will be used to record reviewer decisions, including reasons for exclusion. Selection criteria ------------------ ### Study design Based on scoping searches, relevant studies are most likely to be (retrospective) cohorts or uncontrolled studies (case series). Inclusion of studies will be limited in the first instance to controlled studies (randomised or non-randomised, prospective or retrospective, concomitant or historical control). It is likely that evidence from controlled studies will be limited (≤10 studies); if this is the case, inclusion will be extended to uncontrolled studies. ### Patient group The patient group will be individuals aged ≥16 with major traumatic haemorrhage with hypotension (as defined by study authors). There will be no restriction on type, mechanism or location of injury, or co-existing conditions. ### Intervention Any blood components likely to be administered in a pre-hospital setting, including the following: whole blood, RBCs alone, RBCs and plasma (any ratio) or plasma alone. ### Comparator (for controlled studies) Any other resuscitative fluid (e.g. crystalloids, colloids) or any of the interventions compared against each other, with or without additional agents such as tranexamic acid. ### Setting The setting is pre-hospital, i.e. administration of fluids at the point of injury or in transit to hospital (ground or air transport). Both civilian and military environments will be eligible. ### Outcomes There will be no restriction by outcome. The primary outcome of interest is survival; however, studies are unlikely to be powered to show any differences. The following (secondary) outcomes will therefore also be considered: ● Lactate concentration ● Blood gases ● Measures of coagulopathy ● Additional blood components (or other fluids) given in hospital ● Organ failure ● Infection ● Adverse events associated with intervention ● Outcomes relating to the feasibility of using blood components in a pre-hospital setting (e.g. wastage and transfusion reactions) Data extraction --------------- Data extraction will be conducted by one reviewer using a standardised, piloted data extraction form and checked by a second reviewer. Disagreements will be resolved through discussion or referral to a third reviewer. For each study, the data required on (but not limited to) the following will be sought: ### Study characteristics ● Country of origin ● Study design ● Setting (civilian or military) ● Sample size ● Length of follow-up ### Population ● Patient inclusion and exclusion criteria (including definition of hypotension) ● Location, mechanism and severity of injury (e.g. as defined by the Injury Severity Score or New Injury Severity Score) ### Intervention/comparator ● Combination/ratio of blood components and quantity administered ● Comparator fluid and quantity administered ● Location of administration (e.g. at the scene or in transit) ● Characteristics/training of individual administering fluids ### Results ● Completeness of follow-up ● Outcome measures ● Statistical methods employed (e.g. for adjusting for confounders) ● Findings ● Effect sizes and associated uncertainty Quality assessment ------------------ Data will be extracted to allow quality assessment of the included studies. Quality assessment will be based on tools specific to a given study design. For randomised and non-randomised controlled trials (should any be identified), quality assessment will be based on the risk of bias tool from the Cochrane Handbook \[[@B25]\]. The Newcastle-Ottawa Scale (NOS) \[[@B26]\] will be used for cohort or case-control studies. Some of the confounding factors likely to arise in non-randomised studies include the following: severity and mechanism of injury, distance between site of injury and hospital, transportation time, co-interventions, characteristics of individual administering fluids and adherence to fluid protocols. When assessing study quality, the extent to which confounders have been given adequate consideration, both in reporting and analysis, will be examined. For uncontrolled observational studies, details will be sought on population characteristics (including representativeness and eligibility criteria), intervention and adequacy of assessment of relevant outcomes and of follow-up (e.g. at least 30 days for survival (civilian population) or up to point of discharge to other healthcare service (military population)). In addition to the methodological criteria listed above, the GRADE \[[@B27]\] framework may be used to consider inconsistency between studies, precision of results, likelihood of publication bias and applicability of results to population(s) of interest. Analysis -------- Narrative synthesis of evidence will be undertaken for all included studies. This will include a narrative description and tabulation of main results across studies and/or for specific outcomes. Visual representation of results in Forest plots without pooling may also be considered. Summary measures for the main outcome of interest, survival, may be in the form of relative risk or hazard ratio (adjusted or unadjusted). For other outcomes (e.g. lactate concentration), mean differences (adjusted or unadjusted) may be reported. Based on preliminary findings, it is unlikely that there will be sufficient (similar) studies to conduct meta-analysis; however, appropriate meta-analytic methods will be employed where possible to combine data from similar studies. Assessment of clinical and methodological heterogeneity will be used to determine whether a fixed or random effects model is the most appropriate, rather than relying on the tests of heterogeneity from a fixed effect model to make such a decision \[[@B28]\]. The *I*^2^ statistic (which gives the percentage of the total variability in the data due to between-study heterogeneity) and the tau-squared statistic (which gives an estimate of the between-study variance) will be reported where appropriate. Where studies have reported time-to-event analyses, meta-analysis using the extracted hazard ratios and their variances will be undertaken, if possible. Evidence from studies of different design will not be quantitatively combined, but presented separately. Any adjusted and unadjusted results will also be presented separately. Presentation of results in Forest plots without a pooled summary estimate will be considered where pooling is not feasible. For each meta-analysis containing 10 or more studies, the likelihood of publication bias will be investigated through the construction of funnel plots and appropriate statistical tests for small-study effects (such as the Peters Test \[[@B29]\]); that is, the tendency for smaller studies to provide more positive findings. It is well recognised that, especially where heterogeneity exists, publication bias may be one of a number of reasons for any small-study effects identified. Subgroup analysis ----------------- Subgroup analysis may include examining studies separately depending on pre-hospital setting (military or civilian) or where there is clear clinical heterogeneity between studies (e.g. in intervention/comparator, patient characteristics etc.). The robustness of any meta-analysis conclusions to the inclusion/exclusion of low quality studies (i.e. those at most risk of bias) may be assessed if feasible. Discussion ========== Emergency services, both in the UK and internationally, are already increasingly incorporating the use of blood components into pre-hospital resuscitation protocols for major traumatic haemorrhage. As far as the authors are aware, however, there have been no attempts to formally review the evidence base, and a systematic review is therefore urgently required. Whilst some studies appear to show benefits of pre-hospital blood components, there are methodological issues associated with many studies, which are mainly of a retrospective observational design. A thorough evaluation of study methodology and risk of bias as part of the proposed systematic review will not only help to make an assessment of the robustness of findings but may also help to inform future study design. Further, an assessment of the extent of transferability of findings from military research to a civilian population will be of interest, both for this systematic review question and potentially for other areas of trauma medicine. Abbreviations ============= HR: haemostatic resuscitation; pRBC: packed red blood cells; RBC: red blood cells. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== MM is the guarantor. JD and IS drafted the manuscript and developed the search and methodological strategy. MM, IS and RJ provided clinical expertise. All authors read, provided feedback and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Sample search strategy.** Sample search strategy to identify relevant primary studies in MEDLINE. ###### Click here for file Acknowledgements ================ This project is funded by the NIHR Surgical Reconstruction & Microbiology Research Centre (<http://www.srmrc.nihr.ac.uk>). The National Institute for Health Research Surgical Reconstruction and Microbiology Research Centre (NIHR SRMRC) is a partnership between The National Institute for Health Research, University Hospitals Birmingham NHS Foundation Trust, the University of Birmingham and the Royal Centre for Defence Medicine. Disclaimer ---------- This article presents independent research funded by the National Institute for Health Research (NIHR). The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, the Department of Health, the University of Birmingham or the Ministry of Defence.

Background {#Sec1} ========== The success rates of endodontic surgery, including root-end resection, are extremely variable. According to the literature, the range spans from 17 to 96% \[[@CR1]--[@CR3]\]. This extreme variability can be put down to the technique-sensitivity of these surgical interventions. However, the results of a meta-analysis suggested that the use of high-power magnification alone can elevate the success rate of endodontic microsurgical interventions to as high as 94% \[[@CR1]\]. Similarly, ultrasonic retrograde preparation with modern root-end filling materials, such as MTA (mineral trioxide aggregate) and bioceramics is superior to shallow cavity preparation with adhesive materials in terms of healing \[[@CR4]\]. All these innovations have been introduced to endodontic surgery in the last few decades, allowing more predictable outcomes. However, the operator factor remains unresolved, and this is a considerable source of error both in root-end resection and the osteotomy that precedes it. The challenge here is to direct the osteotomy in a way that allows the removal of the desired section as accurately as possible \[[@CR5]\] - a goal that is extremely difficult to achieve solely via mental navigation. For this reason, cone-beam computed tomography (CBCT) is considered to be essential before periapical surgical procedures \[[@CR6]\]. Indeed, CBCT is a great help, but the accuracy of the procedure still depends on how accurately the surgeon can mentally register the three-dimensional image with the actual clinical appearance of the corresponding structures. This leaves plenty of room for error, and the profession has been on the search for further and better navigation aids for some time. In 2007, Pinsky et al. were the first to report on the computer-assisted design and manufacturing of surgical templates for endodontic application \[[@CR7]\]. Comparing the guided approach to freehand surgery, they found the former to be significantly superior. The recent years have seen a renewed interest in surgical guides (templates) for endodontic surgery, possibly because stereolithographic manufacturing (i.e. 3D printing) has become widely available and development in this direction has become a real possibility \[[@CR8]--[@CR10]\]. Patel et al., in a case report, described the use of a 3D printed custom retractor for endodontic surgery \[[@CR11]\]. Strbac and colleagues published a case report, where a stereolithographically fabricated surgical template was used to help the osteotomy and the root resections \[[@CR12]\]. In these cases, though, the templates were not used to guide the osteotomy itself, as is usual in dental implantology. It is becoming widely accepted that the placement of dental implants is more predictable and accurate when using 3D printed surgical templates \[[@CR13]\]. Studies have shown that implant placement through a guide allowed a more accurate implementation of the virtual plan to the surgical site than freehand insertion \[[@CR14]\]. Based on the results of their randomized controlled trial, Younes and colleagues suggest that fully guided surgery (i.e. all osteotomies and implant placement through guide) should be considered the gold standard approach instead of freehand surgery in dental implantation \[[@CR15]\]. The application of this idea to endodontic surgery was, in fact, a logical next step. The first case that can be considered a truly guided endodontic surgery case was described in 2018, by Giacomino et al. \[[@CR16]\]. The authors concluded that guided endodontic microsurgery is useful for osteotomy and root-end resection when precise control of depth, diameter, and angulation of osteotomy are necessary. Further case reports followed. Ye et al. operated a left maxillary lateral incisor and canine with a 3D printed model to help the localization of the apices \[[@CR17]\]. Ahn et al. used a surgical template to localize the apices in a mandibular molar with a thick buccal bone plate \[[@CR18]\]. Interestingly, the authors used the template according to what implantology would call "the pilot protocol"- that is, only the initial osteotomy ("pilot" osteotomy) was performed with the help of the template, the rest of the procedure was done without it. All in all, template-based guided surgery is becoming recognized as an option for endodontic surgery, but still, there is a relative scarcity of studies on this subject. To test the validity of these observations, we carried out a prospective exploratory study in 2018--2019 in 11 patients. In the study, we resected 14 root ends with our template-and-trephine method, utilizing tooth-supported, stereolithographically fabricated surgical templates. The aim of the study was to give an approximation of the clinical safety and accuracy of this method. We hypothesized that a) intra- and postoperative complications would be no more frequent and severe than what is usual in freehand cases; b) the method would allow the resection of the root with the trephine in all cases, so no further manipulation to this end would be necessary; c) by utilizing this method, the vertical error of root-end resection and the error of osteotomy depth would not be greater than ±1 mm; and c) the angular accuracy of the osteotomies would be close to that of template-guided dental implantation. Materials and methods {#Sec2} ===================== Patients {#Sec3} -------- Eleven patients were enrolled (mean age: 48.9 ± 12.4 years). Seven of these patients were women (mean age: 45.4 ± 11.8 years), and four were men (mean age: 55.0 ± 11.0 years). The demographic and baseline clinical characteristics of the study population are given in Table [1](#Tab1){ref-type="table"}. Lesion sizes - to give an approximation of the severity of the periapical process - were calculated as recommended by Kim et al. \[[@CR19]\]: the mesiodistal (width), buccolingual (depth) and apicocoronal (height) dimensions were measured. All patients were referred for endodontic surgery by general dental practitioners to the Department of Operative and Esthetic Dentistry, Faculty of Dentistry, University of Szeged (Szeged, Hungary). Only front teeth were included with healthy periodontium and restored clinical crown. The inclusion criteria were persisting periapical lesion and pain with or without swelling, impossible or previously failed root canal revision, age between 18 and 75 years, and signed informed consent. Relative and absolute contraindications of endodontic surgery counted as exclusion criteria, as well as any other condition that would have put the patient at unacceptable risk during or after surgery. Patients with non-restorable clinical crown or damaged periodontium were also excluded. The study conformed to the Declaration of Helsinki "Ethical Principles for Medical Research Involving 'Human Subjects", adopted by the 18th World Medical Assembly, Helsinki, Finland, June 1964, as amended by the 64th World Medical Assembly, Fortaleza, Brazil, October 2013. Furthermore, the study observed the principles of Good Clinical Practice. The protocol was approved by the National Institute of Pharmacy and Nutrition of Hungary (Approval No. OGYÉI/43796--6/2018). Table 1Demographic and baseline clinical characteristics of the study populationCasePatientSexAgeToothLesion size (mm) width x height x depthSwellingFistula11F29223.42 × 2.74 × 3.13--+22F32123.06 × 3.58 × 2.94----33F48113.02 × 2.83 × 5.24+--43F48216.37 × 6.46 × 5.47+--54M401116.33 × 12.48 × 10.08--+65M49115.13 × 4.46 × 4.18----76F47123.30 × 4.66 × 3.23--+87F52224.51 × 3.23 × 4.18----98M64444.91 × 7.61 × 5.88+--109M67342.40 × 4.30 × 2.42+--1110F43143.60 × 3.90 × 4.51+--1211F67114.37 × 2.43 × 5.54----1311F67223.69 × 4.59 × 3.18----Thirteen teeth were treated in 11 patients, resulting in altogether 14 root end removals. The lesion sizes were calculated utilizing CBCT scans, as proposed by Kim et al. \[[@CR16]\]: the maximum diameter of the lesions was measured in 3 directions parallel to the standardized axes: mesiodistal (L~x~), apico-coronal (Ly) and buccolingual (Lz) Presurgical procedures {#Sec4} ---------------------- Cone beam computed tomographies were acquired (iCAT Next Generation, Imaging Sciences-Kavo, Hatfield, PA, USA) with standard settings for all patients (120 kV, 5 mA, 9 s, voxel size: 250 μm, FOV: 110 mm; all scans in this study were done with these specifications). A bite block was used to ensure non-occlusion and a correct head position. A silicone impression (Zetaplus, Zhermack, Italy) was taken in a plastic tray (Hi-Tray, Zhermack, Italy), and it was scanned separately. The acquisition was always performed by an experienced radiologist according to the recommendations of the guide manufacturer (dicomLAB Dental, Ltd., Szeged, Hungary), with the minimum exposure necessary for adequate image quality \[[@CR20]\]. The images were reconstructed as a volume (i-Cat Vision, Imaging Sciences International, Hatfield, PA, USA), and saved as DICOM files to provide input for surgical planning. The two scans were sent online to the template manufacturer, where they were registered, and sent back to the surgeon for planning. For 3D surgical planning, SMARTGuide 1.25 (dicomLAB Dental, Ltd., Szeged, Hungary) was used. For the planning of the surgeries, a virtual cylinder of the same dimensions as the actual trephine was used (Fig. [1](#Fig1){ref-type="fig"}). The only difference between the model and the trephine was that the model was rounded at the distal end, but this did not confound apical deviation calculations, as the axial lengths were the same, and for the calculations, two properly aligned models were compared (see below). This cylindrical model was positioned in a way that its axis was perpendicular to the tooth axis. The planned drilling length was 20 mm from the outer margin of the guiding sleeve in all cases. The surgical plans were prepared with the intention to resect 3 mm of the apical portion of the root. In cases with previous apicoectomy in history, only 1.3 mm was planned to be resected, always keeping in mind that enough root surface should be left to provide sufficient retention. All planning was performed by the same experienced surgeon, familiar with both implant and endodontic surgeries. The surgical templates were fabricated according to these plans, using a 3D printer (3D Systems ProJet MD 3510, USA). As a final step, to enhance the fit of the trephine in the guide, metal guiding sleeves of an inner diameter of 4.25 mm were inserted into the guiding tunnels of the templates. This diameter was wide enough to allow the rotation of the trephine of 4.21 mm outer diameter (see below) but narrow enough to allow only negligible lateral deviation. After fabrication, the templates were tried on the patients' plaster casts to check correct and reproducible fitting (Fig. [2](#Fig2){ref-type="fig"}). A final check was performed right before each surgery, on the patients' dentition. The insertion of the templates into the patients' oral cavity was always preceded by disinfection, as per the manufacturer's instructions. Fig. 1Surgical plan in the planning software (orovestibular view). **a** guiding sleeve; **b** virtual model to represent trephine; **c** the angulation of the planned osteotomy; **d** the planned depth of the osteotomy; **e** the planned length of the piece to be resected Fig. 2Left: the surgical setup demonstrated on a gypsum cast. **a** surgical template **b** guiding tunnel with metal sleeve; **c** trephine. Right: intraoperative image Surgical procedure {#Sec5} ------------------ The surgeries were performed under local anesthesia (Ultracain D-S Forte 1:100000; Sanofi-Aventis GmbH). Full-thickness flaps were raised, the size and shape were always determined by the anatomical properties and accessibility of the current case. As the surgical guide was placed, it defined the exact osteotomy site and the angle at which the osteotomy would be performed. For the osteotomy, a bone trephine was used with an outer diameter of 4.21 mm (Hager & Meisinger, Neuss, Germany) under copious irrigation. The trephine was applied through the guiding sleeve of the template (Fig. [2](#Fig2){ref-type="fig"}). After the combined osteotomy and apicoectomy, periapical curettage was performed if necessary, and retrograde preparation was performed using a piezosurgery unit (Piezomed, W&H, Austria). For the retrograde filling, MTA was used (Cerkamed, Poland). We applied methylene blue to visualize the ramifications. Before retrograde fillings, ferric-sulfate was used to ensure a bloodless working area. The retrograde preparation and filling were performed under high-power magnification (Opmi Pico, Zeiss, Germany). The flaps were closed and sutured with 5.0 monofilament sutures (Ethicon, USA). Within a month after the surgeries, a follow-up CBCT scan was made, with the same unit and settings as before the surgery. The sutures were removed 7 ± 1 days following the surgery, and follow-ups were scheduled at 6 months and 12 months. Analyses {#Sec6} -------- The frequency and severity of intraoperative and postoperative complications were recorded, as well as the frequency of osteotomies when the root end was resected in the same step (i.e. no further manipulation was necessary for this purpose). Frequencies and percentages were calculated. The angular deviation was analyzed in Amira 5.4.0 (Thermo Fisher Scientific, USA) with dedicated algorithms. Pre- and postoperative CBCT scans of the given patient were transformed into the same coordinate system. For this registration, the region of interest was narrowed down to the analyzed bone (i.e. maxilla or mandible) to avoid inaccuracy stemming from differences in mouth opening. The bony tunnel formed by the trephine was manually segmented via a slice-by-slice method and transformed into a three-dimensional virtual model. As a next step, the cylindrical model used for planning was aligned with the model of the actual tunnel along their principal axes. The corresponding surgical plan was then extracted from the database of the planning and manufacturing system of the surgical guide and applied to a copy of the same cylindrical model. This way, it became possible to compare the result of the osteotomy with the plan in terms of the deviation of the principal axes (Fig. [3](#Fig3){ref-type="fig"}). The angular deviation was defined as the angle closed by the principal axes of the aligned models in degrees. The procedure was repeated three times for each case, and the mean of the three measurements was used for further analyses. The results were calculated as median (95% CI) as we found it more meaningful and informative in such a small sample than the usual mean (SD). Fig. 3Analysis of angular deviation in Amira (blue: planned, red: realized). This figure does not depict the analysis of any of the actual cases, it is for illustration purposes only The length of the given tooth was measured in both the preoperative and postoperative CBCT images (i-CAT Vision, i-CAT, USA). This allowed the calculation of the length of the resected piece, which was subtracted from the planned length to be resected and so apex resection error (ARE) was calculated. Osteotomy depth error (ODE) was calculated similarly, by subtracting the actual depth from the planned depth (Fig. [4](#Fig4){ref-type="fig"}). The calculations were performed three times for each case, and the mean of the three measurements was used for further analyses. The results were calculated as median (95% CI), for the same reasons as given above. Fig. 4Explanation of the 2D measurements. Left: preoperative, Right: postoperative; **a**: coronal reference point; **b**: apical reference point (end points of the axis); **c**: axial length before surgery **d**: axial length after surgery; **e**: planned length of removal; **f**: actual resected length; **g**: planned depth of osteotomy; **h**: actual depth of osteotomy (for the measurements, the missing cortical was substituted by a straight line connecting the remaining cortical edges). Calculations: ARE = **e-f**; ODE = **g-h** All statistical calculations were done in SPSS 23.0 (IBM, USA). Results {#Sec7} ======= Thirteen teeth were treated in 11 patients, resulting in altogether 14 apicoectomies. The root end was successfully and completely resected by the trephine in all cases, and in 11 cases, the resected piece was also removed with the trephine (Fig. [5](#Fig5){ref-type="fig"}). In 3 cases, the root end was removed with a periotome. No intraoperative complications were observed in any of the cases. In all cases, the surgery resolved the preoperative swelling and pain, and the patients were free of symptoms indicating recurrence or complications at the 6-month follow-up. In two of the cases, a key-hole extension of the trephine tunnel had to be performed in order to allow enough vertical space for the piezo tip for the retrograde preparation. In three cases, due to the extensive lesion and excochleation during the operation, the digital segmentation of the cavity was not possible as the borders could not be properly defined. Accordingly, in these cases, the angular deviation was not calculated. Some postoperative radiographs showed overpenetration (Fig. [6](#Fig6){ref-type="fig"}). The median angular deviation was 3.95° (95% CI: 2.1--5.9) (Table [2](#Tab2){ref-type="table"}). The median apex removal error in the vertical plane (ARE) was 0.19 mm (95% CI: 0.03--0.07). The highest overcut was 0.93 mm, and the shortest cut fell behind the plan by 0.94 mm. In one case, exactly the planned length was cut. In 10 cases (71.4%), a longer piece of the apex was cut than planned, by a median of 0.37 mm (95% CI: 0.06--0.76). In 4 cases, the resected piece was shorter, by a median of 0.19 mm (*n* \< 5, 95% CI not possible). The median osteotomy depth error (ODE) was 0.37 mm (95% CI: 0.15--1.35). Of the 13 remaining depth values, 9 (69.2%) indicated shallower osteotomy than planned by a mean of 0.71 mm, while the rest of the osteotomies were deeper than planned, by a mean of 0.31 mm. The highest overpenetration was 0.51 mm, while the shallowest penetration fell behind the plan by 1.56 mm. Fig. 5Bone cylinders removed with the trephine containing the resected root ends Fig. 6Overpenetration: note the trephine markings in the palatinal cortical (arrows) Table 2Results of the measurementsCaseToothADAREODE122NA−.033+ 0.292124.0−0.08+ 1.353116.3−0.70+ 0.134212.1− 0.76− 0.15511NA− 0.29+ 0.566113.9−0.06− 0.067125.3−0.03+ 1.438224.0+ 0.15+ 0.57944NA−0.41+ 1.5610342.40.00−0.531114^a^1.5−0.93+ 0.231114^a^1.5−0.15+ 0.2312112.9+ 0.94+ 0.2313225.9+ 0.23−0.51median3.950.190.3795%CI2.10--5.900.03--0.700.15--1.35min-max2.10--6.30−0.93-0.94−0.51-1.56*AD* angular deviation (degrees), *ARE* apex resection error (mm; -: longer than planned; +: shorter than planned), *ODE* osteotomy depth error (mm; -: deeper than planned; +: shallower than planned). In Case \#11, two roots of the same tooth were treated. NA: in these cases, the given parameter could not be measured (see Results). ^a^Two roots of the same tooth were treated. Medians and corresponding confidence intervals were calculated from absolute values to express the degree of error regardless of its direction Discussion {#Sec8} ========== Research in endodontic microsurgery unequivocally suggests that modern microsurgical approaches yield much higher success rates than traditional ones \[[@CR21], [@CR22]\]. Despite this, relatively little has been said in the literature about the use of such modern methods for root end localization and resection. The few available studies agree that guided root-end resection is efficient and more accurate than freehand surgery \[[@CR23], [@CR24]\]. To our knowledge, we are the first to report a series of clinical cases where osteotomy and root-end resection were carried out at the same time, with the same instrument (a bone trephine), using 3D printed surgical guides. Our first hypothesis was that intra- and postoperative complications would be no more frequent and severe with the studied method than what is usual in freehand cases as reported by the literature and as shown by our own clinical experiencee. As we observed no complications at all during surgeries, we consider this hypothesis confirmed. The second hypothesis was that the method would allow the resection of the root with the trephine in all cases, so no further manipulation to this end would be necessary. This hypothesis was also confirmed; the root end was successfully resected in all cases, and in most of the cases it was also removed with the trephine. This is practically important because this way the root end resection and removal procedure can be carried out in one simple step. Our third hypothesis was that by utilizing this method, the vertical error of root-end resection and the error of osteotomy depth would not be greater than ±1 mm. This hypothesis was only partially confirmed, regarding the error of root end resection, which remained within 1 mm in both directions in the vertical plane (− 0.93 mm to + 0.94 mm). This indicates that the guidance was quite efficient (as reflected also by the angular deviation measurements, see below). The error of the depth of the osteotomy, however, exceeded the ±1 mm limit in three instances, which indicates that the method was less accurate in the horizontal dimension. These deviations indicated underpenetration (the fact that this did not affect the success of resection indicates that the planned depth was excessive in these cases). At the same time, overpenetration was also a recurrent finding, even if within the ±1 mm limit (Fig. [6](#Fig6){ref-type="fig"}). What these results suggest in general is that repeated depth check with a periodontal probe and observing the markings on the trephine are not enough to make the surgeon confident about this dimension. We propose that trephines with a stop (like implant drill bits) could resolve this problem. Our last hypothesis was that the angular accuracy of the osteotomies would be close to that of template-guided dental implantation. We formulated this hypothesis like this because ours is the first study to assess this parameter for guided trephination, so we had nothing else to compare our results against. At the same time, this is a very important parameter, as it can determine if the root-end resection is successful (i.e. if a large enough part is resected to get rid of all the accessory canals, the very aim of the procedure). This last hypothesis was also confirmed. Tahmaseb et al., in their meta-analysis on guided implantation with tooth-supported guides, reported an overall angular deviation of 3.5° (studies of full and partial edentulousness included) \[[@CR25]\]. Endodontic surgical guides cannot be considered entirely tooth-supported guides, Of course, it is the teeth that the template rests on, but the direction of the operation is not perpendicular to the occlusal plane, which would help to keep the guide in place. Rather, the operation happens perpendicular to the soft tissue, which adds instability to the system. The poorer accuracy of mucosa-supported guides is a known problem in guided implantology \[[@CR26]\]. Following from these, one would expect the angular deviation to be slightly poorer than but still comparable to that of implant guides. The median deviation of 3.95° we found confirms that expectation. Beyond addressing our hypotheses, we would like to point out a practical difficulty we often faced. Obviously, to have the guiding sleeve at the level of the apex, the impression (that serves as the model for the template) must contain information on the tissues at that level. Otherwise the planned sleeve falls outside the template, and the guide cannot be produced. In other words, the achievable depth of the impression is a limitation of digital planning, and this is probably true for digital impressions as well. Although we used orthodontic impression trays, in some cases the impressions had to be retaken as they were not deep enough. Sometimes this did not help either. In those cases, a compromise had to be made, and the trephination was planned not exactly at the originally intended 90°. That is, the planned trephination path was not exactly perpendicular to the axis of the tooth. Naturally, such a compromise is allowable only if it does not risk the aim of the surgery, and the decision requires careful consideration. Another aspect of the same problem is sometimes the patient's lip had to be retracted almost to an extreme degree to allow access with the trephine through the guide in the proper direction. We find that our study offers valuable insight into the accuracy and clinical characteristics of the studied method, but we must mention two limitations that have to be kept in mind when interpreting the results. First, as our study is the first of its kind, no comparison with the literature is possible. While there is a multitude of accuracy studies for guided dental implant surgery, the interest in guided root-end resection is quite new, and research into this topic is in the exploratory phase. Still, we consider it important to share our experience, as the literature shows that the method is becoming widespread. Our data indicate that the method is safe and accurate enough for the purpose. Second, while there is an established methodology for the assessment of accuracy of dental implantation, including the assessment of coronal and apical deviations and various other measures, there is no such recognized methodology for the assessment of trephination. Therefore, we had no option but to choose our measures ad hoc. Of the methods of measurement we had access to, we chose the ones we considered to be the clinically most relevant. This is, however, not to say that the measures presented here are the best to characterize trephination accuracy. Conclusions {#Sec9} =========== We conclude that our results support the use of guided trephination for root-end resection. However, the research is in an early stage, and there is ample room for the improvement of both the method and its assessment. ARE : Apex removal error CBCT : Cone Beam Computed Tomography CI : Confidence Interval DICOM : Digital Imaging and Communication in Medicine FOV : Field of view MTA : Mineral trioxide aggregate ODE : Osteotomy depth error **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. N/A MA contributed to conception, design and interpretation, drafted and critically revised the manuscript and performed the surgeries. EN contributed to conception, design and interpretation, drafted and critically revised the manuscript. GB contributed to conception, design and interpretation, drafted and critically revised the manuscript and performed calculations. MF contributed to conception, design and interpretation, drafted and critically revised the manuscript. JP contributed to conception, design and interpretation, drafted and critically revised the manuscript and supervised the work. All authors gave final approval and agree to be accountable for all aspects of the work. Authors' information {#FPar1} ==================== MA has specialization in oral surgery and also endodontics. All operations were performed by MA. EN is clinical dentist and MF is head of the Department of Operative and Esthetic Dentistry at the University of Szeged. GB is chief researcher at dicomLAB Dental Ltd., Szeged, Hungary. JP is head of Department for Maxillofacial Surgery at the University of Szeged. N/A The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. This study was approved by the National Institute of Pharmacy and Nutrition of Hungary (Approval No. OGYÉI/43796--6/2018) and followed the Declaration of Helsinki on medical protocols and ethics. Written informed consent was obtained from all patient. Written informed consents were obtained from the patients for publication of the technical notes and images. Dr. Gábor Braunitzer is chief researcher at dicomLAB, Ltd.

In [Figure 1](#pntd-0003340-g001){ref-type="fig"}, the PCR negative carnivores are omitted. Please see the corrected [Figure 1](#pntd-0003340-g001){ref-type="fig"} here. {#pntd-0003340-g001}

Introduction {#Sec1} ============ Since Hans Selye's definition of stress as "the nonspecific response of the body to any demand",^[@CR1]^ many studies have revealed the negative influence of an overload of stress on health and wellbeing. Observational data suggest associations between psychological stress and depression, cardiovascular disease, sudden death, and myocardial infarction.^[@CR2],[@CR3]^ Early detection and prevention of the adverse consequences of stress are therefore of utmost importance, and require personalized prevention and treatment strategies that take individual variability into account, as is suggested in the Precision Medicine Initiative.^[@CR4]^ Towards precision medicine, digital phenotypes are a new paradigm to extend our assessment of human illness beyond traditional examinations.^[@CR5]^ They represent a subject's interactions with digital technologies such as connected health devices and smartphones to generate longitudinal, individual health profiles. Leveraging data-driven approaches, these data can fundamentally change our understanding of disease prognoses and provide new insights towards disease prevention and early detection.^[@CR5]^ The most widely used method and current gold-standard to assess stress is by means of questionnaires, e.g., the Perceived Stress Scale (PSS).^[@CR6]^ However, these questionnaires are qualitative, time-consuming, and reflect responses collected during spot-checks only. This limits the accurate monitoring of stress and the use of just-in-time interventions to reduce stress. Therefore, research has focused on finding continuous and quantitative physiological markers of stress,^[@CR7]^ by exploiting measurable functioning of the sympathetic nervous system's fight-or-flight response,^[@CR8]^ such as skin conductance (SC), the electrocardiogram (ECG), the electromyogram (EMG), blood pressure (BP), and skin temperature (ST).^[@CR9]^ These have shown to be reliable indicators of stress in laboratory conditions.^[@CR9]^ In recent years, the growing availability of wearable sensors has led to increased research towards continuous, ambulatory monitoring of stress. So far mainly small-scale studies of 20--50 subjects have been conducted.^[@CR10]--[@CR12]^ Multiple findings suggest that physiological responses to stress tend to be person-dependent.^[@CR13],[@CR14]^ Therefore, large datasets are essential to grasp subject-to-subject variability and develop personalized risk profiles. Furthermore, in the majority of ambulatory trials, subjects' demographics and psychological baseline profiles (e.g., self-reported anxiety and depression levels) are either overlooked or not assessed.^[@CR10],[@CR11]^ Context information is often not measured,^[@CR10],[@CR12]^ although it can also provide actionable insights.^[@CR7],[@CR11]^ Here we present the SWEET study (Stress in the Work EnvironmEnT): a comprehensive, cross-sectional study on an office workers' population of 1002 healthy volunteers, who were monitored continuously for 5 consecutive days. We collected baseline psychological information, 5 consecutive days of free-living physiological data through wearables and smartphone-based contextual measurements, self-reported stress through ecological momentary assessments (EMAs) and physiological responses to an application-based stress test. We show strong associations between physiology, contextual information, and behavior, highlighting the benefit of multi-sensor information to improve our understanding of stress in daily life. We show that physiological signals differ significantly according to reported stress levels and identified stress digital phenotypes, characterized by self-reported poor health and high depression, anxiety and stress scores, that are associated with blunted physiological stress responses. Building on preventive health, these findings and this comprehensive dataset of physiology, context, and stress in the daily life, allow for multi-sensor models to detect stress in daily life. Results {#Sec2} ======= Dataset quality and compliance {#Sec3} ------------------------------ A comprehensive summary of the intake questionnaire, including subject demographics and psychological baseline information is available in [S1 Table](#MOESM1){ref-type="media"}. The measurements lasted 5 days per subject, starting on Thursday morning and ending on Monday evening. An overview of self-reported stress responses through EMAs and compliance is available in [S2 Table](#MOESM1){ref-type="media"}. Subjects provided on average 25 self-reported stress responses, with highly imbalanced data, containing 50.4% no stress and only 0.3% extremely high stress, reported on a five-point Likert scale. Therefore, the three highest stress levels were merged, representing 14.3% of the data, so that three, instead of five, levels of stress (S1 = no stress, S2 = light stress, S3 = high stress) were considered. An overview of smartphone and wearable sensor data quality is available in [S3 Table](#MOESM1){ref-type="media"} and [S4 Table](#MOESM1){ref-type="media"}, respectively. High quality physiological data (the chest patch, measuring ECG and acceleration (ACC), see Methods, had on average 86.4 ± 8.2% good quality data; the Chillband, measuring SC, ST, and ACC, had on average 96.3 ± 2.2% good quality data), was complemented with smartphone-based context information such as location and audio features (see Methods). Using high quality physiological signals, 18 features were calculated (6 ECG features, 8 SC features, and 4 ST features) in a window of 5 min, with 4 min overlap. An overview can be found in material and methods and [S5 Table](#MOESM1){ref-type="media"}. Further, we compared questionnaire-based lifestyle (e.g., practicing sports, smoking habits) and health indicators (e.g., sleep quality, depression levels). Our findings represent a large-scale verification supporting previous work and confirm the value of data sampled with validated questionnaires ([S1 Fig](#MOESM1){ref-type="media"}. and [S1 Text](#MOESM1){ref-type="media"}). Associations between physiology, context, and behavior {#Sec4} ------------------------------------------------------ We investigated correlations between physiology, context, and behavior in order to improve our understanding of stress in daily life. Through EMAs we daily asked questions related to stress, activity, food and beverage consumption, sleep quality, and gastro-intestinal symptoms. Based on fixed day and night intervals, circadian rhythms of the physiological signals can be identified, namely lower average HR during the night and higher SC and ST (mean HR~day~: 74.6 ± 12.7, mean HR~night~: 63.0 ± 10.2, Wilcoxon ranksum *p* \< 0.001; mean SC~day~: 1.7 ± 2.7, mean SC~night~: 2.8 ± 3.4, Wilcoxon ranksum *p* \< 0.001; mean ST~day~: 31.4 ± 2.1, mean ST~night~: 33.1 ± 2.6, Wilcoxon ranksum *p* \< 0.001, day = 06 am--23:59 pm, night = 00--06 am). During weekdays (i.e., Thursday, Friday, and Monday) consumption of caffeinated beverages or breakfast corresponded to higher stress levels (caffeine: 1.84 ± 0.81, breakfast: 1.87 ± 0.81, average: 1.77 ± 0.83, Wilcoxon ranksum *p* \< 0.001), while dinner or alcohol consumption, corresponded to lower stress levels (dinner: 1.51 ± 0.71, alcohol: 1.30 ± 0.64, average: 1.77 ± 0.83, Wilcoxon ranksum *p* \< 0.001). During the weekend (i.e., Saturday and Sunday), the consumption of alcohol was associated with lower stress levels (alcohol: 1.34 ± 0.66, average: 1.45 ± 0.71, Wilcoxon ranksum *p* = 0.001), other reported consumptions did not show significant differences. A possible confounder here could be time of the day since breakfast is consumed most in the morning (82% of reports between 6 and 10 h), and alcohol and dinner most in the evening (alcohol: 61% of reports between 18 and 22 h, dinner: 65% of reports between 18 and 22 h), caffeine was reported equally throughout the day, but less during the evening (32% of reports between 6 and 10 h, 37% between 10 and 14 h, 23% between 14 and 18 h, and 7% between 18 and 22 h). Further, linear mixed effects models were computed to investigate associations between repeated measures. A significant negative association between self-reported stress and self-reported pleasure (based on the Self Assessment Manikin (SAM)) was observed, with higher levels of self-reported stress corresponding to decreasing levels of pleasure (Table [1](#Tab1){ref-type="table"}). It can be speculated that rating of high self-reported stress is likely associated with the feeling of distress (negative stress) rather than eustress (positive stress). The standard deviation of the magnitude of acceleration (ACC SD), was associated with intensity of movement as ACC SD was higher during self-reported high-intensity activities (low-intensity, i.e., lying, sitting, and standing: 0.0175 ± 0.0089, high-intensity, i.e., walking, running, biking, driving car, and other activities: 0.0189 ± 0.0096; Wilcoxon ranksum *p* \< 0.001) and HR and SC features increased with ACC SD (Table [1](#Tab1){ref-type="table"}) while ST decreased with ACC SD (Table [1](#Tab1){ref-type="table"}). This illustrates that physical activity is associated with changes in physiology, highlighting the challenge of differentiating physiological changes caused by physical activity from those caused by stress.^[@CR12],[@CR15]^ Therefore, to account for the confounding effect of physical activity on physiology and stress, we excluded segments of high activity in the subsequent analysis. Finally, increasing activity levels (ACC SD), decreased the quality of physiological signals (Table [1](#Tab1){ref-type="table"}), an issue inherent to the free-living nature of the study.Table 1Effects of repeated measuresFormulaB ± SEInferential statisticsTest statistic and df*p*-ValueSignificanceStress\~Pleasure + (1 \| subject)−0.22 ± 0.01*X*^2^(2,3) = 1353.5\<0.001\*SC mean\~ACC SD + (1 \| subject)0.93 ± 0.02*X*^2^(3,4) = 2448.6\<0.001\*HR mean\~ACC SD + (1 \| subject)12.04 ± 0.02*X*^2^(3,4) = 488,051\<0.001\*ST mean\~ACC SD + (1 \| subject)−5.61 ± 0.02*X*^2^(3,4) = 87,378\<0.001\*SC Quality\~ACC SD + (1 \| subject)−0.23 ± 0.0008*X*^2^(3,4) = 90,652\<0.001\*ECG Quality\~ACC SD + (1 \| subject)−0.33 ± 0.0009*X*^2^(3,4) = 115,642\<0.001\*Results of the linear mixed effects models for repeated measures. For each model, the fixed effect coefficient is presented with standard error (B ± SE) and inferential statistics on the significance of the effect, which were calculated by testing the change in model performance (based on Akaike's information criterion) when a given predictor (e.g., pleasure) was excluded from the model using an ANOVA test Five days of measurements of physiological data and acceleration are shown in Fig. [1a](#Fig1){ref-type="fig"}, indicating daily variations in a healthy population. Behavior and mood self-annotations captured through EMA questionnaires and smart-phone sensor data are shown for a single representative subject in Fig. [1b](#Fig1){ref-type="fig"}.Fig. 1Physiology and context timeline. **a** Healthy-population physiological data over 5 days of measurements depicting smoothed daily profiles of high quality (Quality \> 0.8) physiological signals (mean HR, SC, and ST) and activity (ACC SD), averaged in 1 min windows. **b** Self-reported annotations (stress, pleasure, activity, consumptions, and wake up/bed times) and location, as indicated with vertical lines when available for a representative subject. Location data are indicated as unique stay locations or commuting locations. An online version of **a** and **b** can be downloaded from <https://drive.google.com/open?id=1Z1q0YLG8cvUllSM84X3mgwmCCBjfO0M7> Associations between physiological signals and self-reported stress levels {#Sec5} -------------------------------------------------------------------------- We aim to use physiological patterns to develop models for psychophysiological stress detection. Therefore, we show here a variety of associations between physiological features and self-reported stress levels, in a healthy population. An overview of all physiological features calculated is presented in [S5 Table](#MOESM1){ref-type="media"}. Mean differences of uncorrelated physiological features (*r* *\<* 0.7), normalized per subject, across self-reported daytime stress levels (S1, S2, and S3) and nighttime (00--06 am, N), included as a baseline rest condition, are shown in Fig. [2](#Fig2){ref-type="fig"}. Population variations (averages and 95% CI) of physiological features across self-reported stress levels and nighttime are presented in Table [2](#Tab2){ref-type="table"}, an extended version, including all physiological features, can be found in [S6 Table](#MOESM1){ref-type="media"}.Fig. 2Associations between physiological features and self-reported stress levels. Each row represents a physiological feature, columns represent the difference of the median of normalized features during the night (N) (00--06 am) and stress levels (S1, S2, and S3). Colors indicate positive (blue) or negative (red) differences. For example, SC phasic is significantly higher (blue) during the night as compared to during all reported stress levels, and significantly lower (red) during S1 as compared to S2 and S3. Symbols: \**p* \< 0.05, \*\**p* \< 0.005, \*\*\**p* \< 0.0005Table 2Physiological features across self-reported stress levelsNight (mean, 95% CI)S1 (mean, 95% CI)S2 (mean, 95% CI)S3 (mean, 95% CI)ECG HF (×10^−3^)0.69 (0.018, 3.6)0.62 (0.016, 2.9)0.66 (0.024, 2.8)0.62 (0.023, 2.7)ECG LF (×10^−3^)1.1 (0.054, 5.0)1.1 (0.074, 4.2)1.1 (0.097, 4.0)1.1 (0.1, 4.0)ECG LFHF3.5 (0.29,14.6)3.8 (0.48, 15.0)3.5 (0.51, 13.8)3.7 (0.53, 14.4)ECG mean HR62.5 (47.3, 83.3)72.1 (52.1,95.1)73.4 (53.6, 95.5)74.6 (56.0, 96.1)SC area0.71 (0, 6.2)1.9 (0, 16.0)1.7 (0, 14.5)2.0 (0, 16.5)SC phasic18.3 (0, 157.5)8.3 (0, 78.9)7.7 (0, 72.5)8.2 (0, 70.4)ST median32.9 (21.0, 36.0)31.5 (27.0, 35.0)31.2 (27.0, 34.0)31.3 (28.0, 34.0)ST slope (×10^−3^)0.061 (−4.6, 4.7)0.34 (-4.4, 4.9)0.30 (-4.3, 4.9)0.31 (−4.2, 4.9)ST SD0.10 (0, 0.50)0.14 (0, 0.50)0.13 (0, 0.50)0.13 (0, 0.50)Population mean and 95% confidence interval (CI) of physiological features during the night (00--06 am) and different stress levels (S1, S2, and S3). For all the time instances, S1--S3 and N, only periods in which the activity level was lower than the empirical threshold (ACC SD \< 0.04)^[@CR16]^ and good quality (Quality \> 0.8) data were considered to exclude artifacts and physiology variations due to physical activity. All features, except ST slope and ST SD (i.e., ST standard deviation, see [S5 Table](#MOESM1){ref-type="media"}), were significantly different during nighttime (N) compared to during daytime self-reported stress levels (S1, S2, S3) (Fig. [2](#Fig2){ref-type="fig"}). ECG LF (i.e., the low frequency component of the RR signal, see [S5 Table](#MOESM1){ref-type="media"}) and LFHF (i.e., the ratio of LF and HF), mean HR, SC area (i.e., the sum of the area of SC responses, see [S5 Table](#MOESM1){ref-type="media"}) were lower at night. The ST median, SC phasic (i.e., power of the phasic SC component, see [S5 Table](#MOESM1){ref-type="media"}), and ECG HF (i.e., the high frequency component of the RR signal, see [S5 Table](#MOESM1){ref-type="media"}) were higher at night. Additionally, mean HR was significantly lower in S1 as compared to S3. Mean HR has a strong negative correlation with RMSSD (i.e., root mean square of the successive RR differences, a time domain HRV feature, see [S5 Table](#MOESM1){ref-type="media"}) (mean HR, RMSSD: *r* = −0.99), which is significantly higher in S1 as compared to S3 (see [S6 Table](#MOESM1){ref-type="media"}). These results confirm findings in laboratory studies reporting an increase in HR and decrease in HRV with increasing stress levels.^[@CR17]--[@CR19]^ The frequency domain HRV features, i.e., the LF signal, HF signal and the ratio of LF and HF signals, did not change significantly during S1 as compared to S2 and S3 and during S2 as compared to S3. In literature, the HF component is thought to represent the cardiac parasympathetic nerve activity, which is active during rest conditions, and the LF component to represent the sympathetic system, which is active during stress conditions.^[@CR20]^ The LF and LFHF components are therefore expected to be higher during stress conditions and the HF component lower.^[@CR20]^ However, varying results have been reported in literature and in general RMSSD has been reported to be more reliable than LFHF,^[@CR20],[@CR21]^ in particular because of the mechanical effects of respiration on HF power and the influence of the prevailing heart rate on LF power.^[@CR20]^ Furthermore, SC area was lower in S1 compared to S3, as reported previously in ref. ^[@CR22]^. SC phasic was lower in S2 compared to S3 and lower in S1 compared to S3, as expected based on previous laboratory research,^[@CR23],[@CR24]^ indicating that higher stress levels are associated with higher power of the phasic SC component. Finally, the ST median was higher in S1 compared to S2 and S3, which indicates that ST amplitude decreases with stress.^[@CR25]^ For most of the features no significant differences were found between S2 and S3. This could either indicate that in general subjects have difficulties in making distinctions between light and high stress levels or that physiological features cannot distinguish between these levels at a population level. Significant differences between S1 and S3, the two most extremes of the stress scale, were found for ST median, SC phasic, SC area, and ECG mean HR. A significant difference between S2 and S3 was found only for SC phasic and between S1 and S2 only for ST median. Overall, the physiological signals measured in daily life showed significant differences between night and different stress levels, in line with previous findings of laboratory studies. These results confirm on a large scale the potential of physiological signals for detecting stress in daily life. Towards digital phenotypes for psychophysiological stress detection {#Sec6} ------------------------------------------------------------------- We used a data-driven approach to uncover digital phenotypes of subjects' daily life stress responses. We developed random forest models using a leave-one-subject-out cross-validation to link physiological features to self-reported stress. We used the classifiers' performances to identify and characterize digital phenotypes representing subjects with similar psychological baseline, physiological responses to stress and health indicators. Only good quality (Quality \> 0.8) and low activity (ACC SD \< 0.04) data were included for 568 subjects, with complete data (i.e., simultaneous continuous recording from wearables and EMAs). The remaining subjects had missing data in one of the two sensors or lacked mobile EMA data, and were not included in this analysis. To account for possible bias we compared baseline psychological questionnaires of the excluded subjects and found no significant difference with the included subjects (PSS included subjects: 14.2 ± 6.1, PSS excluded subjects: 14.6 ± 6.1, Wilcoxon ranksum *p* = 0.19). The classification performance, as calculated using the average F1-score across all subjects, was 0.43 (95% CI: 0.05--0.86), which is slightly better than the F1-score of 0.36, obtained when all samples are classified as the majority class (i.e., S1). The most important features across all subjects were ST median, SC phasic and SC diff2. An overview of feature importance is presented in [S2 Fig](#MOESM1){ref-type="media"}. Subjects were categorized in groups of low performance (*n* = 216), with F1-score \< 0.33 (performance as good as random), medium performance (*n* = 249), with 0.33 \< F1-score \< 0.66 and high performance (*n* = 103), with F1-score \> 0.66. We compared three aspects of each group: self-reported stress imbalance, physiological dynamic range and demographics, and psychological background information. Subjects in the high performance group had on average a more imbalanced dataset (86% no stress, 12% light stress, and 2% high stress), compared to the low performance group (26% no stress, 45% light stress, and 29% high stress). This imbalance could provide an explanation for the difference between low and high performance. Additionally, on average subjects in the low performance group reported 26 times (SD~low~ = 14) their stress levels on the EMA's, whereas for the medium and high performance groups subjects reported their stress levels on average 31 times (SD~medium~ = 13, SD~high~ = 12). Although the difference is small, the response rate for the low performance group was significantly lower (p \< 0.001) as compared to the medium and high performance groups. However, we also found that for 15 out of 18 features, the high performance group had a higher dynamic range (i.e., a larger average difference per physiological feature between low and high stress) as compared to the low performance group. In [S2 Text](#MOESM1){ref-type="media"} we show that this effect is significantly different as compared to random assignment in three groups. Examples for mean HR, phasic SC, and median ST, are shown in Fig. [3a--c](#Fig3){ref-type="fig"} respectively; a complete summary for all features is provided in [S3 Fig](#MOESM1){ref-type="media"}.Fig. 3Comparison of subjects with low, medium and high classification performance. In **a**--**c** average features ECG mean HR, SC phasic, and ST median are shown respectively for low (red), medium (yellow), and high performance (green) groups and compared with the entire population average (black) in phases of no, light and high stress. In **d**--**f** baseline psychological information of subjects in low, medium, and high performance groups are compared To account for possible confounders we further investigated subjects' demographics and psychological information, based on the intake questionnaire, in the three groups. There was no difference in gender in all three groups (*X*^2^ low-high performance: *p* = 0.62, *X*^2^ low-medium performance: *p* = 0.41, *X*^2^ medium-high performance: *p* = 0.92). On average subjects in the high performance group reported a healthier lifestyle and lower baseline depression, anxiety and stress levels than subjects in the low performance group (Fig. [3d--f](#Fig3){ref-type="fig"}). They report to eat less take-out (low performance group: 1.1 ± 1.3 times per week, high performance group: 0.8 ± 0.9 times per week, Kruskal--Wallis *p* = 0.04), to practice more sports (low performance group: 26% does not practice sports, high performance group: 18% does not practice sports, *X*^2^ = 0.01), they have higher sleep quality based on the Pittsburgh Sleep Quality Index (PSQI scores higher than 5 indicate worse sleep quality; low performance group: 5.3 ± 2.5, high performance group: 4.1 ± 2.3, Kruskal--Wallis *p* \< 0.001) and score lower on depression scale (Depression Anxiety Stress Scale (DASS)---depression scale; low performance group: 3.5 ± 3.4, high performance group: 1.4 ± 2.1, Kruskal--Wallis *p* \< 0.001), anxiety scale (DASS---anxiety scale; low performance group: 2.6 ± 2.9, high performance group: 1.0 ± 1.7, Kruskal--Wallis *p* \< 0.001) and stress scales (DASS---stress scale; low performance group: 6.5 ± 3.9, high performance group: 3.1 ± 3.2; PSS; low performance group: 17.1 ± 5.6, high performance group: 10.5 ± 5.5, Kruskal--Wallis *p* \< 0.001) as compared to subjects in the low performance group. Subjects in the high performance group are also significantly older (low performance group: 38.6 ± 10.0, high performance group: 41.7 ± 10.0, Kruskal--Wallis *p* = 0.007). Discussion {#Sec7} ========== To assess stress we collected a dataset of 1002 subjects during five consecutive days, including a wide variety of subject background information, physiological data in ambulatory settings and smartphone-based self-reports and contextual information. We found significant differences between physiological features for ECG, SC, and ST between different stress levels and nighttime baseline, confirming laboratory findings and indicating the potential of psychophysiological stress detection in daily life on a large-scale population. Additionally, we compared digital phenotypes based on wearable and self-reported data emerging from a data-driven analysis. Although the classification performance of the generalized models (F1-score = 0.43) does not allow use in practice yet, we found that physiological responses to stress strongly differ among subjects, distinguishing groups with small and large dynamic ranges of the physiological features. These results highlight the need for future research to focus on personalized models as subjects differ in the magnitude and type of their physiological stress response. These groups are also characterized by different psychological baselines and demographics, where the group with a more blunted physiological stress-reactivity (small dynamic range) tend to report a less healthy lifestyle and higher depression, anxiety and stress scores than the more responsive group (large dynamic range). These findings suggest that self-reported poor health and high depression scores are negatively correlated to physiological reactivity. Similar findings have been reported previously in laboratory research,^[@CR26]^ but to date no studies have investigated this relationship in real-life ambulatory physiological recordings. In the current study, a general sample of a healthy population was included, where 'healthy' was broadly defined as being able to go to work. Although several subjects scored high on the DASS or PSS scales, they were not diagnosed with any clinical disorder as per the DSM-V guidelines. The questionnaires merely provide a quantitative measure of distress along the axes of depression, anxiety and stress, not a categorical measure of clinical diagnoses. In future research, it would be interesting to compare the results of this baseline population with those of subjects with a clinical diagnosis, such as depression. Although this dataset provides a wide variety of demographic profiles, caution for bias is needed when analyzing and interpreting the results. For example, subjects were mostly educated employees with sedentary jobs. It is possible that highly educated persons show different stress profiles compared with lower educated persons, or that highly educated persons would be better at wearing the wearable devices, which would bias the results. Also, to translate these results to persons with more active jobs or for example high performance athletes, additional experiments are needed. Identifying the concept of stress in ambulant conditions is challenging as the gold standard is based on self-reports, which could lead to bias and reduced classification accuracies as compared to controlled laboratory studies in which stressors are artificially induced. This raises the question whether indeed stress is detected or rather an activation of the autonomic nervous system (ANS). From a data analytical perspective, one could argue stress is detected since the models are trained based on a, self-reported, stress reference, however this can be also biased due to subjective perception. From a psychophysiological perspective, it is not clear whether the physiological sensing models can differentiate between actual stress and arousal or an ANS activation. Further research should compare the link between physiological signals and self-reported stress-responses on the one hand and self-reported pleasure, arousal and control levels based on the SAM on the other hand to better differentiate stress and arousal levels. This manuscript focused on the prediction of stress using physiological parameters. In the future, it could be investigated how context information (e.g., location, noise levels, ambient light), combined with physiological data, could be used to improve the performance of stress detection models. Further, in our study physiological data during high physical activity was excluded. It could be investigated if accelerometer data could be used to improve signal quality or to improve model performance, by incorporating the accelerometer signal itself for stress prediction and physical activity as an additional class. Additionally, the links between physiology, sleep quality and gastro-intestinal symptoms (Leuven Postprandial Distress Scale) for different psychological profiles (e.g., high versus low depression, high versus low stress), need to be investigated. The results of this study provide a baseline for large-scale ambulatory population monitoring to uncover blunted physiological responses to stress. Furthermore, these findings have important implications related to stress modeling strategies, indicating that stress detection models should be tailored to phenotypes by including multi-sensor data sources, as subjects with different physiological responses to stress, display different health statuses. This study exemplifies how large-scale, data-driven analytics can be used to derive digital phenotypes and generate new insights into stress detection and disease interception in general. Continuous stress detection will form the basis to enable highly personalized, just-in-time interventions for preventive health. Methods {#Sec8} ======= Experiment {#Sec9} ---------- This observational study was approved by the Research Ethical Committee of UZ Leuven. The trial was conducted with 1002 subjects (484 male, 451 female, 67 did not fill in the questionnaire correctly), aged 39.4 ± 9.8, recruited in 11 technology-oriented, banking, and public sector companies. Subjects were included if they were active employees at the time of the study, no other inclusion or exclusion criteria were applied. Subjects did not receive any compensation for participating in the study apart from having a chance at winning a restaurant or travel voucher. The experiment was conducted over a time span of 2 years. The measurements lasted 5 days per subject, starting on Thursday morning and ending on Monday evening (Fig. [4](#Fig4){ref-type="fig"}). All subjects signed the informed consent before participating in the study.Fig. 4Study protocol. **a** Protocol timeline: starting with online intake questionnaires, followed by a 5-day trial, ending with a follow-up questionnaire just after the experiment and 1 year later. **b** Ecological momentary assessments (EMAs): once per subject the Montreal Imaging Stress Task is performed containing a series of mental arithmetic challenges. Once per day a sleep diary and gastro-intestinal symptoms diary are filled in and 12 times a day stress levels are recorded. **c** Physiological recordings: Chillband and chest patch to measure SC, ST, ECG, and acceleration. **d** Smartphone sensor data: overview of the data recorded Before the start of the experiment subjects completed an intake questionnaire ([S1 Table](#MOESM1){ref-type="media"}). The first part inquired personal information such as age, gender, health problems, work situation, and lifestyle. Thereafter, four psychological questionnaires were used to assess baseline stress, depression, anxiety, sleep, and general health levels. These were the PSS,^[@CR6]^ the PSQI,^[@CR27]^ the DASS,^[@CR28]^ and the RAND-36.^[@CR29]^ On Thursday morning, the subjects received two wearable devices, along with a user manual and a USB-stick containing instructions on how to apply them. Two wearables were used to capture three physiological signals unobtrusively: the ECG, SC, and ST. Both the devices also measure 3D ACC, which signal was used for estimating intensity of physical activity and control for movement artifacts. Although EMG and BP are also frequently used in laboratory settings, these are less feasible to measure continuously in daily life. The first wearable was a chest patch (Fig. [4c](#Fig4){ref-type="fig"}), which received regulatory approval and is able to measure the ECG and ACC at a sampling rate of 256 and 32 Hz respectively. The second wearable was the imec's Chillband (Fig. [4c](#Fig4){ref-type="fig"}), a wrist-worn device, designed to measure and record SC, ST, and ACC, sampled at 256, 1, and 32 Hz, respectively. Subjects were advised to wear the Chillband the entire day, but could take it off during the night, and to wear the chest patch the entire day and night. Subjects were asked to remove the Chillband while taking a shower and to remove Chillband and chest patch during vigorous physical activities. The battery life of both sensors exceeded the duration of the experiment. Data were recorded and stored on the devices' internal SD cards and uploaded to a central data platform at the end of the experiment. A custom-made smartphone application was used to trigger subjects to fill-out the EMAs (Fig. [4b](#Fig4){ref-type="fig"}). Previous research has shown correlations between stress and sleep efficiency^[@CR30]^ and between stress and digestive diseases (e.g., irritable bowel syndrome).^[@CR31]^ Therefore, the sleep quality of the previous night was inquired every morning and gastro-intestinal symptoms experienced during the day were inquired every evening with the Leuven Postprandial Distress Scale.^[@CR32]^ Twelve times per day, the smartphone application requested the subjects to indicate their level of stress. The requests were sent at random times and at least 30 min apart. The request consisted of four brief questions (Q1--Q4 in Fig. [4b](#Fig4){ref-type="fig"}) pertaining to the past hour: first, the SAM^[@CR33]^ was used as a visual scale to assess pleasure, arousal and dominance (i.e., level of control), i.e., affective emotions related to stress. The pleasure level could be used to differentiate "good" stress from "bad" stress, i.e., eustress versus distress, where eustress reflects the transition of the body to a lower allostatic load (i.e., "the price the body pays for being forced to adapt to unfavorable psychosocial or physical situations"^[@CR34]^) and distress to a higher allostatic load.^[@CR34]^ Second, the maximum stress level was annotated on a 5-point Likert scale, i.e., not at all, slightly, moderately, very and extremely stressed. Since eating and drinking behavior and physical activity can influence physiology,^[@CR13],[@CR35]^ the third and fourth questions were used to indicate food and beverage consumption (i.e., caffeine, alcohol, soft drinks, breakfast, lunch, dinner, snack, or none) and activity levels (i.e., lying down, sitting, standing, walking, running, biking, driving the car, or something else), for which subjects could select multiple answers. To assess individual physiological stress responses to a known common stressor, the Montreal Imaging Stress Task,^[@CR36]^---based on the well-known Trier Social Stress Test---was included in the smartphone application (Fig. [4b](#Fig4){ref-type="fig"}). Each subject underwent this stress test during the first day of the experiment (Thursday) at a suitable moment (i.e., given enough time and in a quiet environment). The test consists of a 5 min rest period (relaxing music and images), a 5 min control period (simple mathematic tasks, no time restrictions or social control), a 5 min stress task (mathematic tasks with time restrictions and social control) and again a 5 min rest period (relaxing music and images). Further, the smartphone application was used, conditional on subjects permission, to collect contextual data, i.e., location, smartphone usage, audio-features, movement, SMS/call/mail logs, and environmental sensors (Fig. [4d](#Fig4){ref-type="fig"}). Location data can be used to investigate the correlation between stress levels and locations, e.g., stress at home versus at work. Many research has already indicated that social support has a large influence on stress and health-related effects caused by stress.^[@CR37]^ Audio features could be used to detect conversations and social interaction. SMS, call, and mail logs could be used as a proxy for workload and environmental sensors could provide a link between physiology and environment, e.g., higher environmental temperatures could be correlated with higher ST and SC.^[@CR38]^ On Monday evening subjects returned the sensors. Finally, subjects completed a questionnaire about sensor comfort. As a follow-up on mental health, a year later a reminder was sent to retake the DASS. Data analysis---preprocessing {#Sec10} ----------------------------- Raw sensor data and subject self-assessments were synchronized using UTC timestamps. Quality indicators and feature extraction algorithms were applied subsequently. Assessing the quality of the signals is necessary since these are prone to artifacts due to motion or incorrect sensor attachment. The ECG quality indicator is based on Orphanidou et al.,^[@CR39]^ which has shown a sensitivity for artifact detection of 94% and a specificity of 97%, and consists out of three rules and a template matching, verified on 10-s segments of ECG data: first, the extracted HR should be within 40 and 180 bpm. Second, the maximum gap between successive R-peaks cannot exceed 3 s. Third, the ratio of the maximum beat-to-beat interval to the minimum beat-to-beat interval within the segment should be less than 2.2. If all rules are satisfied, an adaptive QRS template matching is performed. The 10-s segment is either classified as of good or of bad quality. In the SC quality indicator,^[@CR40]^ the ratio of lost versus overall signal is calculated for each 5-s window. The signal is deemed lost if its value is below 0.001 µS. If this ratio is above 0.9, the signal is classified as of bad quality. Next, the algorithm searches for anomalies. For each second the maximum increase of a signal value is set to 20% and the maximum decrease to 10%, as suggested by Boucsein et al.^[@CR41]^ If SC values within the segment do not satisfy these conditions the segment is classified as of bad quality. Previous research defined the ST range at the wrist between 20 and 40 °C.^[@CR42]^ Therefore, ST values outside this range are classified as of bad quality. Data analysis---physiological feature extraction {#Sec11} ------------------------------------------------ Eighteen physiological features of interested as investigated previously in refs ^[@CR10],[@CR11],[@CR23],[@CR43]--[@CR49]^ were included in our study: 6 features for ECG, including mean HR and time and frequency domain HRV features, 8 SC features, including tonic and phasic features, and 4 ST features. For accelerometer-based activity we included the standard deviation of the accelerometer magnitude (ACC SD).^[@CR16]^ A complete list of all features is available in [S5 Table](#MOESM1){ref-type="media"}, a codebook including the Python code to compute the features is included in [S3 Text](#MOESM1){ref-type="media"}. All features were calculated in a window of 5 min with 4 min overlap. This is the minimum window required to calculate HRV features such as the root mean squared difference of successive RR intervals (RMSSD)^[@CR50]^ due to the inherent regulation periodicity.^[@CR51]^ The 4 min overlap was set to obtain a resolution of smoothed processed data of one sample per minute. Statistical analysis {#Sec12} -------------------- Statistical tests were performed using the nonparametric Wilcoxon ranksum test for comparisons of continuous variables. To assess differences of continuous variables across multiple demographic groups we used the Kruskal--Wallis test. The *X*^2^ test was used for comparisons of categorical variables. Two-sided *p*-values of \<0.05 were considered statistically significant. All statistical tests requiring multiple comparisons were corrected based on the Benjamini--Hochberg procedure with a false discovery rate of 0.05. Associations between longitudinal data (e.g., questionnaires presented 12 times per day, or continuous wearable data) were assessed using linear mixed effects models, using the lme4 R package,^[@CR52]^ with self-reported pleasure or continuous wearable feature data as fixed effects and the subjects as random effect. A Gaussian family was used to model continuous variables (e.g., ACC SD), while a Poisson family was used to model stress responses. An ANOVA test was used to assess whether model parameters differed significantly from zero by comparing the change in model performance (Akaike's information criterion) when a fixed effect (e.g., pleasure) was excluded from the model. Correlations between stationary data (e.g., questionnaires with single responses) were calculated using the Spearman correlation coefficient (*r*). Location data were anonymized based on a random translation and rotation. Locations were clustered as unique stay locations, i.e., average location in more than 60 min within a radius of 1 km and commuting. Only good quality physiological data (good QI in ≥80% of data points in the 5 min window) were used and features were normalized (*z*-normalization) per subject. Redundant features were removed based on correlations (max *r* = 0.7), resulting in a reduced feature set. Since self-reported stress responses (based on the maximum stress during the last hour, i.e., Q2 in Fig. [4b](#Fig4){ref-type="fig"}) were highly imbalanced ([S2 Table](#MOESM1){ref-type="media"}), the three highest stress levels were merged, representing 14.3% of the data, so that three, instead of five, levels of stress (S1 = no stress, S2 = light stress, S3 = high stress) were considered. Based on these data, associations between physiological features and self-reported stress levels were investigated. For each stress level the median of the normalized features across the entire population was calculated. Additionally, the median during the night (N) (00--06 am) was included as baseline. For each feature, the differences between medians of different states were computed: N--S1, N--S2, N--S3, S1--S2, S2--S3, and S1--S3. A Wilcoxon-test was performed to investigate significant differences and corrected for multiple comparisons. A machine learning model was developed to predict stress levels based on physiological responses. Subjects reporting only one stress level (e.g., only "no stress") were discarded. Since self-reported stress levels reflect the situation of the last hour, the stress value reported was registered for the 60 data points pertaining to that entire hour. We included only data for windows of at least 10 min of good quality and low physical activity (ACC SD ≤ 0.04, based on ref. ^[@CR16]^ and adapted according to subject's self-reported activity levels). A false discovery rate supervised feature selection was applied on the training set on uncorrelated features, according to the Benjamini--Hochberg procedure (Python scikit-learn, alpha = 0.05). We trained Random Forest models in a leave-one-subject-out approach. This means a model was trained based on the data of all subjects but one and tested on the data of that subject. This procedure was repeated until all subjects were tested exactly once. We used the F1-score, a weighted average between precision and recall, to evaluate the model's performance on the left-out-subject. As comparison, we also calculated the F1-score for all subjects if the Random Forest model classified all samples as the majority class, i.e., S1. We further evaluated subject's physiological response, demographics, and psychological information based on individual model performance. Subjects were categorized in groups of low performance, with F1-score \< 0.33 (performance as good as random), medium performance, with 0.33 \< F1-score \< 0.66 and high performance, with F1-score \> 0.66. For each group we evaluated three characteristics: first, we evaluated the imbalance of the self-reported stress levels, as a higher imbalance (e.g., mainly reporting S1), could lead to a higher classification performance. Second, we investigated the average dynamic range of each group, where the dynamic range represents the average difference per physiological feature of each group between low (S1) and high (S3) self-reported stress levels. A higher dynamic range could be beneficial for model performance, as the feature can better differentiate between low and high stress. Third, we investigated subject's demographics and psychological information based on the intake questionnaire. A Wilcoxon ranksum test was performed to investigate significant differences across low and high performance groups, we corrected for multiple comparisons. All data analyses were performed using Python (version 2.7). Code availability {#Sec13} ----------------- All analyses were performed using Python (version 2.7) and scikit-learn (version 0.18.1). Detailed information on the functions that were used are listed in the Methods section. A detailed codebook on feature calculation is presented in [S3 Text](#MOESM1){ref-type="media"}. Other code can be shared upon request to the authors. Electronic supplementary material ================================= {#Sec14} Supplemental material **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Elena Smets, Emmanuel Rios Velazquez, Giuseppina Schiavone. Supplementary material ====================== **Supplementary information** accompanies the paper on the *npj Digital Medicine* website (10.1038/s41746-018-0074-9). The research was partly funded by a Ph.D. grant of the Flanders Innovation & Entrepreneurship agency (VLAIO). Conceptualization: Elena Smets, Giuseppina Schiavone, Emmanuel Rios Velazquez, Walter De Raedt, Chris Van Hoof. Data curation: Elena Smets, Imen Chakroun, Ellie D'Hondt, Emmanuel Rios Velazquez. Formal analysis: Elena Smets, Giuseppina Schiavone, Emmanuel Rios Velazquez, Olivier Janssens. Investigation: Elena Smets, Jan Cornelis. Methodology: Elena Smets, Giuseppina Schiavone, Stephan Claes, Ilse Van Diest. Project administration: Elena Smets. Resources: Stephan Claes, Chris Van Hoof. Supervision: Walter De Raedt, Chris Van Hoof. Visualization: Elena Smets, Giuseppina Schiavone, Emmanuel Rios Velazquez. Writing---Original Draft Preparation: Elena Smets, Giuseppina Schiavone, Emmanuel Rios Velazquez. Writing---Review & Editing: Elena Smets, Giuseppina Schiavone, Emmanuel Rios Velazquez, Imen Chakroun, Ellie D'Hondt, Walter De Raedt, Jan Cornelis, Olivier Janssens, Sofie Van Hoecke, Stephan Claes, Ilse Van Diest, Chris Van Hoof. Guarantor: Chris Van Hoof. The data that support the findings of this study are available on request to the corresponding author. The data are not publicly available due to them containing information that could compromise research subject privacy. The authors declare no competing interests.

**Core tip:** We constructed a liver fibrosis rat model with carbon tetrachloride (CCl~4~). The Sprague-Dawley rats were randomly divided into four groups: control group, morin group, CCl~4~ group, and morin + CCl~4~ group. α-SMA, collagen I, collagen III, NF-E2-related factor 2 (Nrf2), heme oxygenase (HO-1), and quinone oxidoreductase 1 (NQO1) were analyzed by real-time PCR and Western blot methods using frozen liver specimens. We found that morin could reduce hepatic fibrosis by inducing the expression of Nrf2 and its downstream antioxidant factors in the CCl~4~-induced rat liver fibrosis model. INTRODUCTION ============ Hepatic fibrosis refers to a series of pathogenic factors and pathological changes in the pathogenesis of a variety of liver diseases with liver extracellular matrix (ECM) metabolic abnormalities\[[@B1]\]. Previous studies have found that the development and progression of liver fibrosis are significantly related to oxidative stress in which a large number of free radicals lead to cell metabolic disorders and subsequent destruction of normal liver cells\[[@B2]-[@B5]\]. Although there is currently no effective therapy for curing liver fibrosis, previous studies showed that the pathological changes in liver fibrosis could be reversed\[[@B6],[@B7]\]. Oxidative stress is closely related to the occurrence of liver disease\[[@B8]\]. A large number of studies have shown that oxidative stress may promote the activation of hepatic satellate cells (HSCs) and increase collagen production\[[@B9]\]. In the past decade, numerous studies proved that NF-E2-related factor 2 (Nrf2) plays a role as an important transcription factor in normal liver cells, and its activation could increase the expression of the downstream specific genes, such as the quinone oxidoreductase 1 (NQO1), heme oxygenase (HO-1), and glutathione, which play a role against oxidative stress\[[@B10],[@B11]\]. Studies have shown that Nrf2 activation could resist oxidative stress caused by hepatic ischemia and injury, liver fibrosis, and drug-induced liver damage\[[@B12]-[@B15]\]. Flavonoids are rich in a variety of fruits, vegetables, and components of herbal-containing dietary agents and play an important role in preventing many kinds of diseases. Morin (3, 5, 7, 2', 4'-pentahydroxyflavone) is a kind of flavonoid that consists of a yellowish pigment found in onion and apple\[[@B16]\], almond (P. guajava L.)\[[@B17]\], fig (*Chlorophora tinctoria*)\[[@B18]\], and other moraceae, including in food and herbal medicines\[[@B19]\] (Figure [1](#F1){ref-type="fig"}). It has been shown that morin possesses biological properties, including antioxidant\[[@B20],[@B21]\], anti-inflammatory\[[@B22]\], anti-apoptosis\[[@B23],[@B24]\], and anticancer\[[@B19]\] activities. Morin also protects various human cells, such as myoblasts\[[@B25]\], hepatocytes\[[@B26]\], and erythrocytes, against oxidative damages\[[@B27]\]. {#F1} Carbon tetrachloride (CCl~4~) intraperitoneal injection is a classical method for establishing an animal model of hepatic fibrosis, and the toxicity of CCl~4~ leads to liver cell necrosis and mitochondrial damage along with aggravating oxidative stress. In addition, the abundant release of inflammatory and fibrogenic cytokines induced by CCl~4~ could further augment the degree of hepatic fibrosis\[[@B28]\]. A previous study demonstrated that morin protected against acute liver damage\[[@B29]\] and ameliorated liver fibrosis\[[@B20]\] induced by CCl~4~, where morin inhibited proliferation and induced apoptosis of activated HSCs by suppressing the Wnt/β-catenin and NF-kB signaling pathways. However, there is no molecular evidence of the effects of morin on the Nrf2 signaling pathway. To our knowledge, *in vivo* investigation of the effect of morin on the Nrf2 signaling pathway and Nrf2 expression in the CCl~4~-induced liver fibrosis model has not been reported. The purpose of this study was to investigate whether morin could reduce hepatic fibrosis by inducing the expression of Nrf2 and its downstream antioxidant enzymes using pathology as a gold standard in a rat model of CCl~4~-induced hepatic fibrosis. MATERIALS AND METHODS ===================== Chemicals and reagents ---------------------- The chemical agents used in this study included CCl~4~ and olive oil (Sinopharm Chemical Reagent Co., Ltd, Shanghai, China) as well as morin (Sigma Chemical Co., St Louis, MO, United States). Serum aspartate transaminase (AST), alanine transaminase (ALT), and alkaline phosphatase (ALP) assay kits were purchased from Nanjing Jiancheng Bioengineering Institute (Nanjing, China). The antibodies against Nrf-2, HO-1, NQO1, collagen I, collagen III, and α-SMA were obtained from Proteintech Group Inc. (Chicago, IL, United States). All other reagents used were in the purest form available commercially. Animals and experimental design ------------------------------- This study was performed in accordance with the Guide for Care and Use of Laboratory Animals published by the National Institutes of Health of China (Guide for the Care and Use of Laboratory Animals, 1996) and was approved by the Animal Care and Use Committee of China Medical University. Twenty male Sprague-Dawley rats with an average body weight of 200-220 g (Changsheng Biotechnology Co., Ltd, Liaoning, China) were used in this study. All rats were fed a standard laboratory diet for a week at room temperature (20-22 °C) with a light/dark cycle of 12 h. Then, the rats were randomly divided into four groups of five rats each, *i.e*., control group, morin group, CCl~4~ group, and morin + CCl~4~ group. The control rats were treated with vehicle only (olive oil) equivalent to the treatment group. The rats in the morin group were treated with morin at a dose of 50 mg/kg (suspended in water as previously described\[[@B30]\]) by oral administration and 2 mL/kg of olive oil by intraperitoneal injection twice a week. The rats in the CCl~4~ group were injected intraperitoneally with CCl~4~ at a dose of 2 mL/kg \[mixed with olive oil (40%, V/V)\] twice a week. The rats in the morin + CCl~4~ group were treated with the same doses of morin and CCl~4~ *via* the same routes as the morin group and the CCl~4~ group. Body weights of animals were recorded twice per week. After 8 wk of treatment, animals were kept fasting for 24 h. Under 10% chloral hydrate anesthesia, the following procedures were performed, including obtaining blood samples from the heart for biochemical tests and resecting the liver and spleen for histopathological analysis. Liver tissues were weighted and cut in 10 mm × 10 mm × 3 mm pieces. Half of the specimen was fixed in 10% formaldehyde for histopathology and the other half was immediately frozen in -80 °C for PCR and Western blot tests. Biochemical analysis -------------------- The blood samples were centrifuged at 3000 *g* for 10 min at 20 °C, and the serum was collected from the supernatant. The values of AST, ALT, and ALP were measured using commercial assay kits according to the manufacturer's protocols. Histopathological assessment ---------------------------- Specimens of the liver were embedded in paraffin and cut into 5-μm-thick sections after 24 h of fixation. Then, the samples were stained with hematoxylin and eosin (HE). The degree of liver fibrosis was analyzed and determined by an experienced pathologist. The liver fibrosis was categorized into five degrees, *i.e*., F0 = no fibrosis, F1 = portal fibrosis without septa, F2 = portal fibrosis with rare septa, F3 = numerous septa without cirrhosis, and F4 = cirrhosis according to reference criteria\[[@B31]\]. Quantitative real-time PCR -------------------------- Total cellular RNA was extracted from tissues using TRIzol (Invitrogen). Reverse transcription of 1 μg of RNA was done using RT regents (TAKARA) following the manufacturer's instructions. Quantitative real-time PCR was done using SYBR Green PCR master mix (Applied Biosystems) in a total volume of 20 μL on the 7900HT fast Real-time PCR system (Applied Biosystems) using the following cycling parameters: 50 °C for 2 min, 95 °C for 10 min, and 40 cycles of 95 °C for 15 s and 60 °C for 60 s. A dissociation procedure was performed to generate a melting curve for confirmation of amplification specificity. GAPDH was used as the reference gene. The relative levels of gene expression were represented as ΔCt = Ct~gene~ - Ct~reference~, and the fold change of gene expression was calculated by the 2^-ΔΔCt^ method. Experiments were repeated in triplicate. The primer sequences are listed in Table [1](#T1){ref-type="table"}. ###### Primer sequences **Name** **Primer sequence** ------------------------ ---------------------------------- Rat *Collagen* I for 5\'-ACTGGTACATCAGCCCAAACCC-3\' Rat *Collagen* I rev 5\'-GGAATCCATCGGTCATGCTCT-3\' Rat *Collagen* III for 5\'-GAGACTCCCCATCATAGATATCGC-3\' Rat *Collagen* III rev 5\'-AGCAAACAGGGCCAATGTCC-3\' Rat α*-SMA* for 5\'-GCTATGCTCTGCCTCATGCC-3\' Rat α*-SMA* rev 5\'-CACGCTCAGCAGTAGTCACGAA-3\' Rat *Nrf2* for 5\'-ACACAGCATAGCCCATCTCGT-3\' Rat *Nrf2* rev 5\'-ACCAACCTGGATGAGCGACAC-3\' Rat *NQO1* for 5\'-CCACGCAGAGAGGACATCATT-3\' Rat *NQO1* rev 5\'-TTCGACCACCTCCCATCCTT-3\' Rat *HO-1* for 5\'-CTTCCCGAGCATCGACAAC-3\' Rat *HO-1* rev 5\'-CTGTCACCCTGTGCTTGACC-3\' Rat *Gapdh* for 5\'-GCTGGTCATCAACGGGAAA-3\' Rat *Gapdh* rev 5\'-CGCCAGTAGACTCCACGACAT-3\' Western blot analysis --------------------- Total proteins from tissues were extracted in lysis buffer (Pierce, United States) and quantified using the Bradford method. A total of 40 μg of protein were separated using 10% SDS-PAGE (80 V -120 V) and then electrophoretically transferred to a PVDF membrane (80 V 100 min) (Millipore, Bedford, MA, United States). The membrane was blocked with 5% dry milk and incubated overnight at 4 °C with antibodies against HO-1 (1:800; Proteintech), NQO-1 (1:1000; Proteintech), Nrf2 (1:800; Proteintch), collagen I (1:800, Proteintech), collagen III (1:1000, Proteintech), α-SMA (1:1000, Proteintch), and GAPDH (1:4000, Proteintech). After washing, the membrane was incubated with a horseradish peroxidase-conjugated secondary antibody (Santa Cruz Biotechnology) at 37 °C for 2 h. Protein bands were visualized by enhanced chemiluminescence (Pierce) and detected using BioImaging Systems (UVP, Upland, CA, United States). The relative protein levels were calculated based on GAPDH protein as a loading control. Western blot images were measured with ImageJ software, and the relative gray values of protein expression were analyzed semi-quantitatively. Statistical analysis -------------------- The experimental data are expressed as the mean ± SD. Statistical analyses were performed using one-way analysis of variance (ANOVA) between groups, and unpaired comparisons were analyzed using the least significant difference method LSD *t*-test. A *P*-value of 0.05 or less was considered statistically significant. All analyses were conducted using SPSS version 17.0 (SPSS, Inc., Chicago, IL, United States) and Prism GraphPad software Version 6.01 (GraphPad Software Inc., San Diego, CA, United States). RESULTS ======= General observation ------------------- A total of four rats died before the end-point of the study, including two in the CCl~4~ group, one in the morin + CCl~4~ group, and one in the morin group. All animals in the control group survived. Normal diet and daily activities were recorded in the control and morin groups, with body weight increasing rapidly. The CCl~4~ group presented poor feeding and daily activities with slow weight growth. The morin + CCl~4~ group presented milder symptoms compared with the CCl~4~ group, with increased body weight, which was, however, lower than that in the control and morin groups (Figure [2](#F2){ref-type="fig"}). {#F2} Histological changes in the liver --------------------------------- The results of HE staining showed that the liver cells appeared with a normal morphology and regular lobular structure in the control and morin groups. The liver tissue of CCl~4~ group rats showed inflammatory cell infiltration, with portal and central veins surrounded by fibrous tissue accompanied by fibrous septa. The lobular structure was fuzzy with clearly visible false lobules. In the morin + CCl~4~ group, the liver tissue demonstrated less hyperplasia of fiber tissue and minimal inflammatory cells compared to the CCl~4~ group (Figure [3A](#F3){ref-type="fig"}-D). {#F3} Liver-spleen ratio and liver weight index ----------------------------------------- Both the CCl~4~ and morin + CCl~4~ groups had increased liver-spleen ratio (LSR) and liver weight index (LWI) compared with the control and morin groups (*P \<* 0.05). The LWI between the CCl~4~ and morin + CCl~4~ groups showed a significant difference (*P \<* 0.05), while no statistically significant difference was found for LSR (*P \>* 0.05) (Table [2](#T2){ref-type="table"}). ###### Comparison of liver-spleen ratio and liver weight index among different groups **Control** **Morin** **CCl~4~** **Morin + CCl~4~** ***F*** ***P* value** ------ -------------- -------------- ------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------- --------- --------------- LSR 12.27 ± 1.92 12.67 ± 1.60 16.43 ± 1.37[a](#T2FN1){ref-type="table-fn"}[c](#T2FN2){ref-type="table-fn"} 15.11 ± 1.99[a](#T2FN1){ref-type="table-fn"}[c](#T2FN2){ref-type="table-fn"} 4.668 0.022 LWI% 2.78 ± 0.25 2.80 ± 0.27 4.77 ± 0.47[a](#T2FN1){ref-type="table-fn"}[c](#T2FN2){ref-type="table-fn"} 4.17 ± 0.39[a](#T2FN1){ref-type="table-fn"}[c](#T2FN2){ref-type="table-fn"}[e](#T2FN3){ref-type="table-fn"} 32.345 \< 0.001 *P* \< 0.05 *vs* control group, *P* \< 0.05 *vs* morin group, *P* \< 0.05 *vs* CCl~4~ group. Liver-spleen ratio (LSR): Liver wet weight/spleen wet weight; liver weight index (LWI): (Liver wet weight/body weight) × l00%. Biochemical findings -------------------- The CCl~4~ and morin + CCl~4~ groups had increased ALT, AST, and ALP levels compared to the control and morin groups (*P \<* 0.05), and CCl~4~ without morin treatment dramatically increased ALT, AST, and ALP values (Table [3](#T3){ref-type="table"}). ###### Serum parameters among different groups **Control** **Morin** **CCl~4~** **Morin + CCl4** ***F*** ***P* value** ------------ ---------------- ---------------- ---------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------- --------- --------------- ALT (IU/L) 101.75 ± 15.46 108.00 ± 48.72 493.33 ± 199.38[a](#T3FN1){ref-type="table-fn"}[c](#T3FN2){ref-type="table-fn"} 291.50 ± 111.92[a](#T3FN1){ref-type="table-fn"}[c](#T3FN2){ref-type="table-fn"}[e](#T3FN3){ref-type="table-fn"} 11.403 0.001 AST (IU/L) 339.25 ± 72.59 257.80 ± 98.22 1027.67 ± 206.60[a](#T3FN1){ref-type="table-fn"}[c](#T3FN2){ref-type="table-fn"} 585.50 ± 131.85[a](#T3FN1){ref-type="table-fn"}[c](#T3FN2){ref-type="table-fn"}[e](#T3FN3){ref-type="table-fn"} 26.280 \< 0.001 ALP (IU/L) 137.75 ± 29.75 160.80 ± 40.90 377.67 ± 41.07[a](#T3FN1){ref-type="table-fn"}[c](#T3FN2){ref-type="table-fn"} 266.50 ± 58.90[a](#T3FN1){ref-type="table-fn"}[c](#T3FN2){ref-type="table-fn"}[e](#T3FN3){ref-type="table-fn"} 22.093 \< 0.001 *P* \< 0.05 *vs* control group, *P* \< 0.05 *vs* morin group, *P* \< 0.05 *vs* CCl~4~ group. ***mRNA expression of*** α***-SMA, collagen ***I**, *collagen*** III**, *Nrf2, HO-1, and NQO1*** Compared with the control and morin groups, significantly higher mRNA expression of α*-SMA*, *collagen *I, and *collagen* III was observed in liver tissues in the CCl~4~ and morin + CCl~4~ groups (*P* \< 0.05). However, the mRNA expression of these molecules in the morin + CCl~4~ group was significantly less than that in the CCl~4~ group (*P* \< 0.05) (Figure [4](#F4){ref-type="fig"}). {#F4} In the CCl~4~ and morin + CCl~4~ groups, mRNA expression values of *NQO1*, *HO-1*, and *Nrf2* were significantly higher than those in the control and morin groups (*P* \< 0.05), while these mRNA values of the morin + CCl~4~ rats were significantly different compared to those of the CCl~4~ group (*P* \< 0.05) (Figure [5](#F5){ref-type="fig"}). {#F5} ***Protein expression of*** α***-SMA, collagen ***I**, *collagen*** III**, *Nrf2, HO-1, and NQO1*** Compared with the control and morin groups, high expression of protein of α-SMA, collagen I, and collagen III in liver tissues in the CCl~4~ and morin + CCl~4~ groups had a statistically significant difference (*P* \< 0.05). However, the morin + CCl~4~ group had less expression of these protein factors compared to the CCl~4~ group (*P* \< 0.05) (Figure [6](#F6){ref-type="fig"}). {#F6} In the CCl~4~ and morin + CCl~4~ groups, the protein expression of Nrf2, HO-1, and NQO1 was statistically higher than that in the control and morin groups (*P* \< 0.05), while these protein factors of the morin + CCl~4~ rats had more expression compared to the CCl~4~ group (*P* \< 0.05) (Figure [7](#F7){ref-type="fig"}). {#F7} DISCUSSION ========== Liver fibrosis is a process of continuous damage to the liver blood vessels and hepatic cells with nodule formation, which may develop into cirrhosis and cancerous lesions. Research of fibrosis at the cellular and molecular levels suggested that the progression of liver injury was closely related to oxidative stress and lipid peroxidation\[[@B32],[@B33]\], leading to cell destruction and inducing hepatic fibrosis. HSCs can be activated by lipid peroxides acting as products of cell damage. After HSC activation, lipid droplets and vitamin A in the cytoplasm could be reduced or exhausted with α-SMA expression, accompanied by liver structural and functional changes resulting from redundant secretion of ECM\[[@B34]\]. However, it is possible to reverse liver fibrosis and early cirrhosis with effective interventions. Previous studies have shown that antioxidants have a protective effect by inhibiting the expression of α-SMA in HSC\[[@B35]\], thus, inhibition of oxidative stress in the liver may reduce and even reverse liver fibrosis\[[@B36]\]. Pathological features of liver fibrosis are reflected by fibrous tissue hyperplasia around the portal area and central vein and forming an interval of destruction of the lobular structure, accompanied by regenerative nodules and even early cirrhosis\[[@B37]\]. The pathological findings in this study showed that liver tissue in the CCl~4~ group had liver cell necrosis, fibrous tissue hyperplasia, interval widening, and pseudolobuli replacing normal lobular architecture. In the morin + CCl~4~ group, the liver tissue showed minimal cell necrosis with less interstitial collagen fibers and lobular structure damage compared with the CCl~4~ group. Thus, morin could effectively protect the liver tissue by reducing inflammation and inhibiting collagen deposition and fiber hyperplasia. There are various enzymes that take part in liver metabolism. The damaged liver cells by pathogenic factors will produce free enzymes that are released into the bloodstream\[[@B20]\]. Liver function and status could be assessed by assaying the contents of serum enzymes. Aminotransferases play an important role in hepatic metabolism. When the liver cells are damaged, the serum ALT and AST levels as well as ALP level will be increased\[[@B38]\]. In this study, in the CCl~4~-induced liver fibrosis rat model, the values of serum ALT, AST, and ALP were reduced with morin administration, which implied that morin can reduce liver cell injury and thus prevent liver fibrosis. This also gives support for morin being able to condition the hepatocytes, protect against membrane frailty, and decrease the outflow of enzymes into circulation. These results are in accordance with previous studies that showed the ability of morin to inhibit hepatotoxicity\[[@B39],[@B40]\]. The amount of collagen accounts for 5%-10% of the total protein in human liver tissue. If the liver injury leads to fibrosis, the collagen content in the liver protein will be significantly increased up to approximately 50%, becoming an important component of ECM\[[@B41]\] and ultimately leading to irreversible cirrhosis changes\[[@B42]\]. Liver fibrosis is a common histological change in liver disease, which is mainly manifested by excessive deposition of ECM, such as type I and type III collagen, and the expression of α-SMA\[[@B43]\]. At present, it is believed that the ECM actively participates in the occurrence and development of fibrosis, which has a great influence on HSC activation\[[@B44]-[@B46]\]. Both *in vitro* and *in vivo* experiments found that ECM synthesis was increased when liver tissue was damaged and further caused the activation of HSCs, which was based on the secretion of type I and III collagen\[[@B47]-[@B49]\], ultimately promoting the occurrence of liver fibrosis. In our study, using both real-time PCR and Western blot methods, it was found that the control and morin groups had only minimal expression of collagen I, collagen III, and α-SMA, which may represent normal physiological function of the liver, while their expression in the CCl~4~ group was significantly increased and had great relevance to the severity of liver fibrosis. With morin intervention reducing the expression of collagen I, collagen III, and α-SMA, the degree of liver fibrosis was relieved, which was evidenced by liver histopathology and serum measurements. All these results suggested that the anti-fibrotic effect of morin may be related to the down-regulation of the expression of collagen I, collagen III, and α-SMA. Nrf2 is a key nuclear transcription factor in the oxidative stress of various cells\[[@B50]\]. Under normal circumstances, Nrf2 and Keapl are in a binding state in the cytoplasm\[[@B51]\]; they will appear dissociated when oxidative stress is occurring\[[@B52]\] and combine with antioxidant components as dimers, which are involved in the synthesis of antioxidase and phase II detoxification enzymes and prevent the occurrence of liver fibrosis by improving the antioxidant capacity of the liver\[[@B53]\]. HO-1 and NQO-1 are well characterized Nrf2-dependent antioxidant defense genes. Studies have suggested that Nrf2 and its downstream antioxidant factors HO-1 and NQO1 may contribute to improvement of liver fibrosis\[[@B54]\]. It has been reported that morin could promote the nuclear translocation of Nrf2 in order to play its biological role and be used as an exogenous agonist of Nrf2\[[@B55]\]. In this study, a CCl~4~ induced liver fibrosis model, along with morin as an intervention, was used to observe the expression of Nrf2 and its downstream products NQO1 and HO-1 in different groups. The results showed that the expression of Nrf2, NQO1, and HO-1 was slightly increased in the CCl~4~ group compared with the control and morin groups (*P* \< 0.05). This might be due to Nrf2 activation acting as a cellular adaptive response against CCl~4~-induced toxicity. Nrf2 activation was initiated as soon as the subjects were challenged by CCl~4~-induced oxidative stress. However, it was unable to completely overcome the toxicity, while the adaptively stimulated Nrf2 might alleviate or delay the deleterious effects of CCl~4~. The expression of Nrf2 and its downstream products NQO1 and HO-1 was evidently increased in the morin-treated group, indicating that morin administration could enhance this effect. Additionally, this supports morin playing an important role in the prevention and treatment of liver fibrosis *via* the Nrf2 pathway. This study has several limitations. First, the sample size was small, which easily led to individual differences and statistical error between the groups. Second, the anti-fibrotic mechanism of morin may be related to activation of the Nrf2 antioxidant pathway and expression of its downstream antioxidases. Further experiments are needed to confirm the specific mechanism of the morin intervention. In summary, our current study showed that morin could play a protective role by inducing the expression of Nrf2 and its downstream antioxidant factors (HO-1 and NQO1) and reducing the expression of α-SMA, collagen I, and collagen III in a rat model of CCl~4~-induced hepatic fibrosis. Although further studies are required, our study demonstrated that morin could effectively alleviate chronic liver damage by activation of the Nrf2 pathway. ARTICLE HIGHLIGHTS ================== Research background ------------------- Previous studies have shown that the pathological changes of liver fibrosis, which refer to a series of pathogenic factors and pathological changes in the pathogenesis of a variety of liver diseases, could be reversed. In the past decade, numerous studies demonstrated that NF-E2-related factor 2 (Nrf2) as a transcription factor plays as an important role against oxidative stress in normal liver cells. Morin possesses biological properties, including antioxidant, anti-inflammatory, anti-apoptosis, and anticancer activities. To our knowledge, *in vivo* investigation of the effect of morin on the Nrf2 signaling pathway and Nrf2 expression in a CCl~4~-induced liver fibrosis model has not been reported previously. Research motivation ------------------- Previous studies demonstrated that morin protected acute liver damage and ameliorated liver fibrosis induced by CCl~4~, and morin inhibited proliferation and induced apoptosis of activated hepatic satellate cells by suppressing the Wnt/β-catenin and the NF-kB signaling pathways. However, there is no molecular evidence about the effects of morin on the Nrf2 signaling pathway. Research objectives ------------------- The purpose of this study was to investigate whether morin can reduce hepatic fibrosis by inducing the expression of Nrf2 and its downstream antioxidant enzymes in a rat model of CCl~4~-induced hepatic fibrosis. Research methods ---------------- Twenty male Sprague-Dawley rats were randomly divided into four groups: control group, morin group, carbon tetrachloride (CCl~4~) group, and morin + CCl~4~ group. At the end-point of the experimental period, serum AST, ALT, and ALP were measured, and the liver specimens were obtained for pathological assessment. α-SMA, collagen I, collagen III, NF-E2-related factor 2 (Nrf2), heme oxygenase (HO-1), and quinone oxidoreductase 1 (NQO1) were analyzed by real-time PCR and Western blot methods using frozen liver specimens. Research results ---------------- Rats in the morin + CCl~4~ group had less hyperplasia of fiber tissues, minimal inflammatory cells, and less body weight loss with favorable liver enzyme measurements compared to rats treated with CCl~4~ only. Additionally, morin-treated rats had significantly lower mRNA and protein expression of α-SMA, collagen I, and collagen III, but significantly higher mRNA and protein expression of Nrf2, HO-1, and NQO1 compared to rats treated with CCl~4~ only (*P \<* 0.05). Research conclusions -------------------- Our study showed that morin could play a protective role by inducing the expression of Nrf2 and its downstream antioxidant factors (HO-1 and NQO1) and reducing the expression of α-SMA, collagen I, and collagen III in a rat model of CCl~4~-induced hepatic fibrosis. Research perspectives --------------------- Although further studies are required, out study demonstrated that morin could effectively alleviate chronic liver damage by activation of the Nrf2 pathway. Manuscript source: Unsolicited manuscript Specialty type: Gastroenterology and hepatology Country of origin: China Peer-review report classification Grade A (Excellent): 0 Grade B (Very good): B, B Grade C (Good): 0 Grade D (Fair): 0 Grade E (Poor): 0 Institutional animal care and use committee statement: This study was performed in accordance with the Guide for Care and Use of Laboratory Animals published by the National Institutes of Health of China (1996), and was approved by Animal Care and Use Committee of the China Medical University (2015038R). Conflict-of-interest statement: The authors declare no conflict of interest. Data sharing statement: No additional data are available. Peer-review started: September 22, 2017 First decision: October 10, 2017 Article in press: November 14, 2017 P- Reviewer: Deepak P, Faerch K S- Editor: Chen K L- Editor: Wang TQ E- Editor: Huang Y [^1]: Author contributions: Sang L, Wang XM and Sang LX conceived and designed the experiments; Sang L, Han Y and Jiang LY performed the experiments; Sang L and Xu DY analyzed the data; Sang L wrote the paper; all authors agreed and approved the final version of the manuscript. Correspondence to: Xue-Mei Wang, MD, Professor, Department of Ultrasound, The First Hospital of China Medical University, No. 155, Nanjing North Street, Shenyang 110001, Liaoning Province, China. <wangxuemei@cmu1h.com> Telephone: +86-24-83282998 Fax: +86-24-83282998

Introduction {#Sec1} ============ High-throughput screening (HTS) is a widely used drug discovery technique which allows researchers to rapidly conduct millions of chemical, genetic or pharmacological experiments^[@CR1],[@CR2]^. HTS technology relies on robotic handling systems, liquid handling systems, data mining tools and control software in order to assess the biological or biochemical activity of a large number of chemical compounds. Using HTS, researchers can discover new active compounds, antibodies or genes modulating a certain biomolecular pathway^[@CR3]--[@CR6]^. Growing needs of the modern pharmaceutical industry have motivated the recent advances in data throughput and data quality of high-throughput screening campaigns^[@CR3]^. Processing hundreds of thousands compounds a day has become routine in screening laboratories worldwide. A typical HTS assay consists of a library of chemical compounds which are screened against the selected biological target in order to discover potential drug candidates, called hits^[@CR2]^. An HTS library, organized according to biological activity (e.g. small molecules, siRNA or shRNA) or target specificity (e.g. enzymes such as kinases, proteases or phosphatases) is arrayed into micro-well plates enabling screening in a miniaturized form -- in 96, 384, 1536 or 3456-well plates. Unfortunately, experimental high-throughput screens are usually affected by spatial bias (i.e., systematic error) which negatively impacts the hit selection process^[@CR7]--[@CR10]^. Various sources of bias include reagent evaporation, cell decay, errors in liquid handling, pipette malfunctioning, variation in incubation time, time drift in measuring different wells or different plates, and reader effects^[@CR8],[@CR10],[@CR11]^. Spatial bias is usually evident as row or column effects, especially on plate edges^[@CR7],[@CR9],[@CR12]^. It produces over or under-estimation of true signals in specific rows or columns within the same plate and/or specific well locations across plates^[@CR8],[@CR9]^. The presence of spatial bias in high-throughput screening assays can lead to an increase in the false positive and false negative rates during the hit identification process (see for example: a colorimetric immunoassay analyzed by Brideau *et al*.^[@CR7]^; McMaster Data Mining and Docking Competition assay described in Elowe *et al*.^[@CR13]^ and analyzed in Dragiev *et al*.^[@CR8]^; RNAi HIV assay analyzed by Carralot *et al*.^[@CR14]^ or LINCS gene expression assay analyzed by Lachmann *et al*.^[@CR10]^). If they are not corrected using appropriate bias minimization methods, biased measurements can be falsely identified as hits, leading to an increase in both the length and the cost of the drug discovery process^[@CR2],[@CR10]^. ChemBank is a public small-molecule screens database which allows life scientists from various institutions to make their experimental screening data accessible to the research community^[@CR15]^. The ChemBank project hosts 4,767 assays (as of November 30^th^, 2016) that cover a wide range of molecules, species and screening technologies. Small molecules are crucial components of a growing drug discovery toolbox used for studying cellular processes and developing effective therapies^[@CR16]^. The three main screening technologies represented in ChemBank are high-throughput screening^[@CR17]^ (HTS), high-content screening^[@CR18]^ (HCS) and small-molecule microarrays^[@CR16]^ (SMM). High content screening encompasses a set of analytical methods based on automated microscopy, multi-parameter image processing and visualization tools to identify substances such as small molecules, peptides or RNAi that alter the phenotype of a cell in a desired manner^[@CR18]^. Thus, HCS can be viewed as a type of phenotypic screen conducted in cells. This technology normally uses fluorescence imaging of samples in a high-throughput format to extract quantitative data, such as spatial distribution of targets, or individual cell and organelle morphology, from cell populations. Microarray technique for screening small molecules is based on the use of machine-printed glass slides with an array of wells^[@CR16]^. SMM is known to be an effective tool for discovering protein-small molecule interactions. In this paper we analyze various data sets from ChemBank and show that most of them are affected by both assay-specific (when a certain bias pattern appears within all the plates of a given assay) and plate-specific (when a certain bias pattern appears within a given plate only) spatial biases. Furthermore, we provide evidence that spatial bias can be either additive or multiplicative^[@CR19]^, depending in part on the screening technology. We show that both plate and assay-specific biases can be effectively identified and corrected. The correction of both of these biases is critical for the selection of quality hits^[@CR20]^ in high-throughput screening campaigns. Results {#Sec2} ======= Simulation study {#Sec3} ---------------- To assess the efficiency of our method, including both assay-specific bias correction by using robust Z-scores and plate-specific bias correction by using the additive and multiplicative PMP algorithms (see the Methods section), we compared its performance with those of the B-score^[@CR7]^ and Well Correction^[@CR11],[@CR21]^ methods, considering synthetic data with known hits and bias rates. The B-score method is the most known plate-specific correction method used in HTS, while Well Correction is an effective assay-specific correction technique removing systematic error from biased well locations. In our simulations, we followed the data simulation protocol applied by Dragiev *et al*.^[@CR8]^ for testing methods minimizing additive spatial bias in HTS. First, 100 HTS assays, each including 50 plates of size (16 × 24), were generated (i.e., this is the most widely-used plate format in Chembank^[@CR15]^) for each parameter combination described below. Inactive compound measurements were sampled from the standard normal distribution. Hit (i.e., active compound) locations were selected independently for each plate using rejection sampling with the following hit percentages: 0.5%, 1%, 2%, 3%, 4% and 5%. Hit measurements were generated from the normal distribution with the following parameters \~*N*(*μ* − 6 *SD*, *SD*), where *μ* and *SD* were respectively the mean and the standard deviation of the inactive compound measurements (i.e., *μ* = 0 and *SD* = 1). Well locations affected by assay-specific spatial bias were selected randomly with probability *p*~*a*~ = 0.29, estimated from the data taken from the ChemBank repository (see Tables [1](#MOESM1){ref-type="media"} and [2](#MOESM1){ref-type="media"} in Supplementary Information). Assay-specific bias was sampled from a normal distribution with parameters \~*N*(0, *C*), where the parameter *C* was taking the values: 0, 0.6 *SD*, 1.2 *SD*, 1.8 *SD*, 2.4 *SD*, and 3 *SD*, and added to the selected well locations. Plate-specific bias was then generated and added independently to rows and columns of each plate of a given assay. A maximum of 8 rows and 12 columns of each plate were allowed to be affected by spatial bias. For each plate of a given assay, the number of biased rows was sampled from a \~*Geometric*(*p* = 0.622) distribution and the number of biased columns was sampled from a \~*Geometric*(*p* = 0.430) distribution. These parameters were obtained using the maximum likelihood approach applied to the distributions of the number of biased rows and columns in the 2441 non-empty plates of the 175 ChemBank assays analyzed in this study (see Supplementary Figure [1](#MOESM1){ref-type="media"} in Supplementary Information). The plate-specific bias model for each plate was selected from: no bias, additive bias model and multiplicative bias model with the following respective sampling probabilities: 0.274, 0.418 and 0.308. These probabilities were also estimated from the considered ChemBank data (here, the no bias model included the probabilities of both undetermined and no bias models discussed in the next sections). The additive plate-specific bias was generated according to the normal distribution with parameters \~*N*(0, *C*) and the multiplicative plate-specific bias according to the normal distribution with parameters \~*N*(1, *C*), where the parameter *C* was taking the following values: 0, 0.6 *SD*, 1.2 *SD*, 1.8 *SD*, 2.4 *SD* and 3 *SD*. Plate measurements affected by additive spatial bias were generated using Equation ([3](#Equ3){ref-type=""}) and plate measurements affected by multiplicative spatial bias were generated using Equation ([4](#Equ4){ref-type=""}). A small Gaussian noise (i.e., random noise *ε*~*ijp*~ in Equations [3](#Equ3){ref-type=""} and [4](#Equ4){ref-type=""}) was generated from the normal distribution with parameters \~*N*(0, 0.1 *SD*) and added to both active and inactive compound measurements of each plate. Four bias correction methods were compared in our simulations: (1) No Correction, (2) B-score^[@CR7]^, (3) Well-Correction^[@CR11]^ and (4) our new method correcting both plate and assay-specific biases (i.e., additive or multiplicative PMP followed by robust Z-scores) as described in the Methods section. Moreover, both the Mann-Whitney *U* test and the Kolmogorov-Smirnov two-sample test, included in our method, were independently executed using the two following significance thresholds: *α* = 0.01 and *α* = 0.05 (this threshold was always the same for both tests). After the data correction, hits were selected according to the *μ*~*p*~ − 3*σ*~*p*~ threshold, where *μ*~*p*~ and *σ*~*p*~ were the mean and the standard deviation of the measurements in plate *p*, respectively. We assessed the methods' performances by comparing their hit detection rates (i.e., true positive rates) as well as by counting the total of false positives and false negatives they provided^[@CR8]^. Two simulations were carried out for both selected values of *α*. In the first simulation, the bias magnitude, *SD*, was fixed to 1.8, while the hit percentage ranged from 0% to 5%. In the second simulation, the hit percentage was fixed to 1%, while the noise magnitude ranged from 0 to 3 *SD*. Fig. [1](#Fig1){ref-type="fig"} shows the average true positive rate and the average total count of false positive and false negative hits per assay obtained for artificially generated HTS screens composed of 384-well plates, yielded by the competing methods (our method was carried out twice, using the significance levels *α* = 0.01 and *α* = 0.05, respectively).Figure 1Average true positive rate (Panels (a) and (c)) and average total number of false positive and false negative hits per assay (Panels (b) and (d)) obtained by No Correction, Well Correction, B-score, and PMP followed by robust Z-scores for the significance levels *α* = 0.01 and 0.05. Panels (a) and (b) show the results obtained for datasets with a fixed bias magnitude (*SD* = 1.8). Panels (c) and (d) show the results for datasets with a fixed hit percentage of 1%. The obtained results suggest that the additive and multiplicative PMP algorithms applied together with robust Z-scores yield the highest hit detection rate and the lowest false positive and false negative total hit count across all examined methods. When the hit percentage varies from 0.5% to 5%, and the bias magnitude is constant at 1.8 *SD* (Fig. [1a](#Fig1){ref-type="fig"}), the true positive rate for all methods decreases with the increase in the hit percentage. A similar trend in the true positive rate can be observed when the bias magnitude increases from 0.0 *SD* to 3.0 *SD* and the hit percentage remains fixed at 1% (Fig. [1c](#Fig1){ref-type="fig"}). It is worth noting that the PMP algorithm, followed by the robust Z-score normalization, has a very similar behaviour for both tested significance levels, *α* = 0.01 and *α* = 0.05 used in the Mann-Whitney *U* test and the Kolmogorov-Smirnov test. A slightly better hit recovery was obtained for the significance level *α* = 0.01 for lower hit percentages, and for the significance level *α* = 0.05 for higher hit percentages (Figs. [1](#Fig1){ref-type="fig"}a and [1b](#Fig1){ref-type="fig"}). When the hit percentage ranges from 0.5% to 5% and the bias magnitude stays constant at 1.8 *SD* (Fig. [1b](#Fig1){ref-type="fig"}), the average total count of false positives and false negatives increases for all methods. The PMP algorithm followed by assay-wise normalization by robust Z-scores provides the lowest total counts in all cases. When the bias magnitude varies from 0.0 *SD* to 3.0 *SD* and the hit percentage stays constant at 1%, our method still outperforms the three other competing approaches. The traditional additive B-score method^[@CR7]^ usually copes well with the recovery of true positive hits (Fig. [1a and c](#Fig1){ref-type="fig"}), but provides a large number of false positives and thus gets outperformed by Well Correction^[@CR11]^ in terms of the total number of false positive and false negative hits (Fig. [1b and d](#Fig1){ref-type="fig"}). In all cases, except the case of unbiased data (i.e., Bias = 0.0 *SD*) and the B-score method (Fig. [1c](#Fig1){ref-type="fig"}), the examined bias correction methods outperform the No Correction approach. Quantifying assay-specific bias in screening technologies {#Sec4} --------------------------------------------------------- To assess the extent of assay-specific bias in the HTS, HCS and SMM technologies, we examined 12 experimental assays from the ChemBank assays repository (i.e., 4 assays per technology were considered; Fig. [2](#Fig2){ref-type="fig"}). First, the data of the selected screens were normalized on a plate-by-plate basis using robust *Z*-scores. Second, the nonparametric Mann-Whitney *U* test was carried out independently for each well location (i.e., a vector of measurements taken across all plates of a given assay which corresponds to a fixed well position (*i,j*), where *i* is the row number and *j* is the column number), comparing its sum of ranks to the sum of ranks of the rest of the assay measurements. The significance level *α* = 0.01 was used in the Mann-Whitney *U* test (see the Methods section). Because the Mann-Whitney *U* test does not make any assumption about the underlying data distribution, its use has been recommended for bias detection purposes in screening technologies^[@CR22]^. If the presence of spatial bias in a particular well location has been supported by the Mann-Whitney *U* test, then the traditional *Z*-score normalization should be carried out across its measurements. Importantly, the traditional *Z*-score normalization carried out across the measurements of a given well location can remove successfully both additive and multiplicative types of spatial bias. The detailed explanation of this property is presented in the Methods section.Figure 2Assay-specific bias detected in 12 experimental assays from the ChemBank database. Here, 4 high-throughput screening assays, 4 high-content screening assays and 4 small-molecule microarrays were examined. The ChemBank IDs of these assays are indicated between parentheses: ABeta42 aggregation inhibitors (1103.0016), Bacterial viability profiling (1064.0002), *E. coli* filamentation (1038.0004), *M. tuberculosis* sulfur assimilation (130.0018), Autophagy cell count (1050.0009), Autophagy EGFP (1050.0111), Toxoplasma invasion imaging screening (141.0027), *C. elegans* assay for anti-infective reagents (1109.0003), HPV-E7 SMM (1049.0001), Male germ cell targets SMM (1154.0015), Male germ cell targets SMM (1154.0009) and NeuroSMM screen on torsin A (1069.0001); for more details see Supplementary Table [1](#MOESM1){ref-type="media"}. Figure [2](#Fig2){ref-type="fig"} shows that although all three small-molecule screening technologies are prone to assay-specific bias, it is much more substantial in HTS assays. According to the Mann-Whitney *U* test, 51.2% of well locations of HTS assays were affected by spatial assay-specific bias, whereas this bias affected only 17.4% and 21.8% of well locations in HCS and SMM screens, respectively. Quantifying plate-specific bias in screening technologies {#Sec5} --------------------------------------------------------- Differences among the bias distribution were also observed at the plate level (Figs [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"}). Plate-specific spatial bias is evident as systematic over or under-estimation of the measurements in specific rows and columns within specific plates. The Mann-Whitney *U* test (with *α* = 0.01) was performed independently on all plates of all assays considered in this study to identify the rows and columns affected by spatial bias (i.e., comparing the row and column measurements to unbiased data of the same plate). If systematic error was detected in one of the rows or columns of the plate, then additive and multiplicative Partial Mean Polish (PMP) bias algorithms^[@CR8]^ were applied independently to correct the plate measurements. The measurements corrected by the additive and multiplicative PMP algorithms were then compared to unbiased plate measurements to select the best-fit correction model for the data at hand. The main advantages of the PMP algorithms are that they modify the values of the biased measurements only and keep the raw and corrected data on the same scale. The Kolmogorov-Smirnov test was performed on the plate surfaces corrected by the additive and multiplicative PMP algorithms and the model providing the smallest significant *p*-value was selected as the most appropriate (see the Methods section).Figure 3Plate-specific bias detected across data of 3 screening technologies and 8 screening categories available in ChemBank - per plate representation; 175 assays were analyzed in total (see Supplementary Table [2](#MOESM1){ref-type="media"} for the complete list of the assays considered); all control wells were ignored.Figure 4Plate-specific bias detected across data of 3 screening technologies and 8 screening categories available in ChemBank - per assay representation; 175 assays were analyzed in total (see Supplementary Table [2](#MOESM1){ref-type="media"} for the complete list of the assays considered); all control wells were ignored. Assays, in which the number of plates containing additive bias was bigger than the number of plates containing multiplicative bias, are reported in the first column of each screening category. Assays with a bigger number of plates affected by multiplicative bias are reported in the second column of each screening category. Assays, in which the number of plates containing additive bias was equal to the number of plates containing the multiplicative bias as well as assays without any biased plate are reported as "Undetermined". Darker portions of bars show the proportion of assays that have a dominant, additive or multiplicative, trend. Lighter portions of bars show the proportion of assays in which the indicated model of bias was present more frequently, but without a clear-cut dominance. Figure [3](#Fig3){ref-type="fig"} shows the *proportion of plates* per screening category affected by additive bias, multiplicative bias, an undetermined type of bias, or when no bias has been detected. An undetermined type of bias has been identified when spatial bias was detected on the plate by the Mann-Whitney *U* test and both *p-*values, corresponding to the additive and multiplicative models, were significant. The presented results suggest that the type of plate-specific bias largely depends on the screening category. For instance, additive bias is dominant in the following screening categories: HTS homogeneous (44.6%), HTS microorganism (51.8%) and SMM (42.1%) assays. The multiplicative bias is dominant in: HTS cell-based (36.5%) and HTS gene expression (45.5%) assays. However, all three HCS categories (area, intensity and cell count) were mostly bias free (Fig. [3](#Fig3){ref-type="fig"}). Figure [4](#Fig4){ref-type="fig"} presents the distribution of additive and multiplicative biases per assay, and per screening category. The *proportion of assays* affected either by additive bias, or by multiplicative bias, or by undetermined type of bias was represented for each screening category. Here, we also determined the proportions of assays in which the detected bias trend was dominant. The additive trend was considered dominant if there were twice or more biased plates better fitting the additive than the multiplicative bias model (and vice-versa for the multiplicative trend). The results presented in Fig. [4](#Fig4){ref-type="fig"} suggest that in most cases one type of spatial bias was dominant in the ChemBank screens, the only exception here being the assays of the HCS count category affected by additive bias. Identifying spatial bias in the McMaster Data Mining and Docking Competition HTS Test assay {#Sec6} ------------------------------------------------------------------------------------------- To illustrate the application of spatial bias correction methods, we carried out the assessment and correction of both assay-specific and plate-specific biases present in the McMaster Data Mining and Docking Competition HTS Test assay examined in several studies^[@CR8],[@CR11],[@CR13]^. This assay was obtained during a screen of 50,000 diverse small molecules against a single target, dihydrofolate reductase enzyme of *E. coli*^[@CR19]^. The McMaster Test assay consisted of a series of 625 plates of size (8 × 12). Each of the assay plates was screened twice. The first and the last columns of each plate contained controls that were used to normalize raw measurements. The remaining 80 wells contained different compounds intended to inhibit the dihydrofolate reductase of *E. coli*. Hit distribution surfaces, representing hit counts per well location, were obtained for raw (Fig. [5a](#Fig5){ref-type="fig"}) and corrected (Fig. [5b](#Fig5){ref-type="fig"}) McMaster measurements using the HTS Corrector software^[@CR11]^. The positional assay-wide pattern can be observed in raw data as the left-sided wells (they are in red) of the raw hit distribution surface overestimate the number of hits, while the right-sided wells (they are in blue) underestimate it. Additive plate-specific bias correction, followed assay-specific well correction, allowed us to remove spatial bias from raw McMaster data and disrupt the original hit count patterns (Fig. [5b](#Fig5){ref-type="fig"}). The *χ*^2^ goodness-of-fit test carried out for the raw McMaster hit distribution surface (Fig. [5a](#Fig5){ref-type="fig"}) at the significance level *α* = 0.01 returned the value of *χ*^2^(79) = 416.23 (with the critical value of 111.14), whereas the value of the *χ*^2^ statistic for the corrected hit distribution surface (Fig. [5b](#Fig5){ref-type="fig"}), computed following both plate-specific and assay-specific bias corrections, was equal to *χ*^2^(79) = 70.81, showing a much more homogeneous hit distribution pattern (where 79 was the number of degrees of freedom). The *χ*^2^ goodness-of-fit test can be used in screening technologies to assess the deviation of the hit distribution surface from the expected (i.e., plane) surface^[@CR11]^.Figure 5Hit maps showing the presence of spatial bias in the McMaster Test assay screened during the McMaster Data Mining and Docking Competition^[@CR13]^: (**a**) hit distribution surface for raw data, (**b**) hit distribution surface corrected both plate and assay-wise, (**c**) Plate 428 raw measurements, and (**d**) Plate 428 corrected measurements. Control columns 1 and 12 are not shown here. Higher hit counts (panels a and b) and intensity levels (panels c and d) are in red; lower hit counts (panels a and b) and intensity levels (panels c and d) are in blue. The hit selection threshold of *µ*-2*σ* was used to compute the hit distribution surfaces. The Mann-Whitney *U* test carried out to detect plate-specific spatial bias suggested that 377 McMaster plates were affected by systematic error, and 873 of them were clean. Error detection was done at the significance level *α* = 0.01. Plate-specific spatial bias was corrected using the additive PMP algorithm, as suggested by our method. The exact values of the raw and corrected measurements of Plate 428 and the raw and the corrected hit distribution surfaces are reported in Supplementary Tables [3](#MOESM1){ref-type="media"}--[6](#MOESM1){ref-type="media"}. Moreover, plate-specific spatial bias can also be observed within certain plates of the McMaster Test assay. For example, a clear underestimation of the true experimental measurements has occurred in row *H* of Plate 428 of this assay (Fig. [5c](#Fig5){ref-type="fig"}). The application of our algorithm presented in the Methods section allowed us to determine that the additive bias model is significant in this case and that it fits the data better than the multiplicative bias model. Using the additive PMP algorithm, we were able to remove systematic error from row *H* without changing the rest of the plate's measurements (Fig. [5d](#Fig5){ref-type="fig"}). Moreover, the Q-Q plots in Fig. [6a](#Fig6){ref-type="fig"} and b show that the spatial bias correction leads to a much better identification of hits and outliers. Figure [6a](#Fig6){ref-type="fig"} presents the Q-Q plot for the raw measurements of Plate 428 of the McMaster Test assay. The hit, located in well (*E*,3) and highlighted by a blue circle in Fig. [6a](#Fig6){ref-type="fig"}, and the outlier, located in well (*E*,11) and highlighted by a red circle in this figure, are not well separated from the rest of the data (see also Supplementary Table [5](#MOESM1){ref-type="media"} reporting the raw measurement values of this plate). Thus, both the hit and the outlier cannot be clearly identified using conventional hit selection methods. After the application of the Mann-Whitney *U* test to raw measurements of Plate 428, row *H* was flagged as biased and then corrected plate-wise using the additive PMP algorithm. Following the plate-specific correction by additive PMP, both the hit and the outlier became much better separated from the rest of the data of this plate (see Fig. [6b](#Fig6){ref-type="fig"} and Supplementary Table [6](#MOESM1){ref-type="media"} reporting the corrected measurement values of Plate 428) and can be now better identified using conventional hit selection methods.Figure 6Q-Q plots for McMaster's Plate 428 from the McMaster Data Mining and Docking Competition HTS Test assay^[@CR13]^ before (**a**) and after (**b**) the correction of additive spatial bias. All control wells (columns 1 and 12 of each plate) were excluded from the analysis. Low values (i.e., false positives) appearing in row *H* prevent a clear-cut identification of the hit located in well (*E*,3) (panel a of the figure; see also Supplementary Table [5](#MOESM1){ref-type="media"} for the exact values of the raw HTS measurements). After the plate-specific correction by additive PMP, the hit appearing in well (*E*,3) becomes much better separated from the rest of the measurements of Plate 428 (panel b of the figure; see also Supplementary Table [6](#MOESM1){ref-type="media"} for the exact values of the corrected HTS measurements). The same trend is maintained for the outlier (i.e., a high measurement value in the McMaster inhibition assay) located in well (*E*,11). Discussion {#Sec7} ========== Over the past decade, a number of novel screening technologies for processing large-scale binding assays have been developed to address the growing needs of drug discovery campaigns. Spatial bias remains one of the major hurdles of most of these technologies, potentially resulting in the identification of false positive and/or false negative hits. Our main objective was to assess the prevalence of spatial bias in experimental data generated by different high-throughput screening technologies. First, we should distinguish between assay-specific and plate-specific types of spatial bias. Assay-specific bias affects all the plates, or a series of continuous plates (i.e., batch effect), of a given assay. Plate-specific bias manifests itself by a spatial pattern that appears within a given plate only. Using experimental data from the largest public small-molecule screens database, ChemBank^[@CR15]^, involving three popular screening technologies -- high-throughput screening, high-content screening and small-molecule microarrays, and eight of their categories, we found that both assay and plate-specific biases affect the data of most of high-throughput assays. We showed that both assay-specific and plate-specific biases can be recognized and removed successfully from experimental screening data using appropriate statistical methods. Second, researchers should pay particular attention to the nature of spatial bias that can be additive, multiplicative or undetermined (i.e., when the presence of bias was detected by a statistical test, but both additive and multiplicative bias models were not significant). The presence of assay-specific bias, either additive or multiplicative, in a given well location can be detected by using the nonparametric Mann-Whitney U test. It is worth noting that Z-score normalization adequately handles both additive and multiplicative types of assay-specific bias when applied to the measurements of a given well location (see the Methods section). However, the case of plate-specific bias is more complicated. The Mann-Whitney U test can be used first to detect the presence of spatial bias in rows and/or columns of a given plate. Then, we can determine whether the detected bias is additive or multiplicative, using the additive and multiplicative versions of the Partial Median Polish (PMP) algorithm (see the Methods section). Afterwards, the Kolmogorov-Smirnov two-sample test should be applied to compare the distributions of unbiased and corrected (by additive and multiplicative PMP) measurements and the *p*-values should be computed for both bias models. The selection of the best-fitting model can be done by comparing the obtained *p*-values. Considering the example of the McMaster Test assay, we showed that both assay and plate-specific types of spatial bias can be detected and eliminated from screening data, thus improving measurement accuracy and minimizing the risk of misidentifying hits and outliers. It is important to note that the presented methodology is of a generic nature and can be used by life scientists involved in the analysis of current or next-generation high-throughput assays consisting of a large series of plates processed in sequence. Our simulation study (see the section Simulation study) indicate that the B-score^[@CR7]^ method is not suitable for removing assay-specific spatial bias and that Well Correction^[@CR11]^ cannot be used to eliminate plate-specific spatial bias. Thus, these methods should be used only when the presence of bias in the data has been confirmed by an appropriate statistical test (e.g., Mann-Whitney *U* test). The main advantage of our method is that it can cope with both of these biases (i.e., assay and plate-specific). However, the proposed method still needs to be used with caution. Before using it for data correction, the researcher must make sure that all tested compounds are randomly distributed both across and within HTS plates. If the randomization condition is not satisfied, some well locations or some areas on the plates can correctly lower or higher signals, in which case data normalization can become detrimental. Alternative approaches could also be used to remove spatial bias from data produced by high-throughput screening technologies. For example, one can consider generalized spatial linear mixed models (GLMMs), which are commonly used in public health, ecological and epidemiological studies dealing with geographical sampling^[@CR23],[@CR24]^. GLMMs explicitly model space-dependent bias through a specific spatial correlation function. In the linear mixed model, all observations are assumed to have a spatial correlation structure. A GLMM effectively postulates that the correlation increases as observations are located closer together, whereas observations at distances greater than the range from one another are uncorrelated. The major reason that GLMMs have not been used for bias modeling in HTS is that only a few columns and rows of a given HTS plate are generally biased, whereas mixed models assume that spatial correlation can be modeled in some way throughout the whole plate surface. Moreover, the biased columns and rows are not necessarily spatially related and may not be located at the same part of a given plate. For example, a column or row effect often results from a robot malfunction that is unique to that column or row^[@CR25]^. Furthermore, in order to accurately model the bias structure in HTS assays, spatial covariates should take into account the dependences between biased rows and columns (i.e., plate-specific bias) as well as between biased well locations (i.e., assay-specific bias) of a given assay. To the best of our knowledge, this type of GLMM still needs to be developed, and this could be an interesting topic for future investigation. It is also known that fitting a GLMM to a large dataset can be computationally expensive (i.e., due to inversion of large matrices growing with the number of sampling units). In contrast, the time complexity of the additive and multiplicative PMP algorithms is linear on the matrix dimensions (see the Methods section). These algorithms require O(*mnI*) time, where *m* and *n* are the plate dimensions and *I* is the number of iterations necessary for convergence. In practice, these algorithms converge after a few iterations. Methods {#Sec8} ======= Correction of assay-specific bias {#Sec9} --------------------------------- Assay-specific bias correction methods rely on both plate-wise and well-wise normalization techniques^[@CR11]^. First, each plate of the given assay is subject to the robust (outlier-resistant) *Z*-score normalization:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{x}}_{ijp}=\frac{{x}_{ijp}-me{d}_{p}}{MA{D}_{p}},$$\end{document}$$where *x*~*ijp*~ is the raw measurement in well (*i,j*) of plate *p*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{x}}_{ijp}$$\end{document}$ is the normalized value of *x*~*ijp*~, and *med*~*p*~ and *MAD*~*p*~ are the median and the median absolute deviation of plate *p*'s measurements, respectively. Second, the Mann-Whitney *U* test can be carried out for each well location, comparing its sum of ranks to the sum of ranks of the rest of the assay measurements. Third, the well locations in which the presence of spatial bias was detected by the Mann-Whitney *U* test (the significance level *α* = 0.01 was used in this work) can be corrected via the traditional *Z*-score normalization, defined as follows:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{x}}_{ijp}=\frac{{x}_{ijp}-{\mu }_{ij}}{{\sigma }_{ij}},$$\end{document}$$where *x*~*ijp*~ is the raw measurement in well (*i,j*) of plate *p*, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{x}}_{ijp}$$\end{document}$ is the normalized value of *x*~*ijp*~, and *μ*~*ij*~ and *σ*~*ij*~ are, respectively, the mean and the standard deviation of the measurements of well location (*i,j*) (computed over all plates of the assay). Note that the traditional *Z*-score normalization ([2](#Equ2){ref-type=""}) adequately handles both additive and multiplicative kind of biases when applied to the measurements of a given well location. Additive biases are removed by the subtraction in the numerator; the standard deviation does not change if the mean is shifted. In case of multiplicative biases, both the numerator and denominator are multiplied by the same value, and thus the computed score remains the same. Hence, *Z*-score can successfully correct both additive and multiplicative types of assay-specific spatial bias over the measurements of a given well location. Importantly, if the presence of assay-specific bias was detected in some well locations of the assay, then all of its well locations should be *Z*-score normalized in order to keep the assay data on the same scale. Correction of plate-specific bias {#Sec10} --------------------------------- Plate-specific biases that affect rows and columns of a given plate *p* can fit either the additive ([3](#Equ3){ref-type=""}) or the multiplicative ([4](#Equ4){ref-type=""}) bias model:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x{^{\prime} }_{ijp}={x}_{ijp}+{r}_{ip}+{c}_{jp}+{\varepsilon }_{ijp},$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x^{\prime} }_{ijp}={x}_{ijp}\times {r}_{ip}\times {c}_{jp}+{\varepsilon }_{ijp},$$\end{document}$$where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$x{^{\prime} }_{ijp}$$\end{document}$ is the observed raw measurement in well (*i,j*) of plate *p*, *x*~*ijp*~ is the real measurement value in well (*i,j*) of plate *p*, *r*~*ip*~ is the value of systematic error (i.e., spatial bias) affecting row *i* of plate *p*, *c*~*jp*~ is the value of systematic error affecting column *j* of plate *p*, and *ε*~*ijp*~ is the random error affecting well (*i,j*) of plate *p*. Partial mean polish, which is a variation of Tukey's median polish, iteratively removes spatial bias from biased rows and columns^[@CR8]^. In our work, the presence of bias in rows and columns of each plate *p* was determined using the Mann-Whitney *U* test (with *α* = 0.01). The biased and unbiased rows and columns are identified during the bias detection procedure. The process starts by assuming that all measurements of a given plate are unbiased. The method then compares in turn the measurements of each row and each column against the rest of the plate's measurements, excluding those of the tested row or column and those that have been already identified as biased using the Mann-Whitney *U* test. The obtained *p*-value is retained for each row and each column; the plate's row or column corresponding to the smallest *p*-value is then considered biased if this smallest *p*-value is larger than the selected threshold. The procedure is applied independently to all plates of a given assay. A similar procedure is carried out to identify biased well locations (see above). Adequate model selection for plate-specific bias {#Sec11} ------------------------------------------------ The goodness-of-fit of the additive and multiplicative bias models can be assessed by comparison of the corrected results with unbiased data. For each plate, the process starts with three data sets: (1) measurements from unbiased rows and columns, (2) measurements corrected using the additive correction, and (3) measurements corrected using the multiplicative correction. Afterwards, we carry out the Kolmogorov-Smirnov two-sample test in order to compare the distributions of unbiased measurements (1) and those corrected using the additive correction model (2), then the distributions of unbiased measurements (1) and those corrected using the multiplicative correction model (3). The obtained *p*-values are then compared to each other to select the best fitting model. The significance level *α* = 0.01 was used in our study. The exact algorithm can be presented as follows: Perform the Mann--Whitney *U* test on each individual plate of the assay (i.e., plate-wise correction) to identify biased rows and columns: If (spatial bias is detected in rows or columns of the given plate), then:Apply the additive PMP algorithm^[@CR8]^ (see also the next section) to correct raw plate's measurements;Apply the multiplicative PMP algorithm^[@CR26]^ (see also the next section) to correct raw plate's measurements;Apply the Kolmogorov-Smirnov two-sample test on the corrected plates yielded by the additive and multiplicative PMP algorithms. Compute the corresponding *p*-values;If both additive and multiplicative *p-*values are lower than the selected significance level *α*, then the bias model for this plate is undetermined. Otherwise, apply the correction algorithm that provides the largest *p*-value (i.e., additive or multiplicative PMP) to remove spatial bias from the measurements of the given plate. Additive and multiplicative Partial Mean Polish (PMP) algorithms {#Sec12} ---------------------------------------------------------------- The additive version of the PMP algorithm is presented in detail in the paper of Dragiev and colleagues^[@CR8]^. Here we show the changes that should be made to the algorithm to adapt it for removing multiplicative biases^[@CR26]^. The main advantages of both additive^[@CR8]^ and multiplicative^[@CR26]^ PMP algorithms are that they correct the biased measurements only and keep the original and corrected data on the same scale, ensuring that the mean of the corrected rows and columns is equal to the mean of unbiased data.Let $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R=\{{r}_{1},{r}_{2},\mathrm{...},{r}_{p}|0\le p\le m\}$$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$C=\{{c}_{1},{c}_{2},\mathrm{...},{c}_{s}|0\le s\le n\}$$\end{document}$ be the sets of biased rows and columns of plate $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$P(m\times n)$$\end{document}$, calculate the mean *µ* of all unbiased measurements of *P*:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mu =\frac{\sum _{i\notin R,j\notin C}{x}_{ij}}{(m-p)(n-s)}.$$\end{document}$$For each biased row *i* in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I=\{1\le i\le p\}$$\end{document}$, compute the mean value, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{{r}_{i}}$$\end{document}$, of row *r*~*i*~: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{{r}_{i}}=\frac{1}{n}\sum _{j=1}^{n}{x}_{{r}_{i}j}$$\end{document}$, and calculate the estimate of the row error, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{e}}_{{r}_{i}}$$\end{document}$, using the equations: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{e}}_{{r}_{i}}={\mu }_{{r}_{i}}-\mu $$\end{document}$(additive model) and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{e}}_{{r}_{i}}=\frac{|{\mu }_{{r}_{i}}|}{\mu }$$\end{document}$(multiplicative model).For each biased column *j* in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J=\{1\le j\le s\}$$\end{document}$, compute the mean value, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{{c}_{j}}$$\end{document}$, of column *c*~*j*~: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{{c}_{j}}=\frac{1}{m}\sum _{i=1}^{m}{x}_{i{c}_{j}}$$\end{document}$, and calculate the estimate of the column error, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{e}}_{{c}_{j}}$$\end{document}$, using the equations: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{e}}_{{c}_{j}}={\mu }_{{c}_{j}}-\mu $$\end{document}$ (additive model) and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{e}}_{{c}_{j}}=\frac{|{\mu }_{{c}_{j}}|}{\mu }$$\end{document}$ (multiplicative model).For all rows affected by spatial bias, adjust their measurements using the error estimates determined in Step 2, i.e., for all *i* in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I=\{1\le i\le p\}$$\end{document}$ and all *j* such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\le j\le n$$\end{document}$ proceed as follows: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{r}_{i}j}={x}_{{r}_{i}j}-{\hat{e}}_{{r}_{i}}$$\end{document}$(additive model) and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{{r}_{i}j}=\frac{{x}_{{r}_{i}j}}{{\hat{e}}_{{r}_{i}}}$$\end{document}$(multiplicative model).For all columns affected by spatial bias, adjust their measurements using the error estimates determined in Step 2, i.e., for all *j* in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$J=\{1\le j\le s\}$$\end{document}$ and all *i* such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1\le i\le m$$\end{document}$ proceed as follows: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{i{c}_{j}}={x}_{i{c}_{j}}-{\hat{e}}_{{c}_{j}}$$\end{document}$(additive model) and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{i{c}_{j}}=\frac{{x}_{i{c}_{j}}}{{\hat{e}}_{{c}_{j}}}$$\end{document}$ (multiplicative model).If $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sum _{i=1}^{p}|\,\,{\hat{e}}_{{r}_{i}}\,|+\sum _{j=1}^{s}|\,{\hat{e}}_{{c}_{j}}| > \varepsilon $$\end{document}$, then go to Step 2, otherwise stop the algorithm. Here *ε* is a small fixed positive threshold. It is worth noting that the median can be used instead of the mean in both additive and multiplicative PMP algorithms. The use of the median usually increases the method's robustness against outliers. Data availability {#Sec13} ----------------- High-throughput screening assays analyzed in this study are available in the ChemBank repository^[@CR15]^. The ChemBank IDs of these assays are indicated in Supplementary Information. The McMaster Data Mining and Docking Competition assay data are available at: <http://www.info2.uqam.ca/~makarenkov_v/HTS/home.php>. Electronic supplementary material ================================= {#Sec14} SUPPLEMENTARY INFORMATION **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Electronic supplementary material ================================= **Supplementary information** accompanies this paper at 10.1038/s41598-017-11940-4. We are thankful to Dr. Iurie Caraus and an anonymous reviewer for critical discussion and helpful comments. This work was funded by Natural Sciences and Engineering Research Council of Canada \[Grant Number 249644\] and le Fonds Québécois de la Recherche sur la Nature et les Technologies \[Grant Number 173539\]. V.M. and R.N. have designed the study. B.M. and V.M. carried out the experimental study. B.M., R.N. and V.M. wrote the article. All authors discussed the results and commented on the manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

1. Introduction {#sec1-sensors-17-00869} =============== There are an estimated 2 million hand amputees in the United States and approximately the same in Europe. More than 200,000 new amputation surgeries are performed each year and approximately 10,000 children receive amputations resulting in a lifelong disability \[[@B1-sensors-17-00869]\]. Amputee patients are supported by a long standing research and development of prosthetic devices, which can be divided in passive and active ones. Passive prostheses have only a cosmetic purpose and do not support any of the hand functionalities. Modern active prostheses are externally powered and feature advanced grasping and control functionalities. The first active prostheses were composed by a split hook able to perform a power grasp, restoring just one Degree of Freedom (DoF). During the last years, the advancement of technology and prosthesis design paved the way for multifinger artificial hands \[[@B2-sensors-17-00869],[@B3-sensors-17-00869]\]. These are poliarticulated systems that independently actuate each finger, reproducing most of the movements of a real hand in daily living activities. They offer high reliability and robustness, but do not provide an intuitive control interface. Electromyography (EMG) is widely used in assistive technologies to sense the muscular activities in the forearm \[[@B4-sensors-17-00869]\] and it is a preferred mode of interaction to control prosthetic devices. However, State-of-the-Art (SoA) systems \[[@B2-sensors-17-00869],[@B5-sensors-17-00869],[@B6-sensors-17-00869]\] use simple thresholds to detect the user's activity and the prosthesis commands are encoded in predefined sequences of extensions and flexions of the residual forearm muscles \[[@B7-sensors-17-00869]\]. This approach does not offer a natural interface, it requires a long learning curve and high cognitive effort during use. Furthermore, several studies compare these SoA controllers against natural control strategies, indicating that the design of intuitive controllers can improve the usability and the functionality of a prosthesis \[[@B8-sensors-17-00869],[@B9-sensors-17-00869],[@B10-sensors-17-00869],[@B11-sensors-17-00869]\]. Massive research efforts are spent to investigate solutions for a more intuitive prosthetic control, using techniques based on soft synergies \[[@B12-sensors-17-00869]\] or pattern recognition \[[@B13-sensors-17-00869]\]. Both approaches exploit EMG sensing of the remaining forearm muscles as the main input for the user-prosthesis interface. A prosthetic hand based on synergies offers only one gesture, i.e., an adaptive power grip, however, it provides a dynamic control of the grasping force and the ability to adapt the hand's grasp to the object in use. On the other hand, the use of pattern recognition and machine learning techniques allows to recognize the intention to perform different gestures, such as pointing and different types of power and precision grips. Development and optimization of gesture recognition techniques for the control of poliarticulated prostheses is today an active research topic, even though the analysis is mainly performed from a machine learning perspective \[[@B14-sensors-17-00869],[@B15-sensors-17-00869],[@B16-sensors-17-00869]\]. Such studies focus on the offline analysis and evaluation of the algorithm's recognition accuracy on all the samples of a given dataset. However, this is not sufficient for the exhaustive evaluation of a complete prosthetic controller, which requires a system approach involving sensing, classification and timely prosthesis actuation on a wearable system for a correct execution of gestures. Even if integration advancement enables single wearable devices to have unprecedented computational power, the case of EMG gesture recognition still presents a processing power challenge to these devices, in particular in the training phase of the classifier and therefore requires to be supported by a gateway device. In this paper, we present the design of a Body Sensor Network (BSN) for the control of a poliarticulated prosthetic hand. This work is the final result of a two year collaboration project between two research partners (University of Bologna and Fondazione Bruno Kessler) and the INAIL prosthetic center, a leading European institute for prosthetics and rehabilitation. The work covers the whole system development, starting from the EMG data collection on healthy and amputee subjects \[[@B17-sensors-17-00869]\] and going through the study and the embedded implementation of an efficient pattern recognition algorithm. We focus on the design of a natural method of interaction between amputees and prosthesis, which is integrated with a robust real-time control strategy and actuation of a multifinger hand. The wearable node is part of the BSN and it is connected via Bluetooth to a personal gateway (e.g., a smartphone) for the personalized training and tuning of the recognition capabilities. The gateway is a second node that can also open the BSN to context-aware information and make the BSN cloud assisted \[[@B18-sensors-17-00869],[@B19-sensors-17-00869],[@B20-sensors-17-00869]\]. Furthermore, the gateway node enables periodic classifier retraining in daily life conditions, which is a process that does not require real-time as the recognition in the wearable node. Such approach is in line with the BSNs and wearables trend of embedding processing near the sensor and combining with cloud technologies to enable big data analysis for the IoT scenarios \[[@B21-sensors-17-00869],[@B22-sensors-17-00869],[@B23-sensors-17-00869],[@B24-sensors-17-00869],[@B25-sensors-17-00869]\]. The performance of the system has been tested in terms of *end-to-end* recognition ratio, evaluating the accuracy and timing of the system to recognize and execute complete hand gestures, as performed by amputees and healthy subjects. The system classifies and executes four hand gestures (open hand, closed hand, precision grasp and point index) with an end-to-end error rate under 2%. It has been tested on five healthy subjects and four amputee patients. Moreover, we validated our approach against the NINAPRO database that represents the largest and most complete dataset for EMG gesture recognition to date \[[@B26-sensors-17-00869]\]. The proposed system is designed with Commercial Of The Shelf (COTS) components and open source development tools, to make it easy to replicate all the results and to provide an useful platform for future research and product development. The reminder of this paper is organized as follows: [Section 2](#sec2-sensors-17-00869){ref-type="sec"} contains background information and related work, [Section 3](#sec3-sensors-17-00869){ref-type="sec"} reports a detailed description of the proposed system and approach and [Section 4](#sec4-sensors-17-00869){ref-type="sec"} presents its experimental evaluation. Finally, in [Section 5](#sec5-sensors-17-00869){ref-type="sec"} we draw the conclusions. 2. Background and Related Work {#sec2-sensors-17-00869} ============================== The hand prosthesis acts as functional replacement of the missing limb in amputees. Technological advancements over the years allowed significant improvements in the restoration of the functionalities of a normal human hand. The first example of these active devices are the body-powered prostheses, capable of restoring basic tasks such as opening and closing a terminal device \[[@B27-sensors-17-00869]\]. In such devices, motion is transmitted to the prosthesis mechanically, controlling the artificial hand with the abduction of the shoulder or the healthy wrist flexion. These devices are often used because they are simple, robust, and relatively inexpensive, even though they do not re-establish a complete functionality of the hand. Electrically powered or active prostheses \[[@B28-sensors-17-00869]\] are advantageous w.r.t. body-powered ones because they require less user effort, as movement is actuated with DC motors \[[@B29-sensors-17-00869]\]. They can be controlled through a variety of means such as force sensors \[[@B30-sensors-17-00869]\], acoustic interfaces \[[@B31-sensors-17-00869]\] and EMG signals \[[@B32-sensors-17-00869]\]. Such devices restore some functionality to amputees, but their control is typically limited to only one or two DoFs. Hence, the research on upper limb prostheses addresses multiple challenges in the development of anthropomorphic prostheses, ranging from the hand design \[[@B33-sensors-17-00869]\] to the fingers' actuation \[[@B34-sensors-17-00869]\], in search of the best tradeoff between the complexity of the design and its usability and reliability. Recently, the biomechatronic design of these prostheses has been refined, and advanced and reliable multifinger prostheses appeared on the market. The Bebionic 3 \[[@B6-sensors-17-00869]\], the Michelangelo Hand \[[@B5-sensors-17-00869]\] and the I-limb \[[@B2-sensors-17-00869]\] are poliarticulated myoelectric hands with embedded control, which feature from 7 to 24 different hand positions. The fingers are independently driven by dedicated DC motors and, exploiting their robust design, these hands can carry up to 45 kg. A dedicated wireless link allows clinicians to monitor these devices and program the user-dependent parameters, such as selection of the control sequences and grip thresholds. These SoA devices are enabling complex functionalities, including proportional grasps and configurable gestures, but their control strategies are still based on non-intuitive codification of gestures in muscle contractions. In this scenario, the deployment of a natural control strategy is a key element; pattern recognition and machine learning techniques are capable to restore a more intuitive control of the artificial hand \[[@B35-sensors-17-00869]\]. Gesture recognition control is based on the assumption that the muscular activation patterns are repeatable among different executions of the same hand gesture. Hence, through the pattern recognition of EMG signals, we can classify hand gestures and grips. The scientific literature proposes a wide analysis of the best sensor configurations \[[@B36-sensors-17-00869]\], classification algorithms \[[@B37-sensors-17-00869],[@B38-sensors-17-00869]\] and actuation controls \[[@B39-sensors-17-00869]\]. Supervised machine learning algorithms, such as Hidden Markov Models (HMM) \[[@B40-sensors-17-00869]\], Linear Discriminant Analysis (LDA) \[[@B41-sensors-17-00869]\], Artificial Neural Networks (ANN) \[[@B42-sensors-17-00869]\] and Support Vector Machines (SVM) \[[@B43-sensors-17-00869]\] provide comparable results of up to 90% of accuracy for the classification of 5 to 10 gestures. EMG-based gesture recognition based on deep learning is starting to be investigated and preliminary results do not show a very strong advancement w.r.t. the SoA \[[@B44-sensors-17-00869]\]. Furthermore, the related classification algorithms are computationally demanding and not suitable for a real-time embedded implementation \[[@B45-sensors-17-00869]\]. One of the main issues of the pattern recognition approach is caused by the variability of the EMG signal during the arm movements. Indeed, differences in the activation patterns caused by the variability of the arm position can cause significant losses in the classification accuracy of the EMG signals. To cope with this, several solutions are proposed in literature \[[@B46-sensors-17-00869],[@B47-sensors-17-00869]\], based on sensor fusion or algorithm tuning. Detailed comparison of acquisition setups and classification algorithms is widely available in literature \[[@B48-sensors-17-00869]\] and it is out of the scope of this work. We focus on the use of the SVM algorithm, which provides an optimal accuracy Vs complexity trade-off and it is suitable for real-time embedded implementation and tight integration with the hand control. Moreover, in our previous work we demonstrated the robustness of the SVM approach against variation of arm postures and electrodes number and positioning \[[@B17-sensors-17-00869],[@B49-sensors-17-00869]\]. Recently, an alternative research solution based on the use of *soft synergies* has been proposed \[[@B12-sensors-17-00869],[@B50-sensors-17-00869]\]. This solution develops the design of a robotic hand with only opening and closing capabilities, but that adapts its grasp on the object being manipulated. Its primary goal is to restore a correct proportional control of the grasping force \[[@B51-sensors-17-00869]\], assuming that the round grip is the most used movement to grasp objects. The hand realized by \[[@B52-sensors-17-00869]\] further refines this approach with the principle of *adaptive synergies* and features 19 joints controlled by a single actuator. This approach focuses on the versatility of the mechanical design of the hand rather than on the EMG-based contraction codification or the gesture recognition It uses the EMG signal for the detection of the hand activation and as a proportional control of the grasping force. This solution is robust and powerful even though the hand can only open and close its adaptive grasp and the finger movements are not independent. Another interesting approach compares linear and non linear regression methods for proportional and simultaneous control of multiple DoF of the prosthesis \[[@B53-sensors-17-00869],[@B54-sensors-17-00869]\]. The aim of these works is to propose a method to overcome the limitations of a controller that manages only one DoF at time. The regression approach differs from the classification mainly because it provides a continuous output instead of a discrete output associated to a recognized gesture. Nevertheless, the presented setup is based on a large array of sensors and the experiments included only flexion, extension and rotation of the wrist. There is a lack of complete devices for embedded gesture recognition and control of prosthetics, which is due to the difficulties in performing system-level development and addressing problems ranging from signal acquisition and processing, embedded system design, applied prosthetics and machine learning. The work described in this paper intends to bridge this gap presenting the development and implementation of an intuitive embedded prosthesis controller, as the result of the collaboration between institutes with multidisciplinary competences. We propose a wearable BSN approach for the implementation of an effective and versatile hand controller: an embedded node is dedicated to real-time sensing, processing and actuation, while a portable personal gateway (e.g., smartphone or tablet) provides advanced functionalities such as training and personalization of the recognition algorithms. Our work analyzes the EMG signal acquisition and integrates the use of SVM-based gesture recognition with a real-time controller to provide accurate and robust actuation of a poliarticulated prosthetic hand. The proposed approach is tailored for the implementation on an embedded microcontroller (MCU) and it is evaluated not only validating the recognition algorithm, but also on its end-to-end and timely gesture actuation capabilities. 3. Materials and Methods {#sec3-sensors-17-00869} ======================== [Figure 1](#sensors-17-00869-f001){ref-type="fig"} presents the architecture of the proposed system, which is composed by: (1) an elastic armband with 4 EMG sensors; (2) a wearable sensing node responsible for data acquisition and classification, prosthesis actuation and Bluetooth communication; (3) a poliarticulated prosthetic hand and (4) a personal gateway for data acquisition, recognition algorithm training and customization of system parameters. This heterogeneous BSN architecture aims to maximize the energy efficiency of the node. The signal acquisition, the pattern recognition algorithm and the closed loop control of the hand are executed in real-time on the wearable node, while the algorithm tuning and the SVM training, which do not require real time execution are provided by the personal gateway. Offline studies of EMG gesture recognition typically report recognition accuracy on all collected samples, regardless of their collocation during a gesture. When actuating a prosthesis, only one output per executed gesture is needed and the gesture decision should be made as soon as possible at the start of its execution, while the classification of subsequent samples becomes unnecessary. However, the transient phase at the onset of a gesture is more difficult to classify when compared to a stable on-going contraction. Hence, a robust implementation of a gesture controller has to cope with the initial uncertainty in the recognition and to provide a timely decision for a correct actuation of the hand. The proposed solution integrates sample-level SVM classification with a high-level Finite State Machine (FSM) to produce accurate and robust control of the prosthesis. For mechanical constraints, the prosthetic hands start every gesture from a reset position, i.e., open hand. Hence, we included this transition to be performed between each executed gesture and used it to improve the robustness and usability of the system. Moreover, during the onset of a gesture, the output of the sample-level recognition is analyzed with a majority voting approach to limit the errors due to the signal transitions and to converge to a decision within a specified time window. 3.1. EMG Signal Acquisition {#sec3dot1-sensors-17-00869} --------------------------- The EMG signal measures the electrical activation of the muscular fibers, generated by the depolarizations of muscular cell membranes named Action Potentials (APs). APs are the result of the propagation of neural impulses from the brain towards the target muscle cells, as a muscular contraction occurs \[[@B55-sensors-17-00869]\]. Surface EMG sensors are composed by two conductive plates connected to the inputs of a differential amplifier, thus sensing the aggregated APs of the underlying muscles. The bandwidth of the EMG signals stays within 2 kHz, while the signal amplitude is contained in the ±10 mV range depending on the diameter of the muscles and on the distance from the sensing elements. In prosthetic applications, EMG acquisition is performed with active analog sensors, to maximize the signal quality reducing noise caused by motion artifacts, fibers crosstalk and floating ground of the human body. Active EMG sensors perform an analog conditioning of the signal using bandpass filtering and an Instrumentation Amplifier (IA) with a high gain stage. In our implementation, we use the Ottobock 13E200, a family of pre-amplified sensors with single ended output. In such sensors, the EMG signal is filtered, amplified and integrated to reach an output span of 0--3.3 V, ideal for acquisition with the single ended stage ADC integrated in an embedded MCU. Furthermore, the active sensors extract the envelope of the raw EMG signal exploiting the HW circuitry and for this reason, in our application no further feature extraction of the EMG signal is required. 3.2. Wearable Controller Node {#sec3dot2-sensors-17-00869} ----------------------------- The main component of the proposed setup is a custom wearable controller board, whose block diagram is illustrated in [Figure 1](#sensors-17-00869-f001){ref-type="fig"} (top). It is designed on a 4-layer PCB and includes a 32 bit MCU, circuitry for EMG signal acquisition, for the actuation and control of the hand prosthesis and a Bluetooth transceiver for the communication with a host device. The board tolerates a single-voltage power supply from 5.5 to 8 V, to easily fit on commercial prosthetic systems that use standard battery cells varying from 7.2 to 8 V. The on-board linear regulators provide stable output voltages of 3.3 V and 5 V employed for the different sub-systems. The system is based on an ARM Cortex M4 MCU from NXP. The presence of two independent 16-bit SAR ADCs (*ADC0* and *ADC1*) and of a dedicated PWM output peripheral allow the control of the proposed prosthetic hand minimizing the need for external components and hence the board complexity. Data are acquired at a frequency 500 Hz, which has been shown to be sufficient for gesture recognition applications \[[@B49-sensors-17-00869]\]. On each channel, an RC low pass filters minimizes the high frequency electrical noise. A further resistive voltage divider protects the ADC's inputs, limiting the signal span to the 0--3.3 V range, while the output span of the Ottobock sensor is 0--5 V. The DC motors powering the finger actuation of the artificial hand are controlled by an integrated H-bridge driver (MC33926 by NXP), connected with the microcontroller, as described in [Figure 2](#sensors-17-00869-f002){ref-type="fig"} (left). The FB signal, which goes from the H-bridge to the MCU, returns a feedback on the current absorption of the DC motor, providing to the ADC a voltage signal that is proportional to the current being drawn. A finger movement is completed when it is completely open, completely closed or if it is halted in grasping an object. In such situations, the DC motor increases its current consumption measured through a shunt resistor and threshold trigger is used to stop the motor. A typical curve of the voltage provided by the H-bridge is reported in [Figure 2](#sensors-17-00869-f002){ref-type="fig"} (right) for different values of the PWM period. The hardware configuration is completed with a Bluetooth (BT) transceiver (SPBT2632C2A by STM), allowing the system to communicate with other devices on the body or in its vicinity. The BT standard was chosen for its easy use, optimal throughput and power consumption tradeoff \[[@B56-sensors-17-00869]\] and its versatility to enable communication with a variety of host devices (body gateways, mobiles phones, tablets or PCs). This bi-directional wireless interface is used to stream the acquired EMG data to the PC or to store on the device customized parameters and settings. Data streaming to a host device is employed in order to test the system and to acquire instances of gestures for offline analysis and for the training of the classification algorithm. The trained recognition models and further system settings are then sent back to the embedded device and stored in the MCU's Flash memory. 3.3. Classification Algorithm {#sec3dot3-sensors-17-00869} ----------------------------- The SVM is a supervised learning algorithm that improves the performance of logistic regression methods through the application of the large margin classification \[[@B57-sensors-17-00869]\]. The offline training phase of the algorithm uses labeled instances of data to calculate the optimal separation hyperplane (*maximum margin* hyperplane) between two classes of data through the solution of a convex optimization problem. Such separation plane is represented by a set of data vectors, named the Support Vectors (SVs), which belong to the borders of the two classes and they are used to classify new data instances. When the two classes are not linearly separable, the input data space can be mapped to a higher-dimensional space through a kernel function to allow an effective separation \[[@B59-sensors-17-00869]\]. Having two classes, denoted as $Cl_{1}$ and $Cl_{2}$, the formula of the decision function to classify a new input instance is: $$f\left( \mathbf{x} \right) = \sum\limits_{i = 1}^{N_{SV}}y_{i}\alpha_{i}K\left\langle \mathbf{x},\mathbf{s}_{i} \right\rangle - \rho\,\,\,\,\,\left\{ \begin{array}{r} {f\left( \mathbf{x} \right) > 0,\mathbf{x} \in Cl_{1}} \\ {f\left( \mathbf{x} \right) < 0,\mathbf{x} \in Cl_{2}} \\ \end{array} \right.$$ where $\mathbf{x} \in \mathbb{R}^{N_{F}}$ is the vector if input features, $\mathbf{s}_{i} \in \mathbb{R}^{N_{F}}$, $i = 1,...,N_{SV}$ are the support vectors, $\alpha_{i}$ are the support values, with $y_{i}$ denoting the class they reference ($y_{i} = + 1$ for $Cl_{1}$, $y_{i} = - 1$ for $Cl_{2}$) and $K\left\langle \, \cdot \,,\, \cdot \, \right\rangle$ denotes the kernel function. In our application, the input for the SVM classifier is the 4-dimensional vector of the EMG signals acquired by the electrodes ($N_{F} = 4$) and we used the Radial Basis Function (RBF) kernel. Since the training of a SVM classifier is computationally demanding for a low power microcontroller and it should be performed only at the setup of the recognition algorithm, it is possible to perform it offline on a PC. This allows the use of a graphical interface to visualize the training data and perform an accurate segmentation without imposing particular limitations on the system architecture. The calculated models are then stored on the MCU where the classification algorithm is executed for a real time recognition of the performed gestures. A diagram of the SVM training and recognition phases is illustrated in [Figure 3](#sensors-17-00869-f003){ref-type="fig"}a. The libSVM \[[@B60-sensors-17-00869]\] is a popular open source multiplatform library implementing the SVM algorithm, which includes the training and classification functions \[[@B60-sensors-17-00869]\]. The library implementation is adapted for the embedded platform, where the dynamic allocation of the memory structures is not recommended in the design of medical systems \[[@B61-sensors-17-00869]\]. Taking advantage of the pointer manipulation and arithmetic functions offered in C, we stored the SVs in a compact memory area to better exploit the structure of the libSVM's prediction function, based on nested *for* cycles iterating through the SVs. [Figure 3](#sensors-17-00869-f003){ref-type="fig"}b shows the MCU's memory structure and the static allocation of the stored parameters. The pseudocode of the execution of the SVM decision function for the classification of new input data is described in Algorithm 1. The first step is the computation of the RBF kernel mapping of the input vector and the SVs, where the products between the class labels and the support values ($C_{i} = \alpha_{i}y_{i}$) are pre-calculated and stored in the flash memory. These pre-computed coefficients allow to execute the complete decision function between two classes just with 2 nested *for* loops. Finally, the predicted label is the one that has totalized the highest number of binary classifications within all the combination of classes. In our body-area embedded implementation, the model parameters and the SV values and coefficients are stored maintaining the dependencies between the relative vector positions to access them in an efficient way. The vectors are correctly addressed exploiting the pointer algebra to substitute the allocation of a dynamic list of structures with the indexing of the Flash sector where the SVs and coefficients are stored. The memory requirements of the algorithm include the contribution of a fixed header of 20 Bytes containing the algorithm's type and configuration parameters and a variable component depending on the number of SVs, data features and recognized classes. Such a variable component has to be multiplied by the size of the used data type to compute the final memory footprint (e.g., 4 Bytes for single precision float). Algorithm 1 Multiclass SVM implementation 1. $C_{i} = y_{i}\alpha_{i}$ 2. $v\left\lbrack N_{Cl} \right\rbrack = \left\{ 0,0,...,0 \right\}$ 3. **function** Class = *svmpredict*($\mathbf{x}$) 4. **for** i = 1 to $N_{S}$ **do** 5.     $K\left\langle \mathbf{x},\mathbf{s}_{i} \right\rangle = exp\left( {- \frac{\left| \middle| \mathbf{x} \right. - \mathbf{s}_{i}{||}_{2}}{2\sigma^{2}}} \right)$ 6. **end for** 7. **for** j = 1 to $N_{Cl}$ **do** 8.     **for** k = j+1 to $N_{Cl}$ **do** 9.         $f\left( \mathbf{x} \right) = \sum_{i = 1}^{N_{S}}C_{i(jk)}K\left\langle \mathbf{x},\mathbf{s}_{i} \right\rangle - \rho$ 10.         **if** $f\left( \mathbf{x} \right) > 0$ **then** 11.            $v\left( j \right)$++ 12.         **else** 13.            $v\left( k \right)$++ 14.     **end for** 15. **end for** 16. Class ← index of $max\left( v \right)$ 3.4. Gateway Application {#sec3dot4-sensors-17-00869} ------------------------ The proposed system is completed with a mobile personal gateway, which is implemented on a smartphone. The personal gateway application, written in Java on Android, connects via BT to the wearable node and allows to perform a set of test and tuning operations. In particular, the embedded system implements three operation modes: Streaming Mode (SM), Update Mode (UM) and Classification Mode (CM). The transmission of a dedicated control string from the app to the board manages the transitions between the operation modes and the CM is the default mode used for stand-alone control of the prosthesis. SM is used to stream EMG data to the gateway device, useful to test the positioning and functionality of the system and to collect example gestures for a personalized training of the SVM algorithm. Finally, UM allows to update the trained SVM model on the device. The personalized setup of the system is performed under the supervision of a clinician and it requires the user to correctly position the sensors, test their activation and to collect a training dataset composed by a few repetition of each gesture to recognize. When the SVM model is computed, the app sets the system in UM and manages the model transfer to the MCU. This operation allows to correctly store the personalized model in the Flash memory of the MCU and use it for real-time classification. The communication protocol sends 3 types of packets containing the following: the general configuration values: SVM type, gamma parameter, the number of classes and the number of features.the model parameters: the $\rho$ parameter, the total number of SVs, and the SVs for each class.the SVs and their coefficients, sending one packet for each SV. When a packet has been sent, the interface waits the Acknowledgment (ACK) message sent from the board before sending a new packet. Each packet has its checksum to check its integrity. [Figure 4](#sensors-17-00869-f004){ref-type="fig"} shows the transmission scheme between the gateway app and the wearable node. During the prosthesis control, the classification algorithm and the prosthetic controller run in real time on the embedded board, hence the interaction with the gateway is not required during continuous use. 3.5. Control Strategy {#sec3dot5-sensors-17-00869} --------------------- For mechanical reasons, prosthetic hands execute the various grasps and movements starting always from a reset state, normally the open hand position. After the execution of a gesture, it is, therefore, necessary to return to the reset configuration before performing another movement. This strategy allows to control the prosthesis movement using only the motor current feedback provided by the integrated driver, since the system always starts from a known configuration. The diagram of the FSM that controls the the system in CM is represented in [Figure 5](#sensors-17-00869-f005){ref-type="fig"}. In the *ADC Acquisition*, ADC peripheral extracts the mean value of 16 consecutive samples of the 4 EMG channels. The total conversion time of the peripheral is 32 $\mathsf{\mu}$s for the 16 consecutive samples and this does not affect the real time requirements of the system. The *Spike Removal* block acts as a time trigger to avoid the activation on spurious spikes due to movement artifacts or external noise. The contraction spikes lasting over an activation threshold for less than 100 ms are filtered because they are not related to voluntary hand movements \[[@B58-sensors-17-00869]\]. If the signal lies over the threshold for longer, the system starts the classification of the incoming samples using the on-board SVM models. Since the onset of a gesture contains the transient phase from the rest to the intended contraction, the recognition of the performed gesture is more difficult at the beginning and gets more reliable as the contraction reaches a steady state. However, a timely response of the prosthesis requires to start the actuation and execute the intended movement as soon as the contraction has started and a natural interaction with the device requires response times below 300 ms \[[@B17-sensors-17-00869]\]. To reach a correct classification within a limited time, in the *SVM Voting* block our system applies the majority voting on 20 consecutive SVM classifications to select the most likely gesture being performed. The gesture with the highest number of classification instances is hence detected as the intended one and the motors are actuated. The hand controller sets up the finger movement parameters according to the intended gesture. For instance, in the power grasp all the motors receive the command to rotate closing the fingers, while in the precision grasp only index and thumb fingers receive the closing command. The MCU stops the motors exploiting the H-bridge current feedback, which indicates when a finger has reached the correct position. After any performed gesture, the device will need to go back to the reset position (i.e., the open hand) before executing further movements. These transitions are managed by the *Gesture Config* Block, in accordance with the current state of the prosthetic hand. Once a gesture is decoded and actuated, the EMG contraction level is acquired again, waiting for a muscular decontraction, that retriggers the FSM controller. This sequence returns a natural control strategy allowing to keep trace of the current hand position without absolute encoders or other positioning sensors and provides robust control of the prosthesis with an intuitive control strategy. [Figure 6](#sensors-17-00869-f006){ref-type="fig"} shows a sequence of gestures with the system outputs to better explain the proposed strategy. The red bold line shows the output of the controller, where the positive values encode the gestures and the value −1 is reserved for the reset (open hand) position. The control strategy can be summarized as follows. The user performs a gesture to be executed and thanks to real-time recognition with majority voting, responsive but accurate actuation of the hand is provided. Once the prosthesis actuated the gesture, the user can relax the contracted muscles, the decontraction retriggers the classifier and the device will hold this configuration until an open hand gesture is detected. This strategy increases the usability of the controller since the user is not required to hold on the muscular contraction to maintain the grasp, decreasing significantly the level of fatigue. Furthermore, since the reset hand movement is the only allowed operation to retrigger the prosthesis and prepare it for a subsequent gesture, it is not strictly necessary to classify it, because the FSM does not allow other possible positions reachable after a gesture has been performed. Hence, it is possible to exclude the open hand gesture from the training set and simply detect any muscle contraction to exit the current gesture and reach the reset position. We evaluated the approach both including and leaving out the open hand gesture from the training dataset and we compared the results in terms of performance and efficiency, as presented in the following section. 4. Experimental Results {#sec4-sensors-17-00869} ======================= With our experiments, we demonstrate that using a proper control strategy on top of the classification algorithm improves greatly the accuracy and the robustness of the final gesture recognition. In prosthetics, a fair comparison between different approaches is not trivial, because differences on the setup have a great impact of the performance. Furthermore, when considering real-time control of a prosthesis, the sample recognition accuracy does not evaluate the complete system, which is better evaluated by its ability to timely perform the intended user movement, with robustness to spikes, false contractions and misclassifications. We can call this metric the *end-to-end* success ratio. To cope with this issues, and compare our solution with a literature benchmark we initially validated our system offline on an EMG dataset, the NINAPRO database, a recently released extensive collection of hand gestures for EMG recognition. The NINAPRO database collects up to 52 different hand gestures from 27 subjects, recorded with an accurate setup using a hand-tracking glove and 10 Ottobock EMG sensors on the forearm. From such dataset, we selected the same gestures as the ones used in our application and we constructed an interleaved dataset inserting open gestures between the others, to have a fair comparison also of the FSM controller. Regarding the number of sensors, we used the 10 EMG channels as the input vectors ($N_{F} = 10$). For the performance evaluation on benchmark datasets, we replicated the functionality of the recognition algorithm and of the controller in a simulated environment using Matlab. Hence, we tested the system with the data from the NINAPRO database, evaluating the number of correctly executed gestures. The end-to-end classification accuracy is evaluated over 10 repetitions of sequences composed by a gesture followed by an open hand gesture. Each user performed 10 repetitions of 3 gestures: power grasp, pointing index and precision grasp. We collected a total of 60 consecutive movements from each user. On this sequence, we evaluated the number of wrong gestures as executed by the proposed control strategy. To have a comparison with the classical classification accuracy, an error of 1 in 10 gestures is equivalent to an error rate of 10% (i.e., 90% accuracy), while an error of 1 in 60 gestures corresponds to an error rate of 1.66% (98.4% accuracy). Then we tested the system in real time on 5 healthy subjects, in our laboratory and on 4 transradial amputees with $1/3$ distal amputation, at the INAIL center. All participants provided informed written consent to take part in the experiment, which was approved by the Ethical Committee of the University of Bologna (Italy). The muscular contractions during the gesture execution is 3 s long and between each contraction there are 3 s of rest. On the healthy subjects, the 4 EMG sensors were placed in correspondence of the *flexor radialis carpi, flexor ulnaris carpi, extensor communis digitorum and extensor carpi ulnaris.* These muscles were targeted as the actuators of the movements analyzed in the experiment, accordingly to \[[@B17-sensors-17-00869]\]. [Figure 7](#sensors-17-00869-f007){ref-type="fig"} shows the real time experiment on the healthy subject with the I-Limb hand. Regarding the amputee subjects, since the muscular structure of the forearm is compromised, the sensors were placed at equal distance between each other, starting from the position of a working or partially working muscle, found with tactile analysis. Since the feature extraction is performed in the analog domain directly on the Ottobock sensors, each sample that comes out from the ADC can be classified by the SVM. To increase the robustness of the system, the majority voting is applied over 20 consecutive samples, allowing a classification output every 40 ms. Results reported in [Table 1](#sensors-17-00869-t001){ref-type="table"} show the end-to-end execution accuracy and the number of SVs in the trained models on NINAPRO (offline) and on healthy and amputee subjects (online). The first column (*Accuracy*) shows the cross-validation recognition accuracy calculated on all the samples of the training dataset, which confirms that the classifier performance is aligned with the SoA. Under the columns labeled *COMPLETE*, we show the number of SVs and the end-to-end error rates using a SVM model that includes the open hand gesture to be recognized. In the *REDUCED* case, we excluded the open hand gesture and simply assume such gesture is performed when a contraction is detected after any other gesture. The evaluation of the proposed control strategy shows similar values of accuracy and of end-to-end classification rates for the *COMPLETE* and *REDUCED* approaches. Our solution with only 4 EMG sensors shows comparable results to the one obtained with the NINAPRO dataset, where EMG acquisition is based on an array of 10 sensors, hence more complex and expensive. Maintaining a low complexity is a key point in the development of an efficient embedded system and its design and implementation should be tailored to optimally employ the hardware resources. The number of EMG channels, and hence the dimensionality of the input data vectors, beside the impact on the system cost has a major impact on the amount of the required memory and on the computational load of the recognition algorithm. The proposed SVM is implemented using single precision floating point representation for the SVs, their coefficients and the input signals. This directly impacts on the hardware resources needed for model storage and real-time execution, which we evaluated on the used MCU. The minimum and maximum memory requirements given by the trained models for the two analyzed setups (our and NINAPRO healthy subjects) are reported in [Figure 8](#sensors-17-00869-f008){ref-type="fig"}a. The two solutions have very similar memory occupation and this is given by their combinations of input features and the resulting SVs. Our solution has a smaller number of input features ($N_{F} = 4$) and results in models with a higher number of SVs ($N_{S} = 178 \div 404$). In contrast, the NINAPRO setup has a higher number of input channels ($N_{F} = 10$) and results in models with a lower number of SVs ($N_{S} = 116 \div 226$). Overall the two have similar memory occupation of up to 12 kB, which can be handled by the proposed MCU equipped with 64 kB of Flash. The run-time computational load of the application is composed by the signal acquisition, the SVM prediction and the control FSM. The SVM prediction dominates the workload and sets the requirements for real-time execution. [Figure 8](#sensors-17-00869-f008){ref-type="fig"}b shows its execution time, expressed in CPU cycle counts, as measured using the MCU's internal cycle counter (CYCCNT) register. We analyzed the two setups, having input vectors of 4 and 10 and we again measured the performance for the best and worst cases (min and max number of SVs). The computational times are again influenced by the number of input features, the number of recognized gestures and the resulting number of SVs, but this time the obtained combinations do not result in similar requirements. In fact, the proposed approach outperforms the NINAPRO setup and leads to a reduced computation time in all combinations. [Figure 9](#sensors-17-00869-f009){ref-type="fig"} shows the quantitative differences of the computational load of the algorithm with healthy subjects, amputees and with NINAPRO data, considering the complete and reduced datasets. To compare the complete and reduced approaches, we consider the mean execution times of the 3 classes of subjects on the proposed MCU. The SVM reaches the best performance with the healthy subjects, because the activation patterns of the different gestures are well separated due to the good condition of the forearm muscles. With the NINAPRO healthy subjects the SVM reaches comparable performance, even though the execution is slightly slower. The test on amputee patients demonstrates that the computation of the classifier's output requires a significant increase of time with the complete dataset, due to the larger SVM model. In fact, the compromised residual muscles of the amputees produce more confused activation patterns and the SVM algorithm needs more SVs to define the separation boundary between the classes. Nevertheless, with the proposed control strategy, it is possible to reduce the number of SVs of the model by the elimination of the open hand gesture. In this case, we obtain a significant reduction of the number of SVs and consequently of the computation time that becomes comparable to the performance obtained on healthy subjects with a redundant sensor setup. The experimental evaluation shows that the proposed hardware and software implementation of this BSN is tailored to meet the requirements of a real time wearable controller for a prosthetic hand through the multimodal approach of the desing, from system architecture to algorithm and control level. 5. Conclusions {#sec5-sensors-17-00869} ============== Recent research efforts explore different solutions for a natural control of poliarticulated hand prostheses, employing advanced gesture recognition techniques or combining myoelectric proportional control with innovative mechanical design of the hand based on adaptive synergies. However, the design of a complete system must draw on knowledge from various research fields and nowadays a considerable gap persists between research and lab evaluation of such solutions and their real-life usability and implementation into successful commercial products. The contribution of the paper is in the complete design and the final deployment of a body area controller for a poliarticulated prosthetic hand, based on natural control strategy. The work described in detail the hardware and software implementation of an efficient embedded pattern recognition algorithm and of a reliable control strategy of the poliarticulated hand. We validated our approach by applying it to data from the NINAPRO database that represents the largest and most complete dataset for EMG gesture recognition analysis, obtaining again minimal errors. Moreover, we presented the online performance of our wearable node in terms of end-to-end recognition ratio, evaluating the accuracy and timing of the system to recognize and execute complete hand gestures, as performed by healthy and amputees subjects. The proposed approach resulted having an average error rate of 1 misclassified gesture every 60, for both healthy and amputee subjects. The innovative complete control strategy has been implemented on the embedded system, which allows its real-time execution. To ensure a robust but timely response of the system, we applied a majority voting approach to the output of the SVM classifier, obtaining a recognition time of 36 ms, in the worst case. Future works will target the consolidation of an advanced prototype to enable a detailed evaluation of usability in real life scenarios and the comparison with SoA solutions. This work was funded by the Swiss National Science Foundation under Grant 162524 (MicroLearn: Micropower Deep Learning), Armasuisse Science & Technology, by the ERC MultiTherman Project (ERC-AdG-291125), and by the OPRECOMP Project founded from the European Union's Horizon 2020 Research and Innovation Programme under grant agreement No. 732631. S.B. and E.G. conceived designed and performed the experiments; S.B. analyzed the data; S.B. developed Hardware and Firmware; B.M. developed the PC interface; S.B., E.F. and L.B. wrote the paper. The authors declare no conflict of interest. {#sensors-17-00869-f001} {#sensors-17-00869-f002} {#sensors-17-00869-f003} {#sensors-17-00869-f004} {#sensors-17-00869-f005} {#sensors-17-00869-f006} {#sensors-17-00869-f007} {#sensors-17-00869-f008} {#sensors-17-00869-f009} sensors-17-00869-t001_Table 1 ###### Gesture recognition accuracy and errors in hand movements decoding of the proposed controller. ---------------------- -------------- ---------------- --------- ---------------- ----- **Healthy Subjects** **COMPLETE** **REDUCED** **Accuracy** **SVs** **FSM Errors** **SVs** **FSM Errors** S1 88.01 178 0 124 0 S2 89.76 312 0 209 0 S3 89.21 229 0 155 0 S4 86.34 404 0 204 0 S5 83.49 311 0 206 0 MEAN 87.37 296 0 179 0 **INAIL patients** **COMPLETE** **REDUCED** **Accuracy** **SVs** **FSM errors** **SVs** **FSM errors** S1 94.86 166 0 55 0 S2 93.65 262 0 38 0 S3 81.38 393 1 290 0 S4 86.43 1316 1 729 1 MEAN 89.09 534 \<1 278 \<1 **NINAPRO data** **COMPLETE** **REDUCED** **Accuracy** **SVs** **FSM errors** **SVs** **FSM errors** S1 90.46 142 0 116 0 S2 93.49 116 4 107 4 S3 92.89 132 0 67 0 S4 92.15 226 0 169 0 S5 90.15 158 0 114 0 MEAN 91.83 155 \<1 112 \<1 ---------------------- -------------- ---------------- --------- ---------------- -----

{#s1} Appropriate use of fluid infusion in cardiac surgery patients is of primary importance in the perioperative period in order to optimize cardiac output and oxygen delivery and to reduce the use of vaspressors and inotropes. Fluid infusion is usually triggered by arterial hypotension, low urine output and signs of inadequate tissue oxygenation (e.g. hyperlactatemia). This general rule, however, may not be applied when the heart is working in the flat part of the Frank Starling curve, where hypervolemia may cause excessive increase of filling pressures and tissue edema. In this light, it is mandatory to carefully dose the exact amount of fluids to administer in order to avoid the risk of volume overload. *In this issue of HSR Proceedings in Intensive Care and Cardiovascular Anesthesia*, Arora and co-workers shed some light on the issue of perioperative fluid administration to cardiac surgery patients and its correlation to mortality \[[@R01]\]. The authors clearly showed how the effects of intravascular filling correlates with mortality, especially if the amount of given fluids exceeded four litres in the perioperative period. This effect remained significant even after adjustment for the presnce of acute kidney injury and/or hypotensive events. The authors did not specify if the nature of infused fluids had a role on patients outcomes nor if specific treatments such as early/aggressive perioperative ultrafiltration might inversely correlate with mortality. More than five years ago, data coming from the Prospective Pediatric Continuous Renal Replacement Therapy registry showed that survival rates in patients with multiorgan dysfunction syndrome were significantly better for patients with less than 20% fluid overload (58% vs 40% survival rate) at continuous renal replacement therapy initiation \[[@R02]\]. Fluid balance is probably underestimated in critically ill adults where a huge fluid volume amount is infused in order to target hypovolemia and organ perfusion. Few clinical investigations, until now, have evaluated the impact that fluid balance has on clinical outcomes in critically ill adults: the Sepsis Occurrence in Acutely Ill Patients (SOAP) study \[[@R03]\] and the PICARD (Program to Improve Care in Acute Renal Disease) study group \[[@R04]\] recently showed how critically ill patients with acute kidney injury and fluid overload experienced significantly higher mortality with respect of patients without fluid overload apart from the need for RRT. The work from Arora and co-authors seems to support, in the specific setting of cardiac surgery patients, the view that there might be a survival benefit from conservative approach to intra-operative intravascular volume expansion. Early initiation of continuous renal replacement therapies to prevent fluid accumulation and overload in critically ill patients, once initial fluid resuscitative management has been accomplished might be an alternative approach \[[@R05]\] in this light, fluid overload is evolving as a primary trigger/indicator for extracorporeal fluid removal, and this may be independent of dose delivery or solute clearance. Another final aspect of the study from Arora must be highlighted: if it is true that extra volume provision was detrimental also in patients without acute kidney injury and/or hemodynamic instability, it must be said that in these last patients correlation between fluid overload and mortality was exponentially higher. If it is evident that counterbalancing fluid accumulation, particularly in patients with oliguria or established acute kidney injury might be beneficial, on the other side it is also clear that more severely ill patients might often miss any active attempt of achieving a negative balance and we do not know if increasing vasopressors dose might really improve survival of such patients. Once a need for increased cardiac output is considered, it is helpful to have an indicator of fluid responsiveness. Central venous pressure and pulmonary artery occlusion pressure have limited predictive value as indicators of fluid responsiveness (with respect to volumetric-echocardiographic estimations) due to the existence of different conditions affecting the distribution of blood volumes \[[@R06]\]. Normally, approximately 70% of the total blood resides in the small venules and veins (unstressed volume). Only the remaining 25-30% (1.2-1.4 l) of the total blood volume (stressed volume) determinates, with the elastic recoil of the vasculature, the mean systemic filling pressure that is, with the right atrial pressure , the main determinant of the venous return, and finally of the cardiac output \[[@R07]\]. Cardiac surgery procedures and cardiopulmonary bypass deeply influence the venous return and the cardiac output by increasing the venous capacitance and right atrial pressure for different reasons (systemic inflammation, anesthetic drugs, mechanical ventilation). An increase in venous capacitance may cause a reduction in stressed volume and an increase in unstressed volume: the final result may be a reduction of cardiac output due to a relative hypovolemia. It seems that some patients (i.e. in case of bleeding) may actually benefit from an increase in mean systemic filling pressure (volume load) whereas in others a venous capacitance reduction (anti-inflammatory strategies, vasopressors, conservative fluid infusion) should be encouraged. More recently, the use of respiratory variations of arterial pressure (systolic pressure variation, stroke volume variation, pulse pressure variation) to predict fluid responsiveness, have shown some interesting data in both operating theatres and intensive care units \[[@R08]\]. Unfortunately, these dynamic indices, robust and reliable under specific conditions (closed chest, controlled mechanical ventilation, sinus rhythm) have not been validated during open chest settings. In this light, low cardiac output in cardiac surgery patients should be managed with a multimodal monitoring (echocardiography, cardiac and vascular filling pressures, dynamic indices of fluid responsiveness) and treatment tailored to the single patient and clinical picture trying to obtain the best balance between fluids, inotropes and vasopressors during the whole intra and post-operative phase.

1. Introduction =============== Cardiovascular disease (CVD) has been the leading cause of morbidity and mortality in developed countries^\[[@R1],[@R2]\]^ and some developing countries, such as China.^\[[@R1],[@R3]\]^ The 2013 Global Burden of Disease Study showed that cardiovascular deaths accounted for almost a third of all deaths globally.^\[[@R1]\]^ In China, CVD is responsible for 41% of all annual deaths, and the increase in CVD mortality in rural residents is greater than that in urban citizens.^\[[@R3]\]^ The American Heart Association (AHA) Strategic Planning Task Force and Statistics Committee^\[[@R4]\]^ developed the concept of "cardiovascular health" along with 7 health metrics, which were categorized as 4 health behaviors (smoking, diet, physical activity, body mass index \[BMI\]) and 4 health factors (smoking, blood pressure, total cholesterol \[TC\], fasting plasma glucose \[FPG\]). Given the importance of abstinence from smoking and smoking cessation for health promotion, smoking is listed in both the health factors and health behaviors.^\[[@R4]\]^ With the use of the 7 metrics, the cardiovascular health status for a whole population is defined as ideal, intermediate, or poor.^\[[@R4]\]^ Epidemiologic evidence indicates that ideal cardiovascular health is associated with a lower risk of CVD,^\[[@R5]--[@R7]\]^ lower mortality rates of CVD and lower all-cause mortality,^\[[@R8]--[@R10]\]^ as well as disease-free survival, better quality of life, and lower healthcare costs.^\[[@R4]\]^ Several studies have determined the status of the AHA-defined cardiovascular health in the USA and European populations.^\[[@R9]--[@R19]\]^ However, the AHA\'s construct of cardiovascular health has not been widely used among populations from developing countries. There are no data available on the status of cardiovascular health, health behaviors, and health factors among rural populations of China. The purpose of this study was to assess the status of cardiovascular health and its associated metrics among rural adults in Northwest China. 2. Methods ========== 2.1. Study population and data collection ----------------------------------------- A population-based cross-sectional study was conducted in the rural areas of Hanzhong, Shaanxi Province, in Northwest China between October 2010 and November 2010 to estimate CVD risk factors among rural adults. Hanzhong is relatively poor and less developed than the coastal, eastern and southern regions of China. A stratified cluster random sampling method was used for sample selection. There are 9 "township" regions in the study area, and an average of approximately 17 (ranging from 15 to 36) villages in each township region. We stratified the study according to the township, that is, each township was a stratum, and 1 village (cluster) was randomly chosen from each township. Using residential registration data, all the available and eligible adults in the chosen villages were informed and invited to participate in the survey several days before the survey. In each chosen village, adults who consented and came to the clinic of the village doctor, where the interview and physical examination were conducted, on a nominated survey day were chosen as subjects of the survey. Information on the participants was collected through an interview that collected their age, sex, education, marital status, family economic level, smoking, physical activity, history of clinical CVD (coronary heart disease, heart failure, myocardial infarction, stroke, or other heart-related conditions), hypertension, diabetes and dyslipidemia, use of antihypertensive, antidiabetic and dyslipidemia medications, and family history of hypertension. Blood pressure was measured by trained doctors from the Hanzhong People\'s Hospital after the subject had rested for at least 5 minutes, using a standard mercury sphygmomanometer with the participant in the sitting position. Two blood pressure measurements separated by a 2-minute interval were obtained, and the mean of the 2 readings was used as the blood pressure value. Height and weight were measured with participants standing without shoes or heavy outer garments, from which BMI (kg/m^2^) was calculated. Dietary intake and alcohol consumption (including grape wine, rice wine, beer, and liquor) within the past year were assessed using a semi-quantified food frequency questionnaire. According to the reference in the Dietary Guidelines for Chinese (≤25 g/d alcohol for men and ≤15 g/d alcohol for women),^\[[@R20]\]^ alcohol consumption was classified into 3 categories: nondrinking (did not consume any type of alcohol or consumed \<1 drink per month), moderate drinking (men consumed ≤25 g/d alcohol and women consumed ≤15 g/d alcohol), and heavy drinking (men consumed \>25 g/d alcohol and women consumed \>15 g/d alcohol). To identify the family economic status of the participants, a wealth index based on communication tools, transportation tools, sources of water, and monthly incomes and expenses of the whole family was derived using a principal component analysis. The first principal component was selected as the wealth index. The participants' family economic level was categorized into 3 groups of low, moderate and high, according to the tertiles of the wealth index.^\[[@R21]\]^ The overnight fasting blood sample was collected from each subject by a qualified nurse. The fasting time was verified before blood sample collection, and the participants who had not fasted for at least 8 hours did not have their blood drawn. FPG and TC levels were analyzed enzymatically. Detailed information on the sampling, data collecting, and blood sample collecting, transporting, storing and analyzing have been described previously.^\[[@R22]\]^ 2.2. Definition of cardiovascular health metrics and cardiovascular health -------------------------------------------------------------------------- In accordance with the AHA\'s definition,^\[[@R4]\]^ the status of each cardiovascular health metric was categorized as "ideal," "intermediate," and "poor" as follows. ### 2.2.1. Health behaviors \(1\) Smoking: never smoking or having quit \>12 months ago was defined as ideal; having quit smoking ≤12 months was defined as intermediate; and current active smoking was defined as poor. (2) BMI: BMI was classified as ideal (\<25.0 kg/m^2^), intermediate (25.0--29.9 kg/m^2^), or poor (≥30.0 kg/m^2^). (3) Physical activity: participants who often performed farm labor or other heavy or moderate manual labor were categorized as ideal; participants who sometimes performed farm labor or other heavy or moderate manual labor were categorized as intermediate; and participants who hardly ever performed farm labor or other heavy or moderate manual labor were categorized as poor. (4) Diet: after adjusting for the level of dietary calorie intake, participants who achieved 4 to 5 diet goals (including ≥4.5 cups per day of fruits and vegetables, ≥ two 3.5-oz servings per week of fish, ≥ three 1-oz equivalent servings per day of fiber-rich whole grains, \<1500 mg per day of sodium, and ≤ 36 oz per week of sugar-sweetened beverages for a 2000-kcal diet) were categorized as ideal; participants who achieved 2 to 3 items were categorized as intermediate; and those who achieved 0 to 1 items were categorized as poor. ### 2.2.2. Health factors \(1\) Smoking: see definition in "health behaviors." (2) TC: TC status was defined as ideal (\<200 mg/dL, untreated), intermediate (200--239 mg/dL or treated to \<200 mg/dL), and poor (≥240 mg/dL). (3) Blood pressure: blood pressure status was classified as ideal (systolic blood pressure \[SBP\] \<120 mm Hg and diastolic blood pressure \[DBP\] \<80 mm Hg, untreated), intermediate (SBP 120--139 mm Hg or DBP 80--89 mm Hg, or treated to SBP \<120 mm Hg and DBP \<80 mm Hg), or poor (SBP ≥140 or DBP ≥90 mm Hg). (4) FPG: FPG status was classified as ideal (\<100 mg/dL, untreated), intermediate (100--125 mg/dL, or treated to \<100 mg/dL), or poor (≥126 mg/dL). ### 2.2.3. Cardiovascular health Considering these health metrics, we defined an overall measure of cardiovascular health as follows: "ideal" if all 7 cardiovascular health metrics are ideal and without a history of clinical CVD; "intermediate" if at least 1 metric is at an intermediate level and none are at a poor level among subjects without a CVD history or if all 7 metrics are at an ideal level among subjects with a CVD history; and "poor" otherwise (at least 1 metric is at a poor level among subjects without a CVD history or at least 1 metric is at an intermediate level among subjects with a CVD history).^\[[@R4],[@R5],[@R23],[@R24]\]^ 2.3. Ethics statement --------------------- The study complied with the Declaration of Helsinki and was reviewed and approved by the Ethics Committee of Xi'an Jiaotong University College of Medicine. Written informed consent had been obtained from each study participant. 2.4. Statistical analyses ------------------------- The Complex Samples Procedure of SPSS 13.0 for Windows (SPSS Inc., Chicago, IL) was used for statistical analyses, accounting for township strata and village clusters. All statistical tests were 2-tailed, and statistical significance was set at *P* *\<* 0.05. Continuous variables are presented as the mean ± SD. Categorical variables are presented as percentages. The proportions of ideal, intermediate and poor categories for each metric, and the overall cardiovascular health were calculated and adjusted for age and/or sex according to the 2010 Chinese National Census Population distribution. Differences between means were compared using general linear models. Chi-square tests and Fisher\'s exact tests were used to compare percentages. Multiple logistic regression was used to evaluate the association between ideal cardiovascular health (dependent variable) and its associated factors. 3. Results ========== 3.1. General characteristics of the study population ---------------------------------------------------- A total of 3016 residents aged 20 to 80 years took part in the survey. However, 2693 participants had complete interview, diet, physical examination and blood sample data, and were included in the analysis. Forty-one participants who had no blood sample data (including 20 who refused to draw blood sample and 21 who did not fast overnight), 276 who had incomplete information on health factors or health behaviors, and 6 who had extreme energy intake (\>5000 kcal/d or \<500 kcal/d) were excluded. The socio-demographic characteristics and history of relevant diseases of the participants in the analysis are shown in Table [1](#T1){ref-type="table"}. There were 899 (33.4%) participants who were men and 1794 (66.6%) who were women. The mean age of the participants was 51.4 ± 11.5 years. Most (99.9%) of the participants were ethnic Han Chinese, and 53.9% of them finished middle school education or above. A total of 92.2% participants were married. No statistically significant differences with respect to socio-demographic characteristics were detected between the 2693 subjects with complete data and the 323 excluded participants with incomplete data. ###### Characteristics of the study participants at the time of survey.  3.2. Distribution of individual cardiovascular health metrics ------------------------------------------------------------- The distribution of individual cardiovascular health metrics in the study population is shown in Table [2](#T2){ref-type="table"}. Most of the participants met the ideal cardiovascular health metrics for smoking, BMI, physical activity, TC, and FPG (68.0%, 80.0%, 85.1%, 81.5%, and 84.3%, respectively), whereas most of the participants' diet and blood pressure levels were classified into intermediate and poor categories. Only 2.8% of the participants fulfilled 4 to 5 items of healthy diet, and 38.8% of the participants' blood pressure values were \<120/80 mm Hg. The percentages of women who met the ideal level for smoking, BMI, physical activity, blood pressure, and FPG were higher than those of men (all *P \<* 0.01). Only for TC, the percentage of men who met ideal level was higher than that of women (*P* *=* 0.006). The percentages of men and women who met the ideal level of healthy diet were both low, and there was no gender difference (2.5% vs 3.0%, *P* *=* 0.406). The most prevalent poor metric was smoking (59.9%) for men and healthy diet (26.1%) for women. ###### Distribution of individual cardiovascular health metrics and overall cardiovascular health status.  Except for sugar-sweetened beverages intake, the healthy diet components were poorly met. Almost all participants (96.3% in total, 95.1% in men, and 97.5% in women) achieved the goal of sugar-sweetened beverages intake. However, only 52.2% (52.4% in men and 52.1% in women), 36.6% (34.2% in men and 39.0% in women), 8.4% (9.2% in men and 7.7% in women), and 7.6% (8.5% in men and 6.6% in women) of the participants achieved the goal of fiber-rich whole grains, fruit and vegetable, sodium, and fish intake, respectively. 3.3. Distribution of cardiovascular health status ------------------------------------------------- As presented in Table [2](#T2){ref-type="table"}, only 0.5% participants achieved ideal cardiovascular health, and the proportion for men was lower than that for women (0.0% vs 0.9%, *P* *=* 0.002). The proportion of intermediate cardiovascular health was also low (33.8%), and it was lower in men than in women (18.0% vs 50.0%, *P* *\<* 0.001). The remaining 65.7% of the participants had poor cardiovascular health. The prevalence of poor cardiovascular health increased with increasing age (*P* \< 0.001 for trend, Fig. [1](#F1){ref-type="fig"}). {#F1} The participants fulfilled, on average, 4.4 (95% confidence interval \[95% CI\]: 4.2--4.7) ideal cardiovascular health metrics, with a decrease across age-groups (from 5.1 \[95% CI: 4.7--5.4\] in participants \<25 years old to 3.9 \[95% CI: 3.8--4.0\] in participants 65 years or older; *P* *\<* 0.001, Fig. [2](#F2){ref-type="fig"}). Women exhibited a higher number of ideal metrics than men across all age groups (4.9 \[95% CI: 4.7--5.1\] vs 3.9 \[95% CI: 3.6--4.2\], respectively, *P* *\<* 0.001). A total of 22.2% of the participants presented with 3 or fewer ideal metrics (33.7% in men vs 10.5% in women, *P* *\<* 0.001). Only 19.4% of the participants presented with 6 or more ideal metrics (4.6% in men vs 34.5% in women, *P* *\<* 0.001, Fig. [3](#F3){ref-type="fig"}). The proportion of subjects who had 6 or more ideal metrics significantly decreased with increasing age, from 34.6% in participants under 25 years old to 3.9% in participants 65 years old or over (*P* *\<* 0.001 for trend). A total of 24.1% of the participants were ideal on all 4 health factors, and only 1.1% were ideal on all 4 health behaviors; the proportions of men were lower than those of women (9.0% vs 39.6% on all 4 health factors, 0.1% vs 2.2% on all 4 health behaviors, both *P* *\<* 0.001). {#F2} {#F3} 3.4. Multiple logistic regression analyses ------------------------------------------ Because of the low percentage (only 0.5%) of the participants who met the ideal cardiovascular health, having 6 or more ideal metrics with no history of clinical CVD was used as one category (ideal cardiovascular health) and having 5 or less ideal metrics or with a history of clinical CVD was used as another category (nonideal cardiovascular health) of outcome in the Complex Samples multiple logistic regression analysis on the associated factors for ideal cardiovascular health. History of hypertension, diabetes, dyslipidemia, and CVD were controlled for in the logistic regression model. The results are presented in Table [3](#T3){ref-type="table"}. Women were 6.03 times more likely to have an ideal cardiovascular health compared with men (95% CI: 3.64--9.96). Compared with adults \<25 years old, participants 35 years old or over were less likely to have ideal cardiovascular health and the ORs decreased with increasing age, from 0.25 (95% CI: 0.10--0.64) to 0.09 (95% CI: 0.02--0.37). Compared with adults without a family history of hypertension, participants with a family history of hypertension were less likely to have ideal cardiovascular health (OR: 0.60, 95% CI: 0.43--0.83). Education level, marital status, alcohol drinking, and family economic status were also explored, but no associations were found with ideal cardiovascular health (all *P* *\>* 0.05). ###### Results of logistic regression analysis of associated factors for the ideal cardiovascular health.  4. Discussion ============= This study is the first population-based study to use the AHA-defined construct of cardiovascular health among the rural population in Northwest China. The results of our study showed that the prevalence of ideal cardiovascular health was extremely low, whereas poor cardiovascular health was prevalent among the rural population in Northwest China. The estimated low prevalence (0.5%) of ideal cardiovascular health among the study population is consistent with the results from the USA (0.1--3.3%),^\[[@R2],[@R5],[@R9]--[@R11],[@R13]--[@R16]\]^ from Spain (0.2%),^\[[@R17]\]^ from Srpska (0.02%),^\[[@R18]\]^ from Italy (1.9%),^\[[@R19]\]^ in urban adults from China (0.5%)^\[[@R25]\]^ and from a health examination population in Southeast China (1.1%).^\[[@R26]\]^ Our results add to previous reports that the prevalences of ideal cardiovascular health are very low in various countries and populations and indicate that effective efforts are needed to improve cardiovascular health levels among the rural population in Northwest China to prevent an epidemic of cardiovascular diseases. Our study showed that women had a better cardiovascular health profile than men. This sex difference might be accounted for by the much higher prevalence of current smoking (59.9% vs 1.2%) whereas a lower prevalence of ideal physical activity (76.3% vs 94.1%) in men than those in women. With economic development and urbanization, physical activities have declined rapidly in Chinese adults, especially in men who do less household work.^\[[@R27]\]^ Efforts should be taken to cease smoking and promote recreational physical activities and exercise among the rural population, especially among men, in Northwest China. Moreover, our study found that the proportion of adults who met the ideal level on all 4 health behaviors (1.1%) was much lower than that of all 4 ideal health factors (24.1%). This finding indicates that education and interventions to promote people\'s lifestyle are much more necessary to improve the cardiovascular health among rural adults in Northwest China. A healthy diet was the least prevalent cardiovascular health metric among the study population. Only 8.4% of rural adults in Northwest China reached the goals for sodium intake. Studies showed that high sodium intake was a major risk factor for hypertension and increased the risks of heart attack, stroke, left ventricular hypertrophy, obesity, and renal disease.^\[[@R28]\]^ Modest reductions in dietary salt could substantially reduce cardiovascular events and medical costs.^\[[@R29]--[@R31]\]^ Du et al^\[32\]^ reported that the Chinese population\'s sodium intake decreased from 6.6 g/d in 1991 to 4.7 g/d in 2009 but remained at double of that recommended. In China, sodium intake was higher in rural areas than in urban areas^\[[@R32]\]^ and higher among Northerners than among Southerners; most of the dietary sodium was from salt added in home cooking.^\[[@R33]\]^ Therefore, population strategies and measures targeted at individuals to reduce salt intake are necessary and urgent for the prevention and control of CVD among the rural population in Northwest China. Anderson and colleagues suggested that efforts to remove excess sodium from diets in rural China should focus on reducing salt in home cooking.^\[[@R33]\]^ Moreover, because of the salt monopoly in China, introducing a low-sodium salt substitute and convincing the salt industry to gradually reduce the sodium content in China\'s entire salt supply may be possible.^\[[@R34]\]^ Physical inactivity is the greatest public health problem of the 21st century.^\[[@R35]\]^ It has a major health effect worldwide, resulting in 6% burden of disease from coronary heart disease, 7% from type 2 diabetes, 10% from breast cancer, and 10% from colon cancer.^\[[@R36]\]^ A decrease in or removal of this unhealthy behavior could improve health substantially.^\[[@R36]\]^ The population of our study was composed of farmers in the rural areas of Northwest China. They usually worked in farming, and in the slack seasons, some of them went to work at rural factories as "blue-collar" workers. Studies showed that in China, more rural residents participated in work-related physical activity than their urban counterparts.^\[[@R37]\]^ Rural residents' intensity of physical activity during working hours was greater than that of urban citizens.^\[[@R38]\]^ However, with China\'s rapid urbanization and economic growth in the past few decades, decreases in physical activity have been reported for several domains, including occupation, transportation and household activity, and work-related physical activity decreased in both rural and urban residents, but the rate of decrease was greater in rural areas than that in urban areas.^\[[@R27],[@R38],[@R39]\]^ Moreover, compared with urban citizens, rural residents are more likely to be physically inactive in leisure time.^\[[@R40]\]^ Previous studies showed that the prevalence of metabolic syndrome (12.8% for men, 17.4% for women, 15.1% overall), according to the modified National Cholesterol Education Program Adult Treatment Panel III criteria, in this rural population was lower than the prevalence among urban adults in China (28.4% for men and 35.1% for women) and much lower than the prevalence among US adults (32.8% for men, 36.6% for women, 34.7% overall).^\[[@R22],[@R41],[@R42]\]^ A meta-analysis also showed that the Chinese people in urban areas (24.9%) were more likely to suffer from metabolic syndrome than those in rural areas (19.2%).^\[[@R43]\]^ Lower physical activity, lower intake of fruits and vegetables, and greater consumption of animal fat among urban citizens than rural residents might account for the difference in prevalence of metabolic syndrome between urban and rural areas. However, the rapid economic development and urbanization in China during the past few decades has increasingly promoted a sedentary lifestyle, an elevated consumption of energy-dense foods, and greater psychological stress, all of which increase the risk of metabolic syndrome, type 2 diabetes and CVD.^\[[@R44]--[@R49]\]^ This has had a greater impact on rural residents, who generally have a lower education level, lower economic status, less awareness and knowledge of risk factors for health and poorer access to health care than urban residents.^\[[@R44],[@R47]--[@R53]\]^ Studies showed that the prevalence of metabolic syndrome and CVD mortality increased more among rural residents than urban residents during the past decade, and the gap between urban and rural residents became narrowed.^\[[@R3],[@R44]\]^ The burden of CVD will shift quickly to the rural people in China. Thus, intervention strategies to promote physical activity, especially physical activity during leisure-time to compensate for declines in occupational and household activity, should be a major health priority in rural China. Fortunately, improving the availability of exercise facilities in communities and implementing fitness projects for rural residents have become a national policy, as highlighted in China\'s Twelfth Five-Year Plan (2011--2015) for Building the Public Sport Infrastructure.^\[[@R54]\]^ This plan advocates the cooperation of central and local governments with civil society and public and private stakeholders to build new exercise facilities, such as community fitness parks and roadside open spaces with exercise and sports equipment, and to enhance the accessibility of existing facilities, for example, by opening the gymnasium and exercise facilities of schools, governments, enterprises, and organizations to the public.^\[[@R54]\]^ Moreover, school education and mass media campaigns are also needed to improve the health behaviors of rural residents. They could protect people\'s health by providing health information, improving health literacy, and promoting physical activity, healthy diets, and other healthy behaviors.^\[[@R55]\]^ 4.1. Limitations ---------------- Our study has some limitations. First, concerning data quality, although most health metrics were measured reliably and objectively, the physical activity data in the study population may be subject to reporting bias because they were based on self-reporting and evaluated according to the type and frequency of physical activity at work and due to housework in the past year. Specific instruments of physical activity were not used and the amount of activity (minutes per week) could not be derived. Second, some factors that may relate to CVD status were not recorded at all. For example, information about the participants' family history of diabetes, stroke, coronary heart disease, and other cardiovascular diseases, as well as medication use for conditions except for hypertension, diabetes, or dyslipidemia were not collected in the study. Thus, we could not analyze the association between these variables and cardiovascular health status in the study population. As a result, the observed differences in health metrics may be confounded by unobserved heterogeneity. Third, all measurements of health behaviors and health factors were obtained from a single visit on the examination day; thus, there is a possibility of misclassification. For example, blood pressure measurement from a single visit usually overestimates hypertension prevalence.^\[[@R56]\]^ However, measurements from a single visit are often used in epidemiologic studies. Fourth, recall bias may be another limitation in the measurement of physical activity, diet, smoking, and other information determined from questionnaires. Therefore, future studies are needed to confirm the cardiovascular health status and its associated factors among the rural population in Northwest China. 5. Conclusions ============== Our study indicated that the prevalence of ideal cardiovascular health was extremely low among the rural population in Northwest China, and most adults, especially men and the elderly, had a poor cardiovascular health status. To improve cardiovascular health among the rural population in Northwest China, efforts, especially lifestyle improvements, education and interventions to make healthier food choices, reduce salt intake, increase physical activities, and cease smoking, will be required at the individual, population and social levels, and should begin at a young age. The authors thank the participants, organizers, and implementers of the survey, particularly Dr. Xianglin Xu, Mr. Yong Ren, and Ms. Shaomei Pan who were responsible for data collection. The authors are grateful to the village doctors for their efforts in the coordination of the fieldwork. They are also grateful to all the investigators for their contributions to data collection. Thanks also go to the editor and reviewers of the manuscript for their constructive comments and suggestions. Abbreviations: 95% CI = 95% confidence interval, AHA = American Heart Association, BMI = body mass index, CVD = cardiovascular disease, DBP = diastolic blood pressure, FPG = fasting plasma glucose, OR = odds ratio, SBP = systolic blood pressure, TC = total cholesterol. Funding: This work was supported by the Science and Technology Research and Development Program of Shaanxi Province, China (Grant No. 2013K14-02-15), the China Medical Board (Grant No. 08-925), and the National Natural Science Foundation of China (Grant No. 81230016). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. The authors have no conflicts of interest to disclose.

Introduction {#s1} ============ Narcolepsy is a chronic disabling sleep disorder with an estimated lifetime prevalence of approximately 1 in 2,000 in the general population and it can severely influence daily activities ([@B1], [@B2]). Clinically, it is characterized by excessive daytime sleepiness (EDS) with sudden sleep attacks, rapid eye movement sleep abnormalities such as cataplexy, sleep paralysis, and hypnagogic or hypnopompic hallucinations, nocturnal dyssomnia with fragmented sleep and awakenings ([@B3]). The symptoms usually begin in adolescence ([@B4]). It has been hypothesized to be caused by a combination of genetic and environmental factors, trigger an autoimmune process that results in hypothalamic destruction, with a loss of hypocretin-1-containing cells, causing the hypocretin deficiency ([@B4], [@B5]). Psychiatric comorbidities with narcolepsy are frequent, and include mood, anxiety, attention deficit hyperactivity, eating disorders, and psychosis ([@B6]--[@B9]). Psychotic disorders, consisting of a group of severe mental disorders, are a leading cause of disability globally ([@B10]). People with psychotic disorders manifest symptoms of delusion, hallucination, and reality impairment. They suffer from significant functional impairment and have elevated risks of suicide, physical morbidity and premature mortality, loss of productivity, and being difficult to care for caregivers ([@B11], [@B12]). Among narcoleptic symptoms, the hallucinatory episodes, which involve hypnagogic and hypnopompic hallucinations, can be mistaken for an active psychotic state of schizophrenia ([@B13]--[@B15]). Sleep disturbances, and particularly poor-quality sleep, are also common in schizophrenia ([@B16]). Furthermore, the development of narcolepsy in patients with schizophrenia may not be identified because their sleepiness may be attributed to their antipsychotic treatment, and cataplexy may be alleviated or masked by antidopaminergic treatment ([@B17]). The overlapping symptoms and similar ages of onset lead to a possibility of misdiagnosis between the two conditions ([@B18]--[@B20]). These difficulties that are faced by clinicians in correctly diagnosing narcolepsy and psychosis may delay well-targeted treatment. Modafinil and methylphenidate (MPH) are two effective treatments for narcolepsy. In 1998, the U.S. Food and Drug Administration approved modafinil, an oral agent, for the treatment of EDS that is associated with narcolepsy ([@B21]). Modafinil is a wakefulness-promoting drug that probably functions by increasing the extracellular concentration of dopamine and enhancing glutaminergic activity with the result of improved alertness ([@B22]). It reacts with histamine and blocks the reuptake of noradrenaline by the noradrenergic terminals on the sleep-promoting neurons from the ventrolateral preoptic nucleus ([@B23]). If modafinil cannot be prescribed, then methylphenidate (MPH) is the second-best option. Methylphenidate is a dopamine and catecholamine reuptake inhibitor that also promotes serotonin transmission. MPH reduces EDS, promotes the onset of sleep; increases REM sleep latency, and reduces the percentage of REM sleep ([@B23]). However, patients with narcolepsy may develop or experience aggravated psychosis as a consequence of stimulant therapy (such as with methylphenidate or modafinil) ([@B6]), although this side-effect seems to be rare even in patients who are treated with high doses of stimulants over a long period ([@B14], [@B24], [@B25]). This nationwide population-based analysis was performed to elucidate the co-occurrence rate of narcolepsy and psychotic disorders. The potential effect of pharmacotherapy for narcolepsy on the risk of developing psychotic disorder is also examined. Materials and Methods {#s2} ===================== Data Source {#s2_1} ----------- This retrospective cohort study used data from the ambulatory claims database of the National Health Insurance Research Database of Taiwan (NHIRD-TW), which includes outpatient, ambulatory, hospital inpatient care and dental service data. National Health Insurance (NHI), a compulsory universal health insurance program, was implemented in Taiwan on March 1, 1995. At the end of the year 2000, it covered the delivery of all health care for 22.3 million people in Taiwan (more than 96% of the national population). The Institutional Review Board of Chang Gung Memorial Hospital reviewed and approved this investigation. In this study, two subjects of the NHIRD-TW, the Longitudinal Health Insurance Database 2000 (LHID2000) and the Longitudinal Health Insurance Database 2005 (LHID2005), were used. They respectively comprise the original claims data for one million beneficiaries that were randomly sampled from the 2000 and 2005 Registries of Beneficiaries of the NHIRD-TW. LHID2000 and LHID2005 include data on all of the medical procedures and prescriptions of two million people (approximately 9% of the population) in Taiwan. A previous study has already demonstrated the reliability of the diagnostic codes in the NHIRD ([@B26]--[@B29]). Study Subjects {#s2_2} -------------- The 347.XX codes of the International Classification of Disease, Ninth Revision (ICD-9), were used to identify cases of narcolepsy. A case of narcolepsy was identified by at least one NHI claim record per visit with a diagnosis of narcolepsy (ICD-9 codes: 347). The LHID2000 and LHID2005 database included 258 such cases of narcolepsy (N = 258) between January 1, 2002 and December 31, 2011. The day when narcolepsy was first diagnosed was set as the index date, and the patients' medical records were traced until a psychotic disorder was diagnosed or until December 31, 2011. A matching control for each patient in the patient group was selected at random from LHID2000 or LHID2005 using the propensity score matching technique (and a ratio of narcolepsy cases to controls was 1:10). The propensity score was obtained using a multivariable logistic regression model with age and sex as covariates. January 1, 2002 was set as the initial observation date, and the 2,580 control subjects were monitored from the entry date until December 31, 2011 or the diagnosis of a psychotic disorder. Variables and Outcomes {#s2_3} ---------------------- The following relevant disorders that were known to be commonly comorbid with narcolepsy were identified in this study ([@B23], [@B30]); attention-deficit hyperactivity disorder (ADHD) (ICD-9-CM code 314.X), autism spectrum disorder (ASD) (ICD-9-CM code 299.X), intellectual disability (ICD-9-CM codes 317 to 319), epilepsy (ICD-9-CM code 345), and alcohol use disorders (ICD-9-CM code 303.X). The results of this study were determined using a diagnosis of any psychotic disorder (ICD-9-CM code 295.X, 297.X, or 298.X). Medications were identified using the Anatomical Therapeutic Chemical (ATC) classification system. According to the Food and Drug Administration of Taiwan, only three drugs were licensed for treating narcolepsy before 2011; they were immediate-release methylphenidate (IR--MPH) (ATC code N06BA04), osmotic controlled-release formulations of methylphenidate (OROS--MPH) (ATC code N06BA04), and modafinil (ATC code N06BA07). These drugs are restricted by the Bureau of NHI in Taiwan, and ideally any prescription for narcolepsy is recorded in an ambulatory care, pharmacy, or hospital care claim. Statistical Analysis {#s2_4} -------------------- To study the distribution of the study population, the characteristics of the narcolepsy group were compared with those of the control group using a chi-square (χ^2^) test or *t*-test. To compare the ages at the diagnosis of a psychotic disorder between the two groups, a Mann-Whitney U test was conducted. A multivariate logistic regression model was used to evaluate the potential comorbidity of a psychotic disorder with narcolepsy across all study subjects. Multivariate logistic regression was also used to investigate the potential effect of a pharmacotherapy prescription for narcolepsy on the risk of a psychotic disorder among narcoleptic patients. The results are presented as an adjusted odds ratio (aOR) and 95% confidence intervals (CIs). All data management and statistical analyses were carried out using the Statistical Package for the Social Sciences (SPSS) Version 16.0 (SPSS Inc., Chicago, IL, USA). A two-tailed value of *p \<* 0.05 indicated statistical significance. Results {#s3} ======= [**Table 1**](#T1){ref-type="table"} presents the characteristics of the narcolepsy group and the control group. No significant difference in age and sex between the two groups. The narcolepsy group was more likely than the control group to exhibit comorbidities among ADHD (8.9%), ASD (2.3%), intellectual disability (2.3%), and epilepsy (8.9%). Of patients with narcolepsy, 41.5, 5.4, and 13.2% were prescribed MPH-IR, MPH-OROS, and modafinil, respectively. In the monitoring period, 8.1% of the cases of narcolepsy were comorbid with a psychotic disorder. In contrast, 1.5% of the control subjects had a psychotic disorder. The age at diagnosis of a psychotic disorder was lower for narcoleptic patients (29.4 years) than for the controls (36.3 years). ###### Characteristics of patients with narcolepsy and control subjects in Taiwan from 2002 to 2011. Characteristics Narcolepsy (N = 258) Controls (N = 2,580) Statistics *P*-value ------------------------------------ ---------------------- ---------------------- ------------ ------------ **Age (years)** 29.3 ± 20.4 30.2 ± 20.9 0.60 ^a^ 0.552 **Gender** 0.47 ^b^ 0.495   Female 118 (45.7) 1,123 (43.5)   Male 140 (54.3) 1,457 (56.5) **Comorbidity**   ADHD 23 (8.9) 24 (0.9) 91.81 ^b^ \< 0.001\*   ASD 6 (2.3) 6 (0.2) 24.40 ^b^ \< 0.001\*  Intellectual disability 6 (2.3) 19 (0.7) 6.78 ^b^ 0.009\*   Epilepsy 23 (8.9) 38 (1.5) 61.76 ^b^ \< 0.001\* **Medication treatment**   Methylphenidate-IR 107 (41.5) 23 (0.9) 883.72 ^b^ \< 0.001\*   ethylphenidate-OROS 14 (5.4) 7 (0.3) 84.86 ^b^ \< 0.001\*   Modafinil 34 (13.2) 0 (0) 344.12 ^b^ \< 0.001\* **Diagnosed a psychotic disorder** 21 (8.1) 39(1.5) 49.79 ^b^ \< 0.001\*   Age at diagnosis (years) 29.4 ± 21.2 36.3 ± 17.5 2.08 ^c^ 0.038\* Data is expressed by n (%) or mean ± SD; Statistic values: ^a^ t using an independent t-test; ^b^ Pearson's χ^2^; ^c^ Z using Mann-Whitney U test; ADHD, attention-deficit hyperactivity disorder; ASD, Autism Spectrum Disorder; Age, age at diagnosis or at recruitment; \*P \< 0.05. The multivariate logistic regression models ([**Table 2**](#T2){ref-type="table"}) indicate that patients with narcolepsy were more likely than the control subjects to be diagnosed with a psychotic disorder (aOR, 4.07; 95% CI, 2.21--7.47). Moreover, subjects who were diagnosed with ASD (aOR, 8.13; 95% CI, 1.74--38.05), intellectual disability (aOR, 4.33; 95% CI, 1.05--17.90), and epilepsy (aOR, 5.15; 95% CI, 2.23--11.93) had a higher risk of being diagnosed with a psychotic disorder. ###### Logistic regression models for the risk of diagnosis with a psychotic disorder among all participants, controlling for gender, age, and comorbidities. Variables Psychotic disorder ----------------------------- -------------------- -------------------- ----------- **Diagnostic group**  Controls (N = 2,580) 39 (1.5) 1  Narcolepsy (N = 258) 21 (8.1) 4.07 (2.21--7.47) \<0.001\* **Age at recruitment** -- 1.01 (1.00--1.02) 0.156 **Gender**  Male (N = 1,241) 33 (2.1) 1  Female (N = 1,597) 27 (2.2) 0.99 (0.58--1.70) 0.965 **ADHD**  Without (N = 2,791) 53 (1.9) 1  With (N = 47) 7 (14.9) 1.74 (0.51--5.87) 0.373 **ASD**  Without (N = 2,826) 55 (1.9) 1  With (N = 12) 5 (41.7) 8.13 (1.74--38.05) 0.008\* **Intellectual disability**  Without (N = 2,813) 55 (2.0) 1  With (N = 25) 5 (20.0) 4.33 (1.05--17.90) 0.043\* **Epilepsy**  Without (N = 2,777) 50 (1.8) 1  With (N = 61) 10 (16.4) 5.15 (2.23--11.93) \<0.001\* ADHD, attention-deficit hyperactivity disorder; ASD, Autism Spectrum Disorder; aOR, adjusted odds ratios; 95% CI, 95% confidence interval; n, number of diagnosed psychotic disorders. \*P \< 0.05. [**Table 3**](#T3){ref-type="table"} provides the effects of pharmacotherapy for narcolepsy on a comorbid psychotic disorder. Prescription of methylphenidate or modafinil for the narcoleptic patients was not significantly associated with a psychotic disorder. However, patients with ASD had a higher risk of having a psychotic disorder (aOR, 10.42; 95% CI, 1.40--77.52). ###### Risk factors of diagnosis with a psychotic disorder in narcoleptic patients. Variables Psychotic disorder ----------------------------- -------------------- --------------------- --------- **Age at recruitment** -- 1.00 (0.98--1.03) 0.755 **Gender**  Male (N = 140) 12 (8.6) 1  Female (N = 118) 9 (7.6) 1.00 (0.38--2.66) 1.000 **ADHD**  Without (N = 235) 16 (6.8) 1  With (N = 23) 5 (21.7) 1.21 (0.26--5.59) 0.812 **ASD**  Without (N = 252) 18 (7.1) 1  With (N = 6) 3 (50.0) 10.42 (1.40--77.52) 0.022\* **Intellectual disability**  Without (N = 252) 19 (7.5) 1  With (N = 6) 2 (33.2) 3.25 (0.38--27.68) 0.280 **Epilepsy**  Without (N = 235) 17 (7.2) 1  With (N = 23) 4 (17.4) 2.02 (0.53--7.67) 0.302 **Methylphenidate**  Without (N = 151) 9 (6.0) 1  With (N = 107) 12 (11.2) 2.39 (0.83--6.86) 0.106 **Modafinil**  Without (N = 224) 19 (8.5) 1  With (N = 34) 2 (5.9) 0.59 (0.11--3.04) 0.528 ADHD, attention-deficit hyperactivity disorder; ASD, Autism Spectrum Disorder; aOR, adjusted odds ratios; 95% CI, 95% confidence interval; n, number of diagnosed psychotic disorders. \*P \< 0.05. We further analyzed the individual effect of IR-MPH, OROS-MPH, and modafinil on psychotic disorders ([**Supplementary Table 1**](#SM1){ref-type="supplementary-material"}). We also examined that association between the risk of psychotic disorders, antipsychotic drugs, antidepressant drugs, and alcohol use disorders. We found that prescription of IR-MPH, OROS-MPH, or modafinil for the narcoleptic patients was not significantly associated with a psychotic disorder. However, prescription of antipsychotic drugs was associated with a higher risk of having a psychotic disorder (aOR, 12.65; 95% CI, 2.90--55.13). Discussion {#s4} ========== The Prevalence of Narcolepsy {#s4_1} ---------------------------- According to our study, the prevalence of narcolepsy in Taiwan is about 12.9 per 100,000, slightly lower than the global mean prevalence of approximately 30 per 100,000 ([@B31]). To our knowledge, reliable evidence about the prevalence of narcolepsy in Taiwan is still lacking currently. In Asia, the prevalence in Korea and China is 15 and 30 per 100,000, respectively ([@B31]). For incidence rates, Dodd et al. ([@B32]) reported that narcolepsy incidence in Taiwan was 0.29 per 100,000 person-years, relatively lower than those in European Union and North America. The lower incidence/prevalence of narcolepsy in Taiwan may be related to under diagnosis or under treatment. Moreover, polymorphism of HLA-DQB1\*0602, H1N1 influenza vaccination, and H1N1 infection itself have been associated with narcolepsy susceptibility ([@B31]) ([@B33]). Various genetic or environment factors may also contribute to the discrepancies in prevalence of narcolepsy across countries. Narcolepsy and the Increased Risk of a Psychotic Disorder {#s4_2} --------------------------------------------------------- This nationwide, population-based cohort study revealed that narcolepsy was associated with a four-fold greater risk of a psychotic disorder. Most relevant studies have discussed the risk of psychosis rather than the psychotic illness among narcoleptic patients. An association between psychosis and narcolepsy has been reported in both adults and children with frequencies in the range 1 to 10% ([@B18], [@B19], [@B25], [@B34], [@B35]). A cross-sectional observational study of 28 narcolepsy and 21 schizophrenia patients found that narcoleptic and schizophrenic patients did not differ with respect to frequency or sensory modality of hallucinations. It also found that the lifetime prevalence of hallucinations did not differ between individuals with schizophrenia and those with narcolepsy ([@B36]). Some reports have found narcolepsy patients with challenging differential diagnoses of psychotic disorders ([@B37], [@B38]). Diagnostic and therapeutic challenges exist for patients with comorbid narcolepsy and psychotic disorder owing to their similar onset times (usually in early life and often in childhood) and their overlapping symptoms ([@B13], [@B14], [@B19], [@B37]--[@B41]). However, our cohort revealed that the age at diagnosis of a psychotic disorder was lower for narcoleptic patients (29.4 years) than for the controls (36.3 years). There were some possible explanations for this. First, patients have dual diagnoses of narcolepsy and schizophrenia have strong biological tendency and tend to onset earlier than patients who had schizophrenia only. The second explanation is that the psychotic disorders among patients who had narcolepsy were diagnosed earlier due to their hypnogogic or hypnopompic hallucination. A cohort study found that narcolepsy is not an under-recognized disease in adult patients with schizophrenia or schizoaffective disorder ([@B42]). It also indicated a possible age-dependent relationship between schizophrenia and narcolepsy, with patients who initially experienced narcolepsy in childhood or early adolescence at a higher risk of developing schizophrenia ([@B6], [@B42]). However, the study did not consider the effects of psychotic disorders on risk of narcolepsy. Additionally, its exclusion of patients who were considered to be unable to answer a questionnaire might have generated bias, such as by the failure to include patients with possibly highly severe functional impairment. Numerous pathophysiological mechanisms potentially explain the high comorbidity of narcolepsy and psychotic disorder. Dopaminergic neurons in ventral tegmental area (VTA) contribute to the regulation of sleep-awake cycle and are involved in the mechanism of cataplexy ([@B43]). Hypocretinergic neurotransmission regulates dopamine firing in ascending midbrain neurons in the ventral tegmental area and prefrontal cortex, and possibly at the level of the nucleus accumbens, which are critical areas in schizophrenia pathophysiology. Therefore, hypocretin, which is known to be an important cause of narcolepsy ([@B4]), may also influence dopamine signaling, causing dysfunction of the mesocorticolimbic and mesocorticostriatal circuits, which are the two main pathways that are implicated in the pathophysiology of psychotic symptoms ([@B44]). Some reports have suggested an overlapping autoimmune pathogenesis between narcolepsy and schizophrenia-like psychosis, associated with both HLA and autoantibodies ([@B34], [@B35], [@B45]--[@B47]). An autoimmune basis of schizophrenia has been suggested ([@B48]), and several studies have examined investigated its association with the HLA region and DQ6 alleles, particularly of the DQB1\*0602 subtype ([@B45]--[@B47]). Narcolepsy is also strongly associated with the HLA DQB1 \* 0602 subtype, which is found in up to 98% of patients ([@B36]). Previous studies have studied a potential autoimmune basis of the relationship between schizophrenia and narcolepsy by assessing NMDAR autoantibodies in either the serum or CSF of patients with anti-NMDA encephalitis, but discordant results have been obtained with respect to narcolepsy ([@B6], [@B34], [@B35]). Some environmental clues about autoimmunity also suggest the overlapping pathogeneses of narcolepsy and psychosis. Several upper airway infections (streptococcal and H1N1 influenza) and Pandemrix vaccinations have been reported as triggering narcolepsy ([@B49]). Prenatal maternal infection with herpes simplex virus type 2, influenza, or cytomegalovirus has been associated with the risk of adult schizophrenia ([@B50]). Early-life infections that are caused by viruses or bacteria have also been associated with the risk of adult psychotic illness ([@B51], [@B52]). A recent review study indicates that severe sleep deprivation itself may cause psychosis ([@B53]). Within 24--48 h of sleep deprivation, perceptual distortions and depersonalization occur. After 48--90 h without sleep, complex hallucinations and delusion develop, yielding psychotic states indistinguishable from acute psychosis or toxic delirium. The underlying biological mechanism may be neuronal instability or a related defect in neural transmission, especially cholinergic, and central chromatolysis ([@B53]). Pharmacotherapy and Psychotic Disorder in Narcoleptic Patients {#s4_3} -------------------------------------------------------------- Our data do not support an association between the use of MPH or modafinil and the development of incident psychotic disorder. Among adolescents and young adults with ADHD who were receiving prescription stimulants, new-onset psychosis occurred in approximately 1 in 660 patients. Amphetamine use was associated with a greater risk of psychosis than MPH ([@B54]). A study revealed that initiation of MPH treatment did not increase the risk of psychotic events in adolescents and young adults, including in those individuals with a history of psychosis ([@B55]). Few investigations have found that a high dose of modafinil may cause mania and psychosis in patients with medical or psychiatric diseases and those who are shift-workers ([@B56], [@B57]). However, in a recent study, modafinil-induced psychosis was identified in only a few case reports ([@B58], [@B59]). In the present study, MPH users among patients with narcolepsy were more likely than others to have a psychotic disorder but not significantly so (OR = 2.39, *P*-value = 0.106). We further analyzed the individual effect of IR-MPH, OROS-MPH, and modafinil on development of psychosis, and there was still no significant association between drug treatment and psychotic disorders. A self-controlled case series revealed no increased risk of incident psychotic events during MPH-exposure relative to periods of non-exposure (incidence rate ratio: 1.02, 95%: 0.53--1.97) ([@B60]). Nonetheless, the prevalence of MPH-induced psychosis has been reported as 18.2% among MPH-treated patients, and this effect should not be considered rare ([@B6], [@B24], [@B25]). Notably, previous studies demonstrated a significantly higher occurrence of psychosis, substance misuse, and psychiatric hospitalizations in patients using high-dose stimulants compared to those using standard doses ([@B24], [@B25]). However, the doses of medication were not available in this study. Further investigations with a larger sample size and comprehensive information about stimulants prescription are warranted to clarify the relationship between stimulants doses and psychotic disorders. One study reviewed data from 49 randomized controlled clinical trials, analyzed the association between psychotic symptoms and the use of ADHD drugs, and found a rate of psychosis/mania events of 1.48 per 100 person-years in the ADHD treatment group. However, only two events were identified in trials of modafinil and four in trials of MPH ([@B61]). We found that prescription of antipsychotic drugs, but not antidepressant drugs or alcohol use disorders, was associated with a higher risk of having a psychotic disorder. This is not surprising because patients who suffered from psychotic symptoms may be prescribed antipsychotic drugs for treating psychotic symptoms. A recent nation-wide study comprised 789 patients with schizophrenia suggested that the use of CNS stimulant (including modafinil and methylphenidate) may reduce in the overall number of psychiatric bed-days ([@B62]). Although CNS stimulant use is not recommended in any treatment guidelines for psychosis, it may be beneficial for cognition, negative symptoms, and the overall function of the patient with schizophrenia. Positive symptoms of schizophrenia are known to be caused by excess dopamine in the midbrain and substantia nigra. CNS stimulants, such as modafinil, may only increase dopamine in the prefrontal cortex. The brain target site influenced by CNS stimulant is different from that for pathophysiology of schizophrenia may count for the possible mechanism that CNS stimulants not seem to enhance psychosis ([@B63], [@B64]). Comorbidities and an Increased Risk of Psychotic Disorder {#s4_4} --------------------------------------------------------- In this investigation, intellectual disability and epilepsy were risk factors for psychotic disorder. Various studies have demonstrated that intellectual disability and epilepsy are risk factors for comorbidity of a psychotic disorder. A cohort study of people with intellectual disability and the general population found an increased diagnosis of psychotic disorders among patients with intellectual disability (OR:10.4) ([@B65]). A systemic review found that individuals with epilepsy have approximately an eight-fold greater risk of psychosis than those without and 6% of them have a comorbid psychotic illness with an even higher prevalence of psychosis in cases of temporal lobe epilepsy (7%) ([@B66]). The neurotoxic effect of epilepsy, a "kindling" process by which acute seizure discharges, a "forced normalization" process, on-going subictal activity in the limbic system, antiepileptic medication, similar structural brain abnormalities and genetic abnormalities in patients with schizophrenia and patients with epilepsy, all may explain the high comorbidity ([@B66]). According to recent studies, ASD seems not to be a common comorbidity of narcolepsy. Our research identified only five cases of concomitant narcolepsy and ASD ([@B67], [@B68]). However, around 2.33% of the patients with narcolepsy had also been diagnosed with ASD in our study. Patients with narcolepsy and ASD were found to have a higher risk of psychotic disorder than those with only narcolepsy. Owing to the few data on the association between narcolepsy and ASD, the pathophysiological pathway of the co-occurrence remains unclear. In narcoleptic patients, ASD may be an additional neurodevelopmental disorder ([@B67]). Although ASD is now believed to be distinct from psychotic disorders, they share multiple phenotypic similarities and risk factors, and reportedly co-occur. Psychotic disorder rates of patients with ASD range from 5 to 16% ([@B69]). Several models have been used explain the association and comorbidity between psychotic disorders and ASD. They include a potential autoimmune basis for a relationship among narcolepsy, ASD and psychotic disorder. Maternal mild acute inflammation that is caused by infection, including influenza, triggers abnormal activation of microglia and overexpression of inflammatory cytokines. This reaction has been observed in animal models. This maternal immune activation generates imbalanced cytokine profiles and characteristics of the offspring that are similar to ASD and schizophrenia ([@B70]). The overlapping pathophysiological mechanisms ([@B49]) potentially explain the association among narcolepsy, ASD, and psychotic disorder. Strengths and Limitations {#s4_5} ------------------------- To the best of our knowledge, this study is the first to investigate the association between narcolepsy and psychotic disorders, and the first to elucidate the relationship between psychotic disorders and pharmacotherapy in narcoleptic patients. The strengths of the investigation include its use of a nationwide population-based database to provide a nationally representative sample, an appropriate observation period, physician-based diagnoses to identify cases of narcolepsy and psychotic disorder, and clear information about the prescription of medications. The use of registration of medical claim data from the NHIRD prevents recall bias and selection bias. However, this investigation has certain limitations. First, misclassifications may arise from the fact that the diagnoses of narcolepsy and psychotic disorders were taken from the registration database rather than mad using validated structural diagnostic instruments. Patients with idiopathic hypersomnia or other hypersomnolence disorders may have been coded as narcoleptics. Besides, the ICD-9-CM does not differentiate between type 1 and type 2 narcolepsy. Moreover, the diagnostic codes for narcolepsy and psychotic disorders in the NHIRD have not been validated, although the accuracy of disease diagnoses in the National Health Insurance system in 1990 to 1991 was reportedly between 54 and 85% ([@B71]), and the validity of diagnosis codes in the NHIRD has been confirmed in previous studies ([@B26]--[@B29]). Second, previous studies have examined many comorbidities of narcolepsy, but this study considered only ADHD, ASD, intellectual disability, and epilepsy. Other confounding factors, such as other organic diseases and psychiatric disorders, may have had an unrecognized effect. Third, the NHIRD is an administrative database that does not contain comprehensive clinical information, such as medication compliance and laboratory examination data. Therefore, our results merely reveal increased associations between psychotic disorders and narcolepsy and no increased associations between psychotic disorders and MPH or modafinil use in narcoleptic patients. This investigation did not analyze whether a psychiatric condition persisted following the diagnosis of narcolepsy or discontinuation of relevant medication. Hence, it could not establish a causal relationship and further longitudinal research is warranted. Finally, this study was carried out among Taiwan's population, and therefore its generalizability to other countries requires further investigation. Besides, the prevalence of narcolepsy varies in different countries and ethnic groups ([@B31]). However, we were unable to control for ethnicity because such information was not registered in the NHIRD. With increasing immigrants in Taiwan, such bias influenced by the different susceptibility to narcolepsy across ethnicity might require other form of study. Conclusion {#s5} ========== This population-based longitudinal study provides evidence that patients with narcolepsy had higher rates of psychotic disorders than a control group. Among narcoleptic patients, the prescription of MPH or modafinil was not associated with a psychotic disorder. However, narcoleptic patients with ASD were found to have a higher risk of a psychotic disorder than those without. Further comprehensive research is warranted to enhance the public awareness of narcolepsy and the possible co-occurrence of psychotic disorders, especially those comorbid with ASD. Continuous efforts must also be made to identify exactly the associated pathophysiological mechanisms and the causal relationship between narcolepsy and psychotic disorders. Data Availability Statement {#s6} =========================== The datasets for this manuscript are not publicly available because: The data are provided by the National Health Insurance Administration, Ministry of Health and Welfare and managed by the National Health Research Institutes. Requests to access the datasets should be directed to nhird\@nhri.org.tw. Ethics Statement {#s7} ================ This study has been approved by the institutional review board at Chang Gung Memorial Hospital. Author Contributions {#s8} ==================== J-YY and Y-CS are joint first authors and contribute equally to this manuscript. J-YY participated in reviewing references and drafting the manuscript. Y-CS executed the statistical analysis. S-YL, S-SY, C-JY, K-CY, T-LL, and C-CS participated in the design of the study. L-JW interpreted the data and revised the manuscript. All authors read and approved the final manuscript and contributed to the drafting and revising of the paper. Funding {#s9} ======= This study was sponsored by the Chang Gung Memorial Hospital Research Projects (CMRPG8D0581, CMRPG2G0071, CLRPG2C0023, and CGRPG2F0021). Conflict of Interest {#s10} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This study is based in part on data from the NHIRD-TW provided by the National Health Insurance Administration, Ministry of Health and Welfare and managed by the National Health Research Institutes (registration number: NHIRD-102-088). The interpretations and conclusions contained herein do not represent those of the National Health Insurance Administration, Ministry of Health and Welfare, or National Health Research Institutes. Supplementary Material {#s11} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpsyt.2020.00205/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Fernando Rodriguez de Fonseca, University of Málaga, Spain [^2]: Reviewed by: Jan Dirk Blom, Parnassia Psychiatric Institute, Netherlands; Francisco Alen, Complutense University of Madrid, Spain [^3]: This article was submitted to Psychopharmacology, a section of the journal Frontiers in Psychiatry

Background {#Sec1} ========== Multiple reports have identified a rising incidence of thyroid cancer over the past several decades, primarily due to the diagnosis of small papillary carcinomas \[[@CR1], [@CR2]\]. According to the World Health Organization, papillary carcinomas that measure 1.0cm or less are diagnosed as papillary thyroid microcarcinomas (PTMs) \[[@CR3]\]. Recent work has suggested that PTMs are the most commonly diagnosed thyroid cancers in individuals who are older than 45 years, and accounts for almost one quarter of new diagnoses \[[@CR1]\]. The clinical significance and recommendations for management of these PTMs is still evolving. In general, PTM is considered a clinically very low risk cancer type that uncommonly causes mortality, with a greater than 90% disease-free survival after long term follow up \[[@CR2], [@CR4], [@CR5]\]. Reported rates for distant metastases and mortality from PTM are \<0.05% \[[@CR6]\]. However, some PTMs exhibit aggressive behaviour, including the development of a regional and distant metastases, and even death \[[@CR7], [@CR8]\]. The rising incidence of PTM, and its associated morbidity, requires risk stratification and guidelines for treatment. Currently, treatment of PTM is based upon stratification of cancer risk for aggressive behaviour, which is based on patient and cancer characteristics \[[@CR2], [@CR4]\]. Cancers considered high risk include those that are: multifocal, have extrathyroidal extension, have vascular invasion, are incompletely resected, have lymph node metastases, especially if multiple nodes are involved, nodes are \> 3cm in largest dimension, or if extra-nodal cancer extension is present, and if distant metastases are present. The presence of these high risk features in a PTM influences the extent of surgery required for treatment \[[@CR9]\]. Small PTM size (\< 5mm versus ≥ 5mm) has also been suggested as being important for risk stratification, with some reports identifying PTMs ≥ 5mm as being more likely to have high risk features \[[@CR8], [@CR10], [@CR11]\]. The objective of this study was to determine if small PTMs (size \< 5mm) are clinically and pathologically different from larger PTMs (size 5mm- ≤ 10mm). Methods {#Sec2} ======= The study protocol was approved by our Research Ethics Board. This study was a retrospective analysis of data collected prospectively between 2002 and 2012 for 1459 sequential patients undergoing thyroid surgery at St. Paul's Hospital, a single tertiary care/academic teaching hospital in Vancouver, Canada. Analysis was limited to cases where pathology identified a PTM of ≤ 1cm *without* the presence of an associated macrocarcinoma (\> 1cm). Applying this inclusion and exclusion criteria resulted in 132 cases for analysis. This cohort was then examined based on four clinical and five histopathological characteristics. Clinical variables evaluated were: age (\< 45 versus ≥ 45 years), gender (male versus female), surgery performed (total thyroidectomy versus thyroid lobectomy), and surgical indication (benign disease versus known or suspected cancer). Surgical indication was based on pre-operative fine need aspiration (FNAB) and/or imaging findings. Diagnosis of suspicion of malignancy was based upon FNAB diagnosis of nodules, or pre-operative imaging suspicious for local or distant metastases. Histopathological variables evaluated were: cancer multifocality, cancer bilaterality, extra-thyroidal cancer extension, and the presence of local and distant cancer metastases, all based on histopathological reports. Cancers were considered multifocal if ≥ 2 foci were found in one or both thyroid lobes. Size in millimeters (\< 5mm vs ≥ 5mm) was the independent variable. The subtype histology of the PTMs were not available for all cases and was not included in the overall analysis. Statistical analysis {#Sec3} -------------------- Statistical differences between groups were determined using the Chi Square and Fisher's exact tests as appropriate, with a *p*-value \< 0.05 considered statistically different and a priori. Analysis was performed in *R*. Results {#Sec4} ======= One hundred thirty-two of 1459 patients (9%) met study inclusion criteria and made up the study patient population. Clinical characteristics {#Sec5} ------------------------ The clinical characteristics and treatment of the study patient population is summarized in Table [1](#Tab1){ref-type="table"}. Of the 132 patients included, there were 105 (80%) women and 27 (20%) men. The mean age at the time of their procedure was 50.2 years (range 23--74 years) with 48 patients (36%) being less than 45 years old and 84 patients (64%) being aged 45 years or older. The surgical procedure was a total thyroidectomy for 63 patients (48%) and thyroid lobectomy for 69 patients (52%). Indication for surgery was for presumed benign disease in 33 cases (25%) and for suspicion of or confirmed malignant disease in 99 cases (75%). The suspicion of malignancy was generally due to an intermediate Fine needle aspiration biopsy of an associated thyroid nodule \> 1cm. Benign pathology included goiter, Hashimoto's thyroiditis, thyroid cyst, hyperplastic nodules, and Hurthe cell adenomas, all with PTM that was identified incidentally. All cases with confirmed malignancy pre-operatively were papillary thyroid cancer.Table 1Papillary Microcarcinoma Patient and Operative CharacteristicsPatient and operative characteristics\< 5mm\ *n* = 75≥ 5mm\ *n* = 57Total *N* = 132Chi square*Age* \<45222648 (36%)*p* = 0.08 ≥45533184 (64%)*Gender* Male111627 (20%)*p* = 0.09 Female6441105 (80%)*Surgery* Total Thyroidectomy372663 (48%)*p* = 0.80 Thyroid Lobectomy383169 (52%)*Surgical Indication* Benign201333 (25%)*p* = 0.76 Malignant554499 (75%)Benign pathology included goiter, Hashimoto's thyroiditis, thyroid cyst, hyperplastic nodules, and Hurthe cell adenomas associated with papillary thyroid microcarcinoma. Malignant pathology (confirmed or suspected) was based on fine needle aspiration biopsy of nodules or pre-operative imaging. All cases with confirmed malignancy pre-operatively were papillary thyroid cancer At least 1 histopathologic high risk feature was identified in 45 patients (34%). Twenty-seven patients (20%) had 1 high-risk feature, 14 patients (11%) had 2 high risk features and 4 (3%) had 3 high risk features. No patients had 4 or 5 high risk histopathologic features. Multifocality was identified in 29 patients (28%), bilaterality in 8 patients (6% of total, 27% with multifocal disease), extrathyroidal cancer extension in 6 patients (5%), lymph node metastases present at diagnosis in 9 patients (7%) and distant metastases present at diagnosis in 1 patient (0.7%). The study population's characteristics are summarized in Table [2](#Tab2){ref-type="table"}.Table 2High Risk Characteristics of Papillary Thyroid Microcarcinoma Patient PopulationHigh Risk Feature\<5 mm\ *n* = 75(%)≥ 5mm\ *n* = 57(%)Total (% of 132)Chi square/Fisher'sCancer Bilaterality6 (8)2 (4)8 (6)0.48Cancer Multifocality21(28)17 (30)38 (29)0.97Extrathyroidal Cancer Extension0 (0)6 (11)6 (5)*p* = 0.005Lymph Node Metastases5 (7)4 (7)9 (7)0.78Distant Metastases1 (1)0 (0)1 (0.7)0.89*Total*242953 Chi square or Fisher's Exact test comparison for patient and surgical variables of interest failed to identify a difference between small or large PTM (Table [1](#Tab1){ref-type="table"}). Examination of high risk features revealed that extrathyroidal cancer extension was present significantly more often in large PTM (*p* = 0.005). Other high risk features were not differentially identified based on cancer size (Table [2](#Tab2){ref-type="table"}). The 9 patients who had lymph node metastases present at the time of their cancer diagnosis were further studied. Six of nine patients (67%) had bilateral disease and two of the six also had extrathyroidal extension (both large PTM), while two others had associated bilateral disease (both small PTM), including one patient with distant metastases from a 3mm cancer. Three of the nine patients (33%) had lymph node metastases in the absence of any high risk features (Table [3](#Tab3){ref-type="table"}).Table 3Patient Population With Lymph Node Metastases*PtAgeSize mmHistologyBilateralMultifocalETELN MetsDis*\ *Mets***7\< 459PTMC----**--**+**--1\< 458PTMCvF--+++--**8**≥ **456PTMC**------**+**--3≥ 455PTMC--+++--2≥ 453PTMCvS++--+--5\< 453PTMC--+--+--6≥ 453PTMCvF++--++4≥ 452.5PTMCvS--+--+--**9**≥ **452PMTCvF**------+--*Pt* patient, Bilateral bilaterality, *Multifocal* multifocality, *ETE* extra-thyroidal extension, *LN* Mets lymph node metastasis, *Dis Mets* distant metastasis, *PTMC* papillary thyroid microcarcinoma, *vF* variant follicular, *vS* variant sclerosing+, present; -, absentBold values, LN metastasis in absence of any other high risk features Discussion {#Sec6} ========== PTM diagnosis is very common with almost half of all new papillary thyroid cancer (PTC) diagnoses being a PTM. Following thyroid surgery, pathologic identification of PTM ranges from 1--24% of surgical specimens \[[@CR2], [@CR5], [@CR10]\], and 6--36% of autopsy specimens \[[@CR2], [@CR5]\]. Analysis of the Surveillance, Epidemiology and End Results database revealed that PTM in patients older than 45 years is now the most commonly diagnosed PTC \[[@CR1]\]. Our observed rate of PTM diagnosed from surgical specimens in the absence of a thyroid macrocarcinoma was 9%, and was consistent with Canadian autopsy (6%) \[[@CR12]\], international autopsy (3--18%) \[[@CR2]\] and pathology data (1--24%) depending on the extent of surgical resection \[[@CR2], [@CR11]\]. Guidelines regarding the surgical management of PTC have continued to evolve and are currently based on clinical and pathological characteristics of the diagnosed individual and their thyroid cancer, respectively \[9}. However, debate still exists regarding whether PTM should be managed in a manner similar to larger PTC \[[@CR4]\]. PTM usually has an excellent prognosis and current American Thyroid Association Guidelines recommend that thyroid lobectomy alone is adequate treatment for unifocal intrathyroiodal cancers in the absence of a prior history of head and neck irradiation, familial thyroid carcinoma, or clinically detectable cervical metastases. As there are no features of PTM that can reliably identify the small subset of patients who will go on to develop clinically significant cancer progression, an active surveillance approach that involves close clinical and ultrasound PTM follow up, has currently not been adopted by the vast majority of centers \[[@CR9]\]. An observational study, with non-surgical management of PTM, reported 3mm of growth in 6% of patients, with 1.4% of patients developing lateral lymph node metastases at 5 years of follow-up. These authors have suggested non-operative management of PTM in the absence of high risk features \[[@CR13]\]. However, other groups have encouraged a surgical approach, which has shifted from thyroid lobectomy to total thyroidectomy, and some groups even recommend nodal dissection of the central neck. Proponents of the latter approach recognize the relationship between high risk features and disease recurrence \[[@CR14], [@CR15]\]. High risk features are often present in PTM and one-third of our patients had at least 1 high risk feature identified after pathological examination. While distant metastases were uncommon in our series (\< 1%), consistent with other reports \[[@CR6]\], multifocality was observed in almost a third of cases (28%) and 27% of these cases had bilateral disease. Reported multifocality rates for PTM have ranged from 15--43% \[[@CR2], [@CR15]\], with contralateral malignancy identified in up to 53% of cases \[[@CR16]\]. Interestingly, multifocality is believed to reflect multiple independent tumors of different clonal origin, as opposed to intraglandular cancer metastases \[[@CR17]\]. While the incidence of PTC multifocality is not related to tumor size \[[@CR18]\] as we observed in the current study, it has been reported to be an independent risk factor for cancer recurrence \[[@CR6], [@CR17]\]. However, the prognostic significance of contralateral PTM is controversial. Given that only 5--10% of the PTCs recur, some investigators have suggested that recurrence can be treated when detected clinically, while other investigators have argued that reoperative surgery carries additional risks that need to be weighted appropriately \[[@CR14]\]. The presence of lymph node metastases at the time of diagnosis also influences cancer recurrence risk and has been reported in 5--51% of cases \[[@CR2], [@CR15]\]. The current study observed a 7% rate of nodal disease at the time of PTC diagnosis is at the lower end of this range, but is consistent with the 9% nodal disease risk that has been recently associated with PTM \[[@CR15]\]. This low rate of lymph node metastases may be an underestimate as the practice of central neck dissection for these cases is variable. Long-term follow-up of PTM has found that the presence of multifocality and lymph node metastases at diagnoses increases the risk of developing subsequent nodal disease, 11% of multifocal PTM recurred compared with 4% of unifocal PTM tumors \[[@CR19]\]. Three of nine patients diagnosed with lymph node metastases at the time of their diagnosis had no other high risk features present. The presence of nodal metastasis in association with PTM is believed to represent different underlying cancer biology when compared with PTM not associated with nodal metastases \[[@CR5]\]. Aggressive biological behavior may be related to molecular characteristics not yet well understood. Recent reports have identified molecular BRAF mutations, present in 17--52% of PTM, as potential early genetic lesions in PTC \[[@CR2], [@CR20]\]. Presence of BRAF mutations are associated with high risk features, including extrathyroidal cancer extension (ETE) and multifocality, and are also predictive of an increased risk of lateral compartment nodal disease \[[@CR20]\]. Lin and colleagues have suggested that pre-operative screening for BRAF mutation using fine needle aspirate biopsy could potentially guide the initial treatment of PTM, with positive BRAF status requiring more aggressive therapy \[[@CR20]\]. Other recent studies have found that combining BRAF mutations with a panel of other molecular prognosticators may improve its clinical utility \[[@CR9]\]. Some investigators have reported that the tumor size of PTM (\< 5 vs ≥ 5mm) may be related to aggressive cancer behavior, with larger PTMs being more likely to have high risk features and nodal metastasis \[[@CR10], [@CR21]\]. Other groups have found no relationship between PTM size and the presence of high risk features \[[@CR22], [@CR23]\]. Our study found that the size of PTM had little relationship with the presence of high risk features. Only ETE was significantly more commonly present in larger than smaller PTM. While the overall rate of ETE was low (5%), these observations were consistent with prior reports (2--21%) \[[@CR2], [@CR5]\]. Kim et al. \[[@CR24]\] found that the presence of ETE was directly related to the PTM size but not with the risk of cancer recurrence. Other groups have evaluated the relationship between cancer size and high risk features concluded that PTM size did not influence patient outcomes \[[@CR10], [@CR21]\]. The current study has several limitations. It is a retrospective analysis from a single institution and results may not be generalizable to all populations. However, the prospective data collection from a multi-cultural catchment area in a publically funded health care system provides support of the representative nature of the sample. The incidental identification of PTC after surgery for benign disease implies that lymph nodes could be under-sampled, and presence of metastatic disease may be higher than our results suggest. As well, outcome and follow-up data is not available and thus no prognostic information can be inferred from this analysis. Conclusions {#Sec7} =========== The diagnosis of PTM after thyroid surgery presents the thyroid cancer management team with a dilemma. Neither PTM size, nor the absence of high risk features, excluded the possibility of synchronous lymph node metastases. While some argue that high risk features should prompt surgical management consistent with PTC \>1cm, further study that includes the evaluation of the role of thyroid cancer molecular prognostications is required in order to identify the optimal management algorithm for this common endocrine malignancy. PTM : Papillary thyroid microcarcinoma PTC : Papillary thyroid cancer ETE : Extrathyroidal cancer extension The authors would like to acknowledge Mr. Jeremy Hamm, Biostatistician, for his assistance with data analysis in this study. Funding {#FPar1} ======= This research study has not received any funding support from any source. Availability of data and materials {#FPar2} ================================== The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Authors' contributions {#FPar3} ====================== NB carried out the literature search, drafted the tables, collected data, analysed data, data interpreted data, and drafted the manuscript. SMW designed the study, collected data, analysed data, interpreted data, and edited the manuscript. All authors read and approved the final manuscript. Authors' information {#FPar4} ==================== Both study authors can be contacted at: Department of Surgery, St. Paul's Hospital & University of British Columbia, Current Address: 1081 Burrard Street, Vancouver, British Columbia, Canada, V6Z 1Y6. Competing interests {#FPar5} =================== No funding was received for this research study and there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported. Consent to publication {#FPar6} ====================== As no details/images/videos that would allow for identification of study participants are presented in this work Consent for Publication is not applicable. Ethics approval and consent to participate {#FPar7} ========================================== This study was approved and consent to participate was obtained by the ethics committee of Providence Health Care Research Ethics Board of St. Paul's Hospital and University of British Columbia. Informed consent was obtained from all the participants. Additional details {#FPar8} ================== This study was presented at the 82^nd^ American Thyroid Association Annual Meeting, September 19--21, 2012, Quebec City, Canada.

Sir, Hyperammonemia as a cause of encephalopathy in a patient with normal liver enzyme levels is unusual.\[[@ref1][@ref2]\] Hence, it remains undiagnosed especially in intensive care (ICU) settings.\[[@ref3]\] We recently had a patient with *Klebsiella* sepsis and hyperammonemic encephalopathy (HE) with normal liver function tests. We report this unique case after taking informed consent from patient\'s kin. A 63-yr-old house-wife presented to our ICU with fever and dysuria for 10 days and impaired level of consciousness with hypotension for 1 day. She was having adequately controlled type-2 diabetes mellitus for last 10 years and hypothyroidism for 3 years. On examination, her best Glasgow coma score was 9 (E~3~V~2~M~4~) with no signs of meningeal irritation. In view of hemodynamic instability trachea was intubated. She was treated with broad spectrum (initially empirical β-lactamase and β-lactamase inhibitor then culture-based carbapenems) antibiotics, mechanical ventilation, enteral nutrition, fluids, and vasoactive agents. On admission liver function test revealed: Albumin 2.8 g/dl, total bilirubin 0.5 mg/dl, AST 37 U/L, ALT 26 U/L, normal INR (maximum1.4), alkaline phosphatase 152 U/l. Her urine culture sent on admission showed growth of *Klebsiella pneumoniae*. Other cultures were sterile. Within 7 days, her urosepsis resolved but her sensorium did not recover. Her renal function, serum electrolytes, arterial blood gases, random blood sugar, perfusion parameters (lactate), thyroid profile were all within normal limits. Repeat liver function tests revealed high alkaline phosphatase (2000 U/l) and g-glutamyl transferase (670 U/l) levels only with negative viral hepatitis markers (initially hepatitis A, B, C, D, and subsequently hepatitis E also). The patient had neither gastrointestinal bleeding nor on intravenous steroids. Abdominal ultrasonography was not suggestive of chronic liver disease. We did brain magnetic resonance imaging (MRI), which suggested (T1 and T2 axial view) age unrelated cerebral atrophy and empty sella \[Figures [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}\]. Cerebro-spinal fluid (CSF) showed a normal study (repeated twice). An electroencephalogram showed continuous generalized slowing suggestive of metabolic encephalopathy. In view of raised alkaline phosphatase and glutamyl transferase with MRI picture (suggesting astrocyte loss) gastroenterologist opined for arterial ammonia level. Her arterial ammonia level (sent on ice and processed within 30 minutes) came extremely high \[265 μmol/L (normal range 9-33 μmol/L)\]. Interestingly, there was no history of any drug intake (like valproate, 5-fluorouracil, salicylate) before and during this hospitalization. Serum M-protein level was also normal. In view of old age and short history, underlying urea cycle enzyme abnormality was not evaluated. We restricted protein intake (0.8 g/kg), initiated dialysis, and started L-carnitine therapy. Her sensorium improved with decreased arterial ammonia level (148 μmol/L). During this period she developed new onset blood stream infection by multi-drug-resistant *K. pneumoniae*. We escalated broad-spectrum antibiotics. Patient, however, developed refractory septic shock and succumbed to her illness. Post-mortem liver biopsy could not be done. {#F1} {#F2} Normally ammonia produced in body is metabolized and eliminated from body by liver. When ammonia production is excessive, portal blood-carrying ammonia can bypass liver leads to hyperammonemia.\[[@ref2]\] As ammonia is freely permeable through blood-brain barrier, astrocytes metabolize excess ammonia to glutamine with the help of enzyme glutamine synthetase.\[[@ref2][@ref3]\] Thus, osmotically active glutamine concentrations increase in brain, leading to astrocytic swelling and destruction. Importantly, astrocytes release inflammatory cytokines (like tumor necrosis factor and interleukins) causing apoptotic astrocyte loss. In HE, neuropathological findings like ventriculomegaly, cerebral cortical atrophy, especially increase in the sulcal depth of the frontal, parietal, regions, and intracranial bleed indicate astrocyte loss.\[[@ref4]\] MRI scans in our case were also suggesting astrocyte loss. HE is usually related with liver dysfunction,\[[@ref3]\] drugs (like valproate, salicylate)\[[@ref5]\] or malignancies (like multiple myeloma).\[[@ref2]\] Due to her old age and short history (as proximal urea cycle disorder is unlikely), we attribute *Klebsiella* sepsis as the most probable reason for HE in our case. *Klebsiella* is a urease-splitting organism. Hyperammonemic coma was described with such urease-splitting organism with anatomical abnormality of lower urinary tract.\[[@ref2]\] *Klebsiella* can cause excessive ammonia production and absorption of ammonia in urinary tract leading to hyperammonemia.\[[@ref6]\] Hence, we suggest early blood ammonia testing in cases of *Klebsiella* sepsis with encephalopathy in intensive care settings.

Background {#Sec1} ========== The management of poor ovarian respond (POR) remains a serious dilemma for fertility clinicians. The incidence of POR, which was associated with very low clinical pregnancy rate (CPR), was reported to be 9--24% \[[@CR1]\]. Among infertile women in vitro fertilization (IVF) cycles. CPR was only 14% and the cycle cancellation rate of no available embryos was as high as 40% for women with fewer than five oocytes \[[@CR2]\]. It was confirmed that the outcomes of IVF/intracytoplasmic sperm injection (IVF/ICSI) would be improved with increasing number of oocytes retrieved, especially for those with poor ovarian response or advanced maternal age \[[@CR3]\]. Various approaches have been used to obtain more mature oocytes and available embryos for POR. Different stimulation protocols \[[@CR4]--[@CR7]\], increasing gonadotropin dosage \[[@CR8], [@CR9]\], combining pretreatment protocol \[[@CR10]--[@CR12]\], and usage of adjuvant \[[@CR13], [@CR14]\] have been tried in IVF/ICSI cycles. But there is no evidence from randomized clinical trials evaluating the effect of dehydroepiandrosterone (DHEA) preconceptional treatment on live birth in POR. Androgen in the ovarian microenvironment plays an important role in early follicular development and granulosa cell proliferation. Administration of androgen was suggested as a promising treatment. Androgens can upregulate the follicular follicle-stimulating hormone (FSH) receptor expression, augment follicular sensitivity to FSH stimulation, and increase the number of antral follicles and mature oocytes \[[@CR15]\]. Clinical observation \[[@CR16]\] indicates that the lack of androgen is associated with low ovarian response and pregnancy rate during IVF. The most widely used androgen in the clinic is DHEA. DHEA has low androgen activity but can transform highly active androgen *in vivo*; meanwhile, the incidence of adverse reactions is low. Numerous reports indicated that DHEA supplementation in patients with decreased ovarian reserve or POR was helpful to improve ovarian reserve parameters \[[@CR17], [@CR18]\], augment ovarian response \[[@CR19]\], reduce age-related aneuploidy \[[@CR20]\], and increase pregnancy rate \[[@CR21]\]. However, most of the studies were based on retrospective or observational data or both. Several randomized controlled trials (RCTs) were criticized for inappropriate study design, the varied definition of POR, and duration of DHEA \[[@CR22]\]. There is an urgent need for large-scale, well-designed confirmatory studies to prove the efficacy of DHEA before it could be recommended for routine use. The purpose of this multicenter randomized placebo-controlled double-blind clinical trial is to evaluate the effect of DHEA treatment for 12 weeks before IVF/ICSI on the live birth rate in infertile POR. Methods/Design {#Sec2} ============== Study design {#Sec3} ------------ This study has a multicenter, randomized active-placebo, double-blind design. Eligible patients will be randomly assigned to active group or placebo group with a 1:1 ratio. Figure [1](#Fig1){ref-type="fig"} shows a flowchart of the study design.Fig. 1Study flowchart Study sites {#Sec4} ----------- This clinical trial involves eight hospitals in China. Live birth rate of active group versus placebo group will be compared in a pool of 710 infertile POR patients undergoing their sequence cycle of IVF/ICSI. The study will be conducted in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki. The study has been approved by the ethics committees at all hospitals and facilities. The trial and study plan will be declared to all participants at their first visit. If patients are eligible and interested in participating, the couples will be asked to sign an informed consent form giving permission to use their results anonymously in future studies. Reporting of the study results will follow the 2010 revised CONSORT (Consolidated Standards of Reporting Trials) statement \[[@CR23]\]. An overview of specific measurements and time points of data collection can be found in the SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) Figure (Fig. [2](#Fig2){ref-type="fig"}, Additional file [1](#MOESM1){ref-type="media"}).Fig. 2Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) figure Training of investigators {#Sec5} ------------------------- All investigators are required to undertake mandatory training in the protocol, Good Clinical Practice, and the use of CONSORT if they had no recent certifications. ### Study population {#Sec6} The study will include patients with POR. Inclusion criteria are (1) women who had tried to become pregnant for at least 1 year, (2) women who will undergo IVF or ICSI, and (3) women with POR diagnosed according to the Bologna consensus \[[@CR24]\]. Exclusion criteria are (1) women who had failed in three or more IVF cycles with ovarian stimulation and oocyte pickup, (2) women with diagnosed uterine malformation or abnormality, (3) women with diagnosed hydrosalpinx in hysterosalpingography, (4) women who had a history of two or more spontaneous miscarriages (biochemical miscarriage does not count in), (5) couples with female or male abnormal karyotype, (6) women who had undergone chemotherapy with cytotoxic agents, (7) women who had undergone pelvic radiotherapy, (8) women who had taken DHEA before enrollment, (9) women with a history of epilepsy, and (10) women who are allergic to DHEA. Determination of the sample size {#Sec7} -------------------------------- The sample size calculation is based on the live birth rate. Data from our hospital indicated that the live birth rate of infertile POR undergoing IVF/ICSI cycles was 20%. In our present study, we plan to test the primary hypothesis of a difference of 10 in the live birth rate for the two randomization arms. A sample size of 294 prospectively enrolled participants in each randomization arm will yield a statistical power of 80 at a *P* value of less than 0.05 to demonstrate a significant difference in live birth rates of 10 between treatment arms. In consideration of a dropout rate of 20%, we will enroll 355 participants each group. Randomization and blinding {#Sec8} -------------------------- Eligible participants will be randomly assigned to one of the two study arms at a ratio of 1:1 according to an Excel table randomization list generated by a data coordinative center (DCC) staff who are independent of the study group, and the allocation will be input into the online database central grouping system. The random number list is kept strictly confidential by the DCC staff, and the researchers who are in charge of the enrollment have no access to the list. Both the placebo and active component are manufactured by the pharmacist who was not involved in the subjects' clinical management. Except for the active ingredients, the rest of the excipient and the appearance and odor are the same. The pharmacist packaged the DHEA and identical placebo capsules according to the randomization list and labeled the drug packs with subject numbers only. After screening the patients eligible for the entry, the researcher will log into the database to obtain the patient's research number. Then the pharmacist will dispense the corresponding drug. The researchers and study participants will be blinded to the allocation until the end of the study. ### Discontinuation criteria {#Sec9} Reasons for discontinuation of treatment may include, but are not limited to, the following: (1) participants become pregnant during the trial; (2) participants who have experienced some complications or serious side effects; (3) participants who do not comply with the DHEA were defined by using less than 80% or more than 120% of the prescribed amount; using prohibited drugs as described in the protocol, such as growth hormone and glucocorticoids; (4) participants who request to withdraw from the trial. ### Intervention {#Sec10} The study will be conducted in following stages. After a recruitment period prior to baseline assessment and randomization, DHEA/placebo will be given to patients three times a day orally from the first day of their menstrual period for 4--12 weeks. After that, participants will undergo their IVF/ICSI cycle. Finally, a live birth rate follow-up will be performed. Controlled ovarian hyperstimulation {#Sec11} ----------------------------------- A short protocol will be suggested to all patients. On cycle day 2, gonadotropin-releasing hormone (GnRH) agonist (Triptorelin; Tiantai Mountain Pharmaceutical Company, Chengdu, China) 0.05\~0.1 mg will be initiated. From cycle day 4, all patients will be stimulated with gonadotrophin (Gn) (FSH/human menopausal gonadotrophin; Livon, Zhuhai, China) 150 or 225 IU/day. Both GnRH agonist and Gn continue until the day of human chorionic gonadotropin (hCG) administration. On stimulation day 6, transvaginal ultrasound scans and serum hormone tests will be performed. The dose of Gn will be adjusted according to follicular development and endocrinometry. Afterward, such monitoring will be performed every other day or every day. When more than two lead follicles reach at least 18 mm, 6000--10,000 IU of hCG will be administered to trigger final follicular maturation. Oocyte retrieval, in vitro fertilization, and embryo culture {#Sec12} ------------------------------------------------------------ Transvaginal ultrasound-guided ovum pickup takes place about 34--36 h after hCG administration. Oocytes will be inseminated by IVF or ICSI according to the quality of sperm. Embryo transfer (ET) is routinely performed on days 3 after fertilization at the cleavage stage, and a maximum of two top quality embryos will be transferred through a catheter by using transabdominal ultrasound guidance (full bladder). If the quality of the embryo is poor, ET will be performed on day 2 or 5. Blastocyst culture and vitrification will be performed for the surplus embryos. All of the operations will be performed by experienced clinicians. Luteal phase support {#Sec13} -------------------- Dydrogesterone tablet (Duphaston, Abbott, Hoofddorp, The Netherlands) 20 mg twice daily orally or vaginal progesterone gel (Crinone gel, 8%, Merck Serono, Geneva, Switzerland) 90 mg once daily will be used for luteal phase support (LPS) from the day after oocyte retrieval for 2 weeks. After a positive pregnancy test (serum β-hCG ≥10 IU/L), LPS will be continued until at least 10 weeks' gestation. If β-hCG is negative or if the pregnancy fails, LPS will be stopped. If a biochemical pregnancy has been achieved, ultrasound scan will be performed at 5--6 weeks after ET to assess clinical pregnancy and repeated at 12 weeks' gestation to confirm ongoing pregnancy. ### Outcome measurement {#Sec14} #### Primary outcome {#FPar1} The primary endpoint of study is live birth, defined as delivery of any neonate at at least 28 weeks' gestation with signs of life after ET. #### Secondary outcomes {#FPar2} Secondary outcomes will include (1) change of ovarian reserve index (basal serum FSH, anti-Müllerian hormone (AMH), and antral follicle counts), (2) indexes of ovarian response (total gonadotropin dosage, estradiol (E2) on the HCG trigger day, and retrieved oocyte number), (3) indicators of oocytes quality (mature oocyte rate, normal fertilization rate, and number of top-quality embryos), and (4) CPR and miscarriage rate. Biochemical pregnancy is defined as a serum β-hCG level of at least 10 IU/L 2 weeks after ET. Clinical pregnancy is defined as detection of intrauterine gestation sac after 3 weeks after a biochemical pregnancy is confirmed. Ongoing pregnancy is defined as detection of a viable fetus with fetal heartbeat at 11--12 weeks' gestation. Cumulative live birth will be defined as each woman having only one allowable live birth from the single IVF cycle. #### Follow-up protocol {#FPar3} First pregnancy follow-up time point is at 12 weeks' gestation when the participants come for their clinic visits. The presence of first-trimester pregnancy complications (Ovarian hyperstimulation syndrome (OHSS), miscarriage, ectopic pregnancy, or gestational trophoblastic disease or a combination of these) will be evaluated by inspecting medical records and be recorded by completing the case report form. Moderate OHSS is diagnosed when ultrasonographic ascites is present in addition to abdominal distension and discomfort with or without nausea, vomiting, or diarrhea or a combination of these. Severe OHSS is diagnosed when there is clinical evidence of ascites and/or hydrothorax or breathing difficulties with or without hemoconcentration, coagulation abnormalities, and diminished renal function. Second pregnancy follow-up time point is at 28 weeks' gestation. The second-trimester pregnancy complications (prenatal diagnosis test results, abortion, gestational diabetes, preeclampsia/eclampsia, incompetent cervix, premature rupture of membrane, or placenta abruption or a combination of these) will be followed up by telephone call. Third pregnancy follow-up time point is at 37 weeks' gestation. The third-trimester pregnancy complications (preterm labor, placenta abruptio, placenta accreta, placenta previa, preeclampsia/eclampisa, intrauterine growth retardation, premature rupture of membrane, or abnormality of amniotic fluid or a combination of these) will be followed up by telephone call. Fourth follow-up time point is at delivery. Participants will be required to inform investigators of the delivery time. The obstetrical and perinatal complications (gestational age, delivery mode, placenta abnormality, or delivery complications or a combination of these) as well as neonatal information (gender, birth weight, and birth defect) will be obtained and recorded to complete designed forms. If possible, copies of the obstetric medical records will be performed for the study chart source documents. The fifth and final follow-up time point is at 6 weeks after delivery. Postpartum complications (postpartum depression, infection, and late postpartum hemorrhage) and neonatal complications (neonatal respiratory distress syndrome, neonatal jaundice, neonatal infection, neonatal death, and neonatal hospitalization) are followed up by telephone call. #### Adverse events {#FPar4} Adverse events are defined as any untoward or unfavorable medical occurrences associated with the subject's participation during the research regardless of whether they are considered related to the study intervention. Serious adverse events are defined as events that are temporally associated with the subject's participation in research: death, life-threatening or severely or permanently disabling events, events requiring in-patient hospitalization or prolongation of existing hospitalization, pregnancy loss after 20 weeks' gestation, neonatal death up to 6 weeks after delivery, congenital anomaly or birth defect, or any events deemed serious by the local principal investigator. All of the adverse events will be recorded in detail. Serious adverse events will be reported to the principal investigator immediately, and appropriate measures will be initiated instantly. The ethics committee will determine whether the adverse event is likely to have been associated with the experimental drug and whether it is necessary to break blinding codes. Data management {#Sec15} --------------- All data will be collected with a standard case report form developed in the web-based data entry system by the double-entry method. Before being input into the database, data will be de-identified. Paper files will be kept in a locked filing cabinet in the treating hospital. Electronic documents will be stored in a password-protected computer, and access will be restricted to the principal investigator. The study site monitor of Shandong University will check the authenticity, accuracy, and integrity of the data regularly to ensure the accuracy of collected data. The data will be stored for at least 5 years after publication. All data will be supervised by independent statisticians from Shandong University. The clinical research must be carried out under the rules on China's New Drug Examination and Approval and Management Standard of the Clinical Test of Drugs (Good Clinical Practice). Data monitors will check and review the quality of data collected in the research accordingly. ### Data analysis plan {#Sec16} Statistical analysis will be performed by using Statistical Package for the Social Sciences Version 22.0 (SPSS Inc., Chicago, IL, USA). Normally distributed continuous variables are expressed as mean ± standard deviation with a Wilcoxon rank-sum test for testing between-group differences. Non-normally distributed continuous variables are expressed as medians and ranges, and categorical data are described as frequency, percentage, and constituent ratio. The primary analysis will use an intention-to-treat analysis approach to examine differences in the live birth rate between the two treatment arms by the Pearson chi-squared test. Secondary efficacy parameters and safety parameters (pregnancy rate, OHSS rate, and so on) will be analyzed by using the Pearson chi-squared test. The Wilcoxon rank-sum test will be performed to compare continuous parameters (retrieved oocyte number, gonadotropin dosage, E2 level on the hCG trigger day, and so on). The analysis of covariance (ANCOVA) tests will be used to compare the indicators (basic FSH, E2 level, and AMH level) pre- and post-treatment between-group, and pre-post paired *t* tests will be used within-group. A secondary per-protocol analysis will be performed according to the actual treatment that subjects received. Results will be presented as point estimates and 95% confidence intervals; the level of significance will be set at 5%. The flowchart of this study is presented in Fig. [1](#Fig1){ref-type="fig"}, and the SPIRIT checklist is included as Fig. [2](#Fig2){ref-type="fig"}. Discussion {#Sec17} ========== This study will compare the efficacy of 12-week DHEA treatment prior to IVF in POR. We plan to enroll 710 subjects from eight academic IVF centers in the Shandong Province. The result of this large multicenter randomized trial will provide level I evidence for the strategy of DHEA treatment for 12 weeks before IVF in poor ovarian responders. DHEA treatment before IVF in POR has attracted increasing attention and has been considered promising for POR \[[@CR25]\]. However, there are still great gaps in the literature to illuminate the risk/benefit ratio of this strategy, which includes multiple steps of treatment. Some research \[[@CR20], [@CR26]--[@CR29]\] showed that DHEA supplementation improved the indicate of ovarian reserve, augmented ovarian response, and increased the clinical pregnancy rate. Meanwhile, the others \[[@CR30]--[@CR32]\] showed there were no such benefits. Some small-scale RCTs \[[@CR28], [@CR30], [@CR31], [@CR33]\] have been conducted to compare the efficacy of DHEA treatment before IVF. However, the varied definition of the patient population and primary outcome may result in the heterogeneity of comparisons. Additionally, the lack of rigorous design strategies, large scale, and high quality also make it difficult to reach any consistent conclusion \[[@CR34]\]. This study is a multicenter randomized placebo-controlled double-blind clinical trial. In this research, we define poor ovarian responder according to the Bologna consensus \[[@CR24]\], which is widely applied in clinical research after the recommendation. In 2003, the European Society of Human Reproduction and Embryology (ESHRE) recommended the singleton live birth rate as the evaluation index of assisted reproductive technology (ART) or non-ART reproductive outcome \[[@CR35]\]. A healthy baby is the purpose of couples. That is the reason we chose live birth rate as the primary endpoint for this study. Despite the lack of convincing evidence, DHEA supplementation is being used by more and more IVF centers around the world for POR. Large-scale, randomized placebo-controlled, blind, clinical trials are urgent to suggest physicians whether or not to use DHEA and how to use it. This study is expected to provide a reliable answer. Trial status {#Sec18} ============ The study was conceived and designed in 2015. Enrollment began in April 2016 and was expected to end in December 2018. At the time of manuscript preparation, more than 500 subjects had been enrolled. Enrollment in this study was ongoing at the time of manuscript submission. Additional file =============== {#Sec19} Additional file 1:SPIRIT (Standard Protocol Items: Recommendations for Interventional Trials) checklists. (DOCX 41 kb) AMH : Anti-Müllerian hormone ART : Assisted reproductive technology CONSORT : Consolidated Standards of Reporting Trials CPR : Clinical pregnancy rate DCC : Data coordinative center DHEA : Dehydroepiandrosterone E2 : Estradiol ET : Embryo transfer FSH : Follicle-stimulating hormone Gn : Gonadotropin GnRH : Gonadotropin-releasing hormone hCG : Human chorionic gonadotropin ICSI : Intracytoplasmic sperm injection IVF : In vitro fertilization LPS : Luteal phase support OHSS : Ovarian hyperstimulation syndrome POR : Poor ovarian respond RCT : Randomized controlled trial SPIRIT : Standard Protocol Items: Recommendations for Interventional Trials We thank all subjects for their voluntary participation in this study and physicians at all recruiting sites. Funding {#FPar5} ======= This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Availability of data and materials {#FPar6} ================================== The datasets generated during the current study are available from the corresponding author on reasonable request. WW, HL, and DW were involved in the study concept and design and in drafting of the manuscript. YS contributed to the study design and critical revision of the manuscript. JL, JZ, JW, and JM contributed to the study design and revision of the manuscript. YS and Z-JC were involved in the study concept and design and in revision of the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar7} ========================================== The study had been approved by the ethics committee of the Reproductive Medical Center of Shandong University, and the institutional review boards in other study sites permitted the use of the approval document of the ethics committee of the Reproductive Medical Center of Shandong University. All patients provided written informed consent prior to participation. Ethics approval has been sought from the ethics committee at the Reproductive Medical Center affiliated with Shandong University (2015 IRB approval no. 20). Written informed consent is obtained from each couple before screening. Consent for publication {#FPar8} ======================= Not applicable. Competing interests {#FPar9} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar10} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1} =============== Proteolytic enzymes are recognized by their catalytic type, that is, aspartic, cysteine, metallo, serine, threonine, and others as yet unclassified. The largest number of proteolytic enzymes are classified as metalloproteases \[[@B1]\]. Almost all metalloproteases contain one or two zinc ions, and several enzymes contain one or two cobalt or manganese ions. The HExxH motif forming an *α*-helix is well conserved in many monozinc enzymes as the active site in which the two histidine residues coordinate with the zinc ion \[[@B2]\]. Some other monozinc proteases have different zinc-binding motifs, for example, HxxE(D)-aa~n~-H in the carboxypeptidase family or HxD-aa~12~-H-aa~12~-H in the matrix metalloprotease family \[[@B2]\]. Dipeptidyl peptidase (DPP) III also has a unique zinc-binding motif, which was classified as family M49 in 1999 by MEROPS (peptidase database) after rat DPP III had been cloned and the HELLGH motif of DPP III was identified as an active site coordinated with a zinc ion \[[@B3], [@B4]\]. Although the motif HELLGH could not be found in any other metalloproteases, it exists in three kinds of monooxygenases (tyrosine, phenylalanine, and tryptophan hydroxylases) as an iron-binding site, as revealed by a search of the NBRF-PIR protein sequence database. Zinc atoms in several zinc metalloproteases, for example, astacin \[[@B5]\], carboxypeptidase A \[[@B6]\], and thermolysin \[[@B7], [@B8]\], have been substituted by other divalent cations to probe the role of the metal for catalysis and structure. Some of these enzymes, for example, DPP III and astacin, were shown to have high metal substitution tolerance by metal substitution studies \[[@B9]\]. However, it is difficult to determine the relationship between the metal tolerance and the metal coordination structure of zinc metalloproteases. Here, we show the metal coordination structure of the unique zinc-binding motif of DPP III, in which the zinc-binding motif is stabilized by several hydrogen bonds with acidic amino acid residues surrounding the zinc-binding motif, in order to clarify the relationship between the metal tolerance and the structure of the zinc-binding domain. The metal tolerances of both DPP III and del-DPP III, whose active site converts into a normal zinc-binding motif (HExxH), are shown here and compared with those reported for other metalloproteases \[[@B10]\]. Finally, we discuss the relationship between the catalytic activities and metal coordination structures of metal-substituted enzymes. 2. Identification of a Zinc-Binding Motif in DPP III {#sec2} ==================================================== We start with the identification of the zinc-binding motif in DPP III, which will be used for further investigation of the relationship between the metal tolerance and the metal coordination structures of DPP III. The deduced amino acid sequences from cDNA for human, rat, and fruit fly DPP IIIs are 723--738 amino acids long and conserve the amino acid sequence HELLGH-aa~52~-E \[[@B3], [@B11], [@B12]\], which resembles the HExxH-aa~n~-E zinc-binding motif conserved in many metalloproteases, such as thermolysin \[[@B13]\] and leukotriene A~4~ hydrolase \[[@B14]\]. Site-directed mutagenesis was performed on rat DPP III in order to testify that the HELLGH-aa~52~-E is a zinc-binding domain. Site-directed mutagenesis studies have clearly shown that the H450Y, H455Y, and E508A mutants, which lack zinc ions, lose their catalytic activity \[[@B4]\]. Replacement of Glu^451^ in these mutants with an alanine or an aspartic acid restores a mol of zinc ion per mol of protein but does not restore catalytic activity \[[@B4]\]. These results show that the H^450^ELLGH-aa~52~-E^508^ motif is a catalytic domain of which His^450^, His^455^, and Glu^508^ are ligands of a zinc ion and of which Glu^451^ is a catalytic amino acid residue, in the same way that the H^142^ExxH-aa~19~-E^166^ motif of thermolysin is a catalytic domain of which His^142^, His^146^, and Glu^166^ are ligands of a zinc ion and Glu^143^ is a catalytic amino acid residue. The 1.95-Å crystal structure of yeast DPP III representing a prototype for the M49 family of metalloproteases was resolved by Baral et al. \[[@B15]\]. It shows a novel protein fold with two domains forming a wide cleft containing the catalytic metal ion. However, the three-dimensional structure of zinc coordination (His^460^, His^465^, and Glu^517^) and the catalytic active (Glu^461^) residues are structurally conserved, similar to those presented in many metalloproteases, such as thermolysin \[[@B13]\]. The HELLGH motif and the third ligand (Glu^517^) of DPP III construct a helix *α*14 and a helix *α*16, respectively \[[@B15]\]. The 3D structure of DPP III is similar to that of thermolysin \[[@B13]\] or leukotriene A~4~ hydrolase \[[@B16]\], the zinc-binding domain of which is constructed of two *α*-helixes, for HExxH (containing two zinc ligands) and xNEx (third ligand). [Figure 1](#fig1){ref-type="fig"} shows the superimposition of the active sites of rat DPP III and thermolysin. The helix *α*14 of DPP III has a slightly larger loop than that of thermolysin, and the glutamic acid on the motif comes close to zinc ion comparing with the glutamic acid on the normal helix of thermolysin \[[@B13], [@B17]\]. 3. Stabilization of the Coordination between Ligands and Metal {#sec3} ============================================================== In the 3D structural model of the zinc-binding domain of many zinc enzymes---neprilysin \[[@B18]\], thermolysin \[[@B13]\], carboxypeptidase A \[[@B19]\], leukotriene A~4~ hydrolase \[[@B16]\], aspzincin \[[@B20]\], and DPP III \[[@B17]\]---the His, His, and Glu residues that coordinate with the zinc ion are engaged in hydrogen bonds with one or two acidic amino acid residues (Glu or Asp) or other carbonyl oxygen atoms ([Table 1](#tab1){ref-type="table"}). 3D structural models of catalytic domains of thermolysin (PDB: 1KEI), peptidyl-Lys metallopeptidase (PDB: 1GE6), and carboxypeptidase A (PDB: 1YME) are shown in [Figure 2](#fig2){ref-type="fig"}. In thermolysin (a), the oxygen atoms of Asp^165^ and Asp^170^ are engaged in hydrogen bonding with the nitrogen atoms of His^146^ and His^142^, respectively. Asp^154^ and Thr^128^ of peptidyl-Lys metalloendopeptidase (b) and Asp^142^ of carboxypeptidase A (c) are also engaged in hydrogen bonding with His^117^, His^121^, and His^69^, respectively. It was proved through the mutational studies of rat DPP III that this network of hydrogen bonds close to the zinc-binding motif plays an important role in stabilizing the coordination of the zinc ion to the protein \[[@B17]\]. The hydrogen bonds surrounding the zinc-binding motif of rat DPP III are shown in [Figure 3](#fig3){ref-type="fig"}, and the kinetic parameters, zinc contents and zinc dissociation constants of the several mutants are shown in [Table 2](#tab2){ref-type="table"}. The replacement of Glu^507^ and Glu^512^, the oxygen atoms of which bind with the nitrogen atoms of His^455^ and His^450^, respectively, increases the dissociation constants by factors of 10\~10^5^ and correlatively reduces the zinc contents and enzyme activities. The hydrogen bonds between acidic amino acid residues and zinc ligands (His, His, and Glu) may stabilize the coordination of the zinc ion with the protein of the metalloprotease. 4. Metal Substitutions of Monozinc Metalloproteases {#sec4} =================================================== Almost all metalloproteases are monozinc enzymes. Some enzymes contain two zinc ions for catalytic domains, for example, human renal dipeptidase \[[@B21]\], and a few are dicobalt or dimanganese enzymes, for example, *Pyrococcus furiosus* methionine aminopeptidase \[[@B22]\] or *Escherichia coli* proline aminopeptidase \[[@B23]\], respectively. The zinc in numerous zinc metalloproteases has been substituted by several divalent cations. The cobalt(II)- or manganese(II)-substituted enzymes showed nearly restored catalytic activity or even excess activity from apoenzyme, as seen in [Table 3](#tab3){ref-type="table"}. Gomis-Rüth et al. \[[@B5]\] demonstrated in the metal substitution studies of astacin that Cu(II)-astacin displays enzyme activity of about 37%, while Ni(II)- and Hg(II)-astacin were almost inactive. In the crystal structure of Cu(II)-astacin, the metal ion is pentacoordinated with His^92^, His^96^, His^102^, Tyr^149^, and H~2~O, as in native Zn(II)-astacin or Co(II)-astacin; however, in the Ni(II)-astacin or Hg(II)-astacin, the metal ion is hexacoordinated with an additional solvent molecule or tetracoordinated with no ordered solvent molecule, respectively \[[@B5]\]. The restoration of catalytic activity in these substituted astacins was shown to be dependent on the metal coordination structure \[[@B5]\]. Meanwhile, almost all Cu(II)-substituted enzymes, such as thermolysin \[[@B7], [@B8]\], carboxypeptidase A \[[@B6]\], aminopeptidase B \[[@B24]\], or endopeptidase from *Lactococcus lactis* \[[@B25]\], show only partial activation or very low activities. The reason why these Cu(II) enzymes do not demonstrate catalytic activities may be that the coordination geometry of Cu(II) is more rigid than that of Zn(II) or Co(II). In the case of DPP III, Co^2+^-, Ni^2+^- and Cu^2+^-DPP IIIs showed comparable catalytic activities to Zn^2+^-DPP III; the kinetic parameters are shown in [Table 4](#tab4){ref-type="table"} \[[@B9]\]. DPP III shows high flexibility of the metal ion for the catalytic activity compared with thermolysin or aminopeptidase B. Thermolysin or aminopeptidase B is a subclan MA (E) metalloprotease containing an HExxH-aa~n~-E motif, and the 3D structure of the active domain is very similar to that of DPP III described above. The zinc ion in a subclan MA (E) metalloprotease or DPP III is tetracoordinated with three ligands (His, His, and Glu) and a water molecule. The metal-substituted (Co^2+^, Cu^2+^, or Ni^2+^) DPP III may have the same tetrahedral coordination structure as Zn^2+^-DPP III, so these enzymes are able to maintain the catalytic activity. The zinc in del-DPP III, whose active site converted into HExxH, was substituted with Co^2+^, Ni^2+^, or Cu^2+^to investigate the grounds for activation of the Cu^2+^-DPP III \[[@B10]\]. The Co^2+^-del-DPP III and Ni^2+^-del-DPP III showed comparable catalytic activity to that of Zn^2+^-del-DPP III, while the Cu^2+^-del-DPP III showed no catalytic activity, as in the case of thermolysin or aminopeptidase B \[[@B8]--[@B10]\]. The EPR (electron paramagnetic resonance) parameters of various Cu^2+^-substituted metalloproteases are shown in [Table 5](#tab5){ref-type="table"}. Each parameter is exactly alike between DPP III and thermolysin, aminopeptidase B, or del-DPP III \[[@B8]--[@B10], [@B24]\]. The results show that the Cu(II) coordination structures of the HExxH-aa~n~-E and HExxxH-aa~52~-E motifs are very similar. In the superimposition of the 3D structure models of active sites of DPP III and del-DPP III, the *α*-helix of DPP III, which is abnormally composed of 5 amino acid residues per one turn of the *α*-helix, is a larger loop than that of del-DPP III, the same as the case for the superimposition of the active sites of DPP III and thermolysin ([Figure 1](#fig1){ref-type="fig"}). The coordination geometries of the two enzymes are similar, while the position of Glu^451^, which is essential for the enzyme activity, is slightly closer to the copper ion in DPP III than in del-DPP III. The distances of oxygen atoms of the Glu^451^ residues of del-DPP III and wild-type DPP III are 4.8 Å and 3.2 Å from the zinc ion, respectively. Zn(II) coordination geometry is flexible, so both wild-type and del-DPP IIIs could have catalytic activity. However, the oxygen atom of Glu^451^ in Cu(II)-del-DPP III is not able to bind to the oxygen atom of the water molecule that is coordinated with the copper ion because the Cu(II) coordination geometry is very rigid. Therefore, the catalytic activity of Cu(II)-del-DPP III was diminished. Some other Cu(II)-substituted enzymes, for example, aminopeptidase Ey \[[@B26]\], vibriolysin \[[@B27]\], hyicolysin \[[@B28]\], and *Legionella* metalloendopeptidase \[[@B29]\], were shown to have enzyme activities. These enzymes are all classified in subclan MA (E), the same as thermolysin or aminopeptidase B. The metal coordination structures of these enzymes have not been shown in detail; however, the catalytic domain may be more flexible than that of thermolysin or aminopeptidase B, in the same way as Cu(II)-substituted DPP III. 5. Conclusions {#sec5} ============== In this paper, we compared metal flexibility with the geometry of metal coordination of metalloproteases, to investigate why DPP III shows metal tolerance. Metal substitution of Zn(II) by Co(II) or Mn(II) on metalloproteases generally maintains catalytic activity, because the metal coordination geometries of Zn(II), Co(II), and Mn(II) are flexible. Most Cu(II)-substituted enzymes could not restore the catalytic activities, because the Cu(II) coordination geometry is very rigid. However, Cu(II)-substituted DPP III showed the same catalytic activity as that of Zn(II)-DPP III. We then studied the metal flexibilities and metal coordination geometries of many metallopeptidases, especially DPP III and del-DPP III, but we could not prove a relation between the metal flexibility and the metal coordination geometry. The metal tolerance of DPP III might depend on the flexibility of the metal-binding motif, not on the metal coordination geometry. By comparison of the 3D structure of active sites of DPP III and del-DPP III, both coordination geometries are seen to be similar, while the positions of catalytic amino acid residues (Glu) on those zinc-binding motifs are slightly different. We conclude that the catalytic site of Cu(II)-DPP III could be flexible enough to form the catalytic complex, with substrate and H~2~O. {#fig1} {#fig2} ![Molecular modeling of the catalytic site of rat DPP III. The model was generated as a template of the human DPP III crystal structure \[[@B40]\]. The zinc ion is shown as a green sphere, and amino acid side chains are shown as sticks colored red for oxygen, blue for nitrogen, and white for carbon. Metal coordinates in light blue and hydrogen bonds in yellow are indicated by dashed lines.](JAA2011-574816.003){#fig3} ###### The zinc coordination residues and the residues that fix the coordination with hydrogen bonds. ---------------------------------------------------------------------------------------------------------------------------------------------------------- Zinc metalloprotease Coordination residues Residues that form the hydrogen bond with the coordination residues PDB no. -------------------------------------------- ----------------------------- --------------------------------------------------------------------- --------- \(1\) Thermolysin type (HExxH- aa~n~-E) *α*-helix-aa~n~-*α*-helix  Thermolysin His^142^, His^146^ Glu^166^ Asp^170^-2.8 Å-His^142^\ 1KEI Asn^165^-2.8 Å-His^146^  Vibriolysin His^345^, His^349^ Glu^369^ Asp^373^-2.8 Å-His^345^\ 3NQX Asn^368^-2.8 Å-His^349^  Staphylococcus aureus  metalloproteinase His^144^, His^148^ Glu^168^ Asn^167^-2.8 Å-His^148^\ 1BQB Asp^172^-2.8 Å-His^144^  Zinc aminopeptidase His^265^, His^269^ Glu^288^ Phe^272^(C=O)-2.9 Å-His^269^ 1Z1W  Leukotriene A4 hydrolase His^295^, His^299^ Glu^318^ Glu^325^-2.8 Å-His^295^\ 1SQM Gly^303^(C=O)-2.6 Å-His^299^  Human thimet oligopeptidase His^473^, His^477^ Glu^502^ Glu^509^-2.6 Å-His^473^ 1SQM  Human neutral endopeptidase  (Neprilysin) His^583^, His^587^ Glu^646^ Asp^650^-2.9 Å-His^583^\ 1DMT Asp^590^-2.7 Å-His^587^ \(2\) Endopeptidase type (HExxH-aa~n~-E or D) *α*-helix-aa~n~-random coil  Peptidyl-Lys metalloendopeptidase His^117^, His^121^ Asp^130^ Asp^154^-2.7 Å-His^117^\ 1GE6 Thr^128^(C=O)-2.8 Å-His^121^ \(3\) Carboxypeptidase A type *β*-sheet-aa~n~-random coil  Carboxypeptidase A His^69^, His^196^ Glu^72^ Asp^142^-2.7 Å-His^69^ 1YME  Putative lysostaphin peptidase His^232^, His^311^ Asp^236^ Glu^315^-2.6 Å-His^311^\ 2GU1 Gly^216^(C=O)-2.8 Å-His^232^ ---------------------------------------------------------------------------------------------------------------------------------------------------------- ###### Kinetic parameters for the hydrolysis of Arg-Arg-NA, zinc contents, and zinc dissociation constants of wild-type and mutated rat DPP IIIs. -------------------------------------------------------------------------------------------------------------------------------- Enzymes *k* ~cat~/*K* ~*M*~ × 10^−4^ \ Zinc content (mol/mol of protein^a^) Zinc dissociation constant (M) (*K* ~*d*~) (M^−1^ s^−1^) ----------- -------------------------------- -------------------------------------- -------------------------------------------- Wild-type 73.6 ± 6.9 1.02 ± 0.15 (4.5 ± 0.1) × 10^−13^ E507D 22.8 ± 1.9 0.65 ± 0.07 (1.0 ± 0.2) × 10^−11^ E507A 4.43 ± 0.41 0.29 ± 0.04 (1.0 ± 0.2) × 10^−8^ E512D 21.0 ± 0.19 0.45 ± 0.06 (1.4 ± 0.1) × 10^−12^ E512A 2.45 ± 0.28 0.08 ± 0.01 (2.6 ± 0.7) × 10^−9^ -------------------------------------------------------------------------------------------------------------------------------- ^a^Values are means ± SD of two separately prepared enzymes with duplicate determinations. ###### Reactivated Co(II) and Mn(II) enzymes substituted from apo-metalloproteases. --------------------------------------------------------------------------------------------------------------------------------------------------------------- Clan Subclan Name of enzyme Replaced Ion Reference ------ --------- --------------------------------- -------------------------------------------- --------------------------------------------------------------- MA E Aminopeptidase Ey Co^2+^, Mn^2+^ Tanaka and Ichishima \[[@B26]\] MA E Aminopeptidase B Co^2+^ Hirose et al. \[[@B24]\] MA E Saccharolysin Co^2+^, Mn^2+^ Achstetter et al. \[[@B30]\] and Büchler et al. \[[@B31]\] MA E Lysyl aminopeptidase Co^2+^, Mn^2+^ Klein et al. \[[@B32]\] MA E Oligopeptidase F Co^2+^, Mn^2+^ Yan et al. \[[@B33]\] and Monnet et al. \[[@B34]\] MA E Mycolysin (Thermolysin) Co^2+^, Mn^2+^ Co^2+^ (200%), Mn^2+^ (10%) Chang and Lee \[[@B35]\], and Holmquist and Vallee \[[@B36]\] MA E Oligopeptidase O Co^2+^\ Tan et al. \[[@B25]\] and Baankreis et al. \[[@B37]\] Mn^2+^ (50%) MA E Hyicolysin Co^2+^ Ayora and Götz \[[@B28]\] ME E Legionella metalloendopeptidase Mn^2+^ (69%) Dreyfus and Iglewski \[[@B29]\] MA A Epralysin Co^2+^ (58%) Diermayr et al. \[[@B38]\] MA M Astacin Co^2+^ (140%) Gomis-Rüth et al. \[[@B5]\] MA M MEP^a^ (*Gf* ^b^ MEP) Co^2+^\ Nonaka et al. \[[@B39]\] Mn^2+^ (200%) MA M *Po* ^c^ MEP Co^2+^ (80%)\ Nonaka et al. \[[@B39]\] Mn^2+^ (30%) --------------------------------------------------------------------------------------------------------------------------------------------------------------- ^a^Peptidyl-Lys metallopeptidase; ^b^ *Grifola frondosa*; ^c^ *Pleurotus ostreatus.* ###### Kinetic parameters for the hydrolysis of Arg-Arg-NA and metal contents of various metal-DPP IIIs. ------------------------------------------------------------------------------------------------------------ Enzyme *K* ~*M*~\ *k* ~cat~\ *k* ~cat~/*K* ~*M*~(×10^4^ M^−1^ s^−1^) Metal content\ (×10^−5^ M) (s^−1^) (mol/mol of protein) ---------------- ------------- ------------ ----------------------------------------- ---------------------- Zn^2+^-DPP III 8.1\ 7.1\ 8.8 0.8\ (±1.0) (±0.2) (±0.1) Co^2+^-DPP III 8.2\ 7.0\ 8.5 1.0\ (±0.9) (±0.1) (±0.1) Cu^2+^-DPP III 9.9\ 10.1\ 10.2 1.1\ (±1.1) (±0.3) (±0.1) ------------------------------------------------------------------------------------------------------------ ###### EPR parameters of Cu^2+^ proteases. *g* ~⊥~ *g~ll~* *A~ll~* (×10^−4^ cm^−1^) ------------------------------ --------- --------- -------------------------- Cu^2+^-DPP III^a^ 2.06 2.27 167 Cu^2+^-del-DPP III^b^ 2.06 2.27 161 Cu^2+^-thermolysin^c^ 2.06 2.26 163 Cu^2+^-aminopeptidase B^d^ 2.06 2.27 157 Cu^2+^-carboxypeptidase A^e^ 2.05 2.33 115 References ^a^\[[@B9]\], ^b^\[[@B10]\], ^c^\[[@B8]\], ^d^\[[@B24]\], and ^e^\[[@B6]\]. [^1]: Academic Editor: Shandar Ahmad

Introduction {#sec1-1} ============ An intrauterine device (IUD) is a small polyethylene device, which contains copper or progesterone. IUDs are a form of long-acting reversible method of contraception, which is the most effective type of reversible birth control \[**[@R1]**\]. Also, IUDs are among the most cost-effective options for reversible contraception \[**[@R2]**\]. Therefore, IUDs are one of the most used and preferred methods of contraception due to their efficacy, reversibility and long term use \[**[@R5]**-**[@R3]**\]. The levonorgestrel-releasing intrauterine system is the most effective IUD available, with a 0.2% pregnancy rate at 1 year of use \[**[@R6]**,**[@R7]**\]. The levonorgestrel-releasing intrauterine system consists of a T-shaped polyethylene frame with a hormone reserve that contains 52 mg of levonorgestrel. The levonorgestrel is initially released at a rate of 20 ug/day and decreases to 11 ug/day after 5 years \[**[@R6]**,**[@R8]**\]. It seems that the mechanism of action of the levonorgestrel-releasing intrauterine system is multifactorial. The levonorgestrel induces an atrophy of the endometrial glands but it also changes the characteristics of the cervical mucus in order to prevent pregnancy \[**[@R6]**,**[@R9]**\]. Plus, the IUD modifies the physiological movement of spermatozoa and zygote, preventing nidation \[**[@R10]**\]. The most common side effects of levonorgestrel-releasing intrauterine system are amenorrhea, spotting and pain \[**[@R6]**,**[@R10]**\]. The last is a frequent symptom of pelvic inflammatory disease, one of the most redundant complications of patients with levonorgestrel-releasing intrauterine system \[**[@R11]**,**[@R12]**\]. Objective {#sec1-2} ========= This study was undertaken in order to determine if antibiotic prophylaxis is mandatory, after the insertion of levonorgestrel-releasing intrauterine system in order to decrease the risk of pelvic inflammatory disease. Materials and methods {#sec1-3} ===================== We prospectively evaluated 44 patients, admitted in Bucharest Emergency Hospital between the 1ⁱ of February 2012 and the 1ⁱ of October 2012, at whom levonorgestrel-releasing intrauterine system was inserted. All patients were informed of the advantages, disadvantages and possible side effects of this type of contraception. An informed consent was then taken from all the subjects of the study. At a visit prior to IUD insertion a preliminary evaluation was performed -- all patients underwent a physical and echo-graphical examination, a PAP smear, screening for common sexually transmitted infections and a complete blood count was sampled. In all subjects a urine pregnancy test was performed -- it was negative in all cases. Following clinical, paraclinical or imagistic criteria all patients suspected to have pelvic inflammatory disease were excluded. The insertion of levonorgestrel-releasing intrauterine system was performed by a experienced physician during the first 7 days of the menstrual cycle. The patients enrolled were divided into two groups. In group A, a number of 22 patients received 875mgAmoxicillin Trihydrate+ 125 mg Potassium Clavulanate, a dose every 12 hours for 5 days after the insertion of levonorgestrel-releasing intrauterine system. Group B was represented by the other 22 patients who did not receive antibiotic prophylaxis. All the patients were reevaluated at 4 and 12 weeks after the insertion of levonorgestrel-releasing intrauterine system. The statistical analysis was conducted by using SPSS version 19. Results {#sec1-4} ======= The global mean age of the evaluated patients was 28.31 years, with a mean age of 29.22 years in group A and of 27.4 years for group B. Of the 44 subjects, 37 (19 from group A and 18 from group B) were first-time users of levonorgestrel-releasing intrauterine system. During the first 4 weeks after the insertion of levonorgestrel-releasing intrauterine system only two patients, one from group A and one from group B were diagnosed with pelvic inflammatory disease. The patient from group A, a first-time user of levonorgestrel-releasing intrauterine system accused severe lower abdominal pain after 14 days past insertion and following clinical and paraclinical criteria she was diagnosed with acute pelvic inflammatory disease; the IUD was extracted and she received specific antibiotic and anti-inflammatory treatment. However, the patient from group B experienced mild pain and leucorrhoea - at the follow up visit after 4 weeks past insertion she was diagnosed with chronic pelvic inflammatory disease and nonspecific vaginosis; she received specific treatment but the IUD was not extracted (**[Fig. 1](#F1){ref-type="fig"},[Fig. 2](#F2){ref-type="fig"}**). {#F1} {#F2} At a second follow up visit -- 12 weeks after the insertion of levonorgestrel-releasing intrauterine system no other patient was diagnosed with pelvic inflammatory disease. Discussion {#sec1-5} ========== The results of our study are similar to those of Milton et al., Zorlu et al. and Walsh et al. -- antibiotic prophylaxis has no effect on the incidence of pelvic inflammatory disease \[**[@R3]**,**[@R13]**,**[@R14]**\]. Multiple studies demonstrated that the risk of pelvic inflammatory disease seems to be the highest in the first 3 weeks, due to the contamination of the upper genital tract during the insertion of the IUD \[**[@R3]**,**[@R15]**,**[@R16]**\]. We incriminate the same mechanism due to the fact that only one patient from the 44 evaluated subjects was diagnosed with acute pelvic inflammatory disease, after only 14 days past the insertion of levonorgestrel-releasing intrauterine system, a participant from group A, who received a prophylactic antibiotic. Also, at the follow up visit -- 12 weeks after the insertion of IUD no other patient was diagnosed with pelvic inflammatory disease, suggesting the same, previously discussed mechanism of appearance of this redundant complication. Conclusion {#sec1-6} ========== Antibiotic prophylaxis is not mandatory, after the insertion of levonorgestrel-releasing intrauterine system in order to decrease the risk of pelvic inflammatory disease. **Disclosure**: None of the authors have a conflict of interest. **Acknowledgement**:This study was supported by the international project "Development of the translational research infrastructure in molecular and imagistic pathology - MOLIMAGEX". Project manager: Prof. Catalin Cirstoiu, MD. Responsible for the Obstetrics-Gynecology Department: Assoc. Prof. Monica Cirstoiu, MD

Background {#Sec1} ========== The use of indoor residual spraying (IRS) and insecticide-treated nets (ITNs) as key vector control interventions has seen a massive scale-up over the last decade \[[@CR1]\]. This scale-up together with other case management and preventive measures were reported to have contributed to a global decline in malaria mortality rates by about 47% between 2000 and 2013, and by 54% in the World Health Organization (WHO) African Region \[[@CR2]\]. It is known that IRS as a vector control intervention can be highly effective when over 80% of the target population are covered \[[@CR3]\]. The effectiveness of IRS has been demonstrated through surveys that show a reduction in entomological indices of transmission \[[@CR4]--[@CR6]\] or significant reduction in malaria prevalence \[[@CR4]\], morbidity and mortality \[[@CR7], [@CR8]\]. However several challenges threaten the efficacy of IRS. Studies have found that the scale up of IRS and ITNs contribute to changes in vector behaviour \[[@CR5], [@CR9]\], and changes in vector species composition or selection for certain traits that may support rapid evolution of insecticide resistance \[[@CR10]\]. In areas where pyrethroid resistance is widespread, non-pyrethroid insecticides can be used to maintain the efficacy of IRS. However, new insecticides and formulations tend to be more expensive than the previously used pyrethroids, and have partly contributed to the scale-down of IRS coverage in some countries, including Ghana \[[@CR2], [@CR11]\]. In Ghana, IRS forms a key component of the national malaria control strategic plan to reduce the malaria burden in the country \[[@CR12], [@CR13]\]. Currently, IRS is being implemented in the Northern Region through support from the United States President's Malaria Initiative (PMI) and in Upper West Region and Obuasi municipality in the Ashanti region supported by the Global Fund. The PMI supported IRS programme in northern Ghana was scaled down from 9 districts in 2012 to 4 districts in 2013 and 2014, as a result of the switch of insecticides from pyrethroid to a more expensive organophosphate. However the national malaria control strategic plan calls for national scale-up of IRS in the country. In view of the sub-national differences in intensity of malaria transmission \[[@CR14], [@CR15]\] and the logistical difficulties of IRS implementation in some areas, there is the need to document the epidemiological and entomological impact of already existing IRS programmes in the different ecological zones of the country \[[@CR15]\] before a rapid country-wide scale-up of this intervention. This is needed to provide evidence for decision-making and guide future programme implementation. Findings of the impact of the IRS operations, as well as IRS withdrawal on the sporozoite and entomological inoculation rates across two sentinel districts of the PMI IRS program in Northern Region, Ghana are discussed in comparison with an unsprayed district. The impact of IRS on parity rates as an indicator of longevity was also measured in sprayed and unsprayed districts. Methods {#Sec2} ======= Study areas {#Sec3} ----------- Three districts in the IRS programme area were selected for entomological monitoring, namely: Savelugu Nanton District (hereafter SND) (9.25° and 10.8°N and 0.33° and 1.00°W), Tolon Kumbungu District (TKD) (9.15° and 10.70°N, 0.52° and 1.23°W), and Tamale Metropolis (TML) (9.16° and 9.34° N and 00.34° and 00.59^o^ W). Three rural communities within each district were used as sentinel sites (Fig. [1](#Fig1){ref-type="fig"}). Tarikpaa, Diare and Nanton communities under SND and Gbullung, Woribugu and Dimabi communities under TKD were purposively selected as entomological sentinel sites. Three rural communities (Kulaa, Tugu and Yong) under the Tamale metropolis with no history of IRS and no near term plans for IRS implementation were selected as control sentinel sites for comparison of trends on longevity, sporozoite infectivity and entomological inoculation rates.Fig. 1Map of study area located within the Northern region of Ghana All the study sites are rural and located in Ghana's northern savannah zone, which experiences a unimodal rainy season that occurs between May and November, with the peak months (July to September) receiving between 150 and 250 mm of rainfall per month. The rainy season is followed by a dry season that lasts from December until March \[[@CR16]\]. The annual range of temperature is between 25 and 30 °C \[[@CR16]\], which appears to be favourable for *Anopheles* larval development. A typical housing complex found in the study area consists mainly of compounds made up of round huts roofed with thatch and scattered over large farmlands. Mean daily rainfall data from the study area was obtained from weather stations of the Ghana Meteorological Services Department in Savelugu and Tamale, and Savannah Agricultural Research Institute (SARI)---Tolon Kumbungu. IRS campaigns and insecticides used {#Sec4} ----------------------------------- IRS was conducted once annually, timed to coincide with the first or second month of the rainy season depending on the year. SND and TKD started IRS in 2008, beginning with the inexpensive pyrethroid insecticide alphacypermethrin (Fendona 5 WP) at 30 mg/m^2^ in 2008 and 2009 which has a residual life of 4--6 months. In 2010 both districts were sprayed with the alternative pyrethroid, deltamethrin (Pali 25 WG) with a residual life of 3--6 months, at a rate of 20 mg/m^2^. Both districts reverted to alphacypermethrin in 2011. In 2012, the two district programmes diverged. Due to emerging insecticide resistance concerns \[[@CR17]\], communities in SND were sprayed with an organophosphate insecticide (pirimiphos methyl CS formulation at a rate of 1 g/m^2^) with a residual life of 4--6 months, whereas communities in TKD were still sprayed with alphacypermethrin. IRS was discontinued in TKD in 2013 due largely to cost considerations, but continued in SND with pirimiphos methyl CS through 2014. Meanwhile the neighbouring control district TML received no IRS throughout. The timeframe for IRS rounds is presented in Table [1](#Tab1){ref-type="table"}.Table 1IRS implementation schedule and insecticide usage for the PMI IRS programme, Northern Ghana 2008--2014DistrictsInsecticide used and spray coverage (%)Calendar year:Pre-20082008200920102011201220132014IRS year:1234567Savelugu-Nanton District (SND)No IRSAlphacypermethrin\ (88.4%)Alphacypermethrin\ (93.6%)Deltamethrin\ (98.7%*)*Alphacypermethrin\ (87.5%)Pirimiphos-methyl CS\ (89.7%)Pirimiphos-methyl CS\ (91.1%)Pirimiphos-methyl CS\ (68.0%)Tolon-Kumbungu District (TKD)No IRSAlphacypermethrin\ (91.3%)Alphacypermethrin\ (93.4%)Deltamethrin\ (98.0%*)*Alphacypermethrin (91.1%)Alphacypermethrin (91.9%)No IRSNo IRSTamale Metropolis (TML)No IRSNo IRSNo IRSNo IRSNo IRSNo IRSNo IRSNo IRS*Entomological monitoring*All districtsNonePeriodicMonthly (post-IRS)MonthlyMonthlyMonthlyMonthlyMonthlyPercentages in parenthesis = yearly IRS coverage of sprayable houses Spray coverage and quality {#Sec5} -------------------------- The programme implemented intensive mobilization and sensitization campaigns. Spray operators were also well trained and closely supervised to make sure the required quality of operation was achieved \[[@CR18], [@CR19]\]. ITN campaigns, coverage and use {#Sec6} ------------------------------- Data on ITN coverage and use was obtained at the regional level from the 2008 and 2014 Demographic and Health Surveys and the 2011 Multiple Indicator Cluster Survey \[[@CR15], [@CR20], [@CR21]\], each of which were conducted nationwide at the end of the rainy season. Adult mosquito collections {#Sec7} -------------------------- Using the human landing catch (HLC) method \[[@CR22]\] monthly mosquito collections were conducted simultaneously across all study sites. In each sentinel community, 8 trained mosquito collectors worked in two teams of four, working in 2 houses each night from 6 p.m. to 6 a.m. In each house 2 collectors worked indoors whilst the other 2 worked outdoors. Collections were done for a total of 4 nights to sample 8 houses in the community per month (24 houses in a district per month). Collectors shifted between indoor and outdoor every hour and were allowed to take 10 min breaks between shifts. Due to logistical challenges, there were occasional gaps in monthly monitoring, but a total of 61 out of 65 months were covered. Mosquito identification and processing {#Sec8} -------------------------------------- ### Vector species identification {#Sec9} Mosquitoes collected were morphologically identified to species level using taxonomic keys \[[@CR23]\]. Mosquito samples of the *Anopheles gambiae* complex were further identified into sibling species using ribosomal DNA-polymerase chain reaction (PCR) \[[@CR24]\] and into molecular forms following the PCR--RFLP (restriction fragment length polymorphism) procedure \[[@CR25]\]. ### Parity dissections {#Sec10} About 30% of unfed *An. gambiae* s.l. identified morphologically was dissected per site per month, except for months when very few numbers of *An. gambiae* s.l. females were captured. Three of every ten unfed *An. gambiae* s.l. identified morphologically from the hourly collections per site were selected for dissection. The selected mosquitoes were dissected and their ovaries examined for parity by observing the degree of coiling in the ovarian tracheoles \[[@CR26]\]. Due to financial and logistical challenges, monthly parity data are available from 2011 to 2014. ### Circumsporozoite assay {#Sec11} The head and thoraces of a subset (\~10%) of *Anopheles* mosquitoes collected were tested using enzyme-linked immunosorbent assay (ELISA) \[[@CR27]\] for the presence of circumsporozoite protein (CSP) of *Plasmodium falciparum*, the major malaria parasite in the study areas. These tests were performed at the Noguchi Memorial Institute for Medical Research (NMIMR) laboratory in Accra, Ghana. Data analysis {#Sec12} ------------- The number of mosquitoes collected per month which belonged to important *Anopheles* vector species (*An. gambiae* s.l. and *An. funestus* group) was used as the denominator to calculate the following ratios:Parity rate*The proportion of parous mosquitoes among unfed and blood meal seeking mosquitoes. Parity was used as a proxy measure for the daily survival rate and average life span of the vector population*Monthly sporozoite rate*Number of mosquitoes found positive for the presence of circumsporozoite proteins/number of mosquitoes tested*The entomological inoculation rates (EIRs) were calculated by the formula$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} EIR{\text{-}}daily &= Daily\;human\;biting\;rates \\ & \quad \times sporozoite\;rate \end{aligned}$$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned} Post \, IRS \, EIRs = \sum \, Monthly \, IRS \, EIRs \, (July{-}December) \end{aligned}$$\end{document}$$ Yearly variations in parity and sporozoite rate for *Anopheles* spp. collected from IRS and unsprayed districts were compared through a z test for differences in proportions. Pearson's correlation was used to assess the relationship between rainfall and EIRs. All tests were performed at 0.05 significance level, using SPSS version 20 and Microsoft Excel^®^. Results {#Sec13} ======= Spray coverage {#Sec14} -------------- The timeframe for IRS implementation, insecticide used and yearly coverage is presented in Table [1](#Tab1){ref-type="table"}. The intensive mobilization and sensitization campaigns implemented by the programme ensured high spray coverage (Table [1](#Tab1){ref-type="table"}). Over 80--90% coverage was achieved in both districts each year, with the exception of 68% coverage in SND in 2014 \[[@CR19]\], attributed to low acceptance among certain town populations. ITN campaigns, coverage and use {#Sec15} ------------------------------- ITN coverage and use was found to be high across all study areas. This reflected the large-scale efforts of the National Malaria Control Programme (NMCP) and international partners since 2006 to promote the ownership and use of ITNs in Northern Region. ITNs were distributed largely through mass door-to-door ITN distribution campaigns in 2010 and 2012, then through school and health facility distributions in 2013--2014 (Fig. [3](#Fig3){ref-type="fig"}) \[[@CR13], [@CR17]\]. ITN ownership in Northern Region increased from 54% in the 2008 Demographic and Health Survey to about 71% by the 2014 DHS survey, with an average of 1.7 ITNs per household in 2014 \[[@CR20], [@CR21]\]. ITN usage among children under age 5 and pregnant women were 58% and 61%, respectively, in 2014 \[[@CR21]\]. ### Anopheles species composition and relative abundance {#Sec16} A total of 192,259 adult female *Anopheles* mosquitoes belonging to 5 different species were collected over the 6 years (2009--2014). *An. gambiae* s.l. (95.3%) was the most abundant species, followed by *An. nili* (2.4%) and *An. funestus* (1.7%). *An. rufipes* and *An. pharoensis* made up 0.2 and 0.4% respectively of the total *Anopheles* mosquitoes collected (Table [2](#Tab2){ref-type="table"}). There was monthly and yearly variation in abundance of the main vectors (*An. gambiae* s.l. and *An. funestus*) in all the areas. The vector abundance in all the areas coincided with the peak of the rainy season (Fig. [2](#Fig2){ref-type="fig"}). The mean biting rates of vectors in the unsprayed communities were higher than biting rates for the IRS areas across all the years. The biting rate of the main vectors from all sites significantly reduced over the duration of the study (2009--2014).Table 2Total number of *Anopheles* caught per study site*Anopheles* species collectedSavelugu-Nanton District (SND)Tolon-Kumbungu District (TKD)Tamale metropolis (TML)Total*An. gambiae* s.l.27,648 (98.9%)63,500 (95.6%)92,111 (94.1%)183,259 (95.3%)*An. funestus*76 (0.3%)1849 (2.8%)1378 (1.4%)3303 (1.7%)*An. rufipes*42 (0.2%)117 (0.2%)259 (0.3%)418 (0.2%)*An. nili*58 (0.2%)854 (1.3%)3632 (3.7%)4544 (2.4%)*An. pharoensis*140 (0.5%111 (0.2%)484 (0.5%)735 (0.4%)27,96466,43197,864192,259 Fig. 2Biting rates of *An. gambiae* s.l. collected from sentinel sites between August 2009 and December 2014 PCR analysis on *An. gambiae* s.l. collected between 2009 and 2013 showed that *An. coluzzii* and *An. gambiae* sensu stricto (s.s.) were present in all the sites in varying proportions with *An. gambiae* s.s. predominating (75--89%) between 2009 and 2013. However, in 2014 *An. coluzzii* was the predominant molecular form making up over 91% of the *An. gambiae* species from all sites (Table [3](#Tab3){ref-type="table"}). Five *An. arabiensis* were identified out of 348 *An. gambiae* s.l. analysed in 2014.Table 3Distribution of molecular forms of *An. gambiae* s.l. caught per study siteYearTotal \# samples analysedSavelugu-Nanton District (SND)Tolon-Kumbungu District (TKD)Tamale Municipality (TML)*An. gambiaeAn. coluzziiAn. gambiaeAn. coluzziiAn. gambiaeAn. coluzzii*M/S hybrid200920049058176970201031268129616982202011135464338395020128018825518602013161540431344702014263058108016972 ### Parity rates of vectors {#Sec17} Higher proportions of parous female *An. gambiae* s.l. were collected from the unsprayed district compared to sprayed districts (Table [4](#Tab4){ref-type="table"}; Fig. [3](#Fig3){ref-type="fig"}) (p \< 0.0001). Whereas a general decrease in the proportion of parous mosquitoes was observed across the IRS sites, parity rates in the IRS withdrawn and the unsprayed district remained high. The proportion of parous *An. gambiae* s.l. collected in SND in 2011 was reduced from 44.8 to 37.4% in 2012 when the district was sprayed with pirimiphos methyl CS (z = 2.55, p = 0.011). The proportion of parous females was further reduced to 27.5% (z = 3.22, p = 0.001) in 2013. Despite a significant reduction in parity rates in 2012 (z = 3.24, p = 0.001) when IRS was being implemented, the withdrawal of IRS in 2013 from TKD resulted in an increase in the proportion of parous (older) mosquitoes from 46.6% in 2012 to 50.4% in 2013 (z = −2.07, p = 0.038). Parity rates in TKD increased to 68.5% in 2014 (z = −9.79, p \< 0.0001). Parity rates in TML remained high, increased somewhat from 64.3% recorded in 2013 to 72.3% in 2014 (z = −6.11, p \< 0.0001) (Table [4](#Tab4){ref-type="table"}).Table 4A comparison of parous *An. gambiae* s.l. collected from IRS and non-IRS districts between 2011 and 2014Study district/year comparedCompared parity rate (%)% ChangePooled sample proportionStandard errorZ test statisticp valueRemarkSavelugu Nanton District (SND) 2011--201244.8 → 37.4%−16.50.4210.0292.5490.011\*Change from ACP to PM 2012--201337.4 → 27.5%−26.50.3240.0313.2220.001\*Sprayed with PM 2013--201427.5 → 28.1%+2.10.2770.034−0.1680.867Sprayed with PMTolon Kumbungu District (TKD) 2011--201253.3 → 46.6%−12.60.4990.0213.2420.001\*IRS continued with ACP 2012--201346.6 → 50.4%+8.00.4900.018−2.0710.038IRS withdrawn 2013--201450.4 → 68.5%+36.00.5670.019−9.7930.000\*IRS withdrawnTamale Metropolis (TML) 2011--201268.6 → 65.8%−4.10.6710.0161.7300.084No IRS 2012--201365.8 → 64.3%−2.20.6490.0150.9910.322No IRS 2013--201464.3 → 72.3%+12.50.6810.013−6.1130.000\*No IRS*ACP* alpha-cypermethrin, *PM* pirimiphos-methyl\* p value significant at 0.05 significance level; −ve% represents percentage reduction; +ve% represents percentage increase Fig. 3Proportion of parous *An. gambiae* s.l. and entomological inoculation rate from IRS and non-IRS districts. Parity data is not available for 2009 and 2010 ### Sporozoite rate of local vector species {#Sec18} Whereas sporozoite infection in SND was reduced to a level that could not be detected in 2013 and 2014, the sporozoite rates in TKD (after IRS withdrawal) and Tamale (unsprayed district) remained high (p \< 0.05) (Table [5](#Tab5){ref-type="table"}) in comparison to the sprayed area.Table 5Entomological parameters of malaria transmission recorded for vector species from all districtDistrict and transmission indicators200920102011201220132014Savelugu Nanton District (SND) Insecticide sprayedPYPYPYOPOPOP Parity rate (\# dissected)----44.8% (806)37.4% (457)27.5% (469)28.1% (285) Sporozoite rates (\# examined)0.4% (550)0.2% (882)0.4% (794)1.9% (530)0.0% (841)0.0% (400) ∑ Monthly EIRs post-IRS season (July--December)2.16.96.33.50.00.0Tolon Kumbungu District (TKD) Insecticide sprayedPYPYPYPYNo IRSNo IRS Parity rate (\# dissected)----53.3% (1140)46.6% (1203)50.4% (2066)68.5% (1089) Sporozoite rates (\# examined)0.5% (582)0.5% (1229)0.5% (1453)3.6% (1048)3.3% (883)4.9% (388) ∑ Monthly EIRs post IRS season (July--December)13.821.024.041.8137.9154.4Tamale Metropolis (No IRS) Insecticide sprayedNo IRSNo IRSNo IRSNo IRSNo IRSNo IRS Parity rate (\# dissected)----68.6% (1625)65.8% (1764)64.3% (2687)72.3% (2395) Sporozoite rates (\# examined)1.6% (939)2.0% (1764)3.2% (1307)1.9% (1275)3.1% (1140)2.2% (1754) ∑ Monthly EIRs post-IRS season (July--December)35.0109.4103.1125.6188.0104.7Mean annual rainfall (mm)149.6120.087.487.977.379.9No ovary dissections done in 2009 and 2010Rainfall data obtained from Ghana Meteorological Services Dept. weather stations in Savelugu and Tamale and Savannah Agricultural Research Institute---Tolon Kumbungu*PY* pyrethroids (alphacypermethrin: 2009, 2011 and 2012; deltamethrin 2010); *OP* organophosphate (pirimiphos methyl) ### Entomological inoculation rate of local vector species {#Sec19} Malaria transmission was highly seasonal across all sites (Fig. [3](#Fig3){ref-type="fig"}), as expected. The highest EIRs were observed during the rainy season (May--October) across all the sites. Bivariate Pearson's correlation analysis of the monthly EIRs with rainfall showed that the mean monthly EIRs had a positive correlation with rainfall across all the sentinel sites: SND (R = 0.447, p \< 0.001), TKD (R = 0.276, p = 0.030) and TML (R = 0.282, p = 0.026). All infections in SND were detected during the rainy season. However, in TKD and TML malaria transmission still occurred in the dry season (Fig. [3](#Fig3){ref-type="fig"}). The entomological inoculation rate (EIR), was lower in the IRS districts sprayed with the organophosphate compared to districts sprayed with pyrethroids, unsprayed and IRS withdrawn districts (Fig. [3](#Fig3){ref-type="fig"}). The EIRs during the post-IRS months (July--December) in the two IRS districts (SND and TKD) were suppressed until 2012 when the post-IRS EIR in TKD doubled from 24 infective bites/person/season (ib/p/s) in 2011 to 41.8 ib/p/s. A year-by-year comparison of the transmission intensity in TKD between July and December revealed that the sum of monthly EIRs during this period increased by about threefold in 2013 to 137.9 ib/p/s when IRS was discontinued and remained high in 2014 (Table [5](#Tab5){ref-type="table"}). On the other hand, the sum of post-IRS monthly EIRs in SND was reduced to about half the levels in 2011 (i.e. from 6.3 ib/p/s in 2011 to 3.5 ib/p/s in 2012) just 1 year after the switch from pyrethroids to organophosphates. Transmission intensity was further reduced to a level that could not be detected by the sampling method used in 2013 and 2014 after the shift to organophosphate insecticide. There was yearly variation in EIRs in all the sites but EIRs in the control district (TML) remained high compared to the sprayed areas. Discussion {#Sec20} ========== The major purpose of IRS is to reduce malaria transmission by reducing the survival of malaria vectors after entering and feeding on humans inside dwellings and consequently preventing transmission of malaria infection to others \[[@CR28]\]. The general significant decline in the parity rates in the IRS district over the years (from 2011 to 2014) can likely be attributed to the effect of the IRS \[[@CR5], [@CR29]\]. Parity rates of vectors in TKD were reduced when IRS was being implemented in 2012 but increased steadily after the withdrawal of IRS in 2013 through 2014. Even though some climatic factors, such as temperature and humidity, are known to affect vector longevity and abundance of parous females \[[@CR30]\], such climatic factors, as well as ITN ownership and use are expected to be similar across all IRS and non-IRS districts in this study, which are adjacent to each other. Therefore, the most likely factor for the low parity rate in the sprayed districts is the IRS operation in these districts, demonstrating the benefit of adding IRS where there is modest to high level ITN coverage. The results confirms the seasonality of malaria transmission in all the study sites as indicated by previous studies in Northern Ghana \[[@CR14], [@CR31], [@CR32]\] and justifies the one spray round per year approach adopted by the PMI IRS programme in the area. The EIR in SND was low at the beginning of the study but increased subsequently in 2010 and 2011. The increase can likely be attributed to pyrethroid resistance in the area. However, the timely change to an organophosphate (pirimiphos methyl CS) resulted in reduction in EIR to undetectable levels in 2013 and 2014. The study found *An. gambiae* s.l. to be the predominant vector species collected from all sentinel sites. This is consistent with studies in other parts of the country that also found *An. gambiae* s.l. to be the major *Anopheles* species feeding on humans in Ghana \[[@CR14], [@CR33]\]. The predominance of *An. gambiae* s.l. could in part be attributed to its high reproductive potential \[[@CR34]\] and its high plasticity \[[@CR35]\]. The relatively high number of *An. funestus* collected from TKD over the period could be attributed to the presence of rice irrigation farms in the area, as some studies \[[@CR36], [@CR37]\] have found that *An. funestus* prefers to breed in semi-permanent waters and sites shaded by vegetation, a condition provided by the irrigation sites in the area. The presence of the rice irrigation farms in the TKD area could have also created several breeding sites for mosquitoes even in the dry season and may account for the relatively high number of *An. gambiae* s.l. collected in TKD district. Distribution of *An. coluzzii* and *An. gambiae* s.s. has been found to be dependent on ecological and geographical factors \[[@CR38], [@CR39]\]. Reduction in mean annual rainfall for the study area between 2009 and 2014 (from 149.6 mm in 2009 to 79.9 mm by 2014) could partly explain the general increase in *An. coluzzii* across all sites in 2014 compared to previous years. It is possible that temporary rainfall dependent larval habitats that could support breeding of *An. gambiae* s.s. \[[@CR39]\] might have also declined over the period with the reduction in amount of rainfall (Table [5](#Tab5){ref-type="table"}), thus restricting potential breeding sites to permanent irrigation fields that have been documented \[[@CR38]\] to support breeding of *An. coluzzii*. As a result of local factors such as irrigation, the EIR rates and entomological indices of transmission in TKD did not drop as low as observed in SND. The continuous use of pyrethroids for IRS in TKD in 2012 might have contributed to development of resistance to pyrethroids in 2012. By the end of 2012 *An. gambiae* s.l. had become highly resistant to alphacypermethrin. The 24 h mortality rates in WHO susceptibility tests conducted in 2012 was 76% compared to 97% recorded in 2011 \[unpublished observations\]. This in-part explains the increased malaria transmission intensity observed in TKD in 2012, bringing to the fore the important role that insecticide choice plays in the effectiveness of an IRS programme within the context of an insecticide resistance management strategy. Two years after the withdrawal of IRS malaria transmission intensity increased by about threefold in TKD, in spite of the high ITN coverage maintained in Northern Region by the door-to-door campaign in 2012 and the school-based campaigns in 2013 and 2014. This suggests that in the event that a decision is taken to discontinue IRS in a similar high burden savannah area, additional efforts such as intensive quality social and behaviour change communication (SBCC) to promote proper use of ITNs amongst all age groups and encourage correct malaria prevention and treatment behaviours at the household level may be warranted. Data from TML (unsprayed district) suggests strongly that IRS in the neighbouring district led to the reductions in transmission in the neighbouring district. The yearly fluctuations and relative reductions in transmission intensity in TML shows that factors other than IRS operational quality and pesticide choice, such as rainfall and ITN usage are going to affect transmission indices. Central to the formulation of malaria control strategies is a more critical understanding of the relationship between malaria risk factors, EIRs and disease outcomes. Therefore collection and analysis of routine health information system data on how the reductions in the entomological indicators of malaria transmission translate to reduction in disease burden in the area would be important. Conclusions {#Sec21} =========== The reduction in malaria transmission in the study districts demonstrates the effectiveness of IRS programmes and the potential benefits of their expansion. However, insecticide resistance and its resulting cost implications will determine impact and feasibility of IRS implementation. The study demonstrated that the shift from the pyrethroid based IRS to organophosphates was effective in reducing the longevity of *Anopheles* vector species and eventually their malaria transmission potential. However, the withdrawal of IRS led to an increase in entomological indicators of malaria transmission, suggesting that maintenance of high local ITN coverage was not adequate to prevent a rebound of entomological indicators. The study shows that if IRS is done optimally, it can contribute to significant reductions in key malaria transmission indices. Therefore collection and analysis of routine health information system data is important to complement entomological indicators of malaria transmission. SC, SKD, AS, YY, DS, PP, JW, MAA and DAB were involved in the study conception and design. DD, MAA, DAB, KG, CF and PP provided methodological advice. SC coordinated the field activities to acquire the data, with assistance from SKD, AS, YY, PM and MAA. SC, SKD, AS, YY and PR analysed the data and drafted the manuscript. All authors thoroughly revised the manuscript. All authors read and approved the final manuscript. Acknowledgements {#FPar1} ================ We are grateful to the PMI team and the Ghana National Malaria Control Programme for their technical support and also thank the regional and district staff of the Ghana Health Services and District Assemblies for supporting the supervision of field surveys. Our gratitude also goes to the Entomology team at the Noguchi Memorial Institute for Medical Research for supporting the fieldwork as well as the laboratory evaluations. We are grateful to Mr. Andrew Deyi Saibu, Ernest Fletcher, Jerry Akordam Karbo, Imoro Nurudeen, Ernest Nsefo, Alhassan Inusah, Osman Mohammed Tawfik, Stephen Hafiz and the mosquito collectors for their valuable contribution towards the fieldwork and data collection. We also thank the Chiefs and people of the sentinel communities who allowed us into their communities and homes to collect data. Competing interests {#FPar2} =================== The authors declare that they have no competing interests. The opinions expressed herein are those of the authors and do not necessarily reflect the views of USAID or other institutions that authors are affiliated with. Ethical consideration {#FPar3} ===================== The study received ethical approval from the Institutional Review Board of the Noguchi Memorial Institute for Medical Research, Ghana. All trained mosquito collectors were recruited from the communities and made to sign an inform consent form before participating in the study. Although, the risk to mosquito collectors was low, any collector that showed symptoms of malaria was treated according to the national guidelines for malaria treatment. Funding {#FPar4} ======= Financial support for this study was provided by the US President's Malaria Initiative. Publisher's Note {#FPar5} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

INTRODUCTION ============ The novel coronavirus SARS-CoV-2, the etiological agent of COVID-19, emerged in Wuhan, China in December 2019 and quickly spread to other countries[@B1] ^,^ [@B2]. Due to the rapid increase in the number of cases, on March 11, 2020, the World Health Organization (WHO) declared it to be a pandemic[@B3]. One month after the declaration, more than two million people worldwide had been infected and 135,000 deaths had been registered across 213 countries[@B4]. Worldwide, health systems faced the need to adapt to a critical overload on services, and a shortage of health care professionals and personal protective equipment[@B5] ^,^ [@B6]. In Brazil, the first case of COVID-19 was confirmed on February 26, 2020, and the first death on March 17, both in the state of São Paulo[@B7]. Community transmission was officially recognized in Brazil on March 20, 2020[@B8]. Through May 5, 2020, there were more than 110,000 confirmed cases and approximately 8,000 deaths, with a mortality rate of 6.9%. The three most affected states were São Paulo (34,053 deaths), Rio de Janeiro (12,391 deaths), and Ceará (11,470 deaths)[@B9]. The state of Ceará in Northeast Brazil was one of the first to confirm sustained transmission. Within 45 days of confirmation of its first case, Ceará had registered the third highest number of deaths in the country. The exponential increase in cases and deaths imposed a series of challenges to meet the demand for care, with a real possibility of a collapse of the health services system. The Brazilian government enacted social isolation regulations on March 19 (Decree 33,519) and a lockdown on May 8 (Decree 33,547). Considerable effort was put into expanding the capacity of emergency services, emergency department care, and laboratory testing, as well as the increasing the number of intensive care (ICU) beds[@B10]. We describe the epidemiological scenario of cases and deaths from COVID-19 and their impact on hospital bed occupancy rate in the first 45 days (February 17 to April 27, 2020) of the epidemic in Ceará, Northeastern Brazil. METHODS ======= Study type ---------- The study used an ecological design to compare confirmed COVID-19 cases and deaths to bed occupancy rates in Ceará. In addition, we describe the actions implemented during the first 45 days of the epidemic. Data sources ------------ Data were collected from six different sources: REDCap - Database in which all suspected and confirmed cases of COVID-19 were recorded from the beginning of the epidemic until April 27, 2020 (45 days after the first known case occurred).SIVEP - Gripe - The National Influenza Epidemiological Surveillance Information System that records all cases of severe respiratory infections and related deaths.e-SUS Notifica - A system developed specifically to meet the high demand for notifications of COVID-19, recording mild and moderate cases of the disease that have undergone laboratory investigation.Ceará state civil registry - The number and cause of verified deaths.Central Laboratory of Public Health of Ceará - Confirmatory laboratory testing results.Unified Health System, Ceará Regulation Center - Hospital admissions. Cases Definitions ----------------- We followed the case definitions below for suspected cases of COVID-19: 1\) An acute respiratory condition characterized by a fever or feverish sensation, even if only reported, accompanied by cough OR sore throat OR runny nose OR breathing difficulty. In the case of children, nasal obstruction was also acceptable in the absence of another specific diagnosis. In the case of the elderly, a reported or diagnosed fever was optional. 2) Specific worsening criteria such as syncope, mental confusion, excessive sleepiness, irritability, and loss of appetite. 3) Dyspnea / respiratory discomfort OR persistent pressure in the chest OR O2 saturation less than 95% in room air OR bluish color of the lips or face. 4) In children, in addition to the previous items, nasal flaring, cyanosis, intercostal circulation, dehydration, or lack of appetite. We followed the case definitions below for a confirmed case of COVID-19: a suspected case with molecular biology (RT-PCR in real time) detection of the SARS-CoV-2 virus OR a positive immunological test for antibody detection (rapid or classic serology) OR a history of close or home contact with a laboratory-confirmed case for COVID-19 within seven days before the onset of symptoms, and for which it was not possible to perform laboratory testing. Study variables and data analysis --------------------------------- The variables used in this study were sex, age group, date of onset of symptoms, whether the subject had been hospitalized, place of hospitalization (public or private), date of hospitalization, the time between first symptoms and hospitalization, whether the patient had been admitted to an intensive care unit, the time between the first symptoms and admission to the intensive care unit, laboratory diagnosis, outcome (discharged with resolved symptoms or death), date and place of death (if occurred), municipality of residence, pre-admission signs and symptoms, the ward occupancy rate, and the number (total and occupied) of ICU beds on the day the patient was admitted. All data were analyzed using Epi Info software version 7.0 (U.S. Centers for Disease Control and Prevention, Atlanta, Georgia). Ethical aspects --------------- All ethical principles provided for in the Resolution of the National Health Council (CNS-translated) No. 466, of December 12, 2012, were respected. The study design was approved by the State Secretariat of Health of Ceará. RESULTS ======= The first confirmed cases of COVID-19 in Ceará were diagnosed on March 15, 2020, with onsets of symptoms on the 10th (two cases) and the 11th (one case). Within 45 days of the country's first known case, 37,268 cases had been confirmed in 85.9% of Brazil's 184 municipalities ([Figures 1](#f1){ref-type="fig"} and [Figure 2](#f2){ref-type="fig"}). Of the confirmed cases, 7,833 (21.0%) were laboratory-confirmed, another 20,791 were under investigation and 1,019 were confirmed COVID-19-related deaths. Epidemic week 20 had the highest number of reported cases and the peak of deaths. FIGURE 1:Number and temporal distribution of COVID-19 cases, by epidemiological week of symptom onset. Ceará, Brazil, 2020. FIGURE 2:Spatial distribution of municipalities with confirmed COVID-19 cases in the first 45 days of the epidemic, by epidemiological week. Ceará, Brazil, 2020. The distribution of cases was initially the most widespread in Fortaleza, Ceará's capital city, later overtaken by metropolitan municipalities in which the highest incidences were identified (over 120 cases per 100,000 inhabitants). The virus also spread through municipalities in the Northern region, which also had incidences above 120 cases per 100,000 inhabitants. During the 45-day period, the Central Laboratory of Public Health of Ceará (LACEN-CE) processed more than 15,000 molecular biology exams, reaching 738 tests in a single day (April 27). The lowest positive rate among the examinations was registered on March 21 (15.4%) and the highest on April 26 (84.8%). In the first 15 days, the average positivity was 21.9%, increasing to 46.2% and 73.6% over the next 15 and 30 days, respectively ([Figure 3](#f3){ref-type="fig"}). FIGURE 3:Number of tests and positivity of COVID-19 tests performed during the first 45 days of the epidemic. Ceará, Brazil, 2020. The first death in Ceará was confirmed on March 23; within four days another 10 had been confirmed. Ten days after that, 57 more deaths had been confirmed. Subsequently, the number of confirmed deaths doubled approximately every seven days, reaching the highest single day occurrence on May 1st (53 deaths). Females made up a higher percentage than males of reported and confirmed cases (54.7%) but deaths occurred in a greater proportion in males (58.1%) ([Table 1](#t1){ref-type="table"}). The same pattern occurred in age groups where a predominance of cases were in people aged 20 to 59 years (70.3%), but the percentage of deaths was highest for those over 70 years of age (52.9%). TABLE 1:Demographic and medical characteristics of COVID-19 cases and deaths in the first 45 days of the epidemic. Ceará, Brazil, 2020.VariablesNotifiedConfirmedDeathsN%N%N%**Gender**Male16 87445.33 66646.859258.1Female20 33254.74 16753.242741.9**Age group**\< 1 year5701.5670.900.01- 9 years1 2743.4931.230.310- 19 years1 0972.91341,720.220-59 years26 03170.35 49270.128728.260-69 years3 72710.194512.118818.4\> 70 years4 31311.71 09113.953952.9**Needed hospitalization**Yes2 6477.11 28816.487891.9No34 62192.96 54583.5778.1**Hospitalization in the Intensive Care Unit (ICU)**Yes7772,14996.447358.0No36 49197.97 33493.634242.0 The average age of the cases that progressed to death was 67 (1-101) years old, with more than half (52.9%) occurring in people over 69 years old ([Table 1](#t1){ref-type="table"}). The main symptoms reported among these cases were dyspnea (86.0%), fever (85.2%), cough (84.7%), respiratory distress (77.1%), sore throat (21.5%), diarrhea (14.1%), and vomiting (7.5%). The most common comorbidities were heart disease (66.5%) and diabetes (58.3%). Most deaths occurred in a hospital setting (91.9%), more than one-half required ICU bed hospitalization (58.0%), and 480 (53.9%) needed ventilator support. The average time between the onset of symptoms and death was 18 (1-56) days. The cases that evolved to death took 6 (0 - 40) days to be hospitalized and among those who were hospitalized, the average time in bed was 8 (1 - 49) days Considering the public hospital network in Ceará, on the 45th day of the epidemic COVID-19 patients occupied 655 ward beds, 421 ICU beds, and 376 beds with respirators. This represented 71.3% (23.8%-100.0%) of all ward beds, 80.5% (40.0%-100.0%) of all ICU beds, and 74.9% (40.0%-100.0%) of beds with mechanical ventilation ([Table 2](#t2){ref-type="table"}). TABLE 2:Evaluation of bed use 45 days after the first confirmed case of COVID-19 in Ceará, in 2020.Health UnitInfirmaryICUMechanical VentilationNOccupied%\*NOccupied%NOccupied%Hospital Leonardo da Vinci666192.41158372.2746689.2Hospital Geral de Fortaleza1117668.5514690.24949100.0Hospital Geral Dr César Cals242291.71010100.010990.0Hospital São José826174.488100.0272281.5Hospital de Messejana5656100.0595796.6574578.9Hospital Infantil Albert Sabin423173.888100.08787.5Hospital Abelardo Gadelha (Caucaia)262180.81212100.0121083.3Hospital MM (Maracanaú)13969.25240.05240.0Hospital São Vicente de Paula (Itapipoca)11981.81010100.088100.0Hospital Batista1248871.07342.93266.7Hospital Regional Norte (Sobral)252496.0362569.4231565.2Hospital Regional SC (Quixeramobim)331751.5302790.0301963.3Hospital Regional do Cariri (Juazeiro)21523.8352262.9351851.4Hospital São Vicente (Iguatu)21523.8352262.9351851.4**Total65548574.042133579.637629077.1**[^3] DISCUSSION ========== The first 45 days of the COVID-19 epidemic in Ceará showed an explosion in the number of cases and deaths, reaching more than 1,000 confirmed deaths. In the first days after the detection of the first cases, infections spread most rapidly in the city of Fortaleza and municipalities with higher HDI, mainly among users of the supplementary health network. In the following month, infections began to spread to the interior of the state, reaching the periphery of large cities and the most vulnerable social and economic populations. These populations have a higher prevalence of comorbidities, and often have living situations that make effective social isolation unfeasible[@B11]. This likely contributed to the early exponential increase in the number of cases and deaths. The symptoms described in this study were similar to those reported for the pandemic for other countries[@B12]. Our findings show a predominance of male deaths, as was also reported in China[@B13]. A recent survey of more than 2,000 people in Ceará showed that a significant percentage of females perceive themselves to be at risk of COVID-19, while males reported greater difficulty adhering to social isolation practices[@B14]. These observations could, in part, explain our findings. At the beginning of the epidemic, due to concerns about the simultaneous increase of seasonal influenza and the possibility of a new dengue epidemic, the Ceará health department advised that medical care should only be sought for severe symptoms such as shortness of breath, difficulty breathing, or cognitive impairment[@B15] ^,^ [@B16]. This initial guidance, together with fears in the population of acquiring infection from attending a health facility, probably contributed to the increase in the number of home deaths. By the beginning of April, there were widespread notices by the Ceará health department that everyone, especially those with known risk factors, should seek basic care at the first sign of symptoms. This change reduced the number of home deaths, but also generated a large demand for basic health centers, contributing to the collapse of the outpatient care network. This study had several limitations. We relied on secondary databases from local surveillance systems. Further, not all cases with laboratory confirmation were by RT-PCR, but those limitations don't invalidate the results. The Brazilian Ministry of Health published Decree 10.211 at the end of January, reactivating the Interministerial Executive Group on Public Health Emergency of National and International Importance (GEI-ESPII). Among other duties, the GEI-ESPII is responsible for implementing preparation measures and the official health surveillance response to the COVID-19 epidemic[@B17] ^,^ [@B18]. It is noteworthy that despite extensive efforts by health services, social isolation measures were not effective in reducing the speed of disease spread to the pace at which the health care network was expanding. This imbalance led to the collapse of the health system[@B19] ^,^ [@B20]. Development and implementation of effective health service responses to COVID-19 has been challenging, especially in poor countries[@B21]. In this scenario, it is essential to identify effective drugs for early stages of the disease and to develop an efficient vaccine[@B22]. LPGC is recipient of a fellowship for research productivity granted by the Brazilian National Council for Scientific and Technological Development (CNPq/Brazil). [^1]: **Authors' contributions: DRQL, SMD, LABGF, MMA, RGG, GPP, JNCF, LXF, ARPC, TBRL, PMCL, LPM, TMC, LPGC:** Conception and design of the study, acquisition of data; **DRQL, SMD, LABGF, MMA, RGG, GPP, JNCF, LXF, ARPC, TBRL, PMCL, LPM, TMC, LPGC:** Conception and design of the study, analysis and interpretation of data, final approval of submitted manuscript; **DRQL, SMD, LABGF, MMA, RGG, GPP, JNCF, LXF, ARPC, TBRL, PMCL, LPM, TMC, LPGC:** Conception and design of the study, analysis and interpretation of data. [^2]: **Conflicts of interest statement:** The authors declare there are no conflicts of interest. [^3]: \*Percent occupied.

In keeping with our recent Paths and Places theme, the column this month provides advice for success as a GI researcher in a setting outside of an academic medical center. Joshua Friedman, who started his career in academia, offers his perspective on the benefits and tradeoffs of carrying out GI research in the pharmaceutical industry. As always, we welcome your feedback and ideas. *REBECCA WELLS, MD, AGAF* *Associate Editor* *CMGH* Joshua R. Friedman, MD, PhDFor many gastrointestinal (GI) research scientists, the prospect of a move from academia to the pharmaceutical industry can be daunting. Their training has qualified them for a range of research-related careers, but in many cases they have been exposed only to those located in academic centers. Their mentors are predominantly academic scientists, so they lack role models on other career paths. It is also common to hear that a colleague working in industry has gone to the Dark Side, reflecting a cultural bias against industry (and a misunderstanding of the Force, but that is a subject for a different essay). In this installment of Pathways to GI Research, I hope to present a view of the pros and cons of industry positions gleaned from my own experience. Training and Independent Research {#sec1} ================================= Over the course of my medical and scientific education (MD/PhD) and clinical training (pediatric gastroenterology) at the University of Pennsylvania, my ideal career comprised conducting sponsored research as faculty in an academic setting. In retrospect, there was a trajectory to my research interests that might have predicted an eventual move to industry. My graduate research was basic in that it focused on protein--protein interactions for a class of transcription factors, it was not related to a specific disease and was conducted almost exclusively in vitro. When I returned to the bench during fellowship in pediatric gastroenterology, my research was devoted to understanding an organ (liver) rather than a class of protein, and the model system was an organism (mouse) rather than cultured cells. However, the central questions related to embryonic development rather than a disease, and the implications for human beings were by inference only. The subsequent 10 years of research continued this trend as I shifted focus to diseases of the gastrointestinal system (biliary atresia, inflammatory bowel disease, eosinophilic esophagitis) and translational studies using human specimens. It is not surprising that continuing on this trajectory would lead me to have a direct impact on patients through drug development. Move to the Pharmaceutical Industry {#sec2} =================================== It often has been observed that major life changes require both a push to change the status quo and a pull toward a new state. Despite my long-standing goal of an academic career, the tenure review process provided an impetus to consider alternatives even though it culminated in promotion. My information gaps and misconceptions regarding the pharmaceutical industry were reduced and I was willing to consider a career change. A second push was the treadmill of applications for research funding; although my laboratory was supported by National Institutes of Health and foundation grants, the uncertainty and time-consuming nature of the process impaired my quality of life as an academic scientist. This is a growing problem in academia[@bib1]; in contrast, I now spend more of my time conducting research, rather than working to obtain the funding to do so. This has increased the odds of research success and my professional quality of life. However, these push factors were only the initial triggers of my interest in industry. The decision to make the transition was driven more strongly by the attractions that I came to appreciate as I learned more. Two impressions struck me early in the process and have been reinforced through experience: the quality of the science and the impact of the work. The bar for admission into the pharmaceutical industry is high, so the resulting community is very strong and its members push each other to excel. The individuals I work with combine deep scientific knowledge with a commitment to rigor. This commitment is more than cultural because all of our work is subject to internal and external scrutiny (eg, by health authorities such as the Food and Drug Administration) that I believe far exceeds that of the academic peer-review process. Moreover, most science within industry must go rapidly from observation into practice in the form of preclinical or clinical validation; simply put, it has to work. The discovery, testing, and marketing of new drugs are a complex enterprise requiring many different areas of expertise, including disease biology, medicinal chemistry, clinical study design and conduct, biomarkers, statistics, medical safety, health care regulation, and more. There is a large gap between National Institutes of Health--funded discovery and the appearance of a beneficial drug on the pharmacy shelf. By bringing together all the moving parts necessary to make drug development possible, the pharmaceutical industry environment allows me to have a much greater impact than I could achieve in my academic laboratory (it also provides a range of career opportunities). For example, I was part of the phase 3 study of the anti--interleukin 12/23 antibody ustekinumab in Crohn's disease, leading to approvals for marketing in the United States and Europe that will directly benefit many patients. Perspective From My Current Position {#sec3} ==================================== I lead a Disease Biology team within the Immunology Therapeutic Area of Janssen Research and Development (Janssen is the pharmaceutical division of the Johnson & Johnson family of companies). The Immunology Therapeutic Area is focused on 3 main diseases: inflammatory bowel disease, rheumatoid arthritis, and psoriasis. The mission of Disease Biology is to advance our understanding of these diseases in areas that will promote drug development. In practice, this includes both internally conducted bench work and external collaborations dedicated to identifying new drug targets or supporting the development of targets already in the portfolio. My experience also illustrates the opportunity to branch out into new (for me) disease areas outside of GI. The focus is strongly translational, using human disease tissues to validate a pathway or a drug candidate, or to derive gene expression networks that can reveal new disease pathways; in this regard, access to large numbers of samples from the company's clinical studies is a great advantage. Despite our internal resources, key scientific expertise and biological specimens are often external to the company, so we encourage collaboration with academic leaders in their fields. The network of my team's collaborations connects us to outstanding scientists across the globe, from the West Coast of the United States to Australia (the long way around), and I find these interactions to be a great perk of the position. Finally, as an industry partner I enjoy being the source of funding (and scientific collaborator), rather than the applicant for funding. The transition from academia to industry also has come with a number of trade-offs. Perhaps the most common concern for scientists considering industry research is a loss of independence. It is true that my team's focus is determined by the company's broader strategic plan. Although persistence is valued in science, in an industry setting, terminating a project that lacks scientific priority or commercial potential is also a virtue. The discontinuation of a project to which one has devoted time and passion can be disappointing. However, initiative and innovation are highly valued, and I have found more opportunity for influencing strategic decisions than I expected. Overall, I have not found industry-related constraints to be very different from those I encountered in academia, where they were based on funding agency priorities and areas considered currently popular in the research community. A second perceived trade-off is a decrease in research publications. Manuscripts describing discoveries made in industry positions are subject to review for intellectual property considerations, although this generally is not an impediment or results in a delay rather than a lack of publication. In addition, there is a range within the pharmaceutical industry in how much priority and time are granted to writing and publishing, so in the worst case these may become extracurricular activities. By the same token, the use of publication-related metrics in academia has been criticized for its negative effects on scientists and publication quality[@bib2]; this pressure is reduced in industry. Job security is an important determinant of work satisfaction for any employee, including scientists. As in any industry, pharmaceutical corporations are subject to cycles of hiring and workforce reductions. However, a track record---and personal network---from prior industry experience is a strong advantage in seeking a new position. In practice, this means that the event of a downsizing may lead to a change in employer or area of scientific focus, rather than a loss of employment. Ultimately, individual scientists must compare that level of risk with the risk of funding loss in the academic sphere. Finally, there are differences in culture between academia and industry, reflecting the common goals of the latter. More time must be invested in communicating to other teams and to leadership. A new set of acronyms and procedures must be learned, some of which are general to industry and some of which are company-specific. Confidentiality must always be considered in external conversations, although this easily is addressed through nondisclosure agreements. However, the corporate culture also provides an opportunity to learn a science that will be new to many biologists: organizational dynamics, the theory and practice of coordinating groups of people in pursuit of shared goals. This can be just as interesting as the coordination of molecules in a living system. Scientists in industry are part of a much larger whole, which can feel different from being an independent scientist in academia. It brings additional burdens but also the benefits of a shared purpose and the potential for large-scale, direct impacts on patients. **Conflicts of interest** The author discloses the following: Joshua R. Friedman is an employee of Janssen Research & Development.

Introduction ============ Here, we report that one suspected dead case and two confirmed, rapid onset and fatal infections caused by a *Klebsiella pneumoniae* ST86 strain occurred in a three-bed surgery ward of a university hospital with 1,800 beds and 40,000 annual admissions in Shanghai, China in 2013 (See **Figure [1](#F1){ref-type="fig"}**). {#F1} The first suspected patient, a 64-year-old male, after resuscitation from acute recurrent myocardial infarction, was sent to the surgery ward at 3 PM, on May 31th. In addition, the patient has the following diseases: coronary atherosclerotic heart disease, cerebral infarction, hypertension, type 2 diabete, and hyperlipoidemia. Since June 1st, the patient was initially treated empirically with moxifloxacin (400 mg daily) and gradually improved. At that time, the WBC count was 9.10 × 10^9^/L with 64.3% granulocytes and 20.7% lymphocytes. However, on June 4th, his WBC count increased to 9.59 × 10^9^/L with unusual 82.0% granulocytes and 8.1% lymphocytes, accompanied by a high temperature, breathing difficulty, and brick red foamy blood sputum, and followed by respiratory distress syndrome. Immediately, the patient was transferred into an intensive care unit (ICU) and died at 4 AM, on June 5th. Since samples were not collected as a result of the patient's sudden death, this case was listed as a suspected case and potential source of infection only. The second patient, a 72-year-old male, was admitted due to chordae tendineae of mitral valve and mitral incompetence on May 16th. At the beginning of admission, the patient showed clear breathing sounds, but no cough, expectoration, and wheezing sound. On May 29th, the patient recovered well in the ICU and was transferred to the same surgery ward. Until August 4th, the chest wound of the patient still didn't heal well and followed by the exudates containing plenty of thick, tan-black colored secretion. The WBC count was 12.91 × 10^9^/L with 85.9% granulocytes and 7.9% lymphocytes. However, microbiological culture of tan-black colored secretion was negative. At 1 AM on June 8th, the patient's body temperature suddenly increased to 39.2°C, accompanied by chest tightness, shortness of breath, coarse breath sounds, brick red foamy blood sputum, expectoration and wheezing as well as coarse rales. At that time, the WBC count decreased to 8.48 × 10^9^/L with 91.0% granulocytes and 4.7% lymphocytes. Subsequently, the patient was transferred into the ICU and immediately treated with linezolid and cefoperazone/sulbactam. However, his condition deteriorated late at night on June 8th and he eventually died from multiple organ failure caused by severe pneumonia and septic shock at 1:30 AM, June 9th. A routine microbiological culture of blood sample remained negative. However, a *K. pneumoniae* strain *Kp523* from a sputum sample was isolated. The third patient, a 74-year-old male, having a long disease history of ischemic cardiomyopathy and old myocardial infarction, was the first one admitted to the ward on April 18th, because of the unsuccessful percutaneous coronary intervention, and daily treated with moxifloxacin (400 mg daily). However, on the night of May 25th, some unexpected symptoms, including chest tightness and shortness of breath, suddenly appeared for large effusions in right pleural and resulted in a restrictive ventilatory dysfunction. And the WBC count was 15.8 × 10^9^/L with 93.8% granulocytes and 3.8% lymphocytes. After discharging more than 1,100 ml dark-red colored effusions in right pleural, the patient condition obviously improved. And microbiological culture of the pleural effusion sample was still negative. Until June 3rd, the patient had almost completely recovered with most parameters of routine blood tests being near normal (6.3 × 10^9^/L WBC with 71.3% granulocytes and 16.0% lymphocytes). Thus antibiotic treatment was stopped. Unexpectedly, at 10:40 PM, on June 8th, the patient experienced acute pneumonia including high temperature (39.0°C), red foamy bloody sputum, breathing difficulty and low oxygen saturation, immediately followed by respiratory distress syndrome. The patient was rapidly transferred to the ICU. At 9 PM on June 9th, despite combination therapy of ganciclovir, imipenem/cilastatin, tigecycline, and hemofiltration, the patient deteriorated and finally died of multiple organ failure caused by severe pneumonia and septic shock at 11:45 PM, June 10th. Subsequent routine microbiological culture of blood sample successfully isolated *K. pneumoniae* strain named *Kp562*. The protocol for this study was approved by Ethics Committee of Ruijin Hospital, Shanghai JiaoTong University School of Medicine. The combination of routine PFGE and MLST typing revealed that the two isolates shared highly similar PFGE profiles as *K. pneumoniae* ST86 (**Figure [2A](#F2){ref-type="fig"}**). Via PCR targeting *wzx* gene of capsule serovar K1, K2,([@B13]) and aerobactin, *allS, kfu, magA*, and *rmpA* ([@B7]). The detailed PCR conditions listed in Supplementary Table [S1](#SM1){ref-type="supplementary-material"}. The two strains were confirmed as serotype K2 and Aerobactin-*kfu-rmpA* positive but *magA-allS* negative (**Figure [2B](#F2){ref-type="fig"}**). The two isolates grew as positive hypermucoviscous phenotype (HV) with a mucoviscous string \> 5 mm in length from the colony on the blood agar plate (**Figure [2C](#F2){ref-type="fig"}**). Moreover, except nitrofurantoin and ampicillin, the two isolates exhibited susceptibility to the other 16 antimicrobial agents including ampicillin/sulbactam, piperacillin/tazobactam, cefazolin, cefotetan, ceftazidime, ceftriaxone, cefepime, aztreonam, ertapenem, imipenem, amikacin, gentamicin, tobramycin, ciprofloxacin, levofloxacin, and trimethoprim/sulfamethoxazole using Vitek 2 automated system (BioMérieux, St Louis, MO, USA) according the Clinical and Laboratory Studies Institute guidelines (CLSI2013). {#F2} Background ========== During the previous decades, infections due to *K. pneumoniae* have primarily been attributed to classic *K. pneumoniae* (cKP) strains. However, since the first hypervirulent *K. pneumoniae* (hvKP) strain causing distinct pyogenic liver abscess (PLA) were reported in Taiwan in 1986, the new phenotypic hvKP strains have been increasingly common worldwide ([@B11]). Compared to the cKP strains, the unique clinical features of the hvKP strains are their ability to cause life-threatening community-acquired invasive infections ([@B11]). The invasive nature of the hvKP strains has been associated with multiple virulence factors (VFs), including K2-specific regulator of mucoid phenotype A gene (*rmpA*), and K1-specific mucoviscosity-associated gene A (*magA*), iron-uptake system (*kfu*) and HTH-type transcriptional activator (*allS*), ([@B13]). In 2011, one report in France indicated that two hypervirulent K2 ST86 strains were more virulent than those of K1 ST23 strains ([@B1]). Actually, in 1991, the first ST86 strain-CG43 causing severe bacteremia and PLA was reported in Taiwan ([@B11]). Recently, two hypervirulent K2 ST86 strains have been reported including strain hvKP1 isolated from a 24-year-old Vietnamese male in New York, which caused a unique PLA accompanied by a metastatic spread to the spleen ([@B10]), while another strain, isolated in South Korea, caused severe bacteraemic infection ([@B3]). Currently, ST86 isolated from five PLA cases throughout Taiwan, Hong Kong, and Singapore have been reported ([@B7]). Here, this is the first report of hvKP ST86 strains causing one suspected and two confirmed fatal cases in one surgery ward in China. Discussion ========== So far, compared to major ST23 of K1, not much has been studies about ST86 strains. In the present study, we have experimentally confirmed the presence of the pivotal VFs closely associated with PLA or hypermucoviscosity phenotype. In the present study, the more distinct difference of the ST86 strains than other K2 strains is the additional presence of *kfu* gene exclusively in our strains. To date, most reported ST86 strains have possessed a HV, *rmpA*, and PLA phenotype but no *kfu*. For example, two recently published studies described that all of the seven hvKP ST86 strains from France, USA, and Taiwan and 11 ST86 strains from Hong Kong, Singapore, and Taiwan, harbored no *kfu* gene ([@B12]; [@B7]). However, the two *Kp523/562* strains exhibited HV and unique K2-Aerobactin*-kfu-rmpA* positive phenotypes, yet no PLA. This may be associated with the rapid, acute onset of disease caused by hypervirulence of ST86 strains; the patients were dead before PLA occurred. Furthermore, the HV-Aerobactin-*rmpA* positive strains of *K. pneumoniae* had previously been shown to be highly lethal to mice and were more virulent than HV-Aerobactin-*rmpA* negative strains ([@B13]). It was well-known that the K1 PLA strains carried *kfu* gene involved in human liver abscesses and metastatic spread infections, though this was generally absent in K2 strains; probably making the K1 strains significantly more prevalent than K2 ([@B13]). Therefore, the acquisition of *kfu* gene by the *Kp523/562* strains might further enhance their hypervirulence ([@B2]). In fact, it was recently found that certain K2 strains from other minor ST types, like ST375 and ST380, also possessed the *kfu* gene, reflecting that the presence of *kfu* gene might be closely associated with different genetic background or geographic origin among serotype K2 strains ([@B7]). It is unknown how K2 strains might obtain the *kfu* gene from K1 strains. Although virulence plasmid acquisition is one of major mechanisms acquiring new gene for hvKP strains, however, as a chromosomal virulence gene, the acquisition of *kfu* gene in K2 strains might occur via horizontal gene transfer mediated by bacteriophage or transposons. Whole genome sequencing of the two strains and comprehensively comparing them with other hvKP strains to further investigate the detailed mechanism of the acquirement of *kfu* in K2 ST86 strain is warrant in the near future. Another important clinical characteristic of the hvKP ST86 strains is the high mortality rates likely due to their increased hypervirulence. Based on these findings above, among all of the 12 ST86-infected cases with detailed clinical information, ST86 strains totally caused five fatal infections, seven severe PLA, and one bacteraemic infection. It seems that mortality rate (41.6%, 5/12) caused by ST86 was obviously higher than that of other hypermucoviscositic strains, such as 7.3% in Canada, ([@B9]), 3.0 and 31% in Taiwan ([@B4]; [@B5]). The significant mortality rate might be primary due to availability of very limited data reporting the prevalence and characteristics of ST86 worldwide. On the other hand, despite the high prevalence of ST86 accounting for 46% PLA cases in other Asia countries/regions ([@B7]), there have only been two recent reports which concluded that only two out of 84 PLA-causing strains were ST86 strains in China. This suggests that both severe onset and rapid death were caused by its hypervirulence without PLA and that the prevalence of ST86 infection in China could be greatly underestimated ([@B6]; [@B8]). Concluding Remarks ================== In this study, for the first time, unusual VF profiles correlating with the hypervirulence of ST86 strain causing two rapid-onset fatal infections were identified. However, regarding very little data of ST86 presently available, thus further molecular and epidemiological characterization studies should be performed to reveal the distribution and clinical importance of the unusual virulence factor profiles of hvKP ST86 strains originating from different geographic regions or countries outside Asia. Based on our finding, despite antibiotic therapy after acute onset, all of the two confirmed and one suspected patients rapidly deteriorated and eventually died. Therefore, combined application of MLST and VF detection implemented immediately to confirm the presence of hvKP ST86 strains should be crucially important for the doctor to early treat the patients with antibiotics if they carry the ST86 strains with Aerobactin-*kfu*-*rmpA* phenotype, even they don't have any clinical signs of this disease at that time. Moreover, when these patients have been clinically improved, keeping on with oral antibiotic treatment for additional 2--3 weeks will be also vital for successfully preventing reinfection or relapse of hvKP strains. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Supplementary material ====================== The Supplementary Material for this article can be found online at: <http://journal.frontiersin.org/article/10.3389/fmicb.2015.00721> ###### Click here for additional data file. [^1]: Edited by: *Evangelos Giamarellos-Bourboulis, University of Athens Medical School, Greece* [^2]: Reviewed by: *George Dimopoulos, ATTIKON University Hospital, Greece; Efthymia Giannitsioti, ATTIKON University General Hospital, Greece* [^3]: This article was submitted to Infectious Diseases, a section of the journal Frontiers in Microbiology

Introduction {#Sec1} ============ The explosive growth of the world's human population and subsequent water and energy demand has led to an expansion of standing surface water (Rosenberg et al. [@CR35]). Many of these standing waters, including small watering ponds for cattle, fishing ponds, irrigation impoundments, and drinking water reservoirs, suffer from an over-enrichment of the water with nutrients (Smith and Schindler [@CR39]). Such ongoing eutrophication may lead to dense cyanobacterial blooms and floating scums (Schindler et al. [@CR36]; Smith and Schindler [@CR39]). These blooms may cause high turbidity, anoxia, fish kills, bad smell (Paerl and Huisman [@CR32]), and serious environmental and human health problems because several cyanobacteria can produce a variety of very potent toxins (e.g., Codd et al. [@CR11]; Dittmann and Wiegand [@CR13]). Climate change is expected to aggravate hazardous blooms (Paerl and Huisman [@CR32]), while safe and aesthetically acceptable water is a growing need in a modern society (Steffensen [@CR41]). Hence, water resource management is challenged worldwide to reduce the vulnerability to the threats of harmful cyanobacterial blooms. Particularly in developing countries where the high costs of chemicals for water and wastewater treatment might limit their application, the development of cost-effective and environmentally acceptable mitigating measures is desirable. In that context, water clarification and disinfection with natural products, such as the seeds of the pan-tropical tree, *Moringa oleifera* Lam, is of particular interest. Crushed *M. oleifera* seeds are highly effective in clarifying very turbid water (Muyibi and Evison [@CR26]; Ndabigengesere and Narasiah [@CR27]). Traditionally, the seeds are used in rural areas of Sudan and Malawi for the clarification of drinking water (Muyibi and Evison [@CR26]; Anwar et al. [@CR4]). Flocculation of raw Nile water revealed that *M. oleifera* seeds could remove up to 97% of the algae present (Shehata et al. [@CR38]). Dehusked *M. oleifera* press cake is efficient in the removal of hydrophobic organic pollutants from water (Boucher et al. [@CR8]), and extracts might remove other pollutants, such as heavy metals and surfactants (Beltrán-Heredia and Sánchez-Martín [@CR5], [@CR6]). *M. oleifera* pods are efficient in absorbing organic pollutants and pesticides (Akhtar et al. [@CR1], [@CR2]). High-quality activated carbon can be prepared from the waste husks of *M. oleifera* (Pollard et al. [@CR33]; Warhurst et al. [@CR46]), which could effectively remove up to 98% of the cyanobacterial microcystin-LR (Warhurst et al. [@CR47]). In addition to the strong water-clarifying properties, *Moringa* seeds have also been reported removing more than 90% of cercariae from the water phase (Olsen [@CR30]). Bacterial numbers can be reduced drastically due to coagulation (Madsen et al. [@CR23]), and a pronounced hygienic effect is caused by a strong antibacterial potential against Gram-negative and Gram-positive bacteria (Suarez et al. [@CR44], [@CR45]). However, so far, no study has examined direct effects of filtrates from crushed *M. oleifera* seeds on growth of cyanobacteria and on their ability in terminating cyanobacterial blooms. The purposes of this research were to (1) examine in the laboratory the effect of filtrates from crushed *M. oleifera* seeds on the growth and Photosystem II efficiency of the most common bloom-forming cyanobacterium in eutrophic freshwater systems worldwide, *Microcystis aeruginosa* Kützing and (2) to test the filtrates on their potential to mitigate high densities of *M. aeruginosa*. Inasmuch as *M. oleifera* seed extract exerts bactericidal activity in vitro against Gram-positive and Gram-negative bacteria (Cáceres et al. [@CR9]; Ali et al. [@CR3]; Oludoro and Aderiye [@CR31]; Suarez et al. [@CR44], [@CR45]) and cyanobacteria are Gram-negative (Stanier and Cohen-Bazire [@CR40]), we hypothesize that *Moringa* seed extracts have an antimicrobial activity against the cyanobacterium *M. aeruginosa*. Materials and methods {#Sec2} ===================== The cyanobacterium *Microcystis aeruginosa* NIVA-CYA 43 was obtained from the Norwegian Institute for Water Research (NIVA, Norway). Stock cultures of this strain were grown in our laboratory in 100-mL Erlenmeyer flasks containing 50 mL of slightly modified Woods Hole Chu (WC) medium (Lürling and Beekman [@CR22]) with vitamins added (H (biotin) and B~12~ (cyanocobalamin**)** at 50 ng L^−1^ and B~1~ (thiamine HCL) at 100 ng L^−1^). The flasks were placed in a Gallenkamp ORBI-SAFE Netwise Orbital Incubator at 22°C in 40 rpm and in a 18:6-h light/dark cycle. The light/dark cycle was programmed that light intensity increased gradually to a maximum of 130 µmol quanta m^−2^ s^−1^ and subsequently decreased again to darkness, which resulted in a daily average light intensity of ∼57 µmol quanta m^−2^ s^−1^. This strain of *M. aeruginosa* is completely uni- and bicellular under the given growth conditions. Seeds of *M. oleifera* were obtained from the Miracle Trees Foundation (Rotterdam, the Netherlands). The seed coat was removed and the pulp crushed in a mortar. Aliquots of the powder were transferred into pre-weighed aluminum boats in fivefold. Dry weights were determined after 24 h at 105°C, and ash weights were determined after placing the boats for 3 h at 550°C. Two grams of the powder was put into 250 mL sterile WC medium, shaken for 5 min, and filtered through a 0.45-µm membrane filter. The filtrate was used in the experiments, had a pH of 7.05--7.10, and was analyzed for dissolved nutrients (ammonium-N, nitrate/nitrite-N, and phosphate-P) using a SKALAR autoanalyzer. Aliquots of the filtrate were analyzed for total dissolved organic carbon content using a TOC analyzer (model 700, OI-Analytical). Turbidity (in NTU) of the filtrate was measured using a Hach 2100P turbidity meter. Absorption was determined over the range 350--750 nm PAR in 1-nm intervals using a 5-cm cuvette that was placed in a Beckman Coulter Du730 Life Sciences UV/VIS spectrophotometer. Experimental procedure {#Sec3} ---------------------- Two experiments were performed to study the effect of filtrate from crushed *M. oleifera* seeds on the growth of *M. aeruginosa* (experiment 1) and on a cultured bloom of *M. aeruginosa* (experiment 2). In the first experiment, filtrate from crushed *M. oleifera* seeds was added to 50 mL autoclaved WC medium in sterile 100-mL Erlenmeyer flasks. To three replicates, the following amounts of filtrate were added: 0, 25, 50, 125, 250, 500, and 1,000 μL, which corresponds to 0, 4, 8, 20, 40, 80, and 160 mg crushed *Moringa* per liter. An inoculum of *M. aeruginosa* was added to each Erlenmeyer flask with an initial chlorophyll-*a* (Chl-*a*) concentration of 13.4 (±0.1) µg L^−1^. The flasks were closed with a cellulose plug and placed at random in a Gallenkamp ORBI-SAFE Netwise orbital incubator under the same conditions as outlined above. Samples were taken initially and after 2, 4, 7, 9, and 11 days and were analyzed on the Chl-*a* concentration and the Photosystem II efficiency (Φ~PSII~) using a PhytoPAM phytoplankton analyzer (Heinz Walz GmbH, Germany). Growth rates were determined by linear regression on natural logarithm transformed chlorophyll-*a* data. Growth rates were statistically evaluated using a one-way ANOVA in the program SPSS version 16.0. Significant differences were determined with a Tukey post hoc comparison test (*P* \< 0.05). In the second experiment, aliquots of *M. aeruginosa* cultures that had been grown for 3 weeks reaching chlorophyll-*a* concentrations of 3,500 (±15) µg L^−1^ were transferred to sterile 100-mL Erlenmeyer flasks and mixed with filtrate from the crushed *Moringa* seeds and WC medium such that the final volume was 50 mL. The initial Chl-*a* concentration was 269 (±4) µg L^−1^, reflecting a bloom of *M. aeruginosa*. The amounts of filtrate tested were 0, 10, 100, 500, and 1,000 μL, which corresponds to 0, 1.6, 16, 80, and 160 mg *Moringa* per liter. Samples were taken initially and after 1, 2, 3, 6, 8, 10, and 14 days and analyzed on the Chl-*a* concentration and the Photosystem II efficiency. In addition, after 2 days, subsamples were analyzed on the number of particles, biovolume, and mean particle volume using a CASY cell counter (Schärfe System Gmbh., Germany). Results {#Sec4} ======= *Moringa* seeds and filtrate {#Sec5} ---------------------------- The crushed pulp of *Moringa* seeds contained 65 (±3) mg water g^−1^, and the organic content was 96.3% (±0.1%, *n* = 5). Mixing 2 g of the pulp powder with 250 mL sterile WC medium yielded a turbid white suspension (\>1,000 NTU). However, filtration through a 0.45-µm membrane filter removed the turbidity. This filtrate of the 8 g L^−1^ suspension had a clear appearance and a turbidity of 1.88 (±0.52) NTU (*n = 4*), while the WC medium had a turbidity of 0.44 (±0.32) NTU (*n* = 4). Where the absorption coefficients in the PAR waveband were higher in the stock filtrate than in WC medium, absorption in controls (WC medium) and the highest dose of filtrate from the crushed *Moringa* seeds were similar (Fig. [1](#Fig1){ref-type="fig"}). Fig. 1Light absorption (per meter) of algal growth medium (WC medium), algal growth medium with the highest dosage of filtrate from crushed *M. oleifera* seeds used in the experiments (160 mg L^−1^; 2.73 mg DOC L^−1^), and of filtrate from the stock (8,000 mg L^−1^; 137 mg DOC L^−1^), over the PAR wave band (350--750 nm) The filtrate from crushed *Moringa* seeds (8 g L^−1^) contained on average 878 (±13) µg PO~4~-P L^−1^ (*n* = 4), 292 (±254) µg NO~3~/NO~2~-N L^−1^ (*n* = 4), and 810 µg NH~4~--N L^−1^ (*n* = 2). The carbon content of the filtrate was 136.7 (±0.5) mg DOC L^−1^ (*n* = 3). Hence, the treatments with filtrates comparable to 0, 4, 8, 20, 40, 80, and 160 mg crushed *Moringa* L^−1^ contained about 0, 0.07, 0.14. 0.34, 0.68, 1.37, and 2.73 mg DOC L^−1^, respectively. Effect of *Moringa* seeds on growth of *Microcystis* {#Sec6} ---------------------------------------------------- In the absence and lowest dosage of crushed *Moringa* seeds (4 mg L^−1^), Chl-*a* concentrations of *M. aeruginosa* increased strongly over time, reflecting good growth (Fig. [2a](#Fig2){ref-type="fig"}). At 8 mg L^−1^, this increase was somewhat reduced, while a decrease was observed at *Moringa* seed concentrations of 20 mg L^−1^ and more (Fig. [2a](#Fig2){ref-type="fig"}). The repeated measure ANOVA revealed that the course of the Chl-*a* concentrations over time was significantly different between treatments (*F*~6,14~ = 700.2, *P* \< 0.001). Tukey's test showed three homogeneous groups: (1) 0 and 4 mg L^−1^, (2) 8 mg L^−1^, and (3) 20, 40, 80, and 160 mg L^−1^. Growth rates, however, were similar and on average 0.50 (±0.01) per day in the 0-, 4-, and 8-mg L^−1^ treatments (Fig. [3](#Fig3){ref-type="fig"}). In exposures of 20- to 160-mg *Moringa* seed per liter, growth rates were not only significantly lower (*F*~6,14~ = 33.2, *P* \< 0.001) but also negative with on average −0.23 (±0.05) per day, indicating that the inoculum did not grow and even declined (Fig. [3](#Fig3){ref-type="fig"}). Fig. 2Effects of filtrates from crushed *M. oleifera* seeds (0--160 mg L^−1^) on the chlorophyll-*a* concentrations (**a**) and Photosystem II efficiency (**b**) of *M. aeruginosa* populations. *Error bars* = SD, *n* = 3Fig. 3Growth rates (per day) of *M. aeruginosa* populations that were exposed to different concentrations of filtrate from crushed *M. oleifera* seeds (0--160 mg L^−1^). *Error bars* = SD, *n* = 3. *Similar letters* (*A*, *B*) indicate homogeneous groups that are not different at the *P* = 0.05 level (Tukey's test) The Φ~PSII~ was high and unaffected in 0, 4, and 8 mg L^−1^, but dropped to zero rapidly in exposures of 20- to 160-mg *Moringa* seed per liter (Fig. [2b](#Fig2){ref-type="fig"}). Where over the course of the experiment the Φ~PSII~ remained zero in the 40- to 160-mg *Moringa* seed per liter treatments, it showed recovery in the 20-mg L^−1^ treatments (Fig. [2b](#Fig2){ref-type="fig"}). The repeated measure ANOVA revealed that the course of the Φ~PSII~ over time was significantly different between treatments (*F*~6,14~ = 881.3, *P* \< 0.001). Tukey's test showed three homogeneous groups: (1) 0, 4, and 8 mg L^−1^, (2) 20 mg L^−1^, and (3) 40, 80, and 160 mg L^−1^. Effect of *Moringa* seeds on *Microcystis* under blooming conditions {#Sec7} -------------------------------------------------------------------- In concentrations of ≤16 mg L^−1^, filtrate from crushed *Moringa* seeds had no effect on the high-density *M. aeruginosa* cultures (Fig. [4](#Fig4){ref-type="fig"}). Exposure to concentrations of 0--16 mg L^−1^ resulted in 2 weeks in more than a tenfold increase in Chl-*a* concentrations (Fig. [4a](#Fig4){ref-type="fig"}). In contrast, in the 80- and 160-mg L^−1^ treatments, Chl-*a* concentrations dropped to 1% and 5% of that of the initial bloom, respectively (Fig. [4a](#Fig4){ref-type="fig"}). A Tukey's test, following a repeated measure ANOVA (*F*~4,10~ = 1123, *P* \< 0.001), revealed three homogeneous groups that were all significantly different from each other: (1) 0 and 1.6 mg L^−1^, (2) 16 mg L^−1^, and (3) 80 and 160 mg L^−1^. Fig. 4The course of chlorophyll-*a* concentrations (**a**) and Photosystem II efficiency (**b**) in *M. aeruginosa* populations in high, blooming densities that were exposed to different concentrations of filtrate from crushed *M. oleifera* seeds (0--160 mg L^−1^). *Error bars* = SD, *n* = 3 The Φ~PSII~ dropped to and remained zero at the highest dosage of 160 mg L^−1^. In 80 mg L^−1^, the Φ~PSII~ also showed and initial decline, but never reached zero and recovered over the course of the experiment (Fig [4b](#Fig4){ref-type="fig"}). In the 0- to 16-mg L^−1^ treatments, the course of the Φ~PSII~ was similar, which was confirmed by a Tukey test indicating that the 0-, 1.6-, and 16-mg L^−1^ treatments formed one homogeneous group (repeated measure ANOVA, between-subjects effects; *F*~4,10~ = 445.8, *P* \< 0.001). In the 80- and 160-mg L^−1^ treatments, the Φ~PSII~ was significantly lower than in 0--16 mg L^−1^; the 80- and 160-mg L^−1^ treatments also differed significantly from each other (Fig. [4b](#Fig4){ref-type="fig"}). Inasmuch as the Chl-*a* concentrations increased in the first 2 days while concomitantly the Φ~PSII~ dropped to zero, subsamples were analyzed with a cell counter. The number of particles and the total biovolume were significantly lower in the 80- and 160-mg L^−1^ treatments (*F*~4,10~ = 45.1, *P* \< 0.001 and *F*~4,10~ = 18.7, *P* \< 0.001, respectively) than in the 0- to 16-mg L^−1^ treatments (Fig. [4](#Fig4){ref-type="fig"}). The mean particle volumes were similar (*F*~4,10~ = 2.13, *P* = 0.151) among controls and treatments (Fig. [5](#Fig5){ref-type="fig"}). Fig. 5Biovolume (*black bars*, µm^3^ mL^−1^), number of particles (*white bars*, mL^−1^), and mean particle volume (µm^3^) in *M. aeruginosa* populations after 2 days of exposure to different concentrations of filtrate from crushed *M. oleifera* seeds (0--160 mg L^−1^). *Error bars* = SD, *n* = 3. *Similar letters* (*a*, *b* and *α*, *β*, *χ*) indicate homogeneous groups that are not different at the *P* = 0.05 level (Tukey's test) Discussion {#Sec8} ========== The results of the current study clearly revealed that filtrate from crushed *Moringa* seeds inhibited the growth of the common cyanobacterium *M. aeruginosa*. Above 20 mg L^−1^ (≈0.34 mg DOC per liter), the Chl-*a* concentrations declined, which was not caused by sedimentation as flasks were shaken. The EC~50~ here was between 8 and 20 mg L^−1^, which is considerably lower than the EC~50~ of 287.5 mg L^−1^ for the green alga *Scenedesmus obliquus* (Ali et al. [@CR3]). In that study, the green alga expressed growth in all concentrations tested and reached, after 10 days, even in 300-mg L^−1^ seed extract Chl-*a* concentrations of around 1,100 µg L^−1^ (Ali et al. [@CR3]). In contrast, in the current study , Chl-*a* concentrations of *M. aeruginosa* populations became less than the inoculum in *Moringa* seed extract concentrations of 20 mg L^−1^ and more. The decline suggests a cyanobactericidal activity, which is corroborated by the Φ~PSII~ measurements that showed a rapid drop to zero. The cyanobacteria died in the treatments where they were exposed to 40 mg L^−1^ or more, but the Φ~PSII~ showed a recovery in the 20-mg L^−1^ treatments that was coincided by a slight recovery in chlorophyll-*a* concentrations (e.g., 1.1 and 3.1 µg L^−1^ on days 7 and 11, respectively). Similar observations on a transitory inhibition have been made in some bacteria that resumed growth after some period of inhibition (Madsen et al. [@CR23]; Suarez et al. [@CR44]; Oludoro and Aderiye [@CR31]). The addition of filtered *Moringa* seed extract to the cyanobacterial cultures implied higher DOC concentrations. DOC can markedly affect the light attenuation in lakes and might be a major light-absorbing component in the blue part of the spectrum, below 500 nm (Markager and Vincent [@CR24]). In most natural freshwaters, DOC concentrations may range between 0.5 and 50 mg L^−1^, but sometimes go up to a few hundreds of milligrams per liter (Steinberg [@CR42]). In this study, the maximum DOC concentration used was 2.73 mg L^−1^, which is not sufficient for having an effect on the light attenuation (Jones [@CR17]). The turbidity of the highest amount of filtrate added was slightly higher than the background turbidity, and spectral analysis clearly revealed that absorption over the PAR wave band was similar in controls and treatments with the highest dosage of filtrate (see Fig. [1](#Fig1){ref-type="fig"}). Hence, physical effects of the DOC can be excluded as a causal factor in the observed inhibition of the cyanobacterium *M. aeruginosa*. Another characteristic of DOC is that it might act as a weak herbicide in a range of 0.025--25 mg DOC per liter by interfering within the photosynthetic electron transport chain and reducing the oxygen release (Steinberg [@CR42]; Prokhotskaya and Steinberg [@CR34]). Also in this study, the filtrate from *Moringa* seed extract had an effect on the photosynthetic apparatus as reflected in the PAM-derived Φ~PSII~. Phenolic compounds are asserted as being major players in these algicidal effects in both natural organic matter and leachates from straw and leaf litter (Steinberg [@CR42]). Especially barley straw extracts have received considerable attention: 50% growth inhibition of *M. aeruginosa* at concentrations between 70 and 230 mg L^−1^ has been reported (Martin and Ridge [@CR25]). Other studies reported only 5% growth of *M. aeruginosa* at 2.57 mg L^−1^ (Newman and Barrett [@CR28]) and toxicity above 35 mg L^−1^ (Waybright et al. [@CR48]). The EC~50~ values (8--20 mg L^−1^) in the current study for *Moringa* seed extracts might be comparable to those of barley straw extracts, but a striking difference seems the lethal effect of *Moringa* seed extracts to *M. aeruginosa* at slightly elevated concentrations (\>40 mg L^−1^), while this is generally not being observed for barley straw extracts. The concentrations at which growth of *M. aeruginosa* was inhibited by *Moringa* are also in the same order of magnitude as those (20--50 mg L^−1^) that have been reported for bacteria (Oludoro and Aderiye [@CR31]). Although reports on the antimicrobial effects of *M. oleifera* are still rare (Anwar et al. [@CR4]), several studies have shown that *M. oleifera* seed extract exerts bactericidal activity in vitro against Gram-positive and Gram-negative bacteria (Cáceres et al. [@CR9]; Ali et al. [@CR3]; Oludoro and Aderiye [@CR31]). This study is the first one reporting a cyanobactericidal activity, which is in line with the findings of others as cyanobacteria are Gram-negative bacteria. Initially, the antimicrobial activity of *Moringa* seeds was attributed to the presence of 4-α-[l]{.smallcaps}-rhamnosyloxy benzyl isothiocyanate, the only known glycosidic mustard oil (Eilert et al. [@CR14]). However, later studies revealed that the *Moringa* seed-derived polypeptide (Flo), which is responsible for the sedimentation, also mediated bacterial disinfection (Suarez et al. [@CR44]). In a follow-up study, Suarez et al. ([@CR45]) showed that the antibacterial activity of Flo did not simply result from a cell aggregation mechanism but that it required additional activities leading to bacterial membrane damage. The current study also showed that a bloom of *M. aeruginosa* could be terminated using high concentrations (≥80 mg L^−1^ or 1.37 mg DOC per liter) of *Moringa* seed extract. However, at 80 mg L^−1^, the Φ~PSII~ did not drop to zero as had been observed in the growth experiment. Most likely, it reflects a cyanobacteria biomass-dependent efficacy of the cyanobactericidal compounds where higher concentrations of cyanobacteria, such as under blooming conditions, might inactivate all the compounds by binding to the membranes, thereby leaving a portion of the cells unexposed. Hence, at higher cyanobacteria densities, more seed extract is needed to obtain a complete wipeout of the bloom. High concentrations of *Moringa* seed extract might, however, come with some serious drawbacks. One drawback of using high amounts of *Moringa* seed extract is that organic matter from the seeds is released into the water that can cause odor, taste, and color problems (Ndabigengesere and Narasiah [@CR27]) and might also facilitate the development of microorganisms after some time (Madsen et al. [@CR23]; Suarez et al. [@CR44]; Oludoro and Aderiye [@CR31]). Another drawback is that *Moringa* seed extracts contain nutrients that might promote cyanobacteria. In this study, filtrate from 1-g crushed *Moringa* seeds per liter yielded approximately 110 µg P L^−1^, which is lower than values of 1.94 mg P g^−1^ (Kalogo et al. [@CR18]) and 1.31 mg P g^−1^ (Ndabigengesere and Narasiah [@CR27]) found in other studies. In the latter study, a dosage of 50 mg L^−1^ caused an increase in phosphate from 0.4 to 1.6 mg L^−1^ (Ndabigengesere and Narasiah [@CR27]). In the current study, in the experimental period of 11 to 14 days, no stimulating effect on the cyanobacteria was detected of the *Moringa*-derived organic carbon and nutrients; however, this aspect warrants further investigation by applying recolonization in the high-dosage Erlenmeyer flasks. The drawbacks of organic carbon and nutrient releases can be overcome using dialysis, delipidation, or ion exchange (Okuda et al. [@CR29]; Ghebremichael [@CR15]). Nonetheless, these methods focused on the coagulant component from *M. oleifera* seeds that might only express part of the antimicrobial activity. Further research is needed on the cyanobactericidal effect of different fractions from the crushed seeds and other parts of the plant (roots, leaves). When dense populations of *M. aeruginosa* were exposed to high concentrations of filtrate from crushed *Moringa* seeds, the Φ~PSII~ of the suspensions dropped to zero rapidly, but Chl-*a* concentrations first increased. Cell counter data, however, revealed that total particle numbers and biovolume were significantly lower in the high-dosage treatments. Apparently, cells died and released pigments in the medium that yielded higher fluorescence values due to the package effect (Kirk [@CR19]). Indeed, Flo-derived peptides have been found to decrease strongly cell viability by permeabilizing or disrupting the bacterial membrane (Suarez et al. [@CR45]). As a consequence, cyanotoxins might also be freed from the cells after cell lyses (Berg et al. [@CR7]), concomitantly deteriorating the water quality as cyanotoxins, such as *M. aeruginosa* hepatotoxin microcystin-LR, may persist for weeks in the water (Lahti et al. [@CR20]). For example, treatment of water infested with the cyanobacterium *Cylindrospermospis raciborskii* in a drinking water reservoir in 1979 in Palm Island (Australia) with copper sulfate, causing cyanobacterial cell lysis, led to an outbreak of hepatoenteritis among 148 consumers (Hawkins et al. [@CR16]). Such cyanobacterial intoxication of humans may occur not only through consumption of cyanotoxin-contaminated drinking water but also through food and dietary supplements; through recreational and occupational contact, and by hemodialysis (e.g., Carmichael et al. [@CR10]; Dittmann and Wiegand [@CR13]; Stewart et al. [@CR43]). It has been demonstrated that 10 mg L^−1^*Moringa* could remove 93% to 98% of the cyanotoxin microcystin-LR (20 µg L^−1^) from the aqueous phase, while 50 mg L^−1^ removed the toxin to levels below the practical detection limit (\<0.08 µg L^−1^) of the HPLC-DAD procedure (Warhurst et al. [@CR47]). In microcystin-contaminated drinking water, the absorbed material should be removed from the water by coagulation, sedimentation, or filtering prior to consumption, whereas in reservoirs, ponds and irrigation impoundments natural sedimentation might yield sufficient removal. Up to now, the effectiveness of *Moringa* seeds in removing cyanotoxins other than microcystin-LR from the water is not known. Yet, *Moringa* seeds have the potential of being cheap biosorbents for cyanotoxins from aqueous media, as they are for the removal of cadmium (Sharma et al. [@CR37]) organic pollutants and methyl parathion (Akhtar et al. [@CR1], [@CR2]). The pan-tropical tree *M. oleifera* is a highly nutritive food with medicinal, phytochemical, pharmacological, and potent water-purifying properties (Anwar et al. [@CR4]). The majority of the studies on water purification with *M. oleifera* seeds have focused on flocculation--coagulation; 97--99% reduction in turbidity can be reached, while 90% of bacteria and 99% of coliforms can be precipitated (Ali et al. [@CR3]; Liew et al. [@CR21]; Oludoro and Aderiye [@CR31]). The results presented here of a clear cyanobactericidal activity of crushed *Moringa* seeds combined with a growing number of more frequent and longer lasting cyanobacterial blooms (de Figueirdo et al. [@CR12]; Paerl and Huisman [@CR32]) show that *Moringa* seed extracts might have a potential as an effect-oriented measure lessening cyanobacterial nuisance, especially at household and community level in tropical regions. We owe special thanks to the youngsters of NXT Generation: Jolie, Marte, Thomas, Joris, Chris, Karlijn, Anne, Maxi, and Marene ([http://www.dse.nl/∼nxtgeneration](http://www.dse.nl/∼nxtgeneration)) for stimulating us to perform the research and for arranging the *Moringa* seeds. **Open Access** This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

INTRODUCTION ============ The glucose transporter (GLUT) protein family contains various important molecules that use glucose diffusion gradients (and those of other sugars) for transport across the plasma membrane. They exhibit various substrate specificities, kinetic properties, and tissue expression profiles, depending on their role ([@nly124-B1]). The GLUT protein family belongs to a larger facilitator superfamily of membrane transporters ([@nly124-B2]); 13 related members of the SLC2A (GLUT) protein family have been identified in humans ([@nly124-B3]). Structurally, these molecules are divided into 3 classes; GLUT1--4 (class 1); GLUT5, 7, 9, and 11 (class 2); and GLUT6, 8, 10, and 12 (class 3) ([@nly124-B2]). In the central nervous system, GLUT1 plays a critical role in cerebral glucose uptake and is the major GLUT isoform expressed in brain endothelial cells and brain astrocytes. It has been suggested that GLUT1 supplies glycolytically derived lactate to neurons as a major fuel source ([@nly124-B4], [@nly124-B5]). The protein is a 492-amino-acid uniporter that is expressed by endothelial cells of the blood-brain barrier. In humans, it is encoded by the *SLC2A1* gene ([@nly124-B6]) and the GLUT1 protein possesses 12 transmembrane segments, a single site of N-linked glycosylation, a relatively large, central cytoplasmic linker domain; it also exhibits a topology where both the N and C terminals are positioned in the cytoplasm ([@nly124-B6]). In 1996, Younes et al reported that GLUT1 expression in colorectal carcinomas associated with high incidence rates of lymph node metastasis ([@nly124-B7]). Subsequently, numerous studies have reported contradictory evidence of the relationship between GLUT1 expression and prognosis in solid human tumors ([@nly124-B8]). Of note is a comprehensive meta-analysis of 2948 patients across 26 different studies that indicated that GLUT1 expression was a promising biomarker for prognosis, consistently showing an association with overall survival at 3 and 5 years ([@nly124-B15]). Among the tumor types evaluated, higher expression of GLUT1 in tumor tissues was also linked to worse overall survival at 3 and 5 years in oral squamous cell carcinomas and breast cancers ([@nly124-B15]). The expression of GLUT1 in astrocytomas and glioblastomas has also been reported to correlate with hypoxia-inducible factor-1 (HIF-1) in pseudopalisades, suggesting a link to hypoxia ([@nly124-B16]). However, the exact role of GLUT1 in tumorigenesis remains to be elucidated in highly aggressive glioblastomas. To address this, we analyzed the expression of GLUT1 in IDH wildtype, World Health Organization (WHO) grade IV glioblastomas and any correlation with patient prognosis. MATERIALS AND METHODS ===================== Patient Samples --------------- Surgically collected tissue samples from 52 patients with glioblastomas were analyzed. All tumor specimens were retrieved from the archives of Kurume University Hospital, Kurume, Japan between 2010 and 2016. The study was performed in accordance with the principles of the Helsinki Declaration and was approved by the institutional ethics committee. All specimens were histologically diagnosed as IDH-wildtype glioblastoma using WHO criteria ([@nly124-B21]). Multivariate survival analysis was used to assess the significance of various prognostic factors, including complete resection, radiotherapy, age (\>18 years), Karnofsky performance status (KPS) scores, and the expression of GLUT1. Immunohistochemistry -------------------- Tissue samples were fixed in 10% buffered formalin, embedded in paraffin, and processed using conventional histological and immunohistochemical methods. Five-micrometer-thick sections were stained with hematoxylin and eosin for histological evaluation. The remaining unstained serial sections were used for immunohistochemistry analysis following heat-induced antigen retrieval. These were stained with the appropriate antibodies and expression was evaluated with immunoperoxidase (ChemMate ENVISION kit/HRP\[DAB\]; DakoCytomation, Carpinteria, CA) using a Dako Autostainer Universal Staining System (DakoCytomation). The primary antibodies used were antiGLUT1 (1:3200; Abcam, Cambridge, UK), anti-endoglin (CD105; 1:50; Novocastra, Newcastle, UK), and antiKi-67 (MIB-1; 1:100; Immunotech, Marseille, France). MIB-1 labeling indices were determined as the percentage of the nuclear area stained in areas of maximal labeling. GLUT1 expression in glioblastomas was evaluated as both staining intensity and the number of stained cells by 2 observers in independent examinations. Samples that varied significantly between the 2 observers were re-evaluated to arrive at a consensus. Fluorescence Immunohistochemical Staining ----------------------------------------- Fluorescence double immunostaining of GLUT1 and other markers (HIF1α, CD34, and Ki67) in tumor tissues was performed. Double staining was performed in a glioblastoma sample with antiGLUT1 (1:100, Thermo Fisher, UK), antiKi67 (1:100, Dako, Glostrup, Denmark), and antiCD34 (1:100, Leica, Milton Keynes, UK) antibodies followed by addition of antirabbit IgG-Texas Red (TR; sc-2780, Santa Cruz Biotechnology, Santa Cruz, CA), and antimouse IgG-Alexa488 (Life Technologies, Carlsbad, CA). Assessment of staining colocalization was performed using the OPAL 7-color fIHC Kit (Perkin Elmer, Waltham, MA) according to the manufacturer's protocols. Fluorescence was measured using a fluorescent microscope (BX51FL, Olympus, Tokyo, Japan) and a charged-coupled device camera (DP71, Olympus). DNA Isolation and Mutation Analysis ----------------------------------- The mutational statuses of *IDH1* and *IDH2* were determined by Sanger sequencing. Briefly, genomic DNA was isolated from the relevant formalin-fixed paraffin-embedded (FFPE) tissue blocks by cutting 10-µm-thick sections, followed by extraction using a QIAamp DNA Micro Kit (Qiagen, Hilden, Germany). *IDH1* exon 4 and *IDH2* exon 4 were amplified by polymerase chain reaction (PCR) with the indicated primers ([Table 1](#nly124-T1){ref-type="table"}). The amplifying conditions for *IDH1* and *IDH2* were both an initial denaturing step of 95°C for 10 minutes, followed by 40 cycles of denaturation at 95°C for 30 seconds, an annealing step at 60°C for 30 seconds, and then extension at 72°C for 30 seconds, with a final extension at 72°C for 10 minutes. PCR products were separated on 2% agarose gels by electrophoresis, excised and then sequenced on an ABI PRISM 310 Genetic Analyzer (Life Technologies, Gaithersburg, MD). Analyses of sequence data were performed using GENETYX software Ver. 10 (GENETYX, Tokyo, Japan) and a reference sequence complementary to each gene. ###### Primer Design and PCR Product Gene Primer Sequence Product -------------------------------------------- -------------------------------------------- --------- IDH1 Forward: 5′-GGTTGAGGAGTTCAAGTTGAAACAAAT-3′ 244 bp Reverse: 5′-CACATACAAGTTGGAAATTTCTGGGCC-3′ IDH2 Forward: 4′-GGGGTTCAAATTCTGGTTGAAAGATGG 289 bp Reverse: 5′-TAGGCGAGGAGCTCCAGTCGGG-3′ RNAscope -------- *SLC2A1* (*GLUT1*) mRNA in situ hybridization was performed using an RNAscope 2.5 HD Assay Kit-BROWN (ACD, Newark, CA), according to the manufacturer's protocols. Tissue sections, 2.5-μm thick, were deparaffinized in xylene and dehydrated through an ethanol series. The sections were then incubated in citrate buffer (10 nmol/L, pH 6.0) and maintained at 100--103°C for 15 minutes using a hot plate. They were then rinsed in deionized water and immediately treated with 10 μg/mL protease (Sigma-Aldrich, St. Louis, MO) at 40°C for 30 minutes in a HybEZ hybridization oven (Advanced Cell Diagnostics, Hayward, CA). Hybridization with target probes, preamplifier, amplifier, label probes, and chromogenic detection were performed as previously described for cultured cells. RNAscope scores and the heterogeneity of *SLC2A1* mRNA signals were estimated using the manufacturer's recommended protocols and a semiquantitative scoring system. The staining scores were as follows: 0, no staining or \<1 dot to every 10 dots; staining score 1, 1--3 dots/cell; staining score 2, 4--10 dots/cell or very few dot clusters; staining score 3, \>10 dots/cell and \<10% positive cells with clusters; staining score 4, \>10 dots/cell and 10% positive cells with clusters. The evaluation was performed by 3 observers (S.K., Y.S., and K.Y.) across independent examinations. Cell Culture ------------ Human glioma cell lines T98, U87, and U251 were obtained from American Type Culture Collection in 2009. Authentication of the cell lines was unnecessary because cells were expanded by culturing them for \<2 passages and stored at −80°C. Low-passage cells were used for experiments within 6 months after resuscitation. They were maintained in Dulbecco's modified Eagle medium supplemented with 10% fetal bovine serum (FBS), 100 U/mL penicillin, and 100 μg/mL streptomycin. Cell Viability and Cell Proliferation ------------------------------------- Cell proliferation was evaluated in triplicates by a colorimetric WST-1 assay (Roche, Mannheim, Germany) according to the manufacturer's protocol. Approximately 1000 cells of each population were seeded in 96-well plastic plates in 200 μL of culture medium supplemented with 0.1% FBS. The plates were incubated for 4 hours at 37°C. Twenty microliters of WST-1 (10% of total volume) was added to the cells, and the cells were incubated. The plate was read using DS2 (Dynex Technologies, Chantilly, VA) by measuring the absorbance of the dye at 450 nm, with 600 nm set as the reference wavelength. Averages of the absorbance values were calculated and plotted. The cells were treated with various concentrations of WZB-117 (Sigma-Aldrich, Munich, Germany), a specific GLUT-1 inhibitor. Statistical Analysis -------------------- To reveal any correlation between the expression of GLUT1 in glioblastomas and patient prognoses, statistical analysis among the relevant groups was performed using JMP13 (SAS Institute Inc., Cary, NC). Survival rates were computed using Kaplan-Meier curves. Patients were excluded if they were lost to follow-up at the time of analysis. Cox proportional hazards regression models were used for multivariate analyses of overall survival. In addition, Kendal tau rank correlation tests were performed between GLUT1 expression, Ki-67 labeling indices, and the number of CD105 (endoglin) positive vessels in glioblastomas. Statistical significance was set at a threshold of p \< 0.05. Gene Expression Database ------------------------ Information regarding SLC2A1 alterations and patient's survival time in glioblastomas was downloaded from the Cancer Genome Atlas (TCGA) Database, an open access database that is publicly available at <http://www.cbioportal.org> ([@nly124-B22], [@nly124-B23]). Six hundred and four glioblastoma cases (TCGA, Provisional) were selected. These cases were divided into 2 groups: SLC2A1 mRNA upregulated or not altered. Gene altered cases were defined as that expression level of SLC2A was observed greater than standard deviation from the mean. The proportion of upregulated cases was calculated and overall survival and progression-free survival were estimated by Kaplan-Meier analysis. Furthermore, co-expression genes with SLC2A1 in glioblastomas were investigated by calculating Spearman's correlation coefficients. We selected vascular endothelial growth factor (VEGF) as the neoangiogenic marker, CD44 as the mesenchymal marker, PDGFRα as proneural marker, transforming growth factor-beta (TGF-β) as the epithelial mesenchymal transformation's marker and HIF1α as the hypoxic marker ([@nly124-B24]). RESULTS ======= Clinical Data ------------- A summary of the clinical information for the 52 patients with glioblastomas is shown in [Table 2](#nly124-T2){ref-type="table"}. The median age at diagnosis was 65.9 (33--86) years and 18 cases (34.6%) were female, 34 cases (65.4%) male. The median KPS at diagnosis was 74.0 (30--90) years. Tumors were localized to the frontal lobe in 21 cases, the temporal lobe in 18 cases, the parietal lobe in 9 cases, and in other regions for the remaining 4 cases. All patients received surgical treatment, followed by chemoradiotherapy with temozolomide. ###### Patients Characteristics Characteristics *n* = 52 ------------------------------------ --------------- Median age, years (range) 65.9 (33--86) Sex, n (%)  Male 34 (65.4%)  Female 18 (34.6%) Localization, n (%)  Frontal 21 (40.4%)  Temporal 18 (34.6%)  Parietal 9 (17.3%)  Others 4 (7.7%) Median KPS at preoperation (range) 74.0 (30--90) Surgical procedure, n (%)  Gross total resection 33 (63.4%)  Partial resection 19 (36.6%) Abbreviations: KPS, Karnofsky Performance Status. Expression of GLUT1 ------------------- Our study showed that GLUT1 was expressed on tumor cell membranes in all 52 cases, supporting previous reports ([@nly124-B13], [@nly124-B19]). The intensity of staining could be divided into 3 categories, mild (grade 1; [Fig. 1B](#nly124-F1){ref-type="fig"}), moderate (grade 2; [Fig. 1C](#nly124-F1){ref-type="fig"}), or strong (grade 3; [Fig. 1D](#nly124-F1){ref-type="fig"}). Six high-power fields in each tissue samples were randomly selected and used to calculate the ratio of grades 2--3 assessments across the tumor sample. We defined a ratio of \>20% between grades 2 and 3 as the high GLUT1 expression group, whereas ratios \<19% were defined as a low expression group. Using these criteria, high expression of GLUT1 was detected in 16 cases and low expression was detected in 36 cases. Using an RNAscope system, *SLC2A1* mRNA was detected in glioblastoma tumor cells as either brown distinct points or clustered patterns ([Fig. 2](#nly124-F2){ref-type="fig"}). A semiquantitative 4-level staining score was established for the RNAscope data. Samples defined as grades 3 or 4 (high expression) in the RNAscope study were the same as the GLUT1 high expression samples. In contrast, RNAscope staining scores of grades 1 or 2 (low expression) were found in the GLUT1 low expression group. These data confirm that both GLUT1 and *SLC2A1* mRNA expression differ in glioblastomas and that there is an association between protein and transcript. {#nly124-F1} {#nly124-F2} Localization of GLUT1 --------------------- We next assessed the localization of GLUT1-positive cells in tissues, revealing that GLUT1 cells were located among perivascular and pseudopalisading cells and in tumor cells at the boundary between tumors and normal brain tissue ([Fig. 3](#nly124-F3){ref-type="fig"}). Of note, GLUT1-positive cells were more commonly found at the perivascular region, in addition to pseudopalisades and tumor boundaries ([Fig. 3D](#nly124-F3){ref-type="fig"}). This suggests that GLUT1 expression is associated with these migratory and hypoxic areas, indicating a link with glycometabolism. To show localized GLUT1 expression, double immunofluorescence was performed. Glioblastoma cells expressing GLUT1 were observed around CD34-positive blood vessels ([Fig. 4A](#nly124-F4){ref-type="fig"}). Furthermore, GLUT1 and HIF1α were co-expressed in pseudopalisading necrosis ([Fig. 4B](#nly124-F4){ref-type="fig"}). These findings supported the specific localization of GLUT1 expression in hypoxic areas of glioblastomas. {#nly124-F3} Association Between GLUT1 Expression, Angiogenesis, and Proliferation --------------------------------------------------------------------- Next, CD105 staining was used as a marker to distinguish between normal vessels and malignant intratumoral neovascularization to assess any correlation between high GLUT1 expression and angiogenesis. Averages of CD105-stained cell counts from 3 areas were recorded as CD105 intratumoral vessel density to compare the degree of angiogenesis. This showed that GLUT1 in the high expression group more frequently associated with CD105-positive vessels relative to the low GLUT1 expression group ([Fig. 5A](#nly124-F5){ref-type="fig"}). The number of CD105-positive vessels positively correlated with the number of GLUT1-positive cells in the high expression tumors (correlation coefficient: 0.61). We also measured if there was any correlation between the expression of GLUT1 and tumor cell proliferation in glioblastomas using MIB-1 labeling indices. MIB-1 staining was immunohistochemically evaluated across continuous sections, revealing that the MIB-1 labeling indices in the high GLUT1 expression group were higher than that in the low expression group (Wilcoxon test: p \< 0.001; [Fig. 5B](#nly124-F5){ref-type="fig"}). Furthermore, double fluorescent immunostainings for GLUT1 and MIB-1 suggested that tumor cells with high proliferative potency expressed GLUT1 ([Fig. 4C](#nly124-F4){ref-type="fig"}). Together, these data indicate a link between GLUT1 angiogenesis and glioblastoma cell proliferation. {#nly124-F4} {#nly124-F5} Cell Proliferation and Viability -------------------------------- GLUT1 expression of U87, U251, and T98 cells was confirmed by Western blot analysis (data not shown). The viability of T98 and U87 with WZB-117 decreased in a concentration-dependent manner ([Fig. 6A, B](#nly124-F6){ref-type="fig"}), whereas that of U251 decreased at 5 μM of WZB-117 ([Fig. 6C](#nly124-F6){ref-type="fig"}). {#nly124-F6} Association Between GLUT1 Expression and Patient Survival --------------------------------------------------------- Finally, Kaplan-Meier survival curve analysis indicated that the high GLUT1 expression group had lower overall survival rates than the low expression group (log-rank test: p = 0.001; [Fig. 6D](#nly124-F6){ref-type="fig"}). There were no differences between progression-free survival rates in the high and low expression groups (data not shown). A further multivariate analysis showed that GLUT1 was an independent predictor for poor prognosis when compared with already known prognostic factors, such as age or KPS ([Table 3](#nly124-T3){ref-type="table"}; hazard ratio 5.59, 95% confident interval 2.22--14.4, p = 0.0003). This suggests that GLUT1 is a novel independent predictor for glioblastoma outcome. ###### Univariate and Multivariate Predictors of Overall Survival Univariate Multivariate ------------------------------ ------------ -------------- ---------- ------ ------------ -------- High expression of GLUT1 7.22 3.05--17.3 \<0.0001 5.59 2.22--14.4 0.0003 KPS \<90 1.93 1.01--3.93 0.04 1.54 0.75--3.32 0.23 Age \>50 2.05 0.91--5.53 0.09 1.86 0.75--5.44 0.18 Except gross total resection 1.13 0.58--2.12 0.69 1.04 0.50--2.06 0.89 Abbreviations: CI, confidence interval; KPS, Karnofsky Performance Status; HR, hazard ratio. Gene Expression Database ------------------------ To validate the result of the present study, data from TCGA database regarding GLUT1 in glioblastomas were analyzed ([Supplementary Data](#sup1){ref-type="supplementary-material"}). Two hundred and forty-nine cases (42%) showed upregulated SLC2A1 in 591 glioblastomas from TCGA database. The overall survival of upregulated cases tended to be shorter than the not altered cases (Log-rank test; p = 0.375, [Supplementary Data Fig. S1A](#sup1){ref-type="supplementary-material"}), and the progression-free survival of upregulated cases was significantly shorter than the not altered cases (Log-rank test; p = 0.01, [Supplementary Data Fig. S1B](#sup1){ref-type="supplementary-material"}). SLC2A1 correlated with VEGF (Spearman's correlation coefficient 0.63, [Supplementary Data Fig. S2A](#sup1){ref-type="supplementary-material"}), TGF-β (Spearman's correlation coefficient 0.35, [Supplementary Data Fig. S2B](#sup1){ref-type="supplementary-material"}), and HIF1α (Spearman's correlation coefficient 0.33, [Supplementary Data Fig. S2C](#sup1){ref-type="supplementary-material"}). CD44 and PDGFRα did not show significant correlation with SLC2A1. These data from genome-wide analyses implied that GLUT1 was related with angiogenesis and hypoxia in glioblastomas overall. DISCUSSION ========== It has previously been shown that glioblastomas are in a relatively hypoxic state with a degree of undernutrition due to their high proliferative potency and invasive capacity ([@nly124-B25], [@nly124-B26]). This assessment has been supported by several studies that have reported hypoxic environments in glioblastomas ([@nly124-B16], [@nly124-B27]). Using these data and a reverse interpretation of the Pasteur effect, it can be inferred that the energy metabolism of glioblastomas under hypoxia would depend on the glycolytic system ([@nly124-B28]). Angiogenesis elicited by such hypoxia would result in an aerobic environment and the glycolytic system of these tumor cells would be enhanced by the Warburg effect ([@nly124-B28]). However, irrespective of whether tumor cells are under an aerobic or anaerobic condition, tumor cells require an enormous amount of glucose to meet their energy requirements. Our study indicates that glioblastoma cells upregulate the GLUT1 transporter to promote glucose uptake in order to meet these high requirements. The present study also suggests a correlation between GLUT1 expression and MIB-1 labeling indices, suggesting a link between cell proliferation and the transporter. GLUT1 expression was observed in high proliferative glioblastoma cells positive for MIB-1 ([Fig. 4C](#nly124-F4){ref-type="fig"}). In our study, GLUT1 expression was evaluated in IDH wild-type glioblastomas using immunohistochemistry (to evaluate protein expression) and RNAscope in situ hybridization (to evaluate transcript expression). RNAscope allows single molecule visualization in individual cells to be achieved through the use of a novel probe design strategy and a hybridization-based signal amplification system that suppresses background noise ([@nly124-B29]). Previously, a number of studies have reported a link between certain mRNAs and cancer. Among them, Wang et al reported the efficacy of the RNAscope technique for examining invasive breast carcinomas that compared it to real-time quantitative PCR (qPCR) and other US Food and Drug Administration-approved methods, including fluorescence in situ hybridization (FISH) ([@nly124-B30]). Both RNAscope and qPCR data matched the FISH analysis in 97.3% of cases, showing that RNAscope is as accurate as qPCR and FISH ([@nly124-B30]). Importantly, the RNAscope approach can be used with archived FFPE tissues on glass slides and can be visualized either under a standard bright-field microscope (with chromogenic labels) or an epifluorescent microscope. A clear advantage of RNAscope is therefore its ability to visualize localized mRNA expression in samples. In the current study, GLUT1 protein expression was found to localize to glioblastoma cells and this was supported by our RNAscope analysis of *SLC2A1* expression ([Fig. 5](#nly124-F5){ref-type="fig"}). Highly stained samples in the GLUT1 high expression group had similarly high *SLC2A1* scores. Conversely, the GLUT1 low expression group had lower *SLC2A1* scores. Our study also revealed that GLUT1 expression was mainly localized to perivascular tumor cells. There was also a positive correlation between GLUT1 expression and neoangiogenesis highlighted by CD105 (endoglin) staining. CD105 is a 180 kDa integral membrane glycoprotein that is an accessory component of the TGF-β receptor complex. Applying analysis of CD105 expression to distinguish between normal vessels and malignant neovascularization has previously been reported, supporting our interpretation ([@nly124-B31]). The protein is predominantly expressed on cellular lineages within the vascular system, although it is more highly expressed by proliferating endothelial cells that participate in tumor angiogenesis. This pattern may emerge due to relatively lower expression in the vascular endothelium of normal tissues ([@nly124-B34], [@nly124-B35]). Markowski et al also reported that the expression levels of *HIF-1α*, *SLC2A1* (*GLUT1*), and *CD105* were higher in a group of patients with bladder cancer when compared with healthy subjects ([@nly124-B33]). The validation of the positive correlation of SLC2A1 and VEGF from TCGA database and GLUT1 expression around CD34-positive vessels suggested GLUT1 was related to tumor neoangiogenesis. Combined, these data indicate that there is a link between GLUT1 and endoglin expression, suggesting that the glycolytic system that emerges during glioblastoma development is accelerated by neoangiogenesis. In addition, our examination of the localization of GLUT1 within tumor tissues revealed an association with pseudopalisade formation and microvascular proliferation, both pathological features of glioblastomas. The generation of pseudopalisades (hypercellular zones surrounding necrotic tissue) often results from the migration of glioma cells outwards from hypoxic areas due to vascular occlusion ([@nly124-B36]). Epithelial-mesenchymal transition (EMT) is the process by which epithelial cells lose cell polarity and adhesiveness and is one of the main mechanisms that exacerbates tumor migration in glioblastomas ([@nly124-B37]). The zinc finger E-box-binding homeobox (ZEB) proteins, ZEB1, and smad1-interacting protein-1 (also known as ZEB2), are members of a family of noteworthy transcription factors that are responsible for the mediation of EMT in numerous types of cancer including glioma ([@nly124-B38]). Chem et al detected the expression of ZEB1 and ZEB2 in 91 cases of GBM with immunohistochemistry ([@nly124-B39]). The percentages of ZEB1 high expression and ZEB2 high expression were 31.9% (29/91) and 41.9% (36/91), respectively. They revealed that high expression of ZEB2 was significantly associated with lower survival rate of GBM patients (p = 0.001). Justin et al revealed hypoxia-induced mesenchymal shift in GBM primary material by showing colocalization of GLUT1, ZEB1 and the mesenchymal marker YKL40 in hypoxic regions of the tumor ([@nly124-B40]). These results suggested that GLUT1 may exist downstream of the ZEB1 pathway; further studies are needed to clarify the relationship GLUT1 and ZEB1. In previous reports, GLUT1 has been found to associate with lymph node metastasis and is also highly expressed by breast cancer cells ([@nly124-B10], [@nly124-B41]). These studies also reported that GLUT1 expression contributed to increased tumor viability by promoting glycolysis. The results showing that GLUT1 is expressed in both hypoxic and proliferative area suggested that glioblastoma cells increased their proliferative potency via hypoxic stress. These studies, in combination with our own analysis showing GLUT1 localization at tumor pseudopalisades, indicate a possible link to EMT. We therefore hypothesize that tumor malignancy in glioblastomas may be increased by GLUT1 through higher migration and invasiveness. Finally, Kaplan-Meier survival curve analysis revealed that survival in the high GLUT1 expression group was lower than the low expression group. Our multivariate analysis also showed that GLUT1 was an independent predictor for worse prognoses compared with already known prognostic factors, including age and KPS. These data suggest that GLUT1 may act as an independent prognostic factor for glioblastoma due to an accelerated glycolytic system, likely through increased neoangiogenesis and the Warburg effect. This would subsequently increase migration and invasiveness by promoting EMT. Previously, Phadngam et al reported that GLUT1 acts downstream of the PI3K-Akt pathway ([@nly124-B28]) and several studies have suggested a link between GLUT1, HIF1α, and GSK3β, although the exact details remain unclear. For example, Azzalin et al reported that various inhibitors of GLUT/SLC2A enhance the activity of the chemotherapeutic agents bis-chloroethylnitrosourea (BCNU) and temozolomide against high-grade gliomas ([@nly124-B42]). Chen et al also reported that specific blockade of GLUT1 using WZB117 resensitizes breast cancer cells to adriamycin ([@nly124-B43]). In the present study, GLUT1 inhibition by WZB-117 decreased cell viability in cell lines. There is a considerable potential for the therapeutic application of GLUT1 inhibitors against glioblastoma, but further experiments, for example, whether WZB-117 potentiate temozolomide and/or irradiation, are required. In conclusion, our study has revealed that GLUT1 is a biomarker and predictor for a worse prognostic outcome in patients with glioblastomas. This will assist the development and application of tools and treatment to better target vulnerable patients. However, recent advancements suggest that GLUT1 also has the potential to increase the efficacy of various anticancer agents. Therefore, our study reveals several important characteristics of GLUT1 expression in glioblastomas. Supplementary Material ====================== ###### Click here for additional data file. We would like to thank Editage ([www.editage.jp](http://www.editage.jp)) for English language editing. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. The authors have no duality or conflicts of interest to declare. [Supplementary Data](#sup1){ref-type="supplementary-material"} can be found at [academic.oup.com/jnen](http://academic.oup.com/jnen).

Introduction {#Sec1} ============ Individuals receiving a diagnosis of multiple sclerosis (MS) face many life-altering challenges. However, over the last decade, improved access to high-quality information and the ability to share experiences with other MS patients through the Internet and social media have helped patients become more knowledgeable about their condition and about MS research, including their diagnosis, treatment options, and clinical trials. This has allowed patients to manage their disease proactively and, in some cases, has afforded the opportunity to influence clinical research. As early as 2007, a survey of 2390 Americans with MS reported Internet use by 93% of participants (compared with 75% of the general population), of whom 53% indicated that it had helped them to become their own MS advocate \[[@CR1]\]. As well as being widespread among patients, the use of social media among Western healthcare professionals (HCPs) has dramatically increased in recent years, rising from 42% of practicing physicians in 2010 \[[@CR2]\] to almost 75% in 2013 \[[@CR3]\]. Among HCPs, 65% use social media for professional reasons, mostly in professional networks and clinical practice, but also for informing and helping patients \[[@CR4]--[@CR6]\]. These developments have empowered patients to act more in partnership with their HCPs in determining what care they receive, shifting away from a traditionally passive role to an active, self-governing one, with participatory decision-making playing a key part in determining treatment \[[@CR7]\]. In this article, we aim to explore the impact of virtually instant global access to information on the interaction between patients with MS, the HCPs who treat them, and the MS community. These interactions are examined in terms of participatory medicine, of the direction of MS research, of the sharing of experiences among patients, and of the privacy and ethical considerations when using social media and the Internet in healthcare communications. Methods {#Sec2} ======= This article offers qualitative case study-based insights from the patient and HCP authors, both of whom are from North America. As in our accompanying article, which examines the impact of social media and the Internet on aspects of patient education, the HCP's perspective in this article is provided by an MS specialist, Daniel Kantor, MD, FAAN, Past President of the Florida Society of Neurology, President of the Medical Partnership 4 MS, Chief Medical Correspondent for MSWorld, and an active member of the Multiple Sclerosis Foundation's Medical Advisory Board and the Multiple Sclerosis Association of America's Healthcare Advisory Council. The patient's perspective is provided by Jeri Burtchell, a clinical trials awareness activist, MS patient advocate, founder of Partners in Research, and Director of Patient Initiatives at HealthiVibe, LLC. These perspectives are drawn together, in the context of published research, by Jeremy Bright, a medical writer at Oxford PharmaGenesis with a special interest in MS. This article also includes some of Jeri's experiences since receiving an MS diagnosis in 1999, 3 months after the onset of symptoms. The patient's perspective was provided as follows: a Novartis employee drafted a set of questions for use in a subsequent interview with the patient author (Jeri Burtchell). During the interview, Jeri's responses to these questions were recorded. Unedited and non-paraphrased quotes were taken directly from the recordings and used as required throughout the manuscript. The HCP's perspective was provided directly by the HCP author, Daniel Kantor. To the best of our knowledge, this "patient--HCP dyad" approach has not been used before to explore patient and HCP perspectives regarding the impact of the Internet and social media on the interaction between members of the MS community. Adopting this novel approach thus enabled us to gain detailed and direct insights into specific issues not previously reported. The patient and HCP perspectives described in this article were provided by single individuals from the USA. As the insights yielded by this approach are specific to the authors' personal experiences, the generalizability of the findings may be limited as they may not be representative of those of the wider MS population and other HCPs. In order to minimize any bias, a separate literature search was conducted to identify relevant articles that contextualized the themes identified by both the patient and HCP authors. Supporting published references and online resources relating to Internet and social media use in MS were identified using MEDLINE and Google Scholar, with the search strings "multiple sclerosis \[Title/Abstract\] AND social media \[Title/abstract\]" and "multiple sclerosis \[Title/Abstract\] AND Internet \[Title/Abstract\]". Compliance with Ethics Guidelines {#Sec3} --------------------------------- This article does not contain any new studies with human or animal subjects performed by any of the authors. Overview of Key Findings {#Sec4} ======================== In exploring the impact that the Internet and social media have had on interactions between members of the MS community, Jeri and Dr. Kantor's perspectives could be grouped into four major themes: (1) participatory medicine and the changing face of patient--HCP interactions; (2) promoting and fostering MS research; (3) sharing experiences and interpersonal contact; and (4) privacy, confidentiality, and ethical considerations. In the following sections, we present Jeri and Dr. Kantor's views on each of these topics and contextualize them in terms of the available literature. Two further sections outlining the future potential of social media and concluding remarks from Jeri have also been included to provide additional information on this topic. Participatory Medicine: Changing Face of Patient--HCP Interaction {#Sec5} ================================================================= Participatory medicine has been defined as a model of cooperative health care that seeks to achieve active involvement by patients, HCPs, caregivers, and others across the continuum of care on all issues related to an individual's health \[[@CR7], [@CR8]\]. Participatory medicine is considered to be an ethical approach to care that may help to improve outcomes, reduce medical errors, increase patient satisfaction, and reduce the cost of care through partnership between patients and their HCPs \[[@CR7], [@CR8]\]. Patient's Perspective {#Sec6} --------------------- In the traditional model of biomedical care (one of medical paternalism), HCPs hold the position of authority while patients assume a passive role, following the instruction of their physicians with little input into the decision-making process. Jeri was diagnosed with MS in 1999, 3 months after the onset of symptoms. Below, Jeri recalls what little involvement she had regarding the treatment she would receive.""At that second visit to my neurologist's office, when he gave me the bad news, we discussed all the available treatment options. They were all injectable drugs, only one of which was not an interferon. This was the drug my doctor chose for me. We didn't choose together because, in my eyes, he was the expert and knew best."" Jeri would typically have three or four relapses annually for the first 7 years, and often had to use a walker or a wheelchair for months at a time. Despite asking if she could switch to a more effective treatment, her HCP said she should remain on her current therapy.""When I asked my doctor about switching \[treatment\], I didn't understand why he thought my drug was working when I was having bona fide relapses requiring steroid use several times per year."" ""At the time there were only four MS drugs and I was taking the only one that wasn't an interferon. He insisted I stay on the therapy I'd been on all along, reassuring me that my MS would be 'so much worse' if I weren't taking it."" ""I had no idea that I could challenge my doctor, or even that I could find a new one. Since he was the only one in my town, I felt I had no choice. I was despondent that I was relapsing so often and felt like things might never get any better."" Central to the success of participatory medicine is the dynamic between HCPs and patients. While physicians bring their training and clinical expertise, patients bring their life experiences, skills, and resilience, their deep investment in their long-term health, and a unique perspective on needs and priorities \[[@CR9]\]. Effective patient--HCP partnerships build trust, help patients feel informed, in control, and empowered, and promote shared decision-making in effective and time-efficient consultations \[[@CR10], [@CR11]\]. Seven years after her initial diagnosis, an opportunity arose for Jeri to engage with a different HCP.""In 2006 my neurologist had a stroke and I was left without a doctor. I would wind up at my PCP's \[primary care physician's\] office every time I had MS symptoms and he finally convinced me in 2007 that I needed to see a new neurologist in a nearby city."" Jeri discussed the advantages and disadvantages of different treatments with her new neurologist. The interaction followed a participatory model, and Jeri's interaction with the online MS community helped her to become more assertive in the management of her care. ""I didn't want to take one of the interferon medications (my only other option) due to the fear I'd developed based on the strong feelings my former neurologist had regarding them, and also from reading so much from patients online who shared their 'flu-like symptoms' and feeling miserable on the medication."" ""Before the Internet, when first diagnosed, I depended on my neurologist. His opinion was the final say about everything from use of steroids to physical therapy to which disease-modifying medicine I would use. The Internet empowered me as a patient to become informed about my condition, to consider my options and the opinions of others, and to take charge of managing my disease in the best possible way for me."" HCP's Perspective {#Sec7} ----------------- Dr. Kantor highlighted how clear, two-way communication is vital to foster successful partnerships between patients and HCPs, something that is also described in the literature \[[@CR10], [@CR11]\]. In a US survey of patients who presented for the first time at an MS clinic between 2003 and 2005, 82% of individuals performed web-based searches in preparation for their initial consultation. Hence, it was perhaps surprising that only 36% of those who had gathered information before their first appointment subsequently discussed it with their HCP \[[@CR12]\]. This is especially surprising given that patients within the 2011 North American Research Committee on Multiple Sclerosis (NARCOMS) registry indicated that their most trusted information source was an HCP, with 98% of patients reporting that they trusted a physician "some" or "a lot" \[[@CR13]\]. Dr. Kantor also noted how in recent years, the importance of shared decision-making in patient--HCP communication has gained increasing recognition, particularly for chronic diseases such as MS for which the treatments are not curative. Indeed, a German study found that between 2001 and 2007, 80% of patients with MS demanded autonomous roles in decisions relating to their treatment \[[@CR14]\]. Given that use of social media has almost certainly led to a patient population that is well informed about current treatment options, it would be reasonable to expect that this has resulted in patients challenging their HCP's prescribing recommendations more often and further influencing their practice patterns. Published evidence examining the effects that social media may have had on the shared decision-making process is scant, but this is an area where more research would be valuable. Although HCPs generally welcome a well-informed patient base, given that a better understanding of treatment may lead to greater adherence and patient engagement, Dr. Kantor acknowledged that some HCPs may feel defensive when patients challenge their medical knowledge or recommendations. He explained that although modern medicine is much less paternalistic than it was several decades ago, the expectation that HCPs' authority should be respected (if not revered) still lingers in some quarters. Attempts have been made in the literature to quantify the influence of social media on the patient--HCP relationship, albeit not specifically relating to MS. In a multidisciplinary survey of 232 Brazilian HCPs carried out in 2011--2012, 57% thought use of the Internet helped the patient--HCP relationship, but 28% thought it interfered with it, and 16% felt it had a negative impact \[[@CR15]\]. A small qualitative multidisciplinary survey of German HCPs, patients, and a patient organization representative published in 2012 largely echoed these findings. Most patients from Germany continued to regard their HCP as the expert, but being a better-informed patient made the relationship with their HCP more balanced, with patients becoming more critical and interrogating recommendations made by their HCPs \[[@CR16]\]. No German HCP reported feeling threatened or undermined by this shift in the relationship, but some noted that consultations could be prolonged when patients attended their consultation ready-armed with a self-diagnosis based on Internet research \[[@CR16]\]. However, findings from UK and US articles have shown that with HCPs becoming increasingly burdened by non-face-to-face commitments (such as inordinate time spent obtaining prior authorizations), a more engaged and self-educated (social media-educated) patient base has the potential to help improve the efficiency of the delivery of care \[[@CR4], [@CR17], [@CR18]\], perhaps by enabling patients to ask more informed questions about their disease, thus saving some of the time that HCPs need to commit to educating patients about their disease process and treatment options. Promoting and Fostering Multiple Sclerosis Research {#Sec8} =================================================== Patient's Perspective {#Sec9} --------------------- Before the availability of the Internet, and even until quite recently, patients would probably have learnt about a trial of a new drug from newsletters (as Jeri did in 2006--2007), or from their HCP.""I was still learning about MS from offline sources at this time and when I received a MOMENTUM magazine from the National MS Society talking about a new pill form of treatment still in clinical trials and recruiting, I decided to ask my new doctor about it. My new neurologist was Dr. Daniel Kantor who, at the time, was lead investigator for the TRANSFORMS trial at the University of Florida. At my first appointment we discussed the pros and cons of being in a study, and we weighed the options."" ""So I took the informed consent home to read, and share with friends and family to get opinions on whether or not I should join."" ""I also tried to find anything online from a patient perspective about participating in a clinical trial. All I could find was from the research side."" Jeri took the decision, which was at the time unusual, to blog her experience of participating in this trial.""My blog would come to be widely accepted as the first start-to-finish blog of a clinical trial from a patient perspective. A pharmaceutical executive once dubbed me the 'godmother of the intersection of social media and research'."" US-based social media and websites (e.g., PatientsLikeMe) are now widely used by patients with MS not only to find information on clinical trials, such as how to participate in them and study results, but also to communicate their on-trial experiences \[[@CR19]--[@CR22]\]. Furthermore, this avenue of communication affords patients the opportunity to provide input on trial design. Indeed, this is also something Jeri has been actively involved in.""Currently I work for HealthiVibe, LLC, a company focused on bringing the patient perspective to clinical trial design and patient-facing initiatives so our pharmaceutical clients can design studies and programs that are more meaningful and patient-friendly."" Concerns have been raised in the literature that large numbers of patients interacting via social media during a clinical trial could have the unwanted consequence of breaking the blinding of a study \[[@CR20]\]. Again, this was something that Jeri identified with and had also experienced directly.""The question of possible unblinding due to patients finding each other via social media has become a real issue. A few of us who remained in contact did reach out to each other once we'd been unblinded to see if the others knew. We were all correct in our suspicions. But we all agreed that regardless of having found each other, it's only natural to wonder which arm you are in and to guess---whether silently to yourself, or out loud to family, friends, or those on social media. Together with a handful of blog followers who had joined the clinical trial, I created a Yahoo! group to have a more private place for us to gather and commiserate. In order to gain access, members had to prove they were in the trial by sending me a photo of their medication bottle. While some people did want to talk about possible side effects and try to figure out if they were on the real drug or not, mostly we talked about normal things like family and jobs."" HCP's Perspective {#Sec10} ----------------- There are several published examples of the successful use of social media to recruit participants to MS studies, including a US trial examining the adverse effects of switching treatment and a large Australian study assessing factors that may contribute to fatigue \[[@CR20], [@CR21], [@CR23]\]. In 2012, the US Food and Drug Administration also approved a "crowdsourced" protocol for an MS trial that was developed with an online community of patients, HCPs, and researchers \[[@CR22]\]. Dr. Kantor also noted that in addition to pharmaceutical company and MS patient advocacy websites that act as gateways to clinical trials, patients with MS write blogs that offer advice on accessing clinical trials, and MS community websites promote participation in studies (e.g., <http://partnersinresearch.org>, <http://mymsteam.com>) and encourage participants to share their research experience (<http://projectdreamnow.org>, <http://patientslikeme.com>). Indeed, in 2009 an online community of US patients with MS was used to develop a self-report questionnaire in order to quantify adherence and to identify any barriers to achieving adherence that are specific to MS disease-modifying treatments \[[@CR24]\]. Social media could also be used by patients for real-time reporting of adverse events \[[@CR25]\]. Sharing Experiences and Interpersonal Contact {#Sec11} ============================================= Patient's Perspective {#Sec12} --------------------- Jeri was a frequent user of online forums, and as time progressed, she began to build up her own network of trusted members and online sources.""After posting on forums for a while, I began to privately message other members and online relationships---some of which continue to this day---began. It was in this way that we developed a network of trusted sources within the larger MS forums, and we would share links to information, or other forums."" It is evident that social media provides an invaluable channel of communication and interaction for people with disabilities \[[@CR26]\] and has the potential to relieve social isolation and improve quality of life by connecting patients to the wider world. For example, among individuals in the 2011 NARCOMS registry, 61% used the Internet for social networking \[[@CR13]\]. The Multiple Sclerosis International Foundation (MSIF) 2014 global survey examining technology use and MS also revealed that social media was used for peer support and for building an MS community, with more than one-third (38%) keeping in touch with other people with MS \[[@CR27]\]. Jeri found the "connecting" aspects of social media particularly valuable, and relationships within her network strengthened as a direct result of being able to connect with each other instantly.""During the time I was blogging, I had my posts set up to automatically show up on Facebook and Twitter. Although I had used Facebook to some degree, I had never used Twitter very much."" ""When I got a smartphone in 2011, however, I became even more involved in social media. The apps for Facebook, Twitter, and LinkedIn allowed me to connect to others even when I wasn't at my computer, which resulted in a much deeper, richer social media experience. I could share with others instantly instead of waiting to blog once I got home."" ""Because I had cultivated a group of friends whose opinions I trusted, the feedback was meaningful and almost always instant."" Examples of the kinds of social networks that have been developed for patients with MS include <http://mymsteam.com>. This US forum was set up specifically to allow patients to talk to each other about the day-to-day realities of living with MS, sharing practical tips and personal experiences, as well as providing support and advice to each other on issues that can only be answered by someone living with the condition \[[@CR28]\]. During a presentation at the 2015 conference of the European Committee for Research and Treatment in Multiple Sclerosis, it was noted how sharing experiences on social media helps patients feel less alone, giving them hope and providing an opportunity to talk about things they felt were too embarrassing or stressful to share with close friends and family (e.g., issues with incontinence) \[[@CR29]\]. The Internet also provides opportunities to explore patients' experiences of living with MS and of different treatment options, such as those in the MS in America study, as reported on the <http://MultipleSclerosis.net> website \[[@CR30]\]. Social media has also transformed how patients and HCPs communicate with each other, especially in regions that have highly developed healthcare systems, such as Canada, Europe, Israel, and the USA. Traditionally, HCPs and patients with MS in these regions have communicated and exchanged information via face-to-face contact during clinic visits as part of routine follow-up. However, as more and more patients use computers and mobile electronic devices to access health-related information, they are also becoming more interested in using digital technologies to complement face-to-face communications with their HCPs \[[@CR31], [@CR32]\]. HCP's Perspective {#Sec13} ----------------- Rather than being spectators to their patients' use of health-related information on the Internet and social media, Dr. Kantor outlined how HCPs have become involved in actively using these channels (a number of reasons that might motivate HCPs to have a social-media presence are summarized in Table [1](#Tab1){ref-type="table"}) \[[@CR33]\]. Studies have shown that social networks afford HCPs the opportunity to communicate with online patient communities, and share health messages that are likely to resonate with patients and be adopted by them \[[@CR6]\], while simultaneously responding to their requests for accessible, interactive, two-way communication \[[@CR34]\]. Evidence suggests that social media may also improve patient outcomes and reduce healthcare resource use \[[@CR35]\]. Indeed, data from a 2015 systematic review and meta-analysis found that social networking site interventions in Australia, the UK, and the USA had a statistically significant positive effect on the promotion of health-related behavioral change \[[@CR36]\].Table 1Reasons why physicians use social media \[[@CR33]\]To connectTo be challengedTo engageTo be supportedTo informTo leadTo reflectTo learnTo shareTo inspire Indeed, US HCPs may use social media, such as Twitter and Facebook, for the purposes of enhancing communication with their patients and the wider healthcare community, including providing patient education and disseminating public health information \[[@CR37]\]. Across regions, the most widely used channels tend to be online communities that allow for catching up on the latest news and developments, and networking and communicating with colleagues on patient issues \[[@CR5], [@CR38]\]. With this in mind, MS HCPs could set up a Facebook page or Google+ circle specifically to distribute disease-based information and guidance to patients, and to provide links to sites with other relevant information \[[@CR4], [@CR37]\]. Such platforms also provide an opportunity to post videos and newsletters and to conduct web-based seminars \[[@CR6]\]. Institutions can play a valuable role in using social media as a vehicle for patient education, but as described in the accompanying article on patient education, institutional use of social media is not always directed in this way nor is it always managed by individuals with knowledge of patient needs in MS. In terms of patient--HCP interaction, Dr. Kantor explained that there is a need to consider the differences between using social media to disseminate information to large groups of patients and using it to interact directly with an individual patient. Evidence suggests that the communication options offered by social media may help to promote efficient use of resources and staff time \[[@CR4]\]. However, despite a rapid increase in its use, many HCPs are reluctant to incorporate social media into routine clinical practice owing to uncertainty surrounding ethical and legal obligations, public access \[[@CR2], [@CR3]\], data security, and privacy regulations (e.g., the Health Insurance Portability and Accountability Act), managing the expectations of social media-savvy patients, and the lack of reimbursement for time spent. Indeed, results from a survey of Australian HCPs published in 2015 revealed that 66% of respondents were hesitant to engage fully in communications with patients via social media \[[@CR3]\]. Notably, 19--35% of US and Australian HCPs have received a "friend" request from a patient via a social networking site, although few respond \[[@CR2], [@CR3]\]. Few of the US HCPs reported sending a "friend" request to a patient or family member (1--5%) \[[@CR2]\], although in a separate US study, some HCPs reported making a conscious decision to "friend" or connect with patients on social networks in an attempt to encourage engagement and to appear approachable \[[@CR39]\]. Privacy, Confidentiality, and Ethical Considerations {#Sec14} ==================================================== Patient's Perspective {#Sec15} --------------------- Maintaining patient confidentiality and privacy is an important issue when considering social media in healthcare settings \[[@CR40]\], and this may influence how patients engage with and benefit from social media. This was a particular concern for Jeri.""With more and more platforms becoming available for patients to connect, I believe security and privacy are being better addressed."" ""The privacy of Facebook groups was hard to understand for some at first. When their posts would appear in their own Newsfeed outside of the group, people became alarmed and raised the question of privacy with the group admins \[administrators\]. I believe nowadays Facebook users have a better understanding of how privacy works within groups. That's not to say there aren't strong suspicions among patient members of these groups that there are 'plants' who are members of pharmaceutical companies trying either to learn about the purpose of any given group, or to correct misinformation about their product, or even to speak badly of a competitor's drug. The group admins try to stay on top of these types of accounts, researching other activity to gauge whether or not the account is that of a real patient."" HCP's Perspective {#Sec16} ----------------- Social media raises privacy issues for HCPs as well as for patients \[[@CR40]\]. As with all patient--HCP contact, patient confidentiality is paramount both ethically and legally (e.g., policies for securing the privacy of individually identifiable health information are enshrined in the US Health Insurance Portability and Accountability Act) \[[@CR41]\], and confidentiality breaches can arise if fragments of information can be assembled from different sources. It is important to remember that certain information can be accessed freely by everyone online; therefore, prudence dictates that the most secure privacy settings should be selected when communicating with patients and that these settings are updated regularly \[[@CR4], [@CR42], [@CR43]\]. Although social media provides a platform that could dramatically change the way that HCPs engage with patients, the boundary between HCPs' private and professional lives can become blurred when using social media \[[@CR44]\]. In the UK and USA, guidelines exist to help HCPs to optimize their use of social media, while ensuring that professional and legal obligations are met and that patients receive appropriate protection \[[@CR42], [@CR43], [@CR45]\]. Dr. Kantor noted that these guidelines highlight the need to maintain professionalism, patient confidentiality, and appropriate patient--HCP relationships, as well as to be aware of legal issues (e.g., defamation, as will be discussed) and conflicts of interest (Table [2](#Tab2){ref-type="table"}). He also pointed out that it is essential to apply the same medical values and principles during online interactions as when meeting a patient face-to-face. Although HCPs should be compassionate and engage with patients, it is imperative to maintain professional boundaries.Table 2Steps healthcare professionals can take to avoid some pitfalls of using social media \[[@CR42]--[@CR45]\]AreaSuggested actionsProfessionalismApply medical ethical values and principles at all times\ Protect your online image\ Be polite and respectful\ Be aware that social media can be monitored by others\ Avoid defamatory comments\ Disclose any conflicts of interestPatient confidentialityPatient confidentiality is an ethical and legal obligation\ Limit access to online content\ Use the most stringent privacy settings available\ Be careful giving individual patients medical advice onlinePatient--HCP relationshipMaintain professional boundaries\ Avoid engaging in non-professional relationships\ Be cautious of accepting "friend" requests*HCP* healthcare professional The UK and US guidelines recommend that HCPs should not accept "friend" requests; however, the advice is inconsistent \[[@CR42], [@CR43], [@CR45]\]. In addition, defamation law applies to interactions via social media, so derogatory or personal comments should be avoided \[[@CR42], [@CR43]\]. It is also important that HCPs are mindful of their online image and how it might reflect on them professionally: social media activity is often monitored by others, including the press, especially in public forums and on micro-blogging sites such as Twitter \[[@CR42], [@CR43]\]. Although social media provides the opportunity to share expertise and information, caution should be exercised when providing personal advice online \[[@CR43]\], and responding to questions sent via social media could expose HCPs to legal action \[[@CR6]\]. It is also important to consider that users with limited social media experience are more likely to make mistakes than those accustomed to using it. Such mistakes could have severe consequences if patient privacy is unwittingly breached \[[@CR46]\]. Future Potential of Social Media {#Sec17} ================================ Social media is helping patients with MS to become more informed about their diagnosis, and thus it is encouraging them to take greater responsibility for their own health care. These "e-patients" are individuals who are equipped, enabled, empowered, and engaged in their own health and healthcare decisions \[[@CR8]\]. They are changing the relationships among HCPs and patients and demanding a more equal partnership, with shared decision-making and responsibility. Social media will continue to drive the cooperative approach of participatory medicine, with patients actively participating alongside HCPs in all aspects of their own health care, with the ultimate aim of improving outcomes and patient satisfaction \[[@CR8]\]. In the future, interfacing with electronic medical records and monitoring disease management in real time will be made possible using mobile electronic devices and the Internet. The role of social media in these developments is unclear. Current guidelines recommend that social media is not used for providing medical advice to individual patients, so wholesale replacement of direct patient--HCP interaction seems unlikely. Final Thoughts: "How Social Media Has Changed Me As a Patient" {#Sec18} ============================================================== Access to the Internet and interaction via social media has broadened horizons for patients, empowering them not only in managing their own treatment but also in helping and supporting others, and in contributing more actively to medical research. This is a sentiment echoed by Jeri.""The Internet empowered me as a patient to become informed about my condition, to consider my options and the opinions of others, and to take charge of managing my disease in the best possible way for me. I realize everyone's experience with MS and medications is different, but the Internet has been the single most powerful influence in my arsenal of tools to combat this disease."" ""I use social media to meet others with MS, and to find ways to make a difference and have a positive impact. Staying connected in this way has given me hope for the future and has given my life purpose. It has made having MS be more of a conduit and less of a burden. Without the Internet and social media I am convinced I would be isolated, introverted, and miserable."" Conclusions {#Sec19} =========== As identified by Jeri and Dr. Kantor, and across different countries in the supporting literature, a wide range of benefits for MS participatory medicine can be gained from the targeted use of the Internet and social media. The rise of the e-patient---an informed and empowered individual who is actively involved in their own healthcare management---is challenging the traditionally passive role that patients used to play. This change in behavior has made the HCP--patient relationship more of an equal partnership, with decision-making and responsibilities now shared between both parties. Use of these platforms also has potential to change dramatically the way HCPs engage with their patients and peers. However, there are a number of important privacy, confidentiality, and ethical issues that must be considered, which may limit certain use of these platforms, especially by HCPs. Also, to maximize adoption by patients and HCPs, it will be important to ensure that these technologies are accessible, cheap, and quick and easy to use. **Enhanced content** To view enhanced content for this article go to <http://www.medengine.com/Redeem/86FCF0604435B798>. Editorial and medical writing support of the manuscript and the journal's article processing charges were funded by Novartis Pharmaceuticals Corporation. The authors acknowledge Ian Williams of Oxford PharmaGenesis Ltd (Oxford, UK), who provided editorial and medical writing support, and Kathleen Hawker, formerly of Novartis Pharmaceuticals Corporation (East Hanover, NJ, USA), who reviewed the article during the early stages of its development and conducted the interview with Jeri Burtchell. In addition, Novartis reviewed the final draft of the manuscript for scientific accuracy only. All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this manuscript, take complete responsibility for the integrity of the work as a whole, and have given final approval for the version to be published. Disclosures {#FPar1} =========== Daniel Kantor has received research support from Novartis and speaking/consulting honoraria, not related to this publication. Jeremy R. Bright is an employee of Oxford PharmaGenesis Ltd. Jeri Burtchell has acted as a consultant for Janssen, Lilly, Louisiana Public Health Institute, Novartis, Rutgers University, and Vanderbilt University, and has received fees for speaking from ERT, Janssen, Lilly, and Novartis. She is the founder of Partners in Research and Director of HealthiVibe, LLC. As an employee of HealthiVibe, LLC, she has worked on projects with many pharmaceutical companies, including Novartis. She also serves on the patient advisory boards of CureClick and MS SoftServe, Inc. Compliance with Ethics Guidelines {#FPar2} ================================= This article does not contain any new studies with human or animal subjects performed by any of the authors. Data Availability {#FPar3} ================= Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study. Open Access {#d29e887} =========== This article is distributed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (<http://creativecommons.org/licenses/by-nc/4.0/>), which permits any noncommercial use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

**What is already known** In contrast to many studies about the diagnostic accuracy of hospital physician-assessed cardiac symptoms for acute myocardial infarction (AMI), the diagnostic accuracy of emergency medical services (EMS) provider-assessed cardiac symptoms for AMI has not been evaluated in previous studies. **What is new in the current study** The sensitivity of EMS provider-assessed cardiac symptom for the final diagnosis of AMI was 73.3% (95% confidence interval \[CI\], 70.8 to 75.7), specificity was 85.3% (95% CI, 85.3 to 85.4), positive predictive value was 3.9% (95% CI, 3.6 to 4.2), and negative predictive value was 99.7% (95% CI, 99.7 to 99.8). We found that EMS provider-assessed cardiac symptoms had moderate sensitivity and high specificity for diagnosis of AMI. INTRODUCTION ============ The mortality rate of acute myocardial infarction (AMI) has declined substantially during the past 30 years \[[@b1-ceem-17-257],[@b2-ceem-17-257]\]. However, AMI is still a leading cause of death in many countries \[[@b3-ceem-17-257]\], and the importance of timely treatment is well established \[[@b4-ceem-17-257]\]. Many AMI patients do not receive timely and proper treatment, and prehospital delay is among the main causes of delayed treatment \[[@b5-ceem-17-257]-[@b7-ceem-17-257]\]. In prehospital areas, various methods, including assessment via 12-lead electrocardiograms (ECGs), activation of the cardiac catheter laboratory, and fibrinolysis to shorten the time to reperfusion, are used to manage AMI patients \[[@b8-ceem-17-257]-[@b13-ceem-17-257]\]. Although prehospital 12-lead ECG is the most studied tool and is a class I recommendation for the prehospital management of AMI \[[@b8-ceem-17-257],[@b14-ceem-17-257]\], whether a prehospital 12-lead ECG reading is taken or not is based on the symptoms of the patient and is yet to be determined. Therefore, emergency medical services (EMS) provider-assessed symptoms have a critical role in evaluating AMI patients in prehospital settings. When AMI is accurately assessed in the prehospital phase, it is possible to provide optimal prehospital management and rapid transport, select a proper receiving hospital, and ultimately reduce first medical contact to device time in AMI patients \[[@b14-ceem-17-257]\]. In contrast to many studies about the diagnostic accuracy of hospital physician-assessed cardiac symptoms for AMI \[[@b15-ceem-17-257]\], the diagnostic accuracy of EMS provider-assessed cardiac symptoms for AMI has not been evaluated in previous studies. We hypothesized that the diagnostic accuracy of EMS provider-assessed cardiac symptoms for AMI would be acceptable and associated with proper management and reduced process times in prehospital and hospital settings. This study aimed to describe the diagnostic accuracy of EMS provider-assessed cardiac symptoms for AMI and to identify the difference in prehospital and hospital processes according to the presence of EMS provider-assessed cardiac symptoms. METHODS ======= Study setting ------------- Korea established a single-tiered and fire-based public EMS system in 1995. The emergency hotline number in Korea is 119, and emergency medical technicians (EMTs) are dispatched during emergency calls. Detailed information on the education and training of Korean EMTs has been described previously \[[@b16-ceem-17-257]\]. EMTs can provide care comparable to that of intermediate EMT level in the US, including intravenous fluid infusion, endotracheal intubation or laryngeal mask airway insertion, and administration of certain medications, including nitroglycerin, under online medical direction \[[@b17-ceem-17-257]\]. An ambulance runsheet is filled out by EMTs for every ambulance dispatch, and all information is electronically recorded in the servers of the headquarters of the 16 provinces. Since a pilot trial of prehospital 12-lead ECG reading by the National Emergency Management Agency in 2009, EMS providers have acquired prehospital 12-lead ECG readings in some regions in Korea. However, it is not yet a widely accepted method for AMI management during emergencies. Study design and data collection -------------------------------- This study was a retrospective observational study using an EMS database, administrative databases of participating emergency departments (EDs), and hospital medical records. The records of 4 large tertiary academic EDs located in urban areas with 40,000 to 80,000 annual patients were reviewed. We acquired the EMS database from the National Emergency Management Agency for this study. Using the EMS database from January 1, 2008 to December 31, 2012, we linked the participating hospitals' administrative data to the EMS database and assessed information on age, sex, visiting date, and visiting time. The hospital medical records of the sampled patients were reviewed by trained researchers. Available ED records, hospital admission records, nursing charts, and coronary angiography reports were reviewed. Study population ---------------- We included patients who visited any of the 4 participating hospitals' EDs by EMS from January 1, 2008 to December 31, 2012. Patients whose EMS data were not linked to the administrative data were excluded. Using the disproportionate stratified sampling method, we planned to enroll 800 patients for our analysis. Sampling and weighting ---------------------- Because patients who had EMS-assessed cardiac symptoms and patients whose final diagnosis was AMI were rare in the EMS database, a disproportionate stratified sample design was used to gain an adequate number of those patients. EMS-assessed cardiac symptoms and a discharge diagnosis of AMI in the administrative database were used for stratification. To gain an evenly distributed sample, the hospital and year of visit were also used for stratification. We planned to sample 10 patients in each stratum for a total of 800 patients. The sampled data were weighted to the probability of selection. Variables and measurement ------------------------- Among the 30 predefined categories of symptoms in the EMS database, we defined EMS-assessed cardiac symptoms as chest pain, dyspnea, palpitation, and syncope. Meanwhile, an ED discharge diagnosis of AMI was defined as an International Classification of Diseases 10th revision (ICD-10) code of I21.0 to I21.9 in the administrative ED database. From the EMS database, we collected data on the patient's age, sex, symptom to call time, response time, scene time, transport time, hospital arriving time, presenting symptoms, prehospital documented shock (at least 1 event of systolic blood pressure less than 90 mmHg as measured by the EMS), and prehospital management, including oxygen therapy, ECG monitoring, intravenous fluid infusion, nitroglycerin administration, and cardiopulmonary resuscitation (CPR). From the participating hospital's administrative database, we collected data regarding patients' age, sex, hospital arriving time, ICD-10 ED diagnosis, and ED disposition status. By reviewing hospital medical records, we collected data on the final diagnosis of AMI or not. For AMI patients, we also collected data on the presence of ST segment elevation, cardiogenic shock before reperfusion therapy, performance of CPR before reperfusion therapy, and the type of reperfusion therapy (thrombolysis, percutaneous coronary intervention, or coronary artery bypass grafting). Data on door-to-needle or balloon times were also collected in patients with ST elevation myocardial infarction (STEMI). Statistical analysis -------------------- Descriptive statistics were used to assess the similarity between the study population and the study sample. The characteristics of patients whose final diagnosis was AMI and STEMI were compared according to EMS provider-assessed cardiac symptoms. Values in the study sample were weighted via the reciprocal of the probability of selection in all analyses. Because we planned to compare weighted values according to EMS provider-assessed cardiac symptoms, design-based statistical tests, including Student's t-test, Wilcoxon rank-sum test, chi-square test, and Fisher exact test, were used as appropriate using a survey package for R software. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) with 95% confidence interval (CI) of each of the EMS-assessed cardiac symptoms as well as all of the EMS-assessed cardiac symptoms combined were calculated for the final diagnosis of AMI. Verification bias-adjusted estimates were calculated by using stratified sampling to weigh results from patients in the EMS database using CompareTests for R software \[[@b18-ceem-17-257]-[@b20-ceem-17-257]\]. Ethics statement ---------------- The study protocol was approved by the Institutional Review Board of the Seoul National University Hospital (1509-120-705). Informed consent was waived by the Institutional Review Board. RESULTS ======= Characteristics of the study participants ----------------------------------------- From January 2008 to December 2012, 121,394 patients who visited the participating hospitals via EMS were identified from the EMS database. Among them, data of 92,353 (76.1%) patients were linked to a participating hospital's administrative ED database. Although stratified sampling of 800 patients was planned, only 775 patients were sampled because the total number of patients in some strata did not exceed the planned number. The medical records of the sampled 775 patients were reviewed. Among them, 276 patients had a final diagnosis of AMI. Notably, no confirmed AMI was observed in patients who did not have a diagnosis of AMI in the administrative ED database ([Fig. 1](#f1-ceem-17-257){ref-type="fig"}). [Table 1](#t1-ceem-17-257){ref-type="table"} shows the baseline characteristics of the study population and sample. Among all patients, 15.1% had any of the EMS provider-assessed cardiac symptoms, and 4.8% had chest pain. The mean age was 52.2 years in the total population, 58.5 years in the unweighted sample, and 49.2 years in the weighted sample. The proportion of men in the total population, unweighted sample, and weighted sample was 52.5%, 59.6%, and 51.5%, respectively. Although baseline characteristics were different between all patients and sampled patients, the weighted values were similar between them. Main outcome ------------ [Table 2](#t2-ceem-17-257){ref-type="table"} shows the sensitivity, specificity, and predictive values of EMS provider-assessed cardiac symptoms for the diagnosis of AMI. The sensitivity and specificity were 73.3% (95% CI, 70.8 to 75.7) and 85.3% (95% CI, 85.3 to 85.4), respectively, for any of the EMS-assessed cardiac symptoms, and 65.1% (95% CI, 62.7 to 67.4) and 95.9% (95% CI, 95.9 to 95.9), respectively, for chest pain. Chest pain was revealed as the most sensitive symptom for diagnosis of AMI, and PPV was also the highest for chest pain (PPV 11.5%; 95% CI, 10.9 to 12.3) ([Table 2](#t2-ceem-17-257){ref-type="table"}). [Table 3](#t3-ceem-17-257){ref-type="table"} shows the prehospital, ED, and hospital characteristics of AMI patients in a weighted sample of EMS-assessed cardiac symptoms. AMI was confirmed in 748 patients (0.8% of total patients). Among all confirmed AMI patients, those whose cardiac symptoms were assessed by EMS had shorter scene time and longer transport time and received more oxygen therapy, ECG monitoring, and nitroglycerine administration than patients who had not. The probability of providing CPR in the prehospital or hospital phase was also lower in patients who had EMS-provider assessed cardiac symptoms than in patients who had not. Although prehospital documented shock was more common in patients who had EMS provider-assessed cardiac symptoms than in patients who had not (10.3% vs. 8.2%), shock in the hospital phase was less common in patients who had EMS provider-assessed cardiac symptoms than patients who had not (2.8% vs. 10.3%). Among STEMI patients, the door-to-balloon time in percutaneous coronary intervention patients was significantly shorter in those who had EMS-assessed cardiac symptoms than in patients who had not (median \[interquartile range\], 72.4 \[27.5 to 87.7\] vs. 107.1 \[68.5 to 205.3\] minutes) ([Table 3](#t3-ceem-17-257){ref-type="table"}). DISCUSSION ========== Because the timeliness of treatment is critical in AMI patients \[[@b4-ceem-17-257]\], accurate symptom assessment in the prehospital phase is important. The effectiveness and efficiency of various aspects of prehospital management, including 12-lead ECG readings, oxygen, drugs and fluids, selection of hospital, and prehospital notification, would be improved if symptoms were assessed accurately in the prehospital setting. In this study, we evaluated the diagnostic accuracy of EMS provider-assessed cardiac symptoms for AMI using an EMS database, administrative ED databases, and hospital records. Regardless of individual symptoms, the specificity of EMS-assessed cardiac symptoms was significantly higher than the sensitivity of EMS-assessed cardiac symptoms. This means that these symptoms are more useful at excluding rather than including AMI. For example, if we use chest pain for identifying AMI, a negative result is reliable at reassuring that a patient does not have AMI (NPV 99.7%), and chest pain as an indicator of AMI correctly identifies 96% of patients who do not have AMI (specificity 95.9%). However, the presence of chest pain is poor at identifying AMI (PPV 11.5%), and further investigation is needed to diagnose AMI. We found that EMS provider-assessed cardiac symptoms had very high NPV. However, because only 0.8% of patients had confirmed AMI, the high NPV of EMS provider-assessed cardiac symptoms was primarily based on the low prevalence of confirmed AMI patients in the EMS database. We also found that the sensitivities of dyspnea, palpitation, and syncope were low. Therefore, the diagnostic value of these symptoms is minimal for identifying AMI. We found that AMI patients who had EMS provider-assessed cardiac symptoms tend to receive more prehospital management and have low prevalence of shock in the ED phase. Active management in the prehospital setting can be associated with early stabilization of AMI patients \[[@b21-ceem-17-257]\]. STEMI patients who had EMS provider-assessed cardiac symptoms had higher prevalence of receiving reperfusion therapy and faster reperfusion time intervals than patients who did not. EMS has been reported to be associated with a wide use of acute reperfusion therapies and fast time intervals in AMI patients \[[@b6-ceem-17-257]\]. Even in EMS users, we found that patients who had EMS provider-assessed cardiac symptoms had a higher proportion of those characteristics than patients who did not. The mechanism of efficient hospital treatment in patients who had EMS-assessed cardiac symptoms was not directly evaluated in our study. Because patients who had EMS-assessed cardiac symptoms had classic symptoms for AMI, the ease of diagnosis can be associated with efficient treatment in the hospital phase. Other EMS provider activities, including prehospital notification and alerts to ED staff when EMS providers hand over patients, can also be associated with those characteristics. Because prehospital 12-lead ECG was not widely used in our setting, the effect of prehospital 12-lead ECG would be minimal in our study. To our knowledge, this is the first study to evaluate EMS provider-assessed cardiac symptoms in view of the final diagnosis of AMI. Previously, the concordance of symptoms of myocardial infarction between paramedic and hospital records was evaluated in one study, and they found that the concordance of paramedic prehospital patient care documentation of symptoms was excellent for almost all symptoms, including chest pain \[[@b22-ceem-17-257]\]. Those findings could suggest that EMS provider-assessed cardiac symptoms could have similar diagnostic accuracy to hospital physician-assessed cardiac symptoms. However, the actual diagnostic accuracy of EMS provider-assessed cardiac symptoms was not evaluated. In one study, the diagnostic accuracy of chest pain assessed by an EMS call center was evaluated \[[@b23-ceem-17-257]\]. In that study, the overall rate of AMI was 12% in patients who called EMS for chest discomfort, and that level was similar to the PPV of chest pain in our study. Compared to previous studies, our study has several strengths. First, rather than include only patients who had chest pain or AMI diagnosis, we included and sampled all patients responded by EMS. Therefore, we could estimate all aspects of diagnostic accuracy, including sensitivity, specificity, and predictive values. If we only included patients who had chest pain and AMI diagnosis, the specificity could not be calculated. Second, because we used disproportionate stratified sampling, we captured a substantial proportion of AMI patients in our sample. Because less than 1% of the total patients had AMI, a sample of more than 27,300 patients would be needed to capture 273 confirmed AMI patients, which is the same number of confirmed AMI patients in our study. Although we captured a substantial proportion of AMI patients using disproportionate stratified sampling, we found that baseline characteristics were similar between all patients and weighted values of sample patients ([Table 1](#t1-ceem-17-257){ref-type="table"}). Third, because other diagnostic tests, including prehospital 12-lead ECG, are seldom used in Korea, the diagnostic accuracy of EMS provider-assessed cardiac symptoms was minimally influenced by other tests. EMS provider-assessed symptoms are critical for evaluation and management of AMI patients. The diagnostic accuracy of EMS provider-assessed cardiac symptoms can be used as baseline data for policymakers and quality indicators in prehospital AMI management and AMI education programs for EMS providers. Studies about the association between the diagnostic accuracy of EMS provider-assessed cardiac symptoms, outcomes of AMI patients, and the additional diagnostic value of other tests, including prehospital 12-lead ECG, for the evaluation of AMI in patients with EMS provider-assessed cardiac symptoms would be helpful to evaluate the importance of the EMS system and each EMS intervention in the management of AMI more objectively. Our study has several limitations. First, it was performed at tertiary hospitals in Korea, and the results may not be generalized to other EMS systems. However, the consistency of findings across five years and four hospitals supports our findings, which may be applicable to different ED settings in these regions. Second, the gold standard of AMI was based on hospital medical review in this study, and there may be some errors. However, many studies on AMI use hospital medical review as the gold standard \[[@b24-ceem-17-257],[@b25-ceem-17-257]\], and because this study was retrospective, there are limited ways to identify real AMI. Third, the effect of EMS provider-assessed cardiac symptoms on the outcomes of AMI patients can be caused by the selection of the designated hospital, but those effects could not be evaluated in our study. Fourth, approximately 24% of EMS databases were not linked to ED databases, and this could affect our findings. Moreover, because one-to-one matching was planned, duplicated matching pairs and ambiguous matching pairs were excluded in our study. Fifth, the diagnostic accuracy of the combinations of symptoms was not evaluated in this study. In addition, the characteristics of those symptoms, including severity of pain, were not also evaluated in this study. In conclusion, each of the EMS provider-assessed cardiac symptoms showed moderate sensitivity, high specificity, low PPV, and very high NPV. We also found that among AMI patients, those who had EMS-assessed cardiac symptoms received more aggressive prehospital management and more efficient in-hospital treatment for AMI. Studies about the association of diagnostic accuracy of EMS provider-assessed cardiac symptoms and outcomes and additional diagnostic value of other tests, including prehospital 12-lead ECG, for the evaluation of AMI in patients with EMS provider-assessed cardiac symptoms are needed. No potential conflict of interest relevant to this article was reported. This study was financially supported by the Woochon Cardio-Neuro-Vascular Research Foundation (2014). {#f1-ceem-17-257} ###### Prehospital and ED characteristics of the study population and sample Variable Total Sample (unweighted) Sample (weighted^[a)](#tfn1-ceem-17-257){ref-type="table-fn"}^) ---------------------------------------------------------------------------- ------------------- --------------------- ----------------------------------------------------------------- Age (yr) 52.2 ± 22.3 58.5 ± 19.4 49.2 ± 22.5 Sex, male 48,526 (52.5) 462 (59.6) 47,577 (51.5) EMS provider-assessed cardiac symptoms  Chest pain 4,413 (4.8) 200 (25.8) 3,470 (3.8)  Dyspnea 7,764 (8.4) 185 (23.9) 8,494 (9.2)  Palpitations 461 (0.5) 9 (1.2) 391 (0.4)  Syncope 2,175 (2.4) 45 (5.8) 2,443 (2.6)  Any cardiac symptom 13,971 (15.1) 375 (51.6) 13,971 (15.1) Prehospital documented shock^[b)](#tfn2-ceem-17-257){ref-type="table-fn"}^ 5,319 (5.8) 76 (9.8) 8,082 (8.8) Prehospital time interval (min)  Response time 7.0 (5.0--9.0) 7.0 (5.0--9.0) 7.0 (5.0--9.0)  Scene time 7.0 (4.0--10.0) 6 (4--10) 7.0 (4.0--10.0)  Transport time 11.0 (7.0--20.0) 11.0 (7.0--19.0) 10.0 (7.0--18.0)  Total prehospital time 28.0 (21.0--38.0) 26.0 (20.0--36.0) 26.0 (20.0--36.0) Prehospital management  Oxygen therapy 22,855 (24.7) 380 (49.0) 25,800 (27.9)  ECG monitoring 16,016 (17.3) 266 (34.3) 13,262 (14.4)  IV infusion 1,247 (1.4) 14 (1.8) 706 (0.8)  Nitroglycerin 295 (0.3) 16 (2.1) 216 (0.2)  Cardiopulmonary resuscitation 1,361 (1.5) 28 (3.6) 817 (0.9) ED diagnosis of AMI 1,001 (1.1) 375 (48.4) 1,001 (1.1) ED disposition  Discharged 58,111 (62.9) 235 (30.3) 55,852 (60.5)  Admitted 25,777 (27.9) 62.6 (18) 27,109 (29.4)  Transferred 3,504 (3.8) 28 (3.6) 5,825 (6.3)  ED death 1,811 (2.0) 18 (2.3) 1,396 (1.5)  Others 1,693 (1.8) 9 (1.2) 2,172 (2.4) Values are presented as mean±standard deviation, number (%), or median (interquartile range). ED, emergency department; EMS, emergency medical services; ECG, electrocardiogram; IV, intravenous; AMI, acute myocardial infarction. Values are weighted in sample. Patients whose systolic blood pressure was less than 90 mmHg in the EMS database at least once. ###### Diagnostic performance of EMS-assessed cardiac symptoms for acute myocardial infarction EMS-assessed symptom Sensitivity Specificity PPV NPV ---------------------- ------------------ ------------------ ------------------ ------------------ Chest pain 65.1 (62.7−67.4) 95.9 (95.9−95.9) 11.5 (10.9−12.3) 99.7 (99.7−99.7) Dyspnea 12.9 (11.3−14.6) 91.6 (91.6−91.6) 1.2 (1.1−1.4) 99.2 (99.2−99.3) Palpitations 0.3 (0.3−0.3) 99.9 (99.9−99.9) 1.9 (1.9−1.9) 99.2 (99.2−99.3) Syncope 1.0 (0.7−1.6) 97.9 (97.9−97.9) 0.4 (0.2−0.6) 99.2 (99.2−99.3) Any cardiac symptom 73.3 (70.8−75.7) 85.3 (85.3−85.4) 3.9 (3.6−4.2) 99.7 (99.7−99.8) Values are presented as % (95% confidence interval). EMS, emergency medical services; PPV, positive predictive value; NPV, negative predictive value. ###### Prehospital, ED, and hospital characteristics of confirmed AMI patients according to EMS provider-assessed cardiac symptom in the study sample^[a)](#tfn3-ceem-17-257){ref-type="table-fn"}^ Variable Confirmed AMI STEMI -------------------------------------------------------------------------------- ------------------ ------------------- --------- -------------------- ------------------ --------- Age (yr) 66.9 ± 13.2 64.6 ± 11.5 0.16 Sex, male 134 (68.4) 362 (67.9) 0.95 80 (76.2) 225 (69.9) 0.43 Symptom to call time (min) 47 (4.0−217.0) 49.0 (12.0−142.5) 0.99 25.8 (1−144) 35.0 (9.5−93.0) 0.96 Response time (min) 7.0 (5.0−9.0) 7.0 (5.0−9.0) 0.56 7.0 (5.0−8.0) 7.0 (6.0−9.0) 0.11 Scene time (min) 6.0 (4.0−9.0) 5.0 (3.0−7.8) 0.03 6.0 (3.8−8.0) 5.0 (3.0−7.0) 0.16 Transport time (min) 10.0 (6.0−17.0) 13.0 (8.0−21.4) 0.01 9.0 (6.0−14.8) 11.8 (8.0−18.0) 0.03 Total prehospital time (min) 24.0 (20.0−33.0) 25.0 (20.0−41.0) 0.22 23.1 (20−29.3) 24.1 (20.0−34.2) 0.18 Prehospital documented shock^[c)](#tfn5-ceem-17-257){ref-type="table-fn"}^ 16 (8.2) 55 (10.3) 0.56 11 (10.5) 38 (11.8) 0.75 EMS treatment  Oxygen therapy 85 (43.4) 353 (66.1) \< 0.01 44 (42.3) 214 (66.5) \< 0.01  ECG monitoring 62 (31.6) 319 (59.8) \< 0.01 33 (31.4) 193 (59.9) \< 0.01  IV line insertion 3 (1.5) 20 (3.8) 0.22 2 (1.9) 10 (3.1) 0.50  Nitroglycerin 0 (0) 50 (9.4) 0.02 0 (0) 29 (9) 0.11  Cardiopulmonary resuscitation 20 (10.3) 8 (1.5) \< 0.01 12 (11.4) 6 (1.9) 0.05 ED status 0.37 0.85  Discharged 0 (0) 0 (0) 0 (0) 0 (0)  Admitted 188 (95.9) 520 (96.6) 102 (97.1) 315 (97.8)  Transferred 0 (0) 7 (1.3) \- 1 (0.3)  Others 8 (4.1) 6 (1.1) 3 (2.9) 6 (1.9) ED death 8 (4.1) 6 (1.1) 3 (2.9) 6 (1.9) Shock before reperfusion therapy^[d)](#tfn6-ceem-17-257){ref-type="table-fn"}^ 20 (10.3) 15 (2.8) 0.02 11 (10.5) 5 (1.6) \< 0.01 CPR before reperfusion therapy 34 (17.3) 15 (2.8) \< 0.01 21 (20.2) 6 (1.9) \< 0.01 Reperfusion therapy  Thrombolysis 7 (3.6) 33 (6.2) 0.34 7 (6.7) 32 (9.9) 0.53  PCI 127 (65.1) 440 (82.6) \< 0.01 84 (80.0) 302 (93.8) 0.02  CABG 12 (6.15) 20 (3.7) 0.42 3 (2.9) 0 (0) 0.08 DtoN time (min)^[a)](#tfn3-ceem-17-257){ref-type="table-fn"}^ NA NA 20.0 (20.0−28.3) 25.3 (16.4−28.0) 0.96 DtoB time (min)^[a)](#tfn3-ceem-17-257){ref-type="table-fn"}^ NA NA 107.1 (68.5−205.3) 72.4 (27.5−87.7) \< 0.01 Values are presented as mean±standard deviation, number (%), or median (interquartile range). ED, emergency department; AMI, acute myocardial infarction; EMS, emergency medical services; STEMI, ST elevation myocardial infarction; EMS Sx-, patients who did not have EMS provider-assessed cardiac symptoms; EMS Sx+, patients who had EMS provider-assessed cardiac symptoms; ECG, electrocardiogram; IV, intravenous; CPR, cardiopulmonary resuscitation; PCI, percutaneous coronary intervention; CABG, coronary artery bypass graft; DtoN time, door-to-needle time in thrombolysis patients; NA, non applicable; DtoB time, door-to-balloon time in primary PCI patients. All values are weighted in sample. Because all confirmed AMI patient's ED diagnosis was AMI, only 2 strata were shown. P-value was calculated via design-based statistical tests, including Student's t-test, Wilcoxon rank-sum test, chi-square test, and Fisher exact test, as appropriate. Prehospital documented shock. Shock before reperfusion therapy.

Introduction ============ ICUs are complex settings, with critically ill patients submitted to invasive care, involving a multidisciplinary team, requiring urgent high-risk decision-making, taking place in an expensive structure with new technologies of increasing complexity. All these conditions facilitate the development of adverse events (AEs). We aimed to determine the occurrence of AEs in four tertiary academic ICUs in Brazil, disclosing their potential risk factors. Methods ======= This prospective cohort was conducted in four medical ICUs of a major academic, tertiary hospital in Brazil, enrolling all adult admissions during June to August 2009. AEs were identified by direct daily monitoring of medical and nursing rounds and chart review. Age, sex, APACHE II scores, length of stay (LOS), and the Nursing Activities Score (NAS) were also registered. The association with the occurrence of AEs was analyzed with logistic regression. Results ======= A total of 180 ICU admissions were included, regarding 176 patients (male/female: 86/90; age: 52.7 ± 1.8 years). The mean LOS, APACHE II scores and NAS were 10.0 ± 0.8 days, 15.7 ± 0.5 points and 69.0 ± 1.5%. Nearly 78% of the admissions (141 admissions) suffered 1,065 AEs. The most frequent AEs were: new dermatitis/pressure ulcers (195 events = 18.3% of events); hypoglycemic episodes not related to insulin use (HENI) (168 events = 15.8%); diagnostic/treatment failures (156 events = 14.6%); and drug AEs (195 events = 12.8%). Those four categories responded for 61.5% of all detected AEs. In the final logistic regression model, three independent variables remained as important risk factors for the occurrence of at least one AE: LOS \>3 days, APACHE scores \>13 points and NAS \>70%, with adjusted OR estimates of 19.5, 3.4 and 3.3, respectively (*P*\< 0.02). Conclusions =========== This prospective study was essential to identify the proportion of our ICU admissions affected by AEs, disclosing their nature. Our AE rates, affecting nearly 78% of admissions, were higher than those previously described. The direct observation of the ICUs contributed to those rates. Six out of 10 AEs corresponded to new cutaneous lesions, HENI, diagnostic/treatment failures and drug AEs. Length of stay, severity on admission and nursing workload were important risk factors for the occurrence of at least one AE. Acknowledgements ================ This study was sponsored by FAPESP.

Background {#Sec1} ========== Water participates in a variety of physiological functions, and is essential for body development and survival \[[@CR1]--[@CR3]\]. Three sources of water input are available: drinking water, water from food and endogenous water. Four ways of water-output includes urine through urinary system, sweat through skin surface, breath through respiratory system, and feces through digestive system. Under normal conditions, water maintains a state of dynamic balance in body, that is, the amount of water-input is approximately equal to the amount of water-output \[[@CR2], [@CR3]\]. However, too much or insufficient water intake disturbs the dynamic water balance, changes the hydration state, and affects body health negatively. When water intake exceeds the regulatory capacity of kidney, it may cause acute water intoxication and hyponatremia. Insufficient water intake may induce a dehydrated state, which reduces cognition ability \[[@CR4]--[@CR6]\] and physical activity ability \[[@CR7]--[@CR10]\] and increases the risk of urinary system diseases (such as kidney stones, urinary tract infections and chronic kidney disease) \[[@CR11], [@CR12]\] and cardiovascular disease \[[@CR13]\]. Therefore, maintaining an optimal hydration status is vital for human health. During pregnancy, physiological changes cause daily water requirements to increase compared with people in normal physiological stages. Blood volume of pregnant women gradually increases from 6 to 8 weeks of pregnancy and reaches a peak at 32 to 34 weeks' gestation. Water is the main component of human tissue, and 83% of blood is composed of water. Many changes occur in the urinary system: the kidneys become slightly larger; renal plasma flow and glomerular filtration rate increase in early pregnancy and remain at the high level throughout the whole pregnancy; and urine volume increases when being in the supine position and during the night \[[@CR14]\]. Urination is the main mode of water output \[[@CR15]\]. In the respiratory system, the ventilation increases by 40% /minute and the tidal volume increases by 39%, thus increasing the amount of water output through expiration. Under normal circumstances, approximately 500 mL of water per day is lost through sweating \[[@CR15]\]. During pregnancy, water loss through sweating increases due to hyperactive adrenal and thyroid functions, an accelerated metabolism and increased cutaneous circulation. Nutrients and energy requirements for pregnant women are also increase, so food intake is increased. Water is the carrier of food metabolism, digestion, absorption, circulation and excretion. The water requirement is 1 mL for every 4.184 kJ of energy consumption, as much, more energy intake requires more water intake correspondingly \[[@CR15], [@CR16]\]. Surveys on fluid intake have shown that pregnant women have insufficient water intake. According to the data of the 1977--1978 Food Consumption Survey conducted by the US Department of Agriculture National, it was found that the average total water intake among 188 pregnant women was 2100 L/d \[[@CR17]\]. In a study in New Zealand in 2014, the average daily water intake among 504 pregnant women was 2200 mL/d \[[@CR18]\]. In a 2016 survey of 20 pregnant women, the average total water intake was almost 2259 mL/d \[[@CR19]\]. In a 2016 study in Indonesia, the average daily water intake among 300 pregnant women was 2332 mL/d, although in approximately 42% of participants, the water intake was less than the recommended level (2048 mL/d) \[[@CR20]\]. In China, only a few surveys on fluid intake among children and adults have been conducted. These studies have shown that only one third of the participants had adequate water intake and that almost one quarter of participants were dehydrated \[[@CR21], [@CR22]\]. However, no surveys has been conducted for water intake among pregnant women in China. Changes occurring in the endocrine system during pregnancy may affect hydration. Prolactin begins to increase gradually at 7 weeks of pregnancy to promote mammary gland development for postpartum lactation. Aldosterone secreted by the outer globular band during pregnancy has a four-fold times. The secretion of estradiol at the end of pregnancy is 100 times greater than that in non-pregnant women \[[@CR16]\]. Water balance is affected by various hormones. Insufficient water intake increases the secretion of vasopressin and aldosterone, which affects the water permeability of the distal renal tubules and collecting ducts. As a result, water reabsorption increases and water output decreases \[[@CR23]\]. Copeptin, a stable peptide derived from the vasopressin precursor, is a potential predictor in various chronic diseases, such as diabetes insipidus and cardiovascular diseases \[[@CR24]\]. Endocrine indexes related to hydration and pregnancy interact; thus, hydration state may be associated with pregnancy complications and maternal and infant outcomes. Few studies have explored the associations of fluid intake and hydration state with pregnancy complications and maternal/infant outcomes. A prospective study in Italy involving 173 pregnant women revealed that total body water and intracellular and extracellular fluid in healthy pregnant women increased gradually throughout pregnancy, but an opposite trend was observed among women with gestational hypertension \[[@CR25]\]. In a study conducted in Canada, the body compositions of 196 women 4 and 12 h postpartum were measured: the total body water of women without arterial hypertension was 44 L, whereas that of women with arterial hypertension was only 18 L \[[@CR26]\]. This suggested that associations might exist between water retention and the development of hypertension. A US study showed that total body water may be a strong predictor of pre-eclampsia \[[@CR27]\]. A Mexican study suggested that total body water in pregnant women with gestational hypertension and severed eclampsia was relatively higher than that in healthy pregnant women. In addition, chronic hypovolemia induced by insufficient water intake may be among the main risk factors in the development of diabetes \[[@CR28]\], and higher intake of boiled water may reduce the risk of diabetes \[[@CR29]\]. Data from the US Health and Nutrition Survey (2009--2012) showed that people without optimal hydration have a greater risk of obesity than those with optimal hydration \[[@CR30]\], suggesting that hydration influences total gestational weight gain and postpartum weight retention. One study showed that the secretion of breast milk in lactating women increased with the amount of water intake \[[@CR31]\]. A case--control study in the United States suggested that insufficient fluid intake was a risk factor for abortion requirements and preterm births, as well as low birth weight \[[@CR32]\]. Other studies have shown that total body water has significant effects on birth weight \[[@CR26], [@CR33]--[@CR36]\]. A US case-control study suggested that insufficient fluid intake was an independent risk factor of low birth weight \[[@CR32]\]. In a randomized controlled trial in California, among 40 women with normal amniotic fluid index at 28 weeks of gestation, the amniotic fluid index increased by approximately 16% in the water intervention group and approximately 8% in the control group \[[@CR37]\]. The relationship between the state of hydration during pregnancy and the physical infant outcomes has not been fully valued and studied, so there are not so many related literatures. More studies are needed to be carried out to explore the associations between water intake, hydration state and on pregnancy complications, maternal-infant outcomes. In order to promote enough water intake, it is necessary to propose recommendations on water intake for pregnant women. Only some countries or organizations have made specific recommendations on water intake for pregnant women: 2.7 L/d is recommended by the American Institute of Medicine and 4.8 L/d is advised by the World Health Organization (WHO) \[[@CR15]\]. A survey of dietary guidelines in 84 countries revealed that only 8 countries had quantitative recommendations for fluid intake \[[@CR38]\]. Of the four editions of *Dietary Guidelines* (1989, 1997, 2007, and 2016) released in China, the first two editions had no recommendation for fluid intake \[[@CR39], [@CR40]\]. In 2007, the recommended amount of daily fluid intake was 1.2 L/d for adults, but no recommendation was given for pregnant women \[[@CR41]\]. In 2016, the recommendation was 2.7 L/d for women and 3.0 L/d for men and pregnant women \[[@CR15]\]. However, the recommendation for water intake for pregnant women was based on data of the water intake survey among pregnant women in other countries, not the data in China. Water requirements vary among pregnant women in different countries due to differences in dietary pattern, environment, and other factors; thus, it is necessary to undertake studies related to water intake during pregnancy in China. Two hypotheses are proposed in this study: one is that some pregnant women are dehydrated due to having insufficient water intake and the other is that dehydration has negative effects on pregnancy and maternal and infant outcomes. The objectives of this study are, firstly, to investigate the fluid intake among pregnant women, secondly, to assess their hydration state, and finally, to investigate the associations with pregnancy complications and maternal-infant outcomes. This preliminary exploratory study on will provides some related reference data for updating the recommendation for adequate water intake among pregnant women. In addition, it will provide some evidences on water-related education. Based on this preliminary study, multicenter studies with larger sample sizes are required for the development of clinical guidelines and recommendations. The ultimate goals are to promote adequate water intake, to maintain optimal hydration state, and to improve mater health and birth outcomes. Methods/design {#Sec2} ============== Study hypotheses {#Sec3} ---------------- One hypothesis is that a certain proportion of pregnant women have insufficient water intake and have risk of being in dehydration state. Another hypothesis is that dehydration state induced by insufficient water intake affects maternal health and birth outcomes. Study design {#Sec4} ------------ A prospective observational cohort study is designed. A total of 380 pregnant women with \< 13 weeks' gestation will be recruited. All items from the World Health Organization Trial Registration Data Set are shown in Table [1](#Tab1){ref-type="table"}. Table 1Items from the World Health Organization Trial Registration Data SetData CategoryInformationRegistration numberChiCTR1800019284, Chinese clinical trial registryRegistration State1,008,001 Prospective registrationPublic titleStudy for the correlation between hydration state and pregnancy complications, maternal and infant outcomes during pregnancyScientific titleThe correlation between hydration state and pregnancy complications, maternal and infant outcomes during pregnancyApproval of ethic committeeEthical Review Committee of the Hainan Medical UniversityEthical approval project identification code2018--4Date of approved by ethic committee20 May 2018Study typeProspective observational studyStudy designProspective observational studyKey inclusion and exclusion criteriaInclusion criteria: Aged between 21\~35; at first trimester of pregnancy (before 13 weeks), in health state, without metabolic disease, oral diseases, and so on. Exclusion criteria: Aged \< 21, or age \> 35; not being at first trimester of pregnancy (before 13 weeks); With metabolic disease, oral diseases, and other diseases.OutcomesUrine and blood osomolality, fluid intake, copeptin, pregnancy complications, maternal and infant outcomesCollecting samplesUrine and bloodRecruitment stateNot initiatedRandomization Procedure (please state who generates the random number sequence and by what method)Not applicable. A convenience sampling method will be used to recriute participants. Sample size calculation {#Sec5} ----------------------- For the calculation of sample size, the variable used is the incidence of new-onset hyperglycemia. Using the following formula, sample size is calculated:$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {n}_1=\frac{\left(1+\raisebox{1ex}{$1$}\!\left/ \!\raisebox{-1ex}{$C$}\right.\right)\overline{p}\ \overline{q}{\left({U}_1-\raisebox{1ex}{$\alpha $}\!\left/ \!\raisebox{-1ex}{$2$}\right.+{U}_1-\beta \right)}^2}{{\left({P}_1-{P}_0\right)}^2} $$\end{document}$; *α* = 0.05, *β* = 0.20. Among these parameters, *p*~*1*~ = 20.9%, indicates the incidence of new-onset hyperglycemia among participants with insufficient fluid intake; *p*~*0*~ = 9.0%, means the incidence of new-onset hyperglycemia among participants with sufficient fluid intake; *c* = 4, indicates the ratio of pregnant women with water intake of \> 0.5 L to pregnant women with water intake of \< 0.5 L. These parameters were set based on the reference to the results of a related study \[[@CR42]\]. A 10% dropout rate is taken into account to obtain the final sample size. For validity, 380 pregnant women will be needed. Participants {#Sec6} ------------ Pregnant women will be recruited from the First Affiliated Hospital of Hainan Medical University, Hainan province of China, using a convenience sampling method. The inclusion criteria are as follows: aged between 21 and 35 years; primigravidas; in the first trimester of pregnancy (before 13 weeks' gestation); and without kidney diseases, diabetes mellitus, digestive system diseases, cardiovascular diseases and other diseases. The exclusion criteria are as follows: smoking; habitual consumption of alcohol (\> 20 g/day) \[[@CR43]\] or intensive physical activity; or with the diseases of kidney diseases, diabetes mellitus, digestive system diseases, cardiovascular diseases and other diseases. Ethics {#Sec7} ------ The study protocol has been reviewed and approved by the Ethical Review Committee of the Hainan Medical University. Ethical approval project identification code is 2018--4. The study will be carried out according to the principles of the Declaration of Helsinki. All participants will read the informed consent form, voluntarily agree to participate in this study and sign the informed consent form prior to the study. Written informed consent will be obtained from each participants before enrolment in the study, and then preserved by researchers. Study procedure {#Sec8} --------------- Participants will be followed throughout pregnancy to childbirth. And their infants will also be followed up to 42 days of age. Relevant indicators and outcomes will be measured and followed at the first trimester of pregnancy (15--17 weeks' gestation), second trimester of pregnancy (20--22 weeks' gestation), third trimester of pregnancy (30--32 weeks' gestation), during childbirth and 42 days after childbirth (Fig. [1](#Fig1){ref-type="fig"}). Fig. 1Technology roadmap of study After recruitment: Maternal socio-economics, socio-demographics and other basic information, such as pregnancy history, childbearing history, family history, disease history, drug use and anthropometric data of the participants will be collected after being recruited and entry into the study. During the first trimester of pregnancy: Survey of fluid intake, dietary intake, psychological state, sleep quality and time, physical activities, history of disease and medication-use will be conducted using corresponding questionnaires at 1 weeks before antenatal care. On the day of antenatal care, anthropometric measurements, such as height, body weight and body composition, will be assessed by obstetricians. The first morning urine samples will be collected in sterile disposable urine sample cup and urine osmolality will be tested immediately by professional laboratory technicians. Antecubital venous blood will be collected to test hemoglobin, osmolality, hydration state regulating endocrine hormones -- copeptin and pregnancy related endocrine hormones, including estradiol, prolactin, and progesterone. Pregnancy complications and other diseases or symptoms including urinary tract infection, anemia, edema and constipation will be assessed and diagnosed by obstetricians in accordance with the relevant physiological indicators. The occurrence time, duration and treatment measures of these outcomes will be recorded in detail as an important factor to be considered in later data analysis. During the second and third trimester of pregnancy: the same procedure as the first trimester of pregnancy. During childbirth and after childbirth: Before childbirth (One day before the expected date of delivery or one day before cesarean section), psychological state, sleep quality and time, physical activities and medication-use will be collected using corresponding questionnaires. On the day of childbirth, the mode of delivery will be recorded, pregnancy complications and other complications during childbirth will be assessed and diagnosed by professional obstetricians. Infant birth weight will be measured and health state will be assessed. Breastfeeding time and the crying behavior of infants before and after breast-feeding will be recorded. At 42 days after childbirth, maternal weight retention will be measured and calculated. Secretion of breast milk and mode of infant feeding will be consulted and infant growth will be assessed. Flow-process diagram of the study was shown in Fig. [2](#Fig2){ref-type="fig"}. Fig. 2Flow-process diagram of the study If participants in this study requires hospitalization for any reason or receives intravenous fluid, or develops vomiting or diarrhea or fever or respiratory tract infections during pregnancy, the occurrence time, duration and treatment measures of these incidents will be recorded in detail. The time for recording fluid intake, food intake and other indicators will be exchanged within the allowable range of corresponding research phase to stagger the occurrence and duration of these situations. The schedule of enrolment, data collections and assessments are shown in Table [2](#Tab2){ref-type="table"}. Table 2The schedule of enrolment, interventions, and assessmentsTimepoint15 April 201915 April 2019--30 September 20201 Octorber 2020Enrolment: Eligibility screen√√ Informed consent√√Collection of basic information√√Assessments of related indicators and outcomes during pregnancy and childbirth√√Close-out√ Hydration related indicators and maternal-infant outcomes {#Sec9} --------------------------------------------------------- Hydration related indicators and maternal-infant outcomes are summarized in Table [3](#Tab3){ref-type="table"}. Table 3Summarize of hydration related indicators and maternal-infant outcomes in this studyRelated indicators and outcomesTrimester of pregnancyFirst trimester (15--17 weeks)Second trimester (20--22 weeks)Third trimester (30--32 weeks)During and after childbirthRelated indicators Fluid intake√√√√ Dietary survey√√√√ Other influencing factors√√√√ Endocrine index√√√ Hydration state√√√√Outcomes Physiological index√√√ Pregnancy complications√√√√ Other diseases or symptoms√√√√ Delivery mode√ Maternal weight intention√ Breast feeding√ Infant weight√ Definition of hydration {#Sec10} ----------------------- Hydration state is determined by the balance between water inputs and outputs and judged by the standard of urine osmolality. Dehydration is judged as urine osmolality of \> 800 mOsm/kg, and it occurs when water inputs is insufficient to replace water outputs \[[@CR44]\]. Middle hydration state is judged as urine osmolality between 500 and 800 mOsm/kg \[[@CR45]\]. Optimal hydration state is judged as urine osmolality of ≤500 mOsm/kg \[[@CR45]\]. Assessment of indicators and outcomes {#Sec11} ------------------------------------- ### Fluid intake {#Sec12} After standardized training by researchers, daily fluid intake of participants will be collected using a 7-day 24-h fluid intake record by themselves for correct use of such records. The amount of fluid intake for each time in seven consecutive days will be measured using a customized cup, and the nearest of cup scale is10 mL. This method has been proved to be reliable in many water intake-related studies \[[@CR21], [@CR46], [@CR47]\]. The volume, type, time and place of water intake will also be recorded by participants. All types of fluid intake will be recorded in detail including plain water, bottled water, tea, sugar-sweetened beverages, and so on. The record will be photographed and sent by participants to the investigators using a mobile phone, and will be reviewed by investigators every day to ensure the accuracy and the integrity of the records. ### Food intake and water intake from food {#Sec13} Food intake and water intake from food will be recorded and estimated using a semi-quantitative food frequency questionnaire (FFQ). The type, frequency and amount of food intake will be recorded. The amount of food intake will be estimated using by participants with the references of food models and photographic models \[[@CR48]\]. Nutrients and water intake from food will be assessed by trained investigators using the *Chinese Food Composition Table* \[[@CR49]\]. ### Psychological status:-anxiety and depression {#Sec14} Self-rated anxiety scale and self-rating depression scale will be used as indicators to assess the psychological status of participants \[[@CR50]--[@CR52]\]. Both scales, which mainly evaluates the frequency of symptoms, have 20 items each. Frequency will be rated on a 4-point scale: 1 indicates "never or seldom," 2 indicates "sometimes," 3 indicates "most of the time," and 4 indicates "all the time". For negative items, scores are graded sequentially according to the order from 1 to 4, and vice versa. The T score = total score for 20 items× 1.25 (the decimal portion is rounded). Standard of assessing anxiety: A T score 50--59 means mild anxiety, 60--69 means moderate anxiety, \> 69 means severe anxiety, and \< 50 means normal. Standard of assessing depression: T score 53--62 means mild anxiety, 63--72 means moderate anxiety, \> 72 means severe anxiety, and \< 53 means normal. ### Sleep quality and time {#Sec15} The Pittsburgh Sleep Quality Index (PSQI) will be used to assess sleep quality and time for participants, which include 19 self-rating items and 5 others-rating items \[[@CR53], [@CR54]\]. Score is calculated including 19 self-rating items. Total 7 factors are measured in the 19 items: subjective sleep quality; sleep latency; sleep persistence; habitual sleep efficiency; sleep disorder; use of sleep drugs and daytime dysfunction. A 4-point scale is used, "0" means "no difficulty"; "2" means "moderate difficulty"; "3" means "severe difficulty"; and "1" means "mild difficulty". The higher the score, the poorer the quality of sleep. ### Physical activity {#Sec16} Information on physical activity will be collected by using the International Physical Activity Questionnaire (IPAQ) \[[@CR55], [@CR56]\]. The IPAQ consists of 27 items, incluing the type, frequency and duration of various physical activities. Participants will be classified into low-intensity, moderate-intensity and vigorous-intensity physical activity according to the similar criteria in some references \[[@CR57], [@CR58]\]. ### Anthropometric measurements {#Sec17} On the day of antenatal care, height and weight will be measured twice by professional obstetricians using the height-weight meter (HDM-300; Huaju, Yiwu, Zhejiang, China) following standardized processes to the nearest 0.1 cm and 0.1 kg, respectively. \[BMI: weight (kg)/height squared (m)\]. ### Indicators related to hydration state {#Sec18} About the evaluation criteria and indicators for hydration state, multiple indicators will be used for comprehensive evaluation, so body fluids, fluid intake, urine and blood osmolality are measured for evaluating hydration state in this study. At present, the internationally recognized and authoritative indicator is urine osmolality, which will be used as the primary determinant of the hydration state in this analysis of this study, and other indicators will be considered to be auxiliary indicators. Intracellular fluid (ICF) and extracellular fluid (ECF): It will be measured by professional obstetricians using a body composition analyzer (Inbody 720; Inbody; Seoul, Korea) with the patients in a fasting state and after having defecated and urinated \[[@CR59]--[@CR61]\]. Urine and blood osmolality: The first morning urine samples will be collected in a sterile disposable urine sample cup, and blood will also be collected in vacuum tubes. Using an osmotic pressure molar concentration meter (SMC 30C; Tianhe, Tianjin, China) with the freezing-point method, osmolality will be determined by laboratory physicians, which will be used to assess the hydration state of participants. Urine-specific gravity (USG): Using an automatic urinary sediment analyzer (FUS-200, Dirui, Changchun, China) with the uric dry-chemistry method, the first morning urine samples will be collected in sterile disposable urine sample cup to determine USG by laboratory physicians. ### Endocrine indexes related to hydration and pregnancy {#Sec19} Copeptin as a hydration regulatory endocrine index and estradiol, prolactin, progesterone as pregnancy-related endocrine indexes will be tested in this study. The elbow venous blood will be collected in the morning and injected into a centrifugal tube containing 0.3 mol/L EDTA-Na2 and 5 × 10^5^ units of aprotinin. The supernatant was centrifuged at 3000 rpm/min for 15 min and stored at − 20 C. Then, these endocrine indexes will be determined with the method of radioimmunoassay with corresponding kits. ### Pregnancy complications, other diseases or symptoms and maternal outcomes {#Sec20} Gestational hypertension: Blood pressure and will be measured twice by obstetricians to the nearest 2 mmHg with the desktop mercury sphygmomanometer (Yuwell, Danyang, Jiangsu, China). Two measurements will be taken after 2-min intervals. Hypertension of pregnancy will be defined as systolic blood pressure of \> 140/90 mmHg and/or diastolic blood pressure of ≥90 mmHg sustained on two measurements. Gestational diabetes mellitus (GDM): Blood glucose will be determined with elbow venous blood using an osmotic pressure molar concentration meter (SMC 30C; Tianhe, Tianjin, China) by laboratory physicians. GDM will be diagnosed based on the results of the 100-g, 3-h oral glucose tolerance test (OGTT) by obstetricians. According to the recommendations of the Committee on Obstetric Practice, a definite diagnosis is made if two or more thresholds be met or exceeded. The threshold of fasting blood glucose is as follows:5.3 mmol/L, 10.0 mmol/L for blood glucose of 1 h, 8.6 mmol/L for blood glucose of 2 h and 7.8 for blood glucose of 3 h \[[@CR62]\]. Preeclampsia or eclampsia: Urine protein will be assessed using an automatic biochemical analyzer (Cobas C501; Roche; Basel, Switzerland) by laboratory physicians. Preeclampsia will be diagnosed at ≥20 weeks' gestation along with new-onset gestational hypertension and new-onset proteinuria \[[@CR63]\]. Eclampsia will be defined as the development of convulsions and/or unexplained coma during or after pregnancy with symptoms of preeclampsia, which will be diagnosed by obstetricians \[[@CR64]\]. Anemia: Hemoglobin of elbow venous blood will be tested using automatic routine blood test device (MC-600, Kubeier, Shenzhen, China) by laboratory physicians. Anemia will be defined as \< hemoglobin 110 g/L. Oligohydramnios: Amniotic fluid will be measured using ultrasonography by professional medical technicians. Oligohydramnios is diagnosed by an amniotic fluid index (AFI) of \< 5 cm or maximum amniotic fluid pool depth of \< 2 cm \[[@CR65]\]. Urinary tract infection: The diagnostic criteria is based on symptoms or laboratory confirmation including white blood cells and bacteria in urine by obstetricians. Routine urine test will be conducted using an automatic biochemical analyzer (Cobas C501; Roche; Basel, Switzerland) by laboratory obstetrician. Spontaneous abortion: It will be recorded by obstetricians. Preterm labor: It is defined as giving birth between 28 weeks' and less than 37 weeks' gestation. It will be diagnosed by obstetricians. Mode of delivery: There are two modes: namely natural childbirth and cesarean section, which will be recorded by obstetricians. Breast milk: The time for early initiation of breastfeeding will be recorded \[[@CR66]\]. And the amount of breast milk will be evaluated based on the performances of infant after breast feeding. Functional constipation: According to the Rome III criteria, it will be defined when participants present at least two of the following symptoms: at least one quarter of defecations for the last 12 weeks: lumpy or hard stools; straining; sensation of incomplete evacuation; sensation of anorectal obstruction; manual maneuvers to facilitate defecation; fewer than three defecations per week. Loose stools are rarely present without the use of laxatives \[[@CR67]\]. Functional constipation will be diagnosed by obstetricians. Total gestational weight gain (GWG) and postpartum weight retention (PWR): Both of them will be calculated based on weight of participants. GWG (kg) = weight before pregnancy (kg) -weight before delivery (kg). GWG will be assessed according to the recommendation of Institute of Medicine (IOM) \[[@CR68]\]. PWR (kg) = weight before pregnancy (kg) - weight at 42 days after childbirth (kg). ### Infant outcomes {#Sec21} Infant birth weight and length: They will be measured using length and weight-measuring devices for infants (HLZ-20; Hualizheng, Tianjin, China) by professional obstetricians while infant is wearing light clothing. The scales for weight and length are calibrated at intervals of 10 g and 0.5 cm, respectively. Low birth weight for newborn infants is defined as a weight of \< 2.5 kg, while macrosomia is defined as a weight of ≥4 kg. The growth of infant will be assessed according to the 2006 WHO Child Growth Standards \[[@CR69]\]. Temperature and humidity of the environment {#Sec22} ------------------------------------------- The temperature and humidity will be recorded according to the report of China Meteorological Administration at 9:00 AM and 3:00 PM each day. Confidentiality and withdrawal {#Sec23} ------------------------------ The information confidential of the participants will be kept carefully in the whole process of study. During and after the trial, the names of participants will be replaced as study ID. Participants can withdraw from the trial freely as they wish. In case of adverse events, an emergency physician will be assigned to deal with them in time. Data entry and statistical analysis {#Sec24} ----------------------------------- *Data entry*: All data will be documented twice by two trained researchers using the software of Epi Data 3.1 to ensure the accuracy of data. *Statistical analysis*: SAS 9.2 (SAS Institute Inc., Cary, NC, USA) will be used. A mixed model of repeated-measures (ANOVA) will be used to analyze the differences of outcomes among participants with different hydration state. Multi-logistic regression analysis will be used to investigate the influencing factors related to maternal health and infant outcomes, such as fluid intake, hydration state, urine osmolality, concentration of copeptin and other indicators. The level of significance will be set at 0.05 (*p* \< 0.05, 2-tailed) with 95% confidence intervals (95% CI). Discussion {#Sec25} ========== Pregnant women may need more fluid intake due to physiological changes in the mother and for fetal growth. Thus, the risk of being dehydration for pregnant women, due to insufficient fluid intake, is relatively high. However, surveys on fluid intake among pregnant women are scarce. Optimal hydration is essential for maintaining health. For pregnant women, the endocrine system shows considerable changes during pregnancy, which may affect water metabolism and balance and thus, hydration state. However, the importance of water and hydration, especially among pregnant women, is often neglected, and more research is required. In the present prospective cohort study, we will evaluate the fluid intake and hydration status of pregnant women and investigate the associations of fluid intake and hydration with pregnancy complications and maternal/infant outcomes. The most considerable challenges in this study is the following up of participants and quality control. Important protocol modifications will be discussed and determined by the researcher group, and will be communicated to investigators, trial participants and other relevant parties by e-mail immediately. In the study, participants will be followed up from the first trimester of pregnancy to 42 days after childbirth. To increase the rate of follow up, effective and convenient methods of communication between participants and researchers will be established, including telephone call, home visits, e-mail, WeChat, short message and so on. The content and purpose of the study will be clearly clarified to establish a relationship of trust and raise awareness of participants concerning the importance of the study. In this study, many methods will also be used to ensure the quality control. The requirements for participants will be fully explained. Participants will also be trained to be familiarized with the content of related questionnaires and study requirements. If a participant failed to follow the requirements, her data will not be included in this study and the reason for dropout will also be recorded. Researchers will be trained on the procedure of this trial, the content of the questionnaires and the technology-related laboratory tests. Some indicators such as height, weight, and blood pressure will be measured twice for accuracy, such as height, weight and blood pressure. Food models and photographic models will be provided to help participants fill the questionnaire related to food and fluid intake after training to enhance estimation accuracy. Physiological indexes, such as blood and urine indicators, will be tested immediately after collection, in order to ensure the accuracy, and will be tested by laboratory physicians with the minimum 5 years of clinical experience to ensure accuracy. Trial conduction will be audited by a competent authority in the hospital every day, and the process will be independent from investigators and the sponsor. All the questionnaires will be regularly checked and stored for researcher rechecks. A data monitoring committee (DMC) is established and is comprised by six members (three clinicians, one statistician, one ethicist and one nutritionists), who are independent of the sponsor and competing interests. DMC will review the procedure and data of the study for once a week. The six members will evaluate the safety concerns and benefits to decide that if the study should be continued or terminated. They will also examine the quality of the data, supervise the quality control, record the follow-up process and supervise the process of data entry and analysis. Four hydration and fluid intake related projects has been completed in China by our team; these provided an authoritative questionnaire and technical supports for the study. There are both strengths and weaknesses for the design of the present study. Referring to strengths, first, to eliminate the potential confounding factors as much as possible, age, BMI, physiological state, sleep quality, food intake and physical activity will be evaluated and included as covariates in the statistical analyses. Second, fluid intake is dependent on environmental factors, such as temperature and humidity. The conduction of this study will be in Hainan, which is an area with a tropical monsoon climate. Thus, temperature and humidity will be recorded clearly to actualize the comparability among different studies in China and with those in other countries. Third, hydration state will be evaluated in many indicators, not only including urine osmolality, but also including body compositions (ICF and ECF), copeptin, fluid intake and blood osmolality. Fourth, water intake from food will be assessed in order to calculate total water intake. Fifthly, the prospective cohort design can avoid recall bias and improve the accuracy of the results. In terms of weakness, the mechanism about the associations of hydration with pregnancy complications, maternal and infant outcomes will not be studied comprehensively. In this study, only the endocrine mechanisms related to hydration and pregnancy will be explored. An observational study will elucidate the associations but is not necessary for the development of recommendations unless confounding factors are carefully controlled. In summary, it is the first time to assess water intake and hydration stare among pregnant women, and it is the first time to investigate the associations of hydration with pregnancy complications, maternal and infant outcomes in China. Our results can provide some scientific reference data for updating the recommendations on fluid take for pregnant women and valuable evidence on the importance of hydration on maternal health during pregnancy. The results may also raise awareness on the importance of adequate fluid intake and optimal hydration in Chinese residents, especially pregnant women. More multicenter studies with larger sample sizes are required for the development of clinical guidelines and recommendations. The final goal is to develop water intake--related interventions to improve maternal health and birth outcomes. ECF : Extracellular fluid FFQ : Food frequency questionnaire GDM : Gestational diabetes mellitus GWG : Total gestational weight gain ICF : Intracellular fluid IPAQ : International Physical Activity Questionnaire OGTT : Oral glucose tolerance test PSQI : Pittsburgh Sleep Quality Index PWR : Postpartum weight retention USG : Urine-specific gravity **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Na Zhang and Fan Zhang contributed equally to this work. We gratefully thank the co-investigators of the survey for volunteer recruitment and determination of physiological indexes, namely, Hainan Medical University and Affiliated Hospital. Standards of reporting {#FPar1} ====================== The manuscript was written adhered to SPIRIT guidelines/methodology. Conceptualization, NZ, FZ, and GM; Methodology, NZ, FZ, JZ and GM; Writing -- original draft preparation, NZ, FZ, SC, FH, GL, YZ, HH, JZ; Writing -- review & editing, NZ, FZ, and GM; Funding acquisition, FZ. All authors revised the manuscript and approved this final version. Danone Institute China provide fund for the project (No. DIC2018--11 and No. DIC2019--06), which is an independent non-profit organization (danone.institute\@danone-institute.org.cn). Danone Institute China do not participate in the design of the study and collection, analysis, and interpretation of data, writing the manuscript and submitting for publication. They do not have ultimate authority over any of these activities. No additional data are available for protocol. The datasets used and/or analyzed of the study will be available from the corresponding author on reasonable request after finishing the study. The study protocol has been approved by the Ethical Review Committee of the Hainan Medical University. Ethical approval project identification code is 2018--4. The study will be conducted according to the guidelines of the Declaration of Helsinki. Prior to the study, all participants will read and sign the informed consent form. The consent obtained from study participants will be written. Not applicable in this section as no personal information is provided in this manuscript. The authors declare that they have no competing interests.

The study of human chromosomes provides valuable insight into processes of clinical diagnosis for many genetic disorders and analysis of chromosome architecture[@b1][@b2][@b3]. For example, more than 400,000 chromosome karyotype analyses for actual human genetic diagnoses are performed each year in the U.S. and Canada. The existing approaches of chromosome imaging tend to work effectively, but depend upon the binding of the fluorescent dyes with the chromosomal DNA. This condition raises a series of questions regarding the affinity of dyes to chromosomal DNA, the number of differently colored fluorescent dyes, and the photostability of the dyes[@b4][@b5][@b6][@b7]. Another important limitation is that, the shape variability of the chromosome, which is caused by the non-rigid nature of the chromosomal structure when placed on microscope slides, directly influences the interaction between the dyes and the DNA. Furthermore, fluorescence *in situ* hybridization (FISH), a fundamental technology in many aspects of genetics, genomics, and cell biology[@b8][@b9][@b10][@b11][@b12][@b13][@b14], is based on a mechanism whereby fluorescently labeled probes recognize and hybridize with the target chromosome DNA by nucleic acid base pairing. The technique enables researchers to identify rapidly the positions of genes and chromosomal aberrations in the clinic and research laboratory by determining the position of the fluorescent probe bound to the chromosomes. The success of FISH, wherein strained DNA is targeted by fluorescently labeled probes and then visualized via microscopy, depends on a specific state of the stained chromosomal DNA. In addition, chromosomal staining is an essential experimental approach for the study of chromosome behavior in dividing cells[@b15][@b16][@b17][@b18][@b19]. Observing the basic stages of cell division by chromosome staining would provide the essential technology to better understand the significant biological process of cell division. Chromosomal staining in the series of stages of prophase of cell division requires a high-affinity binding between the dye and chromosomal DNA to provide clear visualizations. The traditional dye molecules used in the study of chromosomal behavior do not form covalent bonds with chromosomal DNA; therefore, it is difficult to monitor chromosome dynamics. To overcome these limitations and further enable the potential of chromosome imaging to be fully exploited in both research and diagnostic laboratories, we have developed a chemistry-based strategy for imaging chromosomal DNA in multicolor in place of the traditional dye pairing. Bioorthogonal chemical reactions have been achieved using the Staudinger ligation[@b20], Diels-Alder reaction[@b21], copper(I)-catalyzed azide--alkyne cycloaddition[@b22][@b23][@b24][@b25][@b26], and strain-promoted azide--alkyne cycloaddition reactions[@b27][@b28][@b29], which has allowed selective labeling of cellular proteins[@b30][@b31][@b32][@b33][@b34][@b35], nucleic acids[@b36][@b37][@b38][@b39][@b40][@b41][@b42][@b43][@b44], lipids[@b45][@b46], and glycans[@b47][@b48][@b49]. Recently, based on the azide--alkyne click reaction, 5-ethynyl-2′-deoxyuridine (EdU) was reported as a thymidine analog to detect DNA synthesis in cells[@b38][@b39][@b40]. After its metabolic incorporation into DNA, EdU can be detected with fluorescent azides by click reaction. This method was used to label DNA in cell level by microscopic analysis[@b39]. An arabinofuranosylethynyluracil derivative (F-ara-EdU) also exhibited selective DNA labeling in cell level[@b38]. Using a click reaction, we successfully found an unusual nucleic-acid structure formed by DNA and RNA[@b50]. Click reaction is a chemical reaction that can proceed in biological systems without interaction with the inside biomolecules or interference on the whole system[@b24][@b51]. The click reaction to form a covalent bond between the chromosomal DNA and fluorescent molecules with highly selective and reactive with in the biological environment, allows that chromosomal DNA imaging is not affected by interaction affinity and structure etc. On the contrary, the traditional dyes work in an affinity-dependent manner with the chromosomal DNA, the intracellular environment may influence the interaction of chromosomal DNA and ligands. A chemical reaction by a covalent bond to connect the chromosomal DNA and fluorescent molecules is believed to overcome this difficulty and is less affected by affinity, environment, or chromosomal structure. Traditional DNA imaging methods have also used fluorescent fusion proteins, staining nucleic acids, immunostaining with antibodies[@b52]. All of these approaches are limited in terms of the low throughput due to their relatively large size of fluorescent proteins, the low cell membrane permeability of antibodies, and large perturbations to native systems. Click reaction, an enzyme-free approach to DNA imaging, would not only eliminate the need for enzymatic reaction, but also readily utilized the azide and alkyne groups by taking advantage of their small size and inertness to most components in a biological environment. Despite advances in DNA labeling, the strategy has an undesirable characteristic related to the additional wash step was required to remove unbound, free fluorescent dyes. Based on the results of the past studies, we have now extended this methodology to image chromosome DNA at an individual chromosome level. We developed a light-up (turn on) reporter strategy to stain chromosomal DNA. We designed and synthesized two new azidocoumarins as the pro-fluorophore that can produce a strong fluorescence in a click reaction. The azido group of the profluorophores quenches the fluorescence, but the azide-alkyne click reaction can eliminate the quenching, resulting in a strong fluorescence. The advantage of having emission dependent on the click reaction is that it allows us to stain chromosomes that do not require extensive wash steps to remove the unreacted fluorescent dyes in the cell. This method provides a high-resolution image allowing the visualization of individual chromosomes. Next, we applied this strategy to FISH to identify telomeres in the end of chromosome. The chemistry-based method allows for determination of telomeres with high sensitivity and specificity. We further extend the utility of this approach to study the chromosome states during cell division. Instead of the traditional dye staining, the click reaction enables direct visualization of several important stages in cell division. Results and Discussion ====================== Chromosome imaging using fluorescent azide molecules at the individual chromosome level --------------------------------------------------------------------------------------- To image chromosomal DNA, an intramolecular-tagging approach is required[@b38][@b39][@b40], because the target DNA sequences are typically not accessible for labeling within cells. We first introduced EdU into chromosomal DNA as a rapid labeling tag for a click reaction. Cells were labeled with 10 μM EdU for 2 h and reacted with Alexa488-azide. We observed very intense nuclear staining by a fluorescent azide, consistent with previous reports. In contrast, cells not labeled with EdU exhibited no detectable staining under the same reaction conditions with Alexa488-azide ([Supplementary Fig. 1](#S1){ref-type="supplementary-material"}). These results provide direct evidence that EdU can be incorporated into cellular chromosomal DNA and that the labeling reaction is highly specific. To explore the possibility of a click reaction to stain individual chromosomes, cells were incubated with 10 μM EdU for 24 h and then stained with the fluorescent azides. We found that the click reaction produced uniformly stained chromosomes that could be visualized at the individual chromosome level with Alexa488-azide (green) and Alexa594-azide (red) ([Fig. 1](#f1){ref-type="fig"}). We next stained chromosomes with multicolor imaging by using a click reaction. First, EdU-labeled chromosomes were reacted with Alexa488-azide. Subsequently, cells were stained with Alexa594-azide. The chromosomes are strongly stained for both the green and red colors ([Supplementary Fig. 2](#S1){ref-type="supplementary-material"}). We demonstrated that the fluorescent azide molecules react specifically and rapidly with the alkyne group of EdU to paint chromosomes. However, We found that for obtaining clear fluorescent image, the additional wash steps were required to remove unbound, free fluorescent dyes. Eliminating the repetitive wash steps to remove the unreacted fluorescent dyes resulted in a high level of background fluorescence and failed to observe nuclear staining ([Supplementary Fig. 3](#S1){ref-type="supplementary-material"}). Preparation of two pro-fluorophore azidocoumarins 1 and 2 for chromosome imaging -------------------------------------------------------------------------------- To improve the labeling method and adapt it to ideal imaging of chromosomal DNA, We designed and synthesized two new pro-fluorophore azidocoumarins **1** and **2** from hydroxybenzaldehyde derivatives by a 3-nitro group conversion from a sodium azide reagent ([Supplementary Scheme 1](#S1){ref-type="supplementary-material"}). The two new pro-fluorophores **1** and **2** have no fluorescence, because of the quenching effect of the electron-rich nitrogen in the azido group. Formation of a triazole ring at its 7-position by the azide-alkyne click reaction can eliminate quenching, resulting in a strong fluorescence ([Fig. 2a](#f2){ref-type="fig"})[@b53][@b54][@b55][@b56][@b57][@b58]. We performed fluorescence microscopy experiments to investigate the fluorescent properties of pro-fluorophores **1** and **2** ([Fig. 2b](#f2){ref-type="fig"} and [Supplementary Fig. 4](#S1){ref-type="supplementary-material"}). When only pro-fluorophores **1** and **2** were present in the solution, almost no fluorescence was observed; whereas, clear fluorescence observed with the naked eye after the addition of EdU via the click reaction ([Fig. 2b](#f2){ref-type="fig"}). The fluorescence spectrum of **1** and **2** exhibits emission around 450 nm and 510 nm, respectively, after the click reaction ([Fig. 2b](#f2){ref-type="fig"}). Importantly, the fluorescent product of **2** exhibits a red-shift in emission (from 450 to 510 nm), than that of **1**, which enables their specific emission wavelengths and eliminates the any background fluorescence in a specimen, a highly desirable feature for chromosome imaging. We applied **1** and **2** to living cells and observed two clear colors, blue and green, in cells using a 360/40 nm excitation filter and a 470/40 nm emission filter for **1** (blue) and an excitation (480/40 nm) and an emission (527/30 nm) filter for **2** (green) ([Fig. 2c](#f2){ref-type="fig"}). These results suggested that **1** and **2** are able to serve as light-up reporters to stain chromosomal DNA by click reaction. Next, we examined the time course of DNA light-up labeling in cells. We observed that the fluorescence intensities of **1** and **2** from the click reaction increased with time, reaching a plateau within 4 and 2 h for **1** and **2** ([Supplementary Fig. 5](#S1){ref-type="supplementary-material"}), respectively, indicating a quick click reaction between **2** and cellular EdU-labeled DNA compared with **1**. Chromosome imaging using the pro-fluorophores 1 and 2 ----------------------------------------------------- Having confirmed the efficient click reaction of **1** and **2** with EdU in cells, we were encouraged to apply this method to stain chromosomes with multicolor imaging. Cells were incubated with EdU and stained with 24 μM of **1** or **2**. We clearly visualized the chromosomes in green and blue colors ([Fig. 3](#f3){ref-type="fig"}). To achieve multicolor painting chromosomes, we first stained cells with **1** for 4 h and then with **2** for 1 h. We obtained similar success in the chromosome visualization by assigning blue and green colors ([Supplementary Fig. 6](#S1){ref-type="supplementary-material"}). To find a more useful application for chromosome staining, we combined **1**, **2**, and Alexa594-azide to image chromosomes ([Fig. 4](#f4){ref-type="fig"}). The three staining patterns display a perfect fluorescent image (blue, green, and red). Multi-labeling offers a way to view images in an overlay mode, which displays a combined visible and fluorescence image. The two-color overlay allows us to visualize clearly the co-localization of two fluorescent probes in a specimen. The combination of **1** (blue) and **2** (green), **2** and Alexa594-azide (red), or **1** and Alexa594-azide is represented in the cyan, magenta, and yellow images ([Fig. 4](#f4){ref-type="fig"}), respectively. Further, the three-color overlay produces (**1**, **2**, and Alexa594-azide) a white fluorescence image. Observing the chromosome in the desired color is possible by changing the fluorescent dye molecules that react with the DNA and achieving an overlay mode, suggesting that the click reaction strategy provides a useful tool for imaging chromosomes. Advantage of the click reaction method compared to traditional method --------------------------------------------------------------------- By using the turn-on fluorescent strategy via the click reaction, having eliminated the background fluorescence which remained a matter of concern from traditional dye staining, we carried out two comparison experiments to further demonstrate the advantage of the click reaction method compared to traditional dye painting. We first used propidium iodide (PI), which was the most widely used traditional dye, to strain chromosomal DNA. Although uniform fluorescence of chromosomal arms was produced, additional wash steps that are required to remove unbound, free fluorescent dyes, can result in loss of signal. We found that the repetitive wash steps to remove unbound dyes (10 times) induce unclear chromosomal DNA signals in traditional method, but the chromosomal DNA strained by click reaction method could be clearly observed even when the wash steps were added to 20 times ([Supplementary Fig. 7a](#S1){ref-type="supplementary-material"}). Next, we compared the click reaction method with traditional method in multicolor chromosome imaging. The chromosomal DNA stained with PI (red) was subsequently bound with Hoechst (a blue fluorescent dye). We visualized the chromosomes in red color with PI, but the clear chromosomes were unable to be observed in blue color with Hoechst ([Supplementary Fig. 7b](#S1){ref-type="supplementary-material"}). The two-color overlay does not also produce a clear chromosome DNA in a desired color ([Supplementary Fig. 7b](#S1){ref-type="supplementary-material"}). The traditional dyes work in a binding-dependent manner with the chromosomal DNA, leading to results that the affinity of dyes to chromosomal DNA is very sensitive to DNA conformation and chromosome state. The dye PI that already binds to chromosome DNA at first binding step is believed to affect the secondary dye (Hoechst) binding with DNA. On the contrary, the click reaction connects the chromosomal DNA and fluorescent molecules by a covalent bond and allows chromosomal DNA imaging that is not affected by affinity, environment, or chromosomal structure. These results provide a new approach to overcome the technical limitations in traditional chromosome staining. Click reaction application in FISH assay ---------------------------------------- Encouraged by these data, we next tested whether highlighting the chromosomes with this method would be effective with the FISH technique. FISH uses fluorescent probes to bind only parts of the chromosome by means of sequence complementarity. FISH is often used to find specific features in DNA for use in genetics, medicine, and species identification, by determining where fluorescent probes are bound to the paired chromosomes. We applied this method to detect human telomeres at the end of the chromosomes, a commonly-used target in FISH[@b59][@b60][@b61]. We directly detected the telomere from **1**, **2**, and Alexa594-azide labeled human chromosomes by FISH ([Fig. 5](#f5){ref-type="fig"}). We observed the red and green fluorescent spots at the ends of each chromosome from **1**, **2**, and Alexa594-azide labeled human chromosomal DNA (blue, green, and red), when fluorescently labeled FISH probes for telomere DNA were fluorescent Cy3 (red) and FAM (green). These results demonstrated that chromosomes stained using click reaction provide a highly applicable substitute for FISH. Click reaction for application in the study of cell division ------------------------------------------------------------ The cell cycle is defined as the series of events that takes place leading up to and including cell division. The cell cycle has two important stages: interphase and mitosis. During interphase, the cell grows in size, doubling its DNA. Mitosis is important for the maintenance of the chromosomal set, in which chromosomes are separated into two identical sets of chromosomes in two daughter cells to maintain the genome's integrity. Mitosis involves several basic stages: prophase, prometaphase, metaphase, anaphase, and telophase. Fluorescent imaging made major contributions to our understanding of the main phases of mitosis during cell division. Chromosome staining is an essential experimental approach for the studies of chromosomal behavior in dividing cells. We applied the click reaction method to visualize mitotic progression in place of the traditional dye-based imaging. We clearly observed the basic steps of the cell cycle by staining chromosomal DNA with a click reaction ([Fig. 6](#f6){ref-type="fig"}). After a period of cell growth in interphase, the cell enters mitosis and chromosomes begin to condense in prophase. In prometaphase, the next step of mitosis, the nuclear envelope breaks down, and chromosomes congression begins. In metaphase, the chromosomes line up in the cell in their most condensed and coiled stage. During anaphase, we observed that the chromosome pairs divided and moved to opposite poles of the cell. In telophase, the chromosomes continued to separate and were cordoned off into the new nuclei. The key stages of cell division can be visualized by click reactions to stain the chromosomes, suggesting that the technique is useful in studies of cell division. Conclusions =========== First, the click reaction connects the chromosomal DNA and commercial dye molecules and allows for the multicolor staining of chromosomes. The observation of chromosomes in a desired color is possible by changing the fluorescent dye molecules that react with the DNA. Next, we developed a turn-on fluorescent strategy based on the click reaction. Two pro-fluorophore moieties served as light-up reporters to stain chromosomal DNA, which can be used to directly visualize the clear chromosomes in multicolor. Multi-labeling also offers a way to view images in an overlay mode by combination of two or three fluorescence images and allows us to visualize clearly the co-localization images in a multi-pattern. In addition to eliminate the background fluorescence using the turn-on fluorescent strategy by the click reaction, a covalent bond formation between the chromosomal DNA and fluorescent molecules by the click reaction allows that chromosomal DNA imaging is less affected by affinity, environment, or chromosomal structure, which remained a matter of concern from traditional dyes. Furthermore, we demonstrated that the chromosomes stained by this approach are effective with the FISH technique for detection of telomere DNA at the ends of chromosomes. We further applied this approach to observe several important stages of cell division. We found that the click reaction can be used to directly visualize the key stages of cell division. These results suggest that the click chemistry approach provides a powerful technique for imaging chromosomes for chromosome analysis and genetic diagnostics. Methods ======= Chemistry --------- ^1^H and ^13^C NMR spectra were measured at 500 MHz on a Bruker AMX spectrometer or 300 MHz on a Bruker (300-AVM) magnetic resonance spectrometer. High-resolution electrospray ionization (ESI) mass spectra were recorded using Exactive Orbitrap mass spectrometer (Thermo Scientific). Data was acquired using Xcalibur software (Thermo Scientific). All samples were dissolved in methanol (LC-MS grade, Wako), and the sample solutions were infused into the ESI source at a flow rate of 20 μL/min by using instrument's syringe pump. In photography experiments, UV irradiation of 365 nm was achieved with a UV Spot Light Source (Hamamatsu Photonics, 200 W) and UV-D36C filter (Asahi Technoglass). Reaction condition: \[EdU\] = 2.5 mM, \[**1** or **2**\] = 2.5 mM, \[CuSO~4~\] = 25 mM, \[Ascorbic acid\] = 125 mM, r.t. 2 h. Fluorescent spectra were measured using a JASCO model FP-8200 spectrofluorometer. JASCO Spectra Manager Software was used for data capture and processing of all spectra. The spectra were recorded using a 1-cm path-length cell. For each sample, at least two spectrum scans were accumulated over a wavelength range from 300--650 nm. DMSO-*d*~*6*~ and CDCl~3~ were used as the solvent. Chemical shifts are reported in parts per million shift (δ value) from Me~4~Si (δ 0 ppm for ^1^H) as an internal standard. Coupling constants (*J*) values are given in Hz and are correct to within 0.5 Hz. Signal patterns are indicated as br, broad; s, singlet; d, doublet; t, triplet; q, quartet; sex, sextet; m, multiplet. All reagents were purchased from Sigma-Aldrich, TCI (Tokyo Chemical Industry) or Wako (Wako Pure Chemical Industries). For organic synthesis, reagents of synthesis grade (\>98% purity tested by GC) from Sigma-Aldrich, guaranteed reagent (GR) grade from TCI, and special grade from Wako were used, respectively. Thin layer chromatography was performed using TLC Silica gel 60 F~254~ (Merck). Synthesis of the pro-fluorophores 1 ----------------------------------- A mixture of 2, 4, 5-trihydroxy benzaldehyde (3.08 g, 20 mmol), *N*-acetylglycine (2.34 g, 20 mmol), anhydrous sodium acetate (4.9 g, 60 mmol) in acetic anhydride (100 mL) was refluxed under stirring for 4 h. The reaction mixture was poured onto ice to give a yellow precipitate. After filtration, the yellow solid was washed by ice water before it was refluxed in a solution of concentration HCl and ethanol (2:1, 30 mL) for 1 hour, then ice water (40 mL) was added to dilute the solution. The solution was then cooled in an ice bath and sodium nitrite (40 mmol) was added. The mixture was stirred for 5--10 min and sodium azide (60 mmol) was added in portions. After stirring for another 15 min, the resulting precipitate was filtered off, washed with water, and dried under reduced pressure to afford a brown solid pro-fluorophore 1; 2.1 g (48% overall yield). Synthesis of the pro-fluorophores 2 ----------------------------------- First, 3-amino-7-dibuthylaminocoumarin was synthesized from 3-nitro-7-dibuthylaminocoumarin (See [Supporting Information](#S1){ref-type="supplementary-material"}). Next, 3-amino-7-dibuthylaminocoumarin (100 mg, 0.43 mmol) was dissolved slowly in HCl aq. (17.2%, 4 mL) at room temperature. Upon cooling to 0--5 °C and addition of a solution of sodium nitrite (30 mg, 0.43 mmol), the reaction mixture was stirred for 1 hour at 0--5 °C. This was followed by the addition of potassium acetate (2 g) in water (5 mL) to adjust the pH of the resulting solution to 4. Sodium azide (57 mg, 0.88 mmol) was added in portions at 0--5 °C, the mixture stirred at 0--5 °C for another 5 h. The precipitated product was rapidly filtered, washed with ice-cold water (10 mL) and dried under vacuum to yield the pro-fluorophore **2** (84 mg, 77%) as a yellow solid. Chromosome imaging ------------------ For biological assay, all reagents of molecular biology grade from all suppliers were used. HeLa cells were grown at 37 °C and 5% CO~2~ in Dulbecco's Modified Eagle's medium (DMEM) containing 10% fetal bovine serum (FBS) and antibiotics (penicillin and streptomycin). 10 μM EdU was added to medium and cells were incubated for 24 h. Colcemid was added to medium and EdU labeled cells were incubated at 37 °C and 5% CO~2~ for 2 h. Cells were washed with PBS, harvested with trypsinization and spinned down. After spinning down the cells, pellet was resuspended in 5 volume of 3:1 (v/v) MeOH/AcOH and incubated at r.t. for 20 min. Cell suspension was fixed on glass slide and treated with 0.5 mg/mL pepsin at r.t. for 10 min. Cells were washed with PBS and were fixed again with 4% of paraformaldehyde (PFA) at r.t. for 10 min. To stain EdU-labeled chromosome, cells were incubated in 24 μM of fluorescent azides (Alexa488-azide and Alexa594-azide) in click reaction buffer (100 mM pH 8.5 Tris-HCl, 1 mM CuSO~4~, and 50 mM ascorbic acid) and incubated at 37 °C and 5% CO~2~ for 30 min under protection from the light. Cells were washed by PBS for several times. To stain chromosomes in multicolor, the cells were stained a second time with 24 μM another fluorescent azide for 30 min. After wash again, chromosomes were observed with AF-6000 (Leica Microsystems) and BZ-9000 Fluorescent microscope. The commercially available reagents and reaction solvents for click reaction were purchased from Invitrogen. Chromosomes were stained by PI and Hoechst 33342 at r.t. for 30 min in dark after washing twice, 10 times, and 20 times by PBS, respectively. Chromosome imaging using 1 and 2 -------------------------------- EdU-labeled cells were reacted with 24 μM of each pro-fluorophore **1** or **2** in click reaction buffer (Click-iT EdU Imaging Kits from Invitrogen) at 37 °C for 6 h. For multicolor imaging of chromosomal DNA, cells were stained with **1** for 4 h and then with **2** for 1 h. Finally, cells were reacted with Alexa594-azide for 30 min and washed by PBS. Chromosomes were observed with AF-6000 (Leica Microsystems) and BZ-9000 Fluorescent microscope. The excitation and emission filters are 360/40 nm and 470/40 nm for blue fluorescence (compound 1), 480/40 and 527/30 for green fluorescence (compound 2 and Alexa 488) and 546/12 and 600/40 for red fluorescence (Alexa 594). Images are acquired using Leica AF6000 and BZ-II analyzer softwares. FISH assay ---------- **1**, **2** and Alexa594-azide stained cells were washed three times with PBS, dehydrated for 5 min in 70% EtOH, 5 min in 95% EtOH and 5 min in 100% EtOH and dried in air. Hybridization buffer (10 mM pH 7.2 Tris-HCl, 0.5% blocking reagent, and 70% formamide) containing 0.1 μM fluorescent labeled PNA (Cy3--(CCCTAA)~3~ for Cy3 labeling and FAM--(CCCTTA)~3~ for FAM labeling) was added to the cells. Then cells were denatured at 80 °C for 3 min and incubated at r. t. for 2 h. After incubation, cells were washed twice for 15 min with washing buffer (10 mM pH 7.2 Tris-HCl, 0.1% BSA, and 70% formamide) and three times for 5 min with washing buffer (0.1 M pH 7.2 Tris-HCl, 0.15 M NaCl, and 0.08% Tween 20). Cells were dehydrated with EtOH in the same procedure above and dried in the air. All procedures were carried out under protection from the light. Labeled cells were observed using AF-6000 (Leica Microsystems) and BZ-9000 Fluorescent microscope. Study of cell division ---------------------- Cells were grown on dish in DMEM supplemented with 10% fetal bovine serum, 1% penicillin, and 1% streptomycin, plated at 30--40% confluence. Remove the culture media, cells were washed twice with 1 × PBS, then added culture media. 10 μM EdU was added to culture media for 24 h. Cell suspensions were fixed on glass dish by 1 mL of 3.7% formaldehyde in PBS for 15 min, washed once with 1 mL of 3% BSA in PBS and treated with 1 mL of 0.005% pepsin at r.t. for 10 min. Cells were washed once and fixed again with 1 mL of 3.7% fixative at r.t. for 10 min. To observe the basic steps of mitosis, chromosomal DNA was stained by click reaction and observed using AF-6000 (Leica Microsystems) and BZ-9000 Fluorescent microscope. Additional Information ====================== **How to cite this article**: Ishizuka, T. *et al.* Fluorescence imaging of chromosomal DNA using click chemistry. *Sci. Rep.* **6**, 33217; doi: 10.1038/srep33217 (2016). Supplementary Material {#S1} ====================== ###### Supplementary Information This work was supported by Ministry of Education, Science, Sports, Culture, and Technology, Japan Grant-in-aid for Scientific Research (B) 26288083. This work was also supported by grants from the Takeda Science Foundation. **Author Contributions** Y.X. designed research. T.I., H.S.L. and K.I. carried out the experiments. Y.X., T.I., H.S.L. and K.I. analyzed data. Y.X. wrote the manuscript. {#f1} {#f2} {#f3} {#f4} {#f5} {#f6}

INTRODUCTION ============ Visceral leishmaniasis (VL) is the most severe vector-borne protozoan disease caused by *Leishmania donovani*, and is transmitted by the bite of female *Phlebotomus* sand flies \[[@b1-epih-39-e2017001],[@b2-epih-39-e2017001]\]. VL is a global health problem, imposing an estimated disease burden of 1.98 million disability-adjusted life years and 20,000 to 40,000 deaths per annum around the world \[[@b1-epih-39-e2017001]\]. Ninety percent of the global burden of VL occurs in India, Bangladesh, Brazil, Sudan, South Sudan, and Ethiopia \[[@b3-epih-39-e2017001]-[@b5-epih-39-e2017001]\]. Although VL is treatable, in sub-Saharan Africa, mortality associated with VL is still high. This is directly or indirectly related to poor treatment outcomes, which occur for a number of reasons \[[@b6-epih-39-e2017001],[@b7-epih-39-e2017001]\]. Ethiopia is a country highly affected by VL, with 1,860 cases of VL reported annually and an estimated annual incidence of VL ranging from 3,700 to 7,400 cases \[[@b8-epih-39-e2017001]\]. In Ethiopia, the northwest lowland areas (Metema and Humera districts) bordering Sudan are the most important endemic foci of VL, accounting for 60% of the VL burden in the country \[[@b9-epih-39-e2017001],[@b10-epih-39-e2017001]\]. This region also has the highest burden of VL/human immunodeficiency virus (HIV) coinfection in the world, with the HIV prevalence among VL patients ranging from 19 to 41% \[[@b10-epih-39-e2017001]\]. Poor treatment outcomes for VL have become a challenging public health problem in Ethiopia, particularly in the northwest region of the country. One of the greatest concerns regarding VL is its high fatality rate, which reaches almost 100% among non-treated symptomatic patients, in contrast to a 10% fatality rate among VL patients who undergo treatment. Data on treatment outcomes and the determinants thereof would enable policymakers, clinicians, and funding organizations to establish priorities for addressing VL incidence and mortality in the northwestern sub-region of Ethiopia. However, although there have been a few studies on VL/HIV coinfection, the proportion of poor treatment outcomes among VL patients and the determinants of treatment failure have not been well investigated in the study area. Hence, this study aimed to assess the proportion and determinants of poor treatment outcomes among VL patients at Kahsay Abera Hospital in northwest Ethiopia. MATERIALS AND METHODS ===================== Study design and setting ------------------------ A hospital-based retrospective study was conducted at Kahsay Abera Hospital in northwestern Ethiopia. This hospital is a general facility where all diseases are treated. There is a leishmaniasis treatment center within the hospital, which is located in Humera, one of the most important VL endemic foci in the country. Humera, at an elevation of 600 m above sea level, is characterized by a very harsh environment, with an annual temprature ranging from 27 to 45°C and rainfall ranging from 900 to 1,800 mm per annum. Humera is one of Ethiopia's most fertile agricultural zones, with large-scale farming of cash crops such as sesame, maize, cotton, and sorghum. Since 1970, Humera has undergone an extensive program of agricultural development, with a consequent influx of migrant workers, which has led to a rapid increase in the number of VL cases. Migrants travel to Humera to work in the sowing and harvest seasons, mainly from October to February and from May to July. Seasonal agricultural work attracts approximately 500,000 migrant laborers each year from non-endemic parts of Tigray and neighboring regions \[[@b11-epih-39-e2017001]\]. A clinical case of VL is defined as a person who presents with fever for more than 2-week and an enlarged spleen (splenomegaly) and/or enlarged lymph nodes (lymphadenopathy), or symptoms such as loss of weight, anemia, or leukopenia, while living in or having recently travelled to a known endemic area for VL. Diagnosis of VL at Kahsay Abera Hospital is performed according to the standardized Médecins Sans Frontières (MSF) and Ethiopian Ministry of Health guidelines using a positive rK39 rapid diagnostic test (DiaMed-IT-Leish, DiaMed, Cressier, Switzerland) and the microscopic examination of aspirates from the spleen \[[@b12-epih-39-e2017001]\]. The diagnosis has a high specificity, but the sensitivity of the microscopy varies (93 to 99% for the spleen, 53 to 86% for bone marrow, and 53 to 65% for lymph node aspirate) \[[@b13-epih-39-e2017001]\]. Treatment is normally provided only after the presence of the disease is confirmed based on a clinical examination and labora2tory tests. Sodium stibogluconate monotherapy is administered through intramuscular injections of 20 mg/kg/d for 30 days. The treatment regimen at the hospital follows the national guidelines. Definitions of terms in visceral leishmaniasis cases ---------------------------------------------------- An initial cure was defined as a patient showing improvement in signs and symptoms after 30 days of standard treatment (fever resolution, hemoglobin increase, weight gain, and spleen size regression), the absence of parasites in smears, and a negative test-of-cure (TOC) culture. In the current study, we were not able to assess whether the cure was definitive at a 6-month follow-up examination, since VL patients are often migrants from rural communities and are difficult to trace once they are discharged from the hospital. Poor treatment outcomes were defined as death, treatment failure, and non-adherence. Initial treatment failure was defined as a positive TOC culture (parasitological failure) and/or clinical signs/symptoms that persisted after 30 days of standard treatment. Functional status was defined as an individual's ability to perform the normal daily activities required to meet basic needs, fulfill his or her usual roles, and maintain health and well-being. Inclusion and exclusion criteria -------------------------------- All VL patients at the Kahsay Abera Hospital Leishmaniasis Treatment Center who underwent anti-leishmaniasis treatment from October 2010 to April 2013 were included in this study. More recent data were not included, as the proper data-handling format was not completely implemented at the hospital after the MSF project was phased out. A total of 890 VL patients were registered during the study period, of whom 595 patients were included. Those who were transferred out to other health institutions for better management and those for whom the available information was incomplete were excluded from the study. Data collection --------------- Demographic, clinical, immunological, and laboratory profile data were collected using a prepared data extraction tool. Four trained health professionals collected the data from the patients' charts after confirming that the records were complete. Treatment outcomes and patient characteristics were collected from the medical registry casebook and VL patient charts, respectively, and transferred to the data extraction tool after the completion of 30 days of standard treatment. A good treatment outcome was considered to be cure of the disease as measured by both a laboratory test (a negative TOC culture at the end of 30 days of standard treatment) and clinical signs and symptoms (weight gain, normal body temperature, and regressed spleen). Non-adherence (patients who were lost to follow-up or did not pick up their drugs for more than 15 days), treatment failure (non-response or failure to decrease the parasitological grade after a 30-day course of the treatment regimen) and death due to visceral leishmaniasis were considered to be poor treatment outcomes \[[@b14-epih-39-e2017001]\]. Data analysis ------------- Data were cleaned and entered into a computer using Epi Info version 7 (Centers for Disease Control and Prevention, Atlanta, GA, USA) and exported to SPSS version 20 (IBM Corp., Armonk, NY, USA) for analysis. Both descriptive and analytical statistical procedures were utilized. Descriptive statistics such as percentages, mean values, and standard deviations were used to present our results. Binary logistic regression was used to identify factors associated with poor treatment outcomes among VL patients. Bivariate analysis was performed to identify the association of each independent variable with VL treatment outcomes. Variables with p-values \<0.2 in the bivariate analysis were entered into the multivariate analysis to identify the determinants of poor outcomes while controlling for the possible effect of confounds. Adjusted odds ratios (aORs) with 95% confidence intervals (CIs) were used to determine the strength of associations, and variables with a p-value \<0.05 in the final model were taken as significant determinants of treatment outcomes. The goodness-of-fit of the model was tested using the Hosmer--Lemeshow test for the full model (p\>0.05). Ethics approval and consent to participate ------------------------------------------ Ethical clearance was obtained from the Ethical Review Committee of the Institute of Public Health, University of Gondar. A permission letter was also obtained from the Tigray Regional State Health Bureau to review pre-existing health records. To ensure confidentiality, patients' names were not extracted from the records. RESULTS ======= Demographic characteristics --------------------------- A total of 595 VL patient records were included in the analysis. The patients' mean age was 25.9±9.6 years, and a majority of them (535, 89.9%) were in the age group of 16 to 45 years. Almost half of them (284, 47.7%) were residents of the district, and one-third (220, 37.0%) were migrants from nearby districts and zones. The majority of patients (555, 93.3%) were males ([Table 1](#t1-epih-39-e2017001){ref-type="table"}). Baseline clinical and laboratory profiles of the patients --------------------------------------------------------- The mean time between clinical manifestations and hospital admission was 25.82 days, ranging from 2 to 365 days. One-fourth of VL patients (141, 23.7%) had a delay in diagnosis and treatment. The mean hospital stay of patients during treatment was 23.6±10.5 days. The mean hemoglobin level of patients was 8.98±4.26 g/dL ([Table 2](#t2-epih-39-e2017001){ref-type="table"}). Determinants of poor visceral leishmaniasis treatment outcomes -------------------------------------------------------------- The proportion of poor VL treatment outcomes was found to be 23.7%, of which 12.4% were death, 5.7% were treatment failure, and 5.6% were non-adherence. Delayed diagnosis and treatment, inability to walk at hospitalization, baseline platelet count, and tuberculosis and HIV coinfections were found to be significantly associated with poor VL treatment outcomes in the univariate analysis. However, in the multivariate logistic regression analysis, only delayed diagnosis and treatment, inability to walk at hospitalization, and VL/HIV coinfection remained significantly and independently associated with poor VL treatment outcomes. VL patients who were admitted to the treatment center late (≥29 days after the onset of symptoms) were 4.34 times more likely to have poor treatment outcomes than patients who were admitted early (\<13 days) (aOR,4.34; 95% CI, 2.22 to 8.46). VL patients who were unable to walk (severely ill) at admission were at a 1.63 times greater risk of poor treatment outcomes than those who were able to walk (moderately ill) (aOR, 1.63; 95% CI, 1.06 to 2.40). Similarly, HIV coinfected VL patients were at a 2.72 times greater risk of poor treatment outcomes than non-coinfected VL patients (aOR, 2.72; 95% CI, 1.40 to 5.20) ([Table 3](#t3-epih-39-e2017001){ref-type="table"}). DISCUSSION ========== Copious data on treatment outcomes and their determinants among VL patients are necessary for national control, prevention, and elimination of VL. In comparison with other similar studies conducted elsewhere in Ethiopia (18.5%) \[[@b15-epih-39-e2017001]\], poor VL treatment outcomes were found to be common (23.7%) in the current study, which was conducted in northwest Ethiopia. This proportion of poor outcomes is also higher than that found in a study conducted in Brazil (2%) \[[@b16-epih-39-e2017001]\]. This could be due to differences in design, setting, and the type of subjects involved in the study. However, the proportion found in this study was lower than that found in a study conducted in another part of Ethiopia (31.6%) \[[@b17-epih-39-e2017001]\]. This discrepancy may be due to differences in the study subjects, as the previous study was performed on HIV-VL coinfected patients, and VL/HIV coinfection increases the likelihood of poor treatment outcomes. However, the results of the present study correspond to those obtained in West Bengal, India (22.7%) \[[@b14-epih-39-e2017001]\] and Peru (24.4%) \[[@b18-epih-39-e2017001]\]. Although the current study used secondary data and was limited in the possible variables it could identify that might have had an effect on treatment outcomes, this study identified late presentation and admission to the treatment center, severe illness as measured by the functional status of patients at admission, and HIV/VL coinfection as determinants of poor VL treatment outcomes. VL patients who were admitted to the treatment center with a delay of more than a month (≥29 days) from the onset of symptoms were four times more likely to have poor treatment outcomes than patients who presented and were admitted earlier. Delays in diagnosis and treatment are a bottleneck challenge in migrant and rural communities working in remote agricultural areas where health facilities are limited and day laborers lack knowledge and awareness of the need for early admission to a health facility after the onset of clinical symptoms. The negative effects of delays in diagnosis and treatment on outcomes have been reported in Peru \[[@b18-epih-39-e2017001]\], Georgia \[[@b19-epih-39-e2017001]\], and South Sudan \[[@b20-epih-39-e2017001]\]. This suggests a need for increased community awareness of the necessity of visiting a health institution as soon as the first clinical symptoms of VL manifest. In addition, further analysis indicated that VL patients who were unable to walk at admission (severely ill) were 1.63 times more likely to have poor treatment outcomes as compared to those who were able to walk (moderately ill). This result is comparable to the findings of a study conducted in Brazil. This association is due to the fact that clinically deteriorated patients may come in at a critical stage, resulting in poor treatment outcomes and a poor prognosis \[[@b5-epih-39-e2017001]\]. This study also found that VL/HIV coinfection was a determinant of poor treatment outcomes, as VL/HIV-coinfected patients were 2.72 times more likely to have poor treatment outcomes than non-coinfected patients. This can be easily explained by the pathophysiology of the two diseases, as both HIV and VL lead to a compromised immune status in the patient and poor treatment outcomes \[[@b5-epih-39-e2017001],[@b16-epih-39-e2017001],[@b21-epih-39-e2017001]\]. This study has several limitations that should be considered when interpreting its results. The data were retrospectively extracted from patients' medical records, and some important variables such as parasite load, total leukocyte count, and CD4 count were inadequately recorded and were not included in the analysis. Moreover, a few patients were discharged and considered cured before the full treatment course was completed and/or without a TOC culture. In addition to this, only 595 of the 890 VL patients were included in the study, while the remaining patients were excluded from the study because they were referred to other facilities or their data files were missing. This may have led to bias and underestimation or overestimation of the associations with treatment outcomes if the patient exclusion conditions were related to treatment outcomes. Poor VL treatment outcomes, defined as death, treatment failure, or non-adherence, were found to be common among VL patients. Late presentation and admission to the treatment center, inability to walk at admission (severe illness), and HIV/VL coinfection were determinants of poor VL treatment outcomes. Hence, VL treatment outcomes would be improved if patients sought care at a health institution at the earliest manifestation of clinical symptoms, and if special attention were given to severely ill and VL/HIV-coinfected patients. In addition, prioritizing treatment for communities in which VL is endemic and VL patient education are advisable. Moreover, a prospective follow-up study design would be helpful for determining other important variables related to VL treatment outcomes. The authors would like to thank the University of Gondar for providing ethical clearance and Kahsay Abera Hospital for material support. The authors are indebted to the Tigray Regional Health Bureau for the letter of permission to Kahsay Abera Hospital. The authors would also like to acknowledge the data collectors and hospital staff members for their dedicated cooperation, which made the study possible. The authors have no conflicts of interest to declare for this study. ###### Demographic characteristics of visceral leishmaniasis patients at Kahsay Abera Hospital, Humera, Ethiopia, 2010-2013 (n=595) Characteristics Frequency \% ------------------ ----------- ------ Age (yr)  ≤15 29 4.9  16-45 535 89.9  ≥46 31 5.2 Sex  Male 555 93.3  Female 40 6.7 Residence  Rural 400 67.2  Urban 195 32.8 Migration status  Migrants 220 37.0  Residents 284 47.7  New settlers 91 15.3 ###### Clinical and laboratory profiles of visceral leishmaniasis patients at Kahsay Abera Hospital, Humera, Ethiopia, 2010-2013 (n=595) Characteristics Frequency \% ----------------------------------- ----------- ------ Time elapsed before diagnosis (d)  \<13 107 18.0  14-28 347 58.3  ≥29 141 23.7 General condition of patients  Able to walk 357 60.0  Unable to walk 238 40.0 Fever  No 28 4.7  Yes 567 95.3 Weight loss  No 64 10.8  Yes 531 89.2 Jaundice  No 537 90.3  Yes 58 9.7 Splenomegaly  No 98 16.5  Yes 497 83.5 Anemia  No 173 29.1  Yes 422 70.9 Hemoglobin (g/dL)  \<7.9 279 46.9  8.0-10.9 102 17.1  ≥11.0 214 36.0 Platelet count (cells/mm^3^)  \<85,000 265 44.5  ≥85,000 330 55.5 BMI (kg/m^2^)  \<18.5 511 85.9  ≥18.5 84 14.1 Tuberculosis  No 534 89.7  Yes 61 10.3 Malaria  No 469 78.8  Yes 126 21.2 HIV status  No 546 91.2  Yes 49 8.2 BMI, body mass index; HIV, human immunodeficiency virus. ###### Bivariate and multivariate logistic regression analysis of the factors associated with poor treatment outcomes among visceral leishmaniasis patients at Kahsay Abera Hospital, Humera, Ethiopia, 2010-2013 (n=595) Variables Good treatment outcome (cure) Poor treatment outcome cOR (95% CI) aOR (95% CI) --------------------------------- ------------------------------- ------------------------ ------------------- ---------------------------------------------------------------------- Age (yr)  ≥46 17 14 3.15 (1.01,9.90) 3.00 (0.80,10.50)  16-45 414 121 1.12 (0.45,2.80) 1.25 (0.45,3.50)  \<15 23 6 1.00 1.00 Duration prior to diagnosis (d)  ≥29 76 65 5.25 (2.78, 9.90) 4.34 (2.20,8.46)^[\*](#tfn3-epih-39-e2017001){ref-type="table-fn"}^  14-28 286 61 1.30 (0.70, 2.40) 1.20 (0.60, 2.20)  \<13 92 15 1.00 1.00 General condition  Unable to walk 167 71 1.74 (1.20, 2.50) 1.63 (1.06, 2.40)^[\*](#tfn3-epih-39-e2017001){ref-type="table-fn"}^  Able to walk 287 70 1.00 1.00 Weight loss  Yes 400 131 1.70 (0.80,3.50) 1.60 (0.70, 3.40)  No 54 10 1.00 1.00 Hepatomegaly  Yes 15 9 2.00 (0.85, 4.66) 1.64 (0.60, 4.20)  No 439 132 1.00 1.00 Hemoglobin (g/dL)  \<7.9 210 69 1.40 (0.89, 2.14) 0.90 (0.50,1.45)  8.0-10.9 71 31 1.80 (1.07,3.16) 1.20 (0.64, 2.15)  ≥11.0 173 41 1.00 1.00 Platelet count (cells/mm^3^)  \<85,000 188 77 1.60 (1.10, 2.40) 1.38 (0.87, 2.20)  ≥85,000 266 64 1.00 1.00 Tuberculosis  Yes 40 21 1.80 (1.03, 3.20) 1.39 (0.73, 2.67)  No 414 120 1.00 1.00 HIV/AIDS  Yes 26 23 3.20 (1.70, 5.80) 2.72 (1.40, 5.20)^[\*](#tfn3-epih-39-e2017001){ref-type="table-fn"}^  No 428 118 1.00 1.00 cOR, crude odds ratio; aOR, adjusted odds ratio; CI, confidence interval; HIV, human immunodeficiency virus; AIDS, acquired immune deficiency syndrome. p\<0.05.

One significant area of research in the multifaceted field of bilingualism over the past two decades, spanning among many others from Green ([@B3]) to Chung-Fat-Yim et al. ([@B2]), has been the demonstration, validation, and account of the so-called "bilingual advantage." This refers to the hypothesis that bilingual speakers have advanced abilities in executive functions (EF) and other domains of human cognition. Such cognitive benefits of bilingualism have an impact on the processing mechanisms active during language acquisition in a way that results in language variation. Within bilingual populations, the notion of language proximity (or linguistic distance) is also of key importance for deriving variation. In addition, sociolinguistic factors can invest the process of language development and its outcome with an additional layer of complexity, such as schooling, language, dominance, competing motivations, or the emergence of mesolectal varieties, which blur the boundaries of grammatical variants. This is particular relevant for diglossic speech communities-bilectal, bidialectal, or bivarietal speakers. The defined goal of the present Research Topic is to address whether the bilingual advantage extends to such speakers as well. Thus, "Linguistic and Cognitive Profiles for Speakers of Linguistically Proximal Languages and Varieties" become an important matter within "Developmental, Modal, and Pathological Variation." The larger issue of cognitive-linguistic representations in bilingual speakers is expressed in [Putnam et al.\'s](https://doi.org/10.3389/fpsyg.2017.02212) model for determining language proximity. Building on Hsin\'s ([@B4]) Integration Hypothesis, the authors sketch a framework in which "bilingual grammars are neither isolated, nor (completely) conjoined with one another in the bilingual mind, but rather exist as integrated source grammars that are further mitigated by a common, combined grammar." Once linguistic distance between the languages of bilingual speakers is measured in computational cognitive architectures, any effects of a bilingual advantage in terms of cognition and memory can be assessed empirically. One such empirical assessment is presented by [Bosma et al.](https://doi.org/10.3389/fpsyg.2017.01453) who investigate whether degree of bilingualism in Frisian-Dutch children influences EF---and if so, whether this effect is sustained over time. To this effect, they analyzed longitudinal data from Frisian-Dutch bilingual children. The results confirm that "cognitive effects of bilingualism are moderated by degree of bilingualism," where the amount of exposure in the minority language (i.e., m the home variety) indirectly affects bilingual children\'s cognitive development. However, as the authors stress, "the findings also demonstrate that the effect of bilingualism on EF is limited and unstable"---a take-home message that is in line with what recent reviews have suggested in relation to the bilingual advantage (Paap et al., [@B6]; Lehtonen et al., [@B5]). A set of three papers further investigates the purported bilingual advantage in combination with sociolinguistic and socio-economic considerations. [Blom et al.](https://doi.org/10.3389/fpsyg.2017.00552) tested whether the sociolinguistic context of language use affects the bilingual advantage. And indeed, bilingual children outperformed their monolingual peers on selective attention, presumably because they focused on different aspects of the task. [Garraffa et al.](https://doi.org/10.3389/fpsyg.2017.01907) explore "the effects of bilingualism in Sardinian as a regional minority language on the linguistic competence in Italian as the dominant language and on non-linguistic cognitive abilities" with adults living in Sardinia. No evidence for a "bilingual advantage" emerged through the task that tapped into the cognitive control of attention, but bilinguals did perform better than monolinguals on working memory tasks. In addition, "\[b\]ilinguals with lower formal education were found to be faster at comprehension of one type of complex sentence," while "bilinguals and monolinguals with higher education showed comparable slower processing of complex sentences." [Meir and Armon-Lotem](https://doi.org/10.3389/fpsyg.2017.01442) explore the influence of socioeconomic status (SES) and bilingualism on the linguistic skills and verbal short-term memory of Hebrew-Russian bilingual preschoolers, half from low SES backgrounds. The authors propose that bilingualism is associated with decreased vocabulary size and lower performance on verbal short-term memory tasks, while SES also impacts verbal short-term memory with lowest linguistic load. They also argue that "an unprivileged background has a negative impact on children\'s cognitive development." Effects of language or linguistic proximity, bi-/multilectal acquisition, and their relevance for the socio-syntax of language development are of particular interest to this Research Topic---that is, apparent sociolinguistic aspects such as formal schooling that may have an effect on the grammatical language development of a child growing up in a bi- or multilectal society. Considering the case of Brazilian (L1) and European (L2) Portuguese bidialectal adults that had moved to Portugal as adults, [Castro et al.](https://doi.org/10.3389/fpsyg.2017.01382) explore possible differences in the interpretation of null and overt object pronouns. They "test the extent to which \[...\] speakers display cross-linguistic influence in either direction." The high degree of typological proximity between the speakers\' linguistic varieties is argued to contribute to L1 attrition and hinder target-like L2 performance at the same time. There are also four contributions that focus on the differences between the two varieties of Greek spoken in Cyprus. When asked to make acceptability judgments, the performance of speakers of non-standard varieties may actually be subject to interference from factors such as prescriptive notions of grammatical correctness and sociolinguistic values typically attached to "dialects." Recognizing the importance of working with corpora of spontaneous speech, [Leivada et al.](https://doi.org/10.3389/fpsyg.2017.01260) investigate variation in the spontaneous productions of adult speakers of the non-standard variety Cypriot Greek. In their corpus, they observed intraspeaker realizations of different values of the same variant within the same syntactic environment; a result that is incompatible with the mainstream "triggering-a-single-value" approach of parametric models. Since the analysis of these conflicting values is ultimately a way of investigating Universal Grammar primitives, the authors further conclude that claims about the alleged unfalsifiability of Universal Grammar are empirically unfounded. [Tsiplakou](https://doi.org/10.3389/fcomm.2017.00017) explores the concept of gradient bilectalism by capitalizing on insights from recent developments in second language acquisition, particularly the suggestion that aspects of the syntax-discourse interface that are not easily accessible to the learner may lead to fossilization, even at end state. Based on quantitative data from a questionnaire survey, she suggests that imperfect acquisition of some structural aspects of the standard language may affect bilectals\' performance in a way that involves a transfer of features from the dialect to the standard. [Themistocleous](https://doi.org/10.3389/fpsyg.2017.01945) investigates the effects of two linguistically proximal Modern Greek dialects, Athenian Greek and Cypriot Greek on the temporal, spectral, and co-articulatory properties of fricatives with the aim to determine the acoustic properties that convey information about these two dialects. The results revealed that Athenian Greek and Cypriot Greek fricatives differ in all spectral properties across all places of articulation. The co-articulatory effects of fricatives on following vowel were different across the two varieties, something that suggests that dialectal information is encoded in the acoustic structure of fricatives. The contribution by [Ayiomamitou and Yiakoumetti](https://doi.org/10.3389/fpsyg.2017.02017) deals with regional linguistic variation and its implications for education by focusing on the Greek Cypriot educational context. The aim of the study was to understand Greek Cypriot primary school pupils\' sociolinguistic awareness via examination of their written production in their home variety. The students were advised to produce texts that reflected their everyday way of talking with family and friends (beyond school boundaries and the formal register this environment may induce). The authors found students\' texts to include many mesolectal features but also "a significant and unexpected number of basilectal features and instances of hyperdialectism," which rendered their texts register-inappropriate. Merging sociolinguistic and neurocognitive insights about language variation, three papers seek to uncover which factors derive variation in the course of language development, that is, how variation in cases of pathological development affects different parts of language and whether the affected markers are manifested in a comparable way. For starters, it is common to find that "minority" languages enjoy fewer (if any) diagnostic tools than "majority" languages. This has repercussions for the detection and proper assessment of children with Specific Language Impairment (SLI) brought up in these languages. With a view to remedy this situation for Catalan, [Gavarró](https://doi.org/10.3389/fpsyg.2017.01865) developed a sentence repetition task to assess grammatical maturity in school-age children. The findings display clear differences with typically developing children providing identical repetition at twice the rate of children with SLI. Moreover, the children with SLI had more deviant productions, both ungrammatical ones and grammatical yet different repetitions. [Saiegh-Haddad and Ghawi-Dakwar](https://doi.org/10.3389/fpsyg.2017.02010) tested phonological and lexical distance between a dialect of Palestinian Arabic spoken in the north of Israel and Modern Standard Arabic on word and non-word repetition in children with SLI and age-matched controls. The authors find that children with SLI underperform on all tasks and point to a "general phonological memory deficit." They also argue that the results "reflect the role of linguistic distance in phonological memory for novel linguistic units in Arabic SLI," which in turn would "support a specific Linguistic Distance Hypothesis of SLI in a diglossic setting." Previous work on linguistic abilities of individuals with Down syndrome (DS) suggests severe impairment of complex syntactic structures in a number of languages. Given difficulties reported with comprehension and production of relative clauses and object clitics in typically developing Greek Cypriot bilectal children, one could hypothesize that the bilectal environment in which children with DS grow up may cause an added difficulty in the acquisition of other complex syntactic structures, such that of the understudied syntactically complex subjunctives. [Christodoulou and Grohmann](https://doi.org/10.3389/fcomm.2018.00019) examine whether Greek Cypriot bilectal children and adolescents with DS evidence an impairment with the comprehension of subjunctive clauses, corroborating arguments for an overall syntactic impairment from past research on DS. Full analysis of the comprehension data evidenced high means of accuracy, with parallel performance across the two groups. The linguistic differences between Cypriot and Standard Modern Greek do not appear to affect the acquisition of subjunctives. As its title suggests, the Research Topic "Developmental, Modal, and Pathological Variation---Linguistic and Cognitive Profiles for Speakers of Linguistically Proximal Languages and Varieties" aimed to approach the topic of language variation from different perspectives. To this end, we brought together studies on typologically different languages (both standard and non-standard), ranging from infancy into adulthood, for speakers with different cognitive phenotypes as well as different language backgrounds (e.g., heritage languages in diaspora). The contributions to this research topic are informative with respect to certain key aspects within current linguistic research such as the bilingual advantage, the passive knowledge of the standard in bi(dia)lectal speakers, aspects of transfer, and the key role of SES in cognitive and linguistic development. As Noam Chomsky has repeatedly argued, in order to understand the human capacity to acquire and use language, we need to know *what options it permits* (Chomsky, [@B1]) through studying language variation, and this Research Topic aims to take a multidisciplinary step into this direction. Author contributions {#s1} ==================== KG wrote a first draft of this editorial, which EL completed and MK edited further. All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** Partial support for this research topic comes from the research project A Cross-Linguistic Investigation of Acceptability Judgment Variation awarded to KG, which is funded through the University of Cyprus by the A.G. Leventis Foundation. EL acknowledges funding support from the European Union\'s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 746652. [^1]: Edited and reviewed by: Manuel Carreiras, Basque Center on Cognition, Brain and Language, Spain [^2]: This article was submitted to Language Sciences, a section of the journal Frontiers in Psychology

{#sp1 .112}

Background {#Sec1} ========== Caring and supporting older people with dementia have become a major public health priority regionally and globally \[[@CR1], [@CR2]\]. The recent report by the World Health Organization and Alzheimer's Disease International \[[@CR1]\] has highlighted the need to raise awareness of dementia as a public health priority, and to advocate actions to be taken at international and national levels. Together with the recently-released World Alzheimer's Report \[[@CR2]\], the society has been alluded to the latest global and regional trends of older adults who are in great need for dementia care in the long-run, and a global epidemic of a predicted unsustainable shortage of informal carers. This report also alarmingly predicts a three-fold increase in the number of dependent older people, with a figure of 101 million in 2010 to 277 million in 2050, of whom one half of them are living with Alzheimer's Disease or other form of dementia. In Hong Kong, the current situation among Chinese older people with dementia is escalating, with a prevalence of 1.2% for the 60--65 age group and 32% for the 85-and-above age group \[[@CR3]\], suggesting that more care and attention will be needed by those with advancing age. Given the rapid growth in the population aged 65 and over in Hong Kong, of a population of 7 million, 13% are older people aged 65 years and older \[[@CR4]\]. It is expected that as older people continue to live longer, the prevalence of those suffering from dementia is expected to increase \[[@CR5]\]. The press release by the Hong Kong Alzheimer's Disease Association \[[@CR6]\] has attempted to make explicit headlines of a diminishing number and lack of support for family members. The focus on the need to plan and address both formal and informal care needs for dementia sufferers is urgently warranted \[[@CR6]\]. Particularly, this study directs specific attention to informal family caregiving of older people with dementia by taking a retrospective method to examine the decision-making experiences and processes of family members of dementia sufferers towards the use of formal CS and RCH services. Dementia is a progressive disease of old age characterized by memory loss over time, changes in behaviour, and the degree to which these older people can continue to perform activities of daily living will depend on the severity of the illness. Although caring for family members with dementia can be positive, it is increasingly clear that caregiving can be overwhelmingly cumbersome and a burden to the family caregivers as they are the sole persons in the family to often shoulder the responsibility of care \[[@CR7]\]. For example, caregivers are known to bear significant psychological, practical and economics strains as the course of the disease advances \[[@CR8]\]. In other studies, caregivers have suffered from sleep disturbance when faced with multilevel stressors \[[@CR9]\]. In addition, during the course of the dementia journey there has been a high degree of uncertainty and lack of confidence among families with coping and handling matters that concern their older relatives \[[@CR10]\]. Indeed, the plights faced by families and the need to overcome the imperative challenges of being a dementia caregiver are compounded by the general lack of timely and relevant information and appropriate support on dementia and dementia care, as supported by our pilot work that examined decision-making potential and challenges faced by family carers of older people with mild-moderate dementia \[[@CR11]\] and by other investigators \[[@CR12]\]. Nevertheless, there is a consensus to suggest that the burden of care can be relieved by using formal services (e.g. dementia-specific day care centres \[[@CR13]\]) to assist family carers to support people with dementia living in the community, and thereby continue with their care for a longer period \[[@CR14]\]. This is in line with the goal of the Hong Kong Government that puts emphasis on the family to provide care in the community through the provision of formal services. As CS and RCH services play a progressively more important role in supporting Chinese older adults with dementia and their family members, it is important to understand the decision-making processes and types of decisions that are made to better support these older people using these services. Although there is a general lack of literature about this phenomenon, studies that make reference to participation, involvement, degree of choice and control in care practices for older people can be used to inform decision-making \[[@CR15]\]. In a very recent study, 20 people with dementia or family caregivers were interviewed to explore the decision-making process about living arrangements and future place of care for the older person with dementia. The results showed that during the decision-making process, they will consider several factors, for example, desire to let the older person stay at home in a familiar environment, safe to be left at home, physical health of the older person with dementia and the carer's health condition \[[@CR16]\]. Apart from the health care/medical decision-making that seems to receive attention for this population group, other types of decisions and how they are made by the family in providing the best possible care for these older adults has not yet been explored. Currently, little is known about the respective roles of family members in decision-making for older people with moderate dementia, their level of involvement in decision-making and the influence they have on decisions that determine the dementia sufferers' lives. This study is an attempt to fill the knowledge gap. Exploring decision-making of family members of dementia sufferers is an imperative area for investigation in this study to gain deeper understanding of the complex demands and challenges associated with dementia caregiving and how families may then find relieve for themselves and for the dementia sufferers through making decisions to use different types of CS and RCH services. It is also important to understand how the families act as decision-makers for their dementia relatives by examining the processes of how decisions are made to receive the required care to meet their needs. Indeed, the literature about this phenomenon on family decision-making for dementia sufferers and the types of decisions that are related to CS and RCH services are scarce. Therefore, the objectives of this project are:▪ To explore the general decision-making experiences and types of decisions that family members usually need to make with/for older persons with dementia;▪ To identify the extent to which current CS-RCH services have helped family members to meet their own needs and challenges when serving their older relatives with dementia;▪ To explore factors influencing the family members' decision to decide on CS and/or RCH services for older persons with dementia;▪ To examine experiences and circumstances that have influenced the family to make a decision to continue using the CS-RCH services for older persons with dementia; and▪ To examine the perspectives and roles family members play and the influences they have in shaping the lives of older relatives with dementia when they use the CS-RCH services. Methods {#Sec2} ======= Study design {#Sec3} ------------ A two-year constructivist grounded theory (conGT) design will be used \[[@CR17]\] to collect the data using semi-structured interviews. This design posits that knowledge is co-created with the participants, and acknowledges them as being the expert. In this case, the family carers become the expert in care, and who are able to share their experiences in the interviews. The inductive approach to managing the data and theory building will unravel co-understanding and interpretations of the multiple meanings surrounding the phenomenon under study. Settings and participants {#Sec4} ------------------------- This study will be conducted in CS and RCH services from non-government organizations (NGOs) and a private aged home. The participants are family members of older persons with moderate dementia. Screening older persons for dementia will include a confirmed diagnosis of dementia, and a MMSE score of 11--20 for moderate cognitive impairment. Recruiting family members include those who have an older relative with moderate dementia, are immediate family members (e.g. daughters, sons, spouses or grandchildren), are identified by staff to be the contact person, and are the elder's primary carer. After the initial purposive sampling of participants, theoretical sampling will follow. In determining the sample size, based on the P.I.'s prior work on decision-making in RCHs that collected 105 qualitative interviews from residents, families and care providers from three residential care homes \[[@CR18]\], the guideline used here was to sample approximately 12 family members per site. Therefore, it is estimated that around 100 family members will be recruited from CS and RCH services in order to reach data saturation of the various service types. Data collection {#Sec5} --------------- The timeframe for this study will be 24 months. Semi-structured individual interviews will be the main data collection method and will be supplemented by field notes. All participants will be interviewed once and will be audio-taped. Examples of broad questions for the family interview were developed. The research assistant (RA) will screen the elders and then recruit family members for the pilot and main study. Pilot interviews will be conducted to revise the interview schedules. For the main study, once an elder with a confirmed diagnosis of dementia has been identified, the person-in-charge and/or their representative will be approached by the RA to identify the elder's primary caregiver and the suitability for inclusion in the study. The person-in-charge will make the initial point of contact by briefly explaining the study to the identified family member based on the information sheet and consent form. Once verbal agreement to participate in the study has been secured, the family member's contact details will be given to the RA to arrange a date, time and place for the interview. The interview schedule revolves around a number of topics as follows:Types of care provided to the older person with dementia.Types of decisions made for this older person in the past.Experiences of how the older relative was helped to make those decisions described.Own needs and challenges of being a family caregiver.Process of deciding which CS and/or RCH service to use to best meet the needs of the older person with dementia.Involvement of family carer in making decisions about the services their older relative will need.Types of decisions made for the older relative now that they are using CS-RCH services.Role and influences family carer have made in shaping the older person's life.Experiences and circumstances that have been influential in continuing to use the CS-RCH services. Ethical considerations {#Sec6} ---------------------- Ethical approval to conduct the study was given by the Research and Ethics Committee of Caritas Institute of Higher Education on (3 October 2014 \[Ref. no. PW/14/E0008\]). Informed written consent will be obtained from the individual participants. Data analysis {#Sec7} ------------- All the interviews will be transcribed verbatim and its accuracy will be checked by rechecking the scripts against the taped-recordings. Data will then be analyzed using constant comparative analysis methods to generate concepts and to develop the theory through an inductive process of defining, categorizing, comparing data, and explaining and seeking relationships in the data \[[@CR17]\]. A systematic approach will be used by the investigators to fully understand the data, compare all the variations in the data, and accounts for the related properties in the categories. 'MAXQDA The Art of Text Analysis' will be used to code and categorize the data and later to facilitate data interpretation and report writing. Rigor {#Sec8} ----- The quality criteria established to judge grounded theory studies include: 'credibility', 'originality', 'resonance' and 'usefulness' in line with Charmaz's work will be followed \[[@CR17]\]. This is a set of evaluative criteria which allow the investigators to explore the broader impact and social relevance of a project. Discussion {#Sec9} ========== This 2-year conGT study on decision-making of family carers towards the use of CS & RCH services for older people with moderate dementia will provide pointers and deeper conceptual understanding to explain the processes and the experiences family members have been through (if in a RCH) or are still at the heart of providing care to their loved ones (if using CS). The final product, the theory, gives an impression that this is all about an academic and impractical endeavour. On the contrary, the resultant theory is highly practical and places emphasis on ensuring the concepts that emerge are closely grounded in the participants' data \[[@CR19]\]; that is, the theory encompasses the stories of our family members who are the experts in caring. However, it needs to be noted that the phenomenon 'decision-making' is an abstract concept to the laymen. Particularly, in this study, some family members of our older adults are themselves in their old age and may encounter difficulty in their understanding of the term. Therefore, inviting them to provide illustration of events and perspectives of decision-making that they have undertaken will ensure that the data obtained will have greater meaning. Deeper understanding of issues including, but not exclusive to, service needs, expectations and hopes among family carers for improving service support to serve dementia sufferers in CS and RCH services will be revealed. Indeed, care that is provided by a spouse to the older person with dementia can be entirely different from that provided to a parent, particularly when the offspring's gender is different from the parent they care for. Therefore, a larger sample size from various sites is planned in this study. Despite this, data collection will stop when data saturation is reached. An anticipated constraint posed upon sampling may be the difficulty of getting a balanced number of male to female carers, especially when female carers have known to dominate. As participants are recruited from multiple sites which will enhance the transferability of the study, the constant comparative analysis method and theoretical sampling will facilitate the investigators to identify the most appropriate participants to interview and thus enhance the credibility of the study and the identification of the decision-making process towards the use of CS and RCH services. There are two limitations identified. First, since participants are interviewed once, details and subtle changes in the decision-making process have to be based on the participants' ability to recall incidents that have happened at different points of time. Second, only participants who have identified themselves as the primary caregivers of the older persons with dementia are interviewed. However, some of them may not be the principal decision-makers for the older adults' major welfare, particularly when these caregivers are financially dependent on other family members. For future studies, it is recommended to interview the primary decision-makers for the older persons' major welfare in addition to the older person's other primary caregiver(s) if they are of different persons. As stated at the outset of this paper, currently there is a lack of support for family caregivers of older persons with dementia. After understanding caregivers' caring experiences and their decision-making process, interventions can be planned to support these caregivers. If caregivers are better supported, the need for CS or RCH services can be deferred to a later stage. Indeed, dementia is a chronic, progressive and irreversible condition. Once caregivers have the feeling of burnout, they may totally leave the care of their older relative in the hands of professional care staff such as opting to transit into RCHs. Consequently, supporting caregivers may facilitate them to continue their care for the older relatives even after they are using formal CS or RCH services. Ultimately, this study will seek to illustrate the practical and strategic aspects of the theory and how it may be useful to transfer its applicability to various service settings to better support those who deliver formal and informal care to the dementia population. CS : Community support RA : Research assistant RCH : Residential care home Not applicable. Funding {#FPar1} ======= This project was fully supported by a grant from the Food and Health Bureau and the Health and Medical Research Fund of the Hong Kong Special Administrative Region, China (Project No.12130661) from 2015 to 2017. Availability of data and materials {#FPar2} ================================== Only materials relating to the study protocol is reported in this paper. There is no raw data reported. Issues pertaining to this paper can be made to the corresponding author on reasonable request. Authors' contributions {#FPar3} ====================== LPLL led the drafting of this manuscript. All authors contributed to the manuscript, and reviewed and approved the final version of the paper. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== Ethical approval to conduct the study was given by the Ethics and Research Committee of the Caritas Institute of Higher Education (Ref. no.: PW/14/E0008). Participants are required to sign a consent form to indicate their willingness to participate. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Potential conflict of interest: M.D., L.W., U.B., and N.V. are listed as coinventors on a patent held by the Academic Medical Center in Amsterdam: "Method for diagnosing IgG4‐related diseases", application no. PCT/EP2013/070384, publication no. WO/2014/ 049177. This research was supported by grants from the German Crohn Colitis Society (section PSC patients) and the American PSC Partners patient organization (to U.B.). E.C. was funded by the Wellcome Trust. E.B. was funded by the Medical Research Council UK, Oxford NIHR BRC, The Jenner Institute, and the Oxford Martin School. M.D. and N.V. were supported by BTCURE, a research project from the Innovative Medicines Initiative Joint Undertaking (grant no. 115142‐2). S.G. was supported by the Netherlands Organization for Scientific Research (Vidi) and the European Research Council (Starting grant). [See Editorial on Page 340](http://onlinelibrary.wiley.com/doi/10.1002/hep.28624/full) AIP : autoimmune pancreatitis ANOVA : analysis of variance AUC : area under the curve BCR : B‐cell receptor CCA : cholangiocarcinoma CA : biliary/pancreatic malignancies cDNA : complementary DNA CP : chronic pancreatitis HISORt : histology, imaging, serology, other organ involvement, response to treatment IAC : IgG4‐associated cholangitis IgG4 : immunoglobulin G4 IgG4‐RD : IgG4‐related disease IQR : interquartile range NGS : next‐generation sequencing PSC : primary sclerosing cholangitis qPCR : quantitative polymerase chain reaction ROC : receiver operating characteristic SSC : secondary sclerosing cholangitis sIgG4 : serum IgG4 ULN : upper limit of normal Immunoglobulin G4 (IgG4)‐related disease (IgG4‐RD) is a multiorgan inflammatory disease that frequently affects the biliary tree (IgG4‐associated cholangitis; IAC) and pancreas (autoimmune pancreatitis; AIP).[1](#hep28568-bib-0001){ref-type="ref"}, [2](#hep28568-bib-0002){ref-type="ref"} Its pathogenesis is enigmatic. Clinical presentations of IAC and AIP show striking similarities to those of primary/secondary sclerosing cholangitis (PSC/SSC), cholangiocarcinoma (CCA), chronic pancreatitis (CP) of other cause, and pancreatic carcinoma (PC). No single diagnostic test can accurately diagnose biliary and pancreatic manifestations of IgG4‐RD. Elevated serum IgG4 (sIgG4) levels are neither sensitive nor specific for IgG4‐RD[3](#hep28568-bib-0003){ref-type="ref"}, [4](#hep28568-bib-0004){ref-type="ref"}, [5](#hep28568-bib-0005){ref-type="ref"}, [6](#hep28568-bib-0006){ref-type="ref"}; up to 30% of IAC/AIP patients have normal sIgG4 levels. Furthermore, increased sIgG4 levels are present in 10%‐20% of patients with PSC and biliary/pancreatic malignancies (CA). Given that these conditions are far more prevalent than IgG4‐RD, the positive predictive value of sIgG4 levels is poor when only moderately elevated. Composite scores, such as the HISORt (histology, imaging, serology, other organ involvement, response to treatment) criteria,[7](#hep28568-bib-0007){ref-type="ref"}, [8](#hep28568-bib-0008){ref-type="ref"} are regarded as diagnostic standards, and a consensus on diagnostic pathological features is available.[9](#hep28568-bib-0009){ref-type="ref"} Still, misdiagnosis is common, and 1 in 3 patients with IgG4‐RD has undergone major interventions because of suspected malignancy before diagnosis.[7](#hep28568-bib-0007){ref-type="ref"}, [8](#hep28568-bib-0008){ref-type="ref"} Therefore, ready‐to‐use tests with high diagnostic accuracy are urgently needed. Moreover, because relapse after tapering of immunosuppressive therapy occurs in 50% of IAC/AIP patients,[10](#hep28568-bib-0010){ref-type="ref"} biomarkers for treatment response and disease activity are highly desired. Recently, we identified dominant IgG4^+^ B‐cell receptor (BCR) clones in peripheral blood of 6 patients with active IAC, but not in healthy or disease controls, using next‐generation sequencing (NGS).[11](#hep28568-bib-0011){ref-type="ref"} This novel technique enables identification of IgG4^+^ clones within the BCR repertoire. Of note, the same clones were encountered in the inflamed tissue.[11](#hep28568-bib-0011){ref-type="ref"} Consequently, this test could represent a novel diagnostic tool. Here, we aimed to validate our previous findings in a prospective, case‐control study. Furthermore, we evaluated the diagnostic accuracy of a novel, more affordable, and widely applicable test based on quantitative polymerase chain reaction (qPCR) and compared it with that of sIgG4 levels. Moreover, we evaluated whether the qPCR test may be used to monitor treatment response. Patients and Methods {#hep28568-sec-0002} ==================== STUDY SUBJECTS {#hep28568-sec-0003} -------------- Consecutive patients meeting the HISORt diagnostic criteria for IAC[7](#hep28568-bib-0007){ref-type="ref"} and/or AIP[8](#hep28568-bib-0008){ref-type="ref"} were prospectively enrolled in this case‐control study. In line with HISORt criteria, three diagnostic pathways were possible: (1) patient had undergone a pancreatic/biliary resection for a presumed malignancy or a core biopsy showing diagnostic pathological features of AIP/IAC (group A)[7](#hep28568-bib-0007){ref-type="ref"}, [8](#hep28568-bib-0008){ref-type="ref"}; (2) patient showed classical imaging for AIP in combination with elevated sIgG4 levels of \>1.4 g/L (group B)[7](#hep28568-bib-0007){ref-type="ref"}, [8](#hep28568-bib-0008){ref-type="ref"}; and (3) patient had suspicion of pancreatic disease and/or strictures of bile ducts together with two or more of the following: elevated sIgG4 levels, suggestive pancreatic imaging findings, other organ involvement, and/or bile duct biopsy with \>10 IgG4^+^ B cells/high‐power field in combination with marked improvement of pancreatic morphology/biliary strictures and biochemical parameters upon 4 weeks of corticosteroid treatment (group C).[7](#hep28568-bib-0007){ref-type="ref"}, [8](#hep28568-bib-0008){ref-type="ref"} Moreover, patients had active disease as confirmed by laboratory and/or radiographical findings. Patients were naïve to immunosuppressive treatment or experienced a recurrence of symptoms with laboratory/radiographic changes requiring a higher dose of immunosuppression. Controls consisted of patients with PSC and CA. PSC patients were diagnosed according to the 2009 European Association for the Study of the Liver Clinical Practice Guidelines "Management of cholestatic liver diseases," based on serological findings (cholestatic serum liver test profile; supportive: atypical perinuclear antineutrophil cytoplasmic antibody), combined with typical findings of multifocal strictures and dilatations on magnetic resonance cholangiopancreaticography and/or endoscopic retrograde cholangiopancreaticography and exclusion of causes for SSC.[12](#hep28568-bib-0012){ref-type="ref"} Diagnosis in patients with CA was confirmed by histopathology (biopsy or resection material) and/or cytology (brush cytology or fine needle aspiration biopsy) together with imaging. In addition, consecutive patients with chronic pancreatitis diagnosed according to the guidelines of the American Pancreatic Association based on suggestive history, abnormal pancreas physiology, and abnormal imaging findings[13](#hep28568-bib-0013){ref-type="ref"} were included. Control patients did not use immunosuppressive treatment. Enrollment took place at the Department of Gastroenterology and Hepatology at the Academic Medical Center (University of Amsterdam) in Amsterdam, The Netherlands, and at the John Radcliffe Hospital (Oxford University) in Oxford, United Kingdom. The study was approved by the local medical ethical committees (MEC AMC 10/007, NL 31142.018.09; Oxford research ethics RECA 10/H0604/51); all study subjects gave written informed consent before inclusion, and the study was performed according to the principles of the Declaration of Helsinki. SERUM IgG4 LEVELS {#hep28568-sec-0004} ----------------- sIgG4 levels (upper limit of normal \[ULN\] = 1.4 g/L) were measured in 122 patients by automated nephelometry in the local laboratory (using BN ProsPec Siemens in Amsterdam and BNII Siemens in Oxford). For 3 patients, a sIgG4 level was only measured after start of treatment, not at time of RNA collection. PERIPHERAL BLOOD SAMPLING AND HANDLING {#hep28568-sec-0005} -------------------------------------- Peripheral blood was collected in PAXGene Blood RNA tubes (catalog no. 762165; PreAnalytiX, Breda, The Netherlands) at the Academic Medical Center in Amsterdam and in Tempus Blood RNA tubes (catalog no. 4342792; Life Technologies, Paisley, UK) at the John Radcliffe Hospital in Oxford and handled according to manufacturer\'s instructions. Simultaneously or at least within 1 week, serum was collected for IgG4 measurement. Processing of blood RNA tubes was performed by a trained technician masked for the origin of the samples or outcome of the sIgG4 levels. In a similar fashion, the person performing the data analysis (M.D.) was blinded. Complementary DNA (cDNA) was synthesized from 1,000 ng of total RNA input using Superscript III RT (Invitrogen Life Technologies, Carlsbad, CA). LINEAR AMPLIFICATION AND NGS {#hep28568-sec-0006} ---------------------------- The linear amplification protocol has been extensively described earlier.[14](#hep28568-bib-0014){ref-type="ref"} Samples were prepared for sequencing according to the manual for amplicon sequencing and sequenced on a Roche Genome Sequencer FLX (Titanium platform; Roche Diagnostics, Almere, The Netherlands). For each sample 10,000 BCR~heavy~ sequences were analyzed. NGS visualizes expanded BCR clones as a deviation in the repertoire because the BCR sequence they carry is determined more frequently than low abundant BCR sequences and thus account for a relatively large proportion of all the sequences found. Moreover, it is not unlikely that plasmacytoid cells are identified in peripheral blood; these cells produce increased amounts of BCR messenger RNA, producing a comparable deviation in the repertoire as expanded B cells. For clarity, we will use the term "dominant clone" to denote these clones. qPCR {#hep28568-sec-0007} ---- IgG and IgG4 primers for qPCR were designed to specifically amplify sequences that encode the constant region of the heavy chain of the receptor and according to common standards for primer design ([Supporting Fig. S1](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)). Additionally, the following rules were applied: (1) two different reverse primers were designed, one for all possible IgG subtypes and one specific for the IgG4 subtype, and (2) the two reverse primers were designed on virtually the same position to allow comparable conditions. After optimization, this resulted in one universal *forward* primer (5′‐GCTGCCTGGTCAAGGACTAC‐3′), one generic IgG *reverse* primer (5′‐TCTTGTCCACCTTGGTGTTG‐3′), and one specific IgG4 *reverse* primer (5′‐CTACGTTGCAGGTGTAGGTCTTC‐3′). Duplicate qPCR reactions were performed in a 10‐μL total volume in the presence of 10 pmol of the forward and reverse primers, cDNA from 50 ng RNA input, and 5 μL of SensiFAST SYBR Lo‐ROX reagent (catalog no. BIO‐94005, Bioline; GC Biotech, Alphen aan den Rijn, Netherlands) for 40 cycles (95°C for 2 minutes, 40 cycles \[95°C for 5 seconds, 60°C for 10 seconds, and 72°C for 20 seconds\], followed by a melting curve \[95°C for 5 seconds, 65°C for 1 minute, and 97°C continuous\]), using the LightCycler 480 system (Roche Diagnostics, Almere, The Netherlands). Both for IgG and IgG4, the starting concentrations in the sample were calculated using LinRegPCR software[15](#hep28568-bib-0015){ref-type="ref"} and used to calculate the percentage of IgG^+^ RNA molecules that were IgG4^+^. Thus, the more expanded IgG4^+^ clones were present in blood, the higher the percentage IgG4 RNA message of total IgG RNA message. BIOINFORMATICS PIPELINE AND DATA ANALYSIS {#hep28568-sec-0008} ----------------------------------------- The bioinformatics pipeline used to obtain the BCR sequences was described in detail[16](#hep28568-bib-0016){ref-type="ref"} and contains four modules: MID sorting; identification of gene segments; CDR3 detection; and removal of artefacts. Immunoglobulin isotype homology was determined using the National Center for Biotechnology Information\'s open‐access Web tool BLASTn (megablast algorithm) and reference sequences for the human immunoglobulin heavy‐chain constant regions, allowing a sequence homology \>97%.[16](#hep28568-bib-0016){ref-type="ref"} STATISTICAL ANALYSIS {#hep28568-sec-0009} -------------------- Following criteria for (non)parametric analysis, values are expressed as mean ± SD or median and interquartile range (IQR). Differences between groups were analyzed using Student *t* test, Mann‐Whitney *U* test, one‐way analysis of variance (ANOVA), Kruskal‐Wallis test of ranks, or χ^2^ test, where appropriate. A receiver operating characteristic (ROC) curve was used to determine cut‐off values for diagnosis of IgG4‐RD. Diagnostic accuracy was expressed in terms of sensitivity and specificity rates with 95% confidence intervals. Graphpad Prism (GraphPad Prism (version 6; GraphPad Software Inc., La Jolla, CA) and PASW Statistics software (version 22; SPSS, Inc., Chicago, IL) were used to perform statistical analyses. Two‐sided *P* values of \<0.05 were considered statistically significant. This study was performed in agreement with the Standards for Reporting of Diagnostic Accuracy. Results {#hep28568-sec-0010} ======= A total of 50 IgG4‐RD patients meeting the HISORt criteria for IAC and/or AIP were prospectively included at two departments of Gastroenterology and Hepatology in Amsterdam and Oxford between December 2011 and December 2015. In total, 15 patients were included in HISORt group A, 13 in HISORt group B, and 22 in HISORt group C. Forty‐eight patients with PSC, 27 with CA, and 10 patients with CP served as disease controls. Patient characteristics are described in [Supporting Tables S1 and S2](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo). Sequencing the BCR repertoire (n = 34 IgG4‐RD patients; n = 34 controls) revealed multiple dominant IgG4^+^ BCR clones within the IgG^+^ repertoire of IgG4‐RD patients (Fig. [1](#hep28568-fig-0001){ref-type="fig"}A) compared to disease controls (Fig. [1](#hep28568-fig-0001){ref-type="fig"}B). This is reflected in a strikingly higher percentage of IgG^+^ BCR clones being IgG4^+^ (Fig. [1](#hep28568-fig-0001){ref-type="fig"}C), a greater fraction of the IgG^+^ repertoire being taken in by IgG4^+^ BCRs (Fig. [1](#hep28568-fig-0001){ref-type="fig"}D), and a higher rank of the most dominant IgG4^+^ BCR clone in IgG4‐RD patients compared to controls (Fig. [1](#hep28568-fig-0001){ref-type="fig"}E), confirming our previous findings.[11](#hep28568-bib-0011){ref-type="ref"} {#hep28568-fig-0001} In order to distinguish IgG4‐RD patients from disease controls on a widely available platform, a qPCR protocol was developed that estimates the fraction of total IgG RNA that encodes IgG4 (qPCR test). Using this qPCR test in blood samples of a Dutch discovery cohort---consisting of 15 IgG4‐RD patients, 7 PSC patients, and 8 CA patients (including patients published earlier for NGS analysis only[11](#hep28568-bib-0011){ref-type="ref"})---resulted in clear differences between the three groups ([Supporting Fig. S2](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)). The cutoff for the qPCR test was determined at 5% of all IgG^+^ RNA molecules being IgG4^+^ using an ROC curve. Subsequently, the cut‐off value was validated in additional cohorts from two clinical sites (Dutch cohort: 26 IAC/AIP, 11 PSC, and 19 CA patients; British cohort: 9 IAC/AIP and 30 PSC; [Supporting Fig. S2](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)). The overall sensitivity and specificity of all cohorts combined was 94.0% and 98.7%, respectively (χ^2^, 109; *P* = 1\*10^−29^; Fig. [2](#hep28568-fig-0002){ref-type="fig"}A). The area under the curve (AUC) value of 0.991 (Fig. [2](#hep28568-fig-0002){ref-type="fig"}B) indicated excellent accuracy. Post‐hoc analysis showed a cut‐off value ranging from 3.5% to 6% had the best clinical utility indices, depending on the aim to avoid false positives or negatives. Additional subanalysis in 40 IgG4‐RD patients with pancreatic involvement and 10 CP patients was performed, and showed comparable results for both BCR repertoire analysis as well as the qPCR test ([Supporting Fig. S3](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)). {#hep28568-fig-0002} Comparing the findings of the qPCR test to a well‐used marker, sIgG4, the latter showed a sensitivity and specificity for sIgG4 of 85.7% and 73.3% (χ^2^, 41; *P* = 9\*10^−11^; [Supporting Fig. S2](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)), respectively. Because sIgG4 is part of the diagnostic criteria for IgG4‐RD (HISORt groups B and C), we performed a subanalysis on the different diagnostic subgroups. This showed that the sensitivity of sIgG4 in patients in whom sIgG4 was not part of the diagnostic criteria was lower (group A: sensitivity, 67%; χ^2^, 9; *P* = 0.005; group B: sensitivity, 100%; χ^2^, 25; *P* = 5\*10^−7^; group C: sensitivity, 91%; χ^2^, 28; *P* = 1\*10^−7^), whereas the qPCR test performed equally in all three groups (group A: sensitivity, 87%; χ^2^, 69; *p* = 9\*10^−13^; group B: sensitivity, 100%; χ^2^, 81; *P* = 1\*10^−14^; group C: sensitivity, 96%; χ^2^, 86; *P* = 5\*10^−19^; [Supporting Fig. S4](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)). Subanalysis of follow‐up samples in 20 Dutch IAC/AIP patients 4 and 8 weeks after start of corticosteroid therapy revealed an IgG4 RNA decrease from 20.3% ± 13.4% of total IgG RNA (mean ± SD) before treatment to 5.8% ± 2.2% after 4 weeks and 1.9% ± 0.8% after 8 weeks (Fig. [2](#hep28568-fig-0002){ref-type="fig"}C; *P* \< 0.0001). All patients responded well and showed clinical improvement with resolution of symptoms, together with a decrease of biochemical markers ([Supporting Fig. S5](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)). In conclusion, analysis in 34 IgG4‐RD cases and 34 disease controls using NGS confirms our former findings[11](#hep28568-bib-0011){ref-type="ref"} that both the number and size of dominant clones are increased in peripheral blood of IgG4‐RD patients, compared to disease controls. The rank of the most dominant IgG4^+^ BCR clone clearly discriminates between IgG4‐RD and disease controls. A simple qPCR test in 50 patients and 85 controls showed comparable diagnostic properties and, in addition, can be used to monitor treatment response. Discussion {#hep28568-sec-0011} ========== The current study in 135 patients confirms recent findings[11](#hep28568-bib-0011){ref-type="ref"} that the presence of dominant IgG4^+^ BCR clones in peripheral blood determined by NGS clearly distinguishes patients with IgG4‐RD of the biliary tree and pancreas from disease controls with similar clinical presentation, such as pancreatobiliary cancer or PSC. Our novel qPCR test adds to these findings by providing an accurate and affordable test for the diagnosis of IgG4‐RD and monitoring of treatment response. The observation that a quarter of the IgG^+^ BCR repertoire is taken up by IgG4^+^ sequences in IgG4‐RD led us to speculate that IgG4^+^ RNA, rather than the released protein, may be used to diagnose IgG4‐RD early and accurately. Therefore, we developed a simple‐to‐use qPCR test that appeared to correlate better with diagnosis than sIgG4 titers, currently the best test available in peripheral blood. Given that the majority of patients in our cohort were included in the absence of a resection or core biopsy, a relatively high proportion of IgG4‐RD patients had elevated sIgG4 levels. A subanalysis in the diagnostic groups revealed that sensitivity of sIgG4 levels in the resection/core biopsy group in whom sIgG4 did not play a diagnostic role was limited (67%), in line with other cohorts. This cannot only be explained by the fact that a definite diagnosis of IgG4‐RD in patients with normal sIgG4 needs histology, but it is also known that sIgG4 levels are lower in IgG4‐RD patients after resection for a presumed malignancy given that at least part of the cells producing the sIgG4 are removed. In disease control groups with PSC and CA, however, sIgG4 levels, but not IgG4 RNA, were elevated, contributing to low specificity of sIgG4 levels. This highlights again the usefulness of this qPCR test as a simple and accurate test for diagnosis and monitoring in an otherwise difficult‐to‐diagnose disorder. Three IgG4‐RD patients with elevated sIgG4 had a qPCR score below the threshold of 5%, and repertoire analysis performed in 2 patients showed no dominant IgG4^+^ clones among the top 10 of IgG^+^ clones. One of the 3 patients was diagnosed with CCA 32 months after fulfilling the HISORt criteria for IgG4‐RD. This emphasizes the need for even more accurate diagnostic criteria than the HISORt criteria and the potential of the presented qPCR test. Although most PSC patients with elevated sIgG4 values had normal IgG4/IgG RNA ratio, their qPCR scores were, on average, higher than in PSC patients with normal sIgG4. It has been suggested that PSC patients with elevated sIgG4 levels have worse clinical outcome in terms of disease severity and time to transplantation compared to PSC patients with normal levels.[18](#hep28568-bib-0018){ref-type="ref"} Whether this also holds true for the IgG4/IgG RNA ratio needs further investigation, but the only false‐positive qPCR score in the PSC group belonged to a patient with severe fibrosis and chronic inflammation on liver histology. Moreover, 1 PSC patient was identified with elevated sIgG4 and an infiltrate of \>30 IgG4^+^ B cells on liver biopsy, a phenomenon that is regularly observed in liver explants of PSC patients with a severe disease course.[19](#hep28568-bib-0019){ref-type="ref"} Indeed, the qPCR score of 3.93% in this patient was relatively high. The patient was recently transplanted because of decompensated cirrhosis and progressive liver failure. Both our previous and current findings on IgG4^+^ BCR clones are supported by a recent study confirming that plasmablasts are present in blood of IgG4‐RD patients.[20](#hep28568-bib-0020){ref-type="ref"} Our previous observation that a vast majority of IgG4‐RD patients were "blue‐collar workers" chronically exposed to solvents, industrial gases, or oil products[21](#hep28568-bib-0021){ref-type="ref"} (confirmed by actual data of [Supporting Table S2](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo)) supports our view of specific immune responses underlying IgG4‐RD and may explain the advanced age and male predominance of IgG4‐RD patients.[21](#hep28568-bib-0021){ref-type="ref"} Moreover, the abnormal response to food and animal antigens in IgG4‐RD deserves attention.[22](#hep28568-bib-0022){ref-type="ref"} In conclusion, identification of dominant IgG4^+^ BCR clones in IgG4‐RD has enabled us to develop an affordable qPCR test, which may allow accurate diagnosis of IgG4‐related disease of the biliary tree and pancreas. {#hep28568-sec-1011} Author names in bold designate shared co‐first authorship. Supporting information ====================== Additional Supporting Information may be found at [onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo](http://onlinelibrary.wiley.com/doi/10.1002/hep.28568/suppinfo). ###### Supporting Information ###### Click here for additional data file. The excellent technical support by Rebecca E. Esveldt is gratefully acknowledged. We express our gratitude to Erik A.J. Rauws, Cyriel Y. Ponsioen, Martin H. Houben, Frans van der Heide, Henk‐Marijn J. De Jonge, A. Boudewijn de Vries, Alaa Alkhalaf, and Marianne J. van Heerde who kindly referred patients for this study. This work was carried out on the Dutch national e‐infrastructure with the support of SURF Foundation. [^1]: These authors contributed equally. [^2]: These authors contributed equally as senior authors.

Introduction {#Sec1} ============ The survivorship curve is a useful visualisation of the frequency distribution of the age classes of a population (Rauschert [@CR30]) and is calculated as *l*~*x*~ = *n*~*x*~/*n*~0~, where *n*~*x*~ is the number of individuals in the study population who survive to the beginning of age category *x* and *n*~0~ is the number of newborns. If the population is stable, then survivorship curves describe how the numbers of individuals of a cohort decline with time. When the logarithm of the number of survivors is plot against age, then three distinct, idealized "types" are distinguished (Type-I, Type-II and Type-III (Deevey [@CR8]); Fig. [1a](#Fig1){ref-type="fig"}). Survival curves, by deduction, give some indication of the rate at which mortality increases with age and, therefore, the rate of senescence of the population. It has also been stated that variation in survival curves reflects species sensitivity to the genetic and environmental factors that have shaped their evolutionary history (Demetrius [@CR9]). The survival curve of modern humans is described as a classic "Type-I", where the probability of survival is high until relative old age, whereby it then declines rapidly, which is typical of many large mammals. A Type-I survival curve is also typical of pre-industrial human populations (e.g., the 1751 Swedish population in Fig. [1a](#Fig1){ref-type="fig"}) and ancient societies, such as hunter-gatherer, forager-horticulturalist and acculturated hunter-gatherer societies, a modern example being the indigenous Hadza population of Tanzania (Gurven and Kaplin [@CR15], Fig. [1a](#Fig1){ref-type="fig"}). Our closest living relative species, the chimpanzee, has a survivorship curve that is variable, depending on whether the population is wild or captive (Thompson et al. [@CR38]), but the wild examples are arguably closer to a Type-II (Hill et al. [@CR18]; Bronikowski et al. [@CR4]), which is described by a constant proportion of individuals dying over time. For illustrative purposes we generated an example Type-II population with a 10% probability of mortality from one age class to the next, which aligns closely with the chimpanzee population (Fig. [1a](#Fig1){ref-type="fig"}). Ancestral human populations studied from archaeological specimens, for example the Libben site skeletal sample, which is radiocarbon dated to between 800 and 1100 CE, have been described as Type-II populations (Lovejoy et al. [@CR23]). However, the remains from such sites are not always representative of the total population (Howell [@CR20]); hence, survivorship curves derived from archaeological specimens are not considered in this study. By comparison, Type-III populations display very high mortality at young ages, but those that do survive to adulthood go on to have a relatively long life expectancy, which is typical of tree and insect species. The difference between the Type-I survivorship curves of hunter-gatherer and pre-industrial populations and the Type-I survivorship curve of modern industrial populations is difficult to quantify and identifies a limitation of this approach when considering the evolutionary demography of a species.Fig. 1**a** Comparison of survivorship curves for six populations. The US females from 2000 (black, Templeton [@CR37]) represent a Type-I population. The Swedish females from 1751 (orange, Human Mortality Database (<http://www.mortality.org>)) and Hadza females (blue, Blurton Jones [@CR3]) represent Type-I populations prior to the increased longevity seen in modern populations. The wild Chimpanzee (purple, Bronikowski et al. [@CR4]) represents our closest living species. The Type-II population (red, simulated (see Methods for details)) is used as a close approximation of the chimpanzee population and the Type-III population (green, simulated (see Methods for details)) is used for completion as a comparator at the other end of the 'type' spectrum. **b** A continuum of lifespan equality (ln(1/*H*)) and life expectancy in six populations presented in **a**. **c** A comparison of 'pace' (life expectancy) and 'shape' (ratio of longevity to life expectancy) for six populations in **a** An early approach to quantifying the information contained within survivorship curves was to use its logarithm, for example Keyfitz's entropy, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H = \frac{{ - \mathop {\smallint }\nolimits_0^\infty \left( {{\rm ln}l_x} \right)l_xdx}}{{\mathop {\smallint }\nolimits_0^\infty l_xdx}} = - \frac{{e^\dagger }}{{e_0}}$$\end{document}$ (Keyfitz [@CR42], Goldman and Lord [@CR14]), described as the ratio of life expectancy lost due to death (*e*^†^) to that of life expectancy (*e*~0~). This more quantitative approach to describing age-distributions has recently been developed to distinguish between the 'pace' and 'shape' of change of populations as they age (Baudisch [@CR1]; Baudisch et al. [@CR2]). The 'pace' of life, or how fast populations age, can be measured as life expectancy (*e*~0~), which captures the average length of life, or the tempo at which organisms survive and reproduce, placing organisms on a fast/slow continuum of ageing (Baudisch [@CR1]; Colchero et al. [@CR7]). On the other hand, the 'shape' describes the direction and degree of change in mortality (Wrycza et al. [@CR41]) and hence captures the rate at which a species senesces (Baudisch [@CR1]). There are various ways of measuring the 'shape' of ageing, all of which are highly correlated (Wrycza et al. [@CR41]). Figure [1b, c](#Fig1){ref-type="fig"} illustrates two methods of measuring 'shape'; one is generally referred to as life table entropy (sometimes lifespan equality), which can be calculated as ln(1/Keyfitz's entropy) (Wrycza et al. [@CR41], Colchero et al. [@CR7]). Another is the ratio of longevity (Ω)/life expectancy (*e*~0~), where Ω is the age at which, for example, 95% of the adults have died. Although there are many measures (Wrycza et al. [@CR41]), we present these two as they have previously been used for cross-species comparisons (Baudisch [@CR1]; Colchero et al. [@CR7]). As can be seen from Fig. [1](#Fig1){ref-type="fig"}, there is some correspondence between these pace/shape metrics and the survivorship curves: the wild chimpanzee population clusters with the Type-II example; the pre-industrial/hunter-gatherer Type-I populations cluster together and the modern US Type-I population stands apart from all of these. For the populations considered here, life expectancy increases from Type-III to Type-II to Type-I, as expected, and shows that population Type corresponds more obviously with notions of the 'pace' of life. There are, however, some important subtleties. For example, the distinct Type-II and Type-III survivorship curves presented in Fig. [1a](#Fig1){ref-type="fig"} have similar 'shape' (Fig. [1b, c](#Fig1){ref-type="fig"}). Because Type-II populations show constant mortality with age, they are representative of populations that show negligible senescence. Baudisch ([@CR1]) points out that long-lived species typically present negligible senescence, for example, long-lived trees and marine invertebrates mostly show Type-III survivorship. For this reason Type-II and Type-III populations should cluster on the 'shape' axis. However, there are clear exceptions to this rule, modern human populations being one as, although they are long-lived, they have a shape score indicative of strong senescence. As helpful as pace and shape are at capturing important aspects of a species life-history, we believe they do not lend themselves so readily to visual comparisons of survival/mortality with age where the corresponding reproductive distribution is being studied. Here we explore the influence of the reproductive distribution, specifically generation time, on the rate of genetic drift for human populations, with current and hypothetical ancestral survival curves. The aim is to identify the influence of population 'type' and generation time on mutation accumulation and, as a consequence, the corresponding changing role of mutation accumulation on the senescence of these populations. The current longevity that modern humans experience came late in human evolution, where for the first time during the Upper Palaeolithic (\~50,000 to 10,000 years ago) there are a larger number of older adults amongst the deceased than there are younger adults (Caspari and Lee [@CR5]). Evidence from throughout the late Archaic up to the Upper Palaeolithic also indicates that mortality patterns for young (pre- 40 years) versus old (post- 40 years) did not alter during this period and that older individuals would have been rare (Trinkaus [@CR39]). If the wild chimpanzee/Type-II survivorships are indicative of our common *ancestral* type, then judging by modern hunter-gatherer societies and the evidence from Upper Palaeolithic humans (Caspari and Lee [@CR5]), it seems reasonable to assume that the Type-I distribution is generally reflective of human populations since the Neolithic period (\~500 generations ago). This has implications for the evolution of ageing in human populations as well as the laboratory models used to study it. We hypothesise that a Type-II survivorship curve displaying a constant mortality rate and relatively few older reproductive adults describes the majority of our ancestral demography, up until the Palaeolithic. A Palaeolithic, demographic shift towards the modern Type-I survivorship curve would then have followed this, with the older, parental age classes now numerically better represented. The greatest shift in modern human evolution may not simply be one of increasing population size, but rather a change in the survivorship curve (Type-II to Type-I), which has implications for the genetic drift of deleterious mutations, and hence mutation accumulation. Standard population genetics theory tells us that the allele frequencies of mutations entering small populations will drift to a greater degree (greater stochasticity) than those entering large populations, as genetic diversity is lost at a rate proportional to 1/2*N*~*e*~, where *N*~*e*~ is the effective population size. A mutation with a negative impact on the fitness of the homozygous genotype (e.g., aa) relative to the wild-type homozygote (e.g. AA), measured as *s*, is effectively removed by natural selection whenever *s* \> 1/ 2*N*~*e*~, else drift dominates the rate of loss (Hartl and Clark [@CR17]). If we consider a snapshot of a population with overlapping generations, then we expect to see a monotonic decline in the number of individuals with increasing age due to the unavoidable causes of mortality that individuals encounter with the passing of time. This then leads to a proportional decline in the effectiveness of selection with age due to the relatively few individuals of older age contributing to the genetics of future generations, known as Hamilton's principle (Hamilton [@CR16]). Because selection removes mutations more effectively when they have a detrimental phenotypic effect early on in life compared with mutations that only affect older individuals, detrimental mutations accumulate (Medawar [@CR25]; Charlesworth and Williamson [@CR6]). Importantly, the negative affect of these mutations may persist in post-reproductive ages, contributing to the ageing phenotype. The effect of drift on late-acting deleterious mutations is one of potential inflation of their frequencies, where older age-classes suffer a greater loss of health relative to the younger age classes and hence where smaller populations age at a faster rate than larger populations. Lohr et al. (2014) explicitly tested this hypothesis using *Daphnia magna* and, consistent with other recent work (Jones et al. [@CR21]) identified a correlation with age at first reproductive output and rate of ageing across numerous wild and model organisms, which is consistent with the expectation that populations with low genetic diversity have accelerated ageing. The main conclusion of Jones et al. ([@CR21]) is that the onset and rate of senescence in both survival and reproduction are associated with generation time, an aspect that is not usually considered with the 'pace' or 'shape' of ageing in populations. Given that generation time influences *N*~*e*~ (Waples and Yokota [@CR40]), where generally *N*~*e*~ ∝ *N*~*nb*~*T*, where *N*~*nb*~ is the number of newborns arriving in each generation and *T* is the generation time (mean age of parents), then we expect effective population size to decline with shorter generation times. It then becomes conceivable that the age of sexual maturity, as well as the survivorship and reproductive span of a species, have consequences for the role of mutation accumulation in ageing phenotypes. The phenotypic effect of a mutation entering a population is dependent upon when the gene is expressed, which can be age-specific (e.g., developmental genes, genes associated with sexual maturity and female menopause), although not always, as when a gene's influence can be cumulative (e.g., IL1RAP and its influence on amyloid plaque accumulation (Ramanan et al. [@CR29])). When the mutation has an age-specific expression, its survival depends upon the size of this age class. As a population declines in size, the age classes that are large enough for selection to dominate drift (i.e., *s* \> 1/2*N*~*e*~) will shift towards the younger classes. Mutations that have a detrimental effect on survival and reproduction are therefore more likely to persist in the older age classes. According to traditional ecology theory, population density governs the optimal age of maturity (Macarthur and Wilson [@CR24]), distinguishing *r*-selected species (typically small organisms producing many offspring, with early maturity and with short lifespans) from *K*-species (reproduce slowly at later ages and with longer life-spans) (Pianka [@CR27]). Most primate species, including humans, are *K*-species. This *r*/*K* categorisation can be quite sensitive to environmental factors. For example, an increase in environmental stochasticity can select for an *r*-type strategy (Engen and Saether [@CR11]). Although this *r*/*K* categorisation of life histories persists in some of the ageing literature (e.g., Reichard [@CR31]), there is now a preference for placing species on a fast--slow continuum, or tempo, of life history. One method is to use the ratio of fertility rate to age at first reproduction, which among other factors is less sensitive to environmental stochasticity (Oli [@CR26]). Which of these *r*/*K* strategies, or position on the fast--slow continuum, a species adopts may be highly constrained by their environment and evolutionary history. Nevertheless, we hypothesise that there will have been situations, e.g., during the Neolithic period and modern industrialization, where the change in culture, environment and reproductive patterns on human survival would have been of sufficient magnitude that the species survivorship curve shifted, resulting in consequent changes in both the 'pace' and 'shape' of their life history. The consequence is that, once a modern-industrialised life-history strategy emerged, the older reproductive age classes once poorly represented in a Type-II population and potentially subject to the consequences of mutation accumulation, become numerically better represented in a Type-I population. Using simulations, we explore the significance of this shift on the rate of genetic drift and the contribution mutation accumulation is likely to make to the ageing, or senescence, of the population. The value in doing so is that any insights relating to the magnitude of influence past demographic shifts have had on the present evolution of ageing in the human population would better inform the choice of model organism employed in its study. Methods {#Sec2} ======= Felsenstein's method of estimating the effective population size of an age-structured population (Felsenstein [@CR12], equation 10) was used to measure the influence of generation time on the rate of drift across populations with Type-I and Type-II survival curves:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N_e = \frac{{N_{nb}T}}{{1 + \mathop {\sum }\nolimits_{x = 1}^k l_xs_xd_xv_{x + 1}^2}},$$\end{document}$$where, for *k* age classes, *N*~*nb*~ is the number of newborns, *d*~*x*~ = *l*~*x*~ -- *l*~*x*+1~ and *v* the reproductive value: $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$v_x = \mathop {\sum }\limits_{i = 1}^k l_im_i/l_x.$$\end{document}$ Because our focus was the hypothetical shift in population type during human evolution, Type-III populations were not considered in any further detail. Following convention, we consider the female constituents of a population of parents and offspring. The number of females of each age (*x*) is denoted as *n*~*x*~ (equivalent to the number of females that survive each age class, *s*~*x*~), with the fecundity of each age class denoted as *m*~*x*~. The probability of surviving *to* each age class from birth is *l*~*x*~, which is simply *l*~*x*~ = *n*~*x*~*/n*~0~. The probability of surviving each age *x* is *P*~*x*~ = *l*~*x*~*/l*~*x*−1~. By convention, *l*~0~ = 1. For simplicity, we consider sampling the number of recruits for the next generation post-breeding. With a birth-pulse model, the effective fecundity of females of age *x* is *f*~*x*~ = *P*~*x*−1~.*m*~*x*~. The net reproductive rate of the population, per generation, is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$R_0 = \mathop {\sum }\nolimits^ l_xm_x$$\end{document}$. Importantly, *R*~0~ is the change in population size by *generation*, *T*, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T = \frac{{\mathop {\sum }\nolimits_{x = 0}^k xl_xm_x}}{{\mathop {\sum }\nolimits_{x = 0}^k l_xm_x}}$$\end{document}$, and where *k* is the total number of age classes. Hence, the intrinsic rate of population growth is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$r = \frac{{{\rm ln}R_0}}{T}$$\end{document}$. Importantly for this discussion, this shows clearly that as *T* increases, the intrinsic rate of population growth (*r*) decreases, hence, for population sizes to remain stable, as they do for our simulations, a change in *T* requires a compensatory change in *m*~*x*~. In order to discriminate between the effect of young and old breeders on the loss of genetic diversity, we explicitly explore the influence of generation time and survivorship by employing a Leslie matrix approach. We use three example survivorship curves to illustrate the influence they may have on the rate of genetic drift. Life history data were obtained from 1) the US female population census from 2000 (Templeton [@CR37]), considered to be typical of a Type-I population where there is high survival until late in life. 2) The Hadza female population (modelled by Blurton-Jones [@CR3]) and, for comparison, 3) a fictional population representative of a typical Type-II population with a constant 10% mortality rate over yearly age classes: *l*~*x*+1~ = 0.9 × *l*~*x*~, designed to be similar to a wild chimpanzee population (Fig. [1a](#Fig1){ref-type="fig"}). The age classes (*x*) are 1 year apart. A Type-III population was also generated for comparison in Fig. [1](#Fig1){ref-type="fig"} as *l*~*x*~ = *x*^--2.126^. The fecundity schedule, which is the tabulation of birth rates (*m*~*x*~), is manipulated similarly to Ryman (Ryman [@CR34]), where the fecundity of each age class of each population is adjusted such that $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop {\sum }\nolimits^ l_xm_x = 1$$\end{document}$. Here our focus is on the variability in survival curves, so the fecundity trajectories for Hadza and Type-II populations are a manipulation of the US female trajectory. The US female fecundity schedule is taken from Templeton ([@CR37]) and simulated as a normal distribution with mean = 27 years and variance = 7 years. The simulated populations have stable age-structure and constant size. Assuming the age-specific survival and fecundity probabilities remain constant across generations (*t*), we track the number of females (*n*~*x*~), starting with age class *x*+1, by multiplication with a Leslie matrix for 100 generations to reach a stable age distribution. We then track the frequency of two alleles, *A* and *B*, at a single locus over a 1000 year time period by multiplying our Leslie matrix **L**, by a mating matrix **M**, such that **n**(*t* *+* *1*) = **LMn**(*t*) (Roughgarden [@CR33]):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{*{20}{l}} {\left( {\begin{array}{*{20}{l}} {n_{_t + 1,x,AA}} \hfill \\ {n_{_t + 1,x,AB}} \hfill \\ {n_{_t + 1,x,BB}} \hfill \\ {n_{_t + 1,x + 1,AA}} \hfill \\ {n_{_t + 1,x + 1,AB}} \hfill \\ \vdots \hfill \\ {n_{_t + 1,k - 1,BB}} \hfill \end{array}} \right)} \hfill & = \hfill & {\left( {\begin{array}{*{20}{l}} {f_{x,AA}} \hfill & {f_{x,AB}} \hfill & {f_{x,BB}} \hfill & {f_{x + 1,AA}} \hfill & {f_{x + 1,AB}} \hfill & \cdots \hfill & {f_{k - 1,BB}} \hfill \\ 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ {P_{x,AA}} \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ 0 \hfill & {P_{x,AB}} \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ \vdots \hfill & \vdots \hfill & \vdots \hfill & \vdots \hfill & \vdots \hfill & \ddots \hfill & \vdots \hfill \\ 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & {P_{_k - 1,BB}} \hfill & 0 \hfill \end{array}} \right)} \hfill \\ {} \hfill & {} \hfill & { \times \left( {\begin{array}{*{20}{l}} {p^2} \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ {2_{pq}} \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ {q^2} \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ 0 \hfill & 0 \hfill & 0 \hfill & 1 \hfill & 0 \hfill & \ldots \hfill & 0 \hfill \\ 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 1 \hfill & \ldots \hfill & 0 \hfill \\ \vdots \hfill & \vdots \hfill & \vdots \hfill & \vdots \hfill & \vdots \hfill & \ddots \hfill & \vdots \hfill \\ 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & 0 \hfill & \ldots \hfill & 1 \hfill \end{array}} \right) \times \left( {\begin{array}{*{20}{l}} {n_{t,x,AA}} \hfill \\ {n_{t,x,AB}} \hfill \\ {n_{t,x,BB}} \hfill \\ {n_{t,x + 1,AA}} \hfill \\ {n_{t,x + 1,AB}} \hfill \\ \vdots \hfill \\ {n_{t,k - 1,BB}} \hfill \end{array}} \right)} \hfill \end{array}.$$\end{document}$$As described in detail in (Roughgarden [@CR33]), the number of newborn females at time *t*+1 is $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$b_{t + 1} = \mathop {\sum }\nolimits_{x,ij} f_{x,ij}n_{t,ij}$$\end{document}$ for alleles *i, j*. The allele frequencies amongst the newborn are simply the result of Mendelian segregation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p_{t + 1} = \frac{{\mathop {\sum }\nolimits_x f_{x,{\rm AA}}n_{x,{\rm AA}} + \left( {1/2} \right)\mathop {\sum }\nolimits_x f_{x,{\rm AB}}n_{x,{\rm AB}}}}{{b_{t + 1}}}.$$\end{document}$$This describes the random union of gametes between parents across all age classes and maintains Hardy-Weinberg genotypic ratios among the newborn, presented in the mating matrix **M**. We performed all calculations and iterations of the matrices in *R* (R Development Core Team [@CR28]). The R Script and input data are available from <https://github.com/AndyOverall/DriftAgeStruct>, along with GNU public license details. The simulation of genetic drift is a two-stage process. The first deals with the random union of gametes to generate the newborn class. The second models the random culling of alleles with time such that the allele frequency distribution in the breeder's age class (e.g., *n*~*t=20,\ x=20*~) is a random subsample of the allele frequency distribution of the age class when they were newborns (*n*~*t=1,\ x=1*~). The number of newborn individuals is kept constant and takes the value of the first age class of the stable age distribution (i.e., the one resulting from 100 iterations of the Leslie matrix multiplication: **n**(*t* = 100) = **Ln**(*t* = 99)). For example, considering the US Female population, where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathop {\sum }\nolimits_{x = 1}^k n_{100,x} = 1000$$\end{document}$, the number of newborns (age class *x* = 1) = *n*~100,1~ = 13, with the numbers of genotypes being in accord with Hardy-Weinberg proportions: *n*~100,1,AA~ + *n*~*1*00,1,AB~ + *n*~100,1,BB~ = 13. The simulation of random genetic drift involves the random sampling of these 13 genotypes using the *R* function **rmultinom**, **rmultinom**(n, size, prob). If *X* is a random variable such that *X* = {*n*~AA~, *n*~AB~, *n*~BB~}, then **X** = **rmultinom**(1,13,*n*~AA~/13,*n*~AB~/13,*n*~BB~/13) regenerates the newborn's genotypes subsequent to drift and from these genotypes the new allele frequencies are obtained. These frequencies then feed directly into the mating matrix **M**. If *P*~*x*~ and *m*~*x*~ remain constant, iterations of this procedure simulate the frequency of the alleles under the influence of drift. The simulation included 1000 repeats, to model 1000 reproductive events, each iterated 1000 times to correspond with the passage of 1000 years. In addition to allele frequency stochasticity occurring in the generation of newborn's genotypes, there is a random cull of genotypes between age classes over generations. For example, with neutral alleles, the number of genotypes of age class *x* = 2 in generation 1 (*n*~*1,2*,AA~; *n*~*1,2*,AB~ and *n*~*1,2*,BB~) each has a probability (*P*~x~) of surviving to the next generation (*n*~*2,2*,AA~; *n*~*2,2*,AB~ and *n*~*2,2*,BB~). This is simulated, in this example, by *n*~*2,2,ij*~ = *P*~1~ × *n*~*1,2,ij*~. Then, *P*~x~% of *n*~*2,2,ij*~ remain the same genotypes, with the remaining 1--*P*~x~% being randomly drawn using the same *R* function **rmultinom**. The simulation of selection involved a modification of the survival probability. For example, for neutral alleles, the survival of the genotypes from one age class (*x*) to the next is equal (e.g., *P*~*x*,AA~ = *P*~*x*,AB~ = *P*~*x*,BB~). Negative selection for individuals that are homozygous for recessive alleles is simulated as *P*~*x*,BB~ \< *P*~*x*,AA~, *P*~*x*,AB~. Figure [2](#Fig2){ref-type="fig"} illustrates the survival curves for the three populations where the total population size is *N* = 1000. The middle dashed curve shows the reproductive distribution for the US female population (where the *y*-axis frequency scale is arbitrary), which corresponds to a generation time $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$T = \frac{{\mathop {\sum }\nolimits_{x = 0}^k xl_xm_x}}{{\mathop {\sum }\nolimits_{x = 0}^k l_xm_x}} = 27$$\end{document}$. Shifting the reproductive distribution either towards younger or older age classes modified the generation times. For example, the dashed curve of Fig. [2](#Fig2){ref-type="fig"} furthest to the left shows the reproductive distribution shifted 20 years younger and, for the US females, *T* = 8.8 (referred to as T-20). When shifted 20 years towards the more elderly individuals, shown by the dashed curve to the right, *T* = 46.8 (referred to as *T* + 20). This manipulation separates out the influence of young versus older breeders, whilst acknowledging that the corresponding plausibility of such an early/late age of reproduction may be unrealistic.Fig. 2Survivorship curves for three populations: US Female population from 2000, Hadza females and a Type-II population with survivorship of *l*~*x*+1~ = 0.9 × *l*~*x*~. Age (*x*) is in years. Central dashed curve shows the reproductive distribution of the US female population, where the frequency scale on the *y*-axis is arbitrary. Left hand curve shows the reproductive distribution shifted 20 years earlier (*T*−20) and the right hand curve shows the reproductive distribution shifted 20 years later (T+20) Results {#Sec3} ======= Setting *N* = 1000 for each of the three populations, Felsenstein's estimate of *N*~*e*~ (Eq.[1](#Equ1){ref-type=""}) results in a simple linear increase in *N*~*e*~ with generation time (*T*) for both Type-I populations (US and Hadza females), but plateaus for the Type-II (Fig. [3](#Fig3){ref-type="fig"}). This comes as no surprise for both Type-I populations as the survivorship curve barely changes over the age range considered here (10--60 years), and so *N*~*e*~ ∝ *T*, as the other parameters in Eq. [1](#Equ1){ref-type=""} change very little with increasing *T*. However, for the Type-II population, after an initial increase in *N*~*e*~ with *T*, the relationship plateaus and *N*~*e*~ becomes largely independent of generation time as a corresponding increase in reproductive value (*v*~*x*~) balances the increase in *N*~*e*~ with *T* (results not shown).Fig. 3The change in *N*~*e*~ with generation time for three population types: US Female population from 2000, Hadza females and a Type-II population with survivorship of *l*~*x*+1~ = 0.9 × *l*~*x*~. Each population has a total population size of *N* = 1000. Generation = generation time calculated as $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$= \frac{{\mathop {\sum }\nolimits_{x = 0}^k xl_xm_x}}{{\mathop {\sum }\nolimits_{x = 0}^k l_xm_x}}$$\end{document}$ The expected loss-of-heterozygosity (*H*) over time is a function of *N*~*e*~, e.g., $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$H_{t + 1} = H_t\left( {1 - \frac{1}{{2N_e}}} \right)$$\end{document}$ (Gillespie [@CR13]). However, for age-structured populations, the number of newborns (*N*~*nb*~) and reproductive values (*v*~*x*~), as well as generation time influence estimates of *N*~*e*~. Although Felsenstein (and others, e.g., (Hill [@CR19]) have formulated estimates of *N*~*e*~ for age-structured populations, it is not immediately obvious how this translates to a loss of heterozygosity over time. For *N* = 1000, the estimate of *N*~*e*~ is larger for the Type-II population than Type-I populations across the generation times considered here (Fig. [3](#Fig3){ref-type="fig"}), leading to the expectation that drift would proceed at a slower rate for the Type-II population. Figure [4a](#Fig4){ref-type="fig"} shows the decline in mean heterozygosity over time for populations of early (*T* − 20) and late (*T* + 20) breeding individuals across the three populations. Figure [4b](#Fig4){ref-type="fig"} presents notched boxplots to summarise the distribution of heterozygosity values at the final time point (Time = 1000) for 1000 replicates. The "notch" of the boxplots span the 95% confidence interval of the median and the box itself spans the interquartile range. Informally, if the notches do not overlap, then it is considered that the medians differ with 95% confidence. These plots show that for the Type-I populations, the earlier breeding populations lose heterozygosity, on average, at a markedly greater rate than later breeding populations and that these earlier breeding populations are highly variable in the rate of this loss. However, for Type-II populations, the distribution of heterozygosity values for populations with early breeders is indistinguishable from that of late breeding populations. For early breeding populations (T−20), the rate of loss is consistent with the Felsenstein *N*~*e*~ estimates presented in Fig. [3](#Fig3){ref-type="fig"}: the smaller the *N*~*e*~, the greater the rate of loss of heterozygosity. However, Fig. [4a](#Fig4){ref-type="fig"} shows that for later breeding populations this is not the case and in fact the rate of drift is not accurately predicted from estimates of *N*~*e*~, with US females drifting the least and Type-II the most.Fig. 4**a** Mean decline in heterozygosity from 1000 age-structured simulations where the starting frequency is *P*~*B*~ = 0.5, *N* = 1000, where reproductive distributions have been shifted towards the young (*T*−20, solid lines) and the old (*T*+20, dashed lines). Time is in years. **b** Notched boxplots summarising each of the 1000 simulations run for each population. The black, blue and red dots indicate the means for the US, Hadza and Type-II populations, respectively With the introduction of negative selection into the simulations, we expect to see heterozygosity lost at a greatest rate in populations, where *s* \> 1/2*N*~*e*~ and to be independent of population type when s ≫ 1/2*N*~*e*~. When the recessive mutation reduces the survival of breeding individuals homozygous for this mutation (*P*~*x*,BB~) by 0.1% (*s*), the rate at which heterozygosity is lost is not altered noticeably (result not shown). However, Fig. [5a](#Fig5){ref-type="fig"} shows the decline in heterozygosity when *s* = 1% for early (*T* − 20) and late (*T* + 20) breeding populations where, as predicted, population type is still having some influence, more noticeably for the *T* + 20 populations where the distribution of heterozygosity values at time point *t* = 1000 are distinct between Type-I and Type-II populations (Fig. [5b](#Fig5){ref-type="fig"}).Fig. 5**a** Mean decline in heterozygosity from 1000 age-structured simulations where the starting frequency is *P*~*B*~ = 0.5, *N* = 1000, where reproductive distributions have been shifted towards the young (*T*−20, solid lines) and the old (*T*+20, dashed lines). Homozygous individuals (BB) have a reduced survival probability (*s* = 0.01). Time is in years. **b** Notched boxplots summarising each of the 1000 simulations run for each population. The black, blue and red dots indicate the means for the US, Hadza and Type-II populations, respectively Discussion {#Sec4} ========== The two main genetic causes of senescence, antagonistic pleiotropy and mutation accumulation, have different expectations with regard to the frequency distribution of mutations that are deleterious to the older age classes of a species (Rodrguez et al. [@CR32]). Here we considered the consequence of population 'type' on the rate of drift in light of this influence on mutation accumulation. We hypothesised that the human survivorship curve has shifted relative to pre-industrialised and hunter-gatherer populations and possibly further back in time from ancestral populations that had survivorship curves more akin to chimpanzees. The literature on the theory underpinning estimates of *N*~*e*~, the rate-determining parameter of drift, is extensive (e.g., Engen et al. [@CR10]), but there has been little attention paid towards the influence of population type (e.g., those that fit the ecological Type-I, II and III survivorship ideals). The potential importance of this lay in the fact that most ageing studies have an understandable bias towards human ageing, but the industrial-age human population type is unusual in being an extreme example of Type-I (see Fig. [1](#Fig1){ref-type="fig"}). Not only is this shift in population type likely to have consequences for the rate of mutation accumulation, there are also consequences relating to the appropriate choice of organism used to model the ageing of human populations. For example, a model organism's evolutionary history shapes its life history, including its survivorship. Ancestral population type may then contribute to the genetic causes of the model organism's senescence, which may be at variance with the evolutionary history of humans. When Type-I and Type-II age-structured populations are of comparable size (e.g., *N* = 1000), then the number of newborns (*N*~*nb*~, age class *x* = 1) in Type-I populations are fewer than the *N*~*nb*~ of Type-II populations (Fig. [2](#Fig2){ref-type="fig"}). The relationship: *N*~*e*~ ∝ *N*~*nb*~*T* (Felsenstein [@CR12]) predicts that Type-I populations will drift at a greater rate than Type-II populations. Figure [3](#Fig3){ref-type="fig"} shows the results of applying Felsenstein's estimate of *N*~*e*~ to the three population types showing that Type-I and Type-II populations differ in how sensitive these estimates of *N*~*e*~ are to changes in generation time, with the allele frequencies in Type-II populations drifting almost independently of generation time throughout the majority of the age range considered here. The reason for this is that the reproductive value of the breeding individuals (*v*~*x*~ in the denominator of Eq. [1](#Equ1){ref-type=""}) differs markedly between the young and elderly of a Type-II population relative to that of a Type-I. For the scenarios presented here, where populations are constrained to remain at a constant population size (e.g., *N* = 1000), an elderly breeder in a Type-II population is required to make a much greater contribution to the next generation in terms of offspring number, relative to an elderly breeder in a Type-I population. Hence, the increase in the reproductive value in Type-II populations balances the increase in *N*~*e*~ expected as a consequence of an increase in generation time (numerator of Eq.[1](#Equ1){ref-type=""}), which is why the relationship between these two parameters (*T* and *N*~*e*~) weakens (Fig. [3](#Fig3){ref-type="fig"}). Figure [4](#Fig4){ref-type="fig"} shows the results of drift simulations for the populations presented in Fig. [3](#Fig3){ref-type="fig"} (modern US females, Hadza females and a Type-II population). For the populations with an early generation time (T -- 20, solid lines, Fig. [4a](#Fig4){ref-type="fig"}) the rate of drift corresponds with estimates of *N*~*e*~ presented in Fig. [3](#Fig3){ref-type="fig"}. However, for the populations with a later generation time (T + 20, dashed lines, Fig. [4](#Fig4){ref-type="fig"}), the mean rates of drift are the opposite of expectations based purely on the relative magnitudes of *N*~*e*~ (Type-II \> Hadza females \> US females), although the distributions of 1000 simulations performed here do overlap (Fig. [4b](#Fig4){ref-type="fig"}). It is not obvious what the cause of this reversal of mean heterozygosity value is. However, the ordering of the populations in terms of their rates of drift does appear to correspond with the number of breeders (*N*~Breeders~), indicated by the overlapping reproductive distributions in Fig. [2](#Fig2){ref-type="fig"}. For example, considering the T − 20 populations, the number of breeders throughout the reproductive distribution (dashed lines, Fig. [2](#Fig2){ref-type="fig"}) is ordered as Type-II \> Hadza females \> US females. However, the number of breeders throughout some of the T + 20 population's reproductive distribution shows a reversed ordering. Taken together, the plots in Figs. [3](#Fig3){ref-type="fig"} and [4a](#Fig4){ref-type="fig"} illustrate the magnitude of difference that population types have on the rate of drift and indicate a few subtleties that may be difficult to predict directly from Felsenstein's estimate of *N*~*e*~. This outcome appears to be a simple consequence of the fact that, despite having equal census sizes (*N*), populations with differing *N*~*nb*~/*N*~breeders~ ratios will drift at differing rates and that population 'type' captures this ratio. For mutations that are selectively neutral with regards to the breeding individual's probability of survival (i.e., those implicated in mutation accumulation), a shift in survivorship from a Type-II to a Type-I, all else being equal, will correspond with an increase in the rate of drift unless this shift also corresponds with an increase in generation time (Fig. [4a](#Fig4){ref-type="fig"}). A corresponding increase in generation time reduces the rate of drift for Type-I populations. Going from a Type-II to Type-I population, the 'pace' of life decreases as life expectancy increases, but the 'shape' steepens indicating an increase in the strength of ageing (Baudisch [@CR1], see Fig. [1b, c](#Fig1){ref-type="fig"}). This shift in population from Type-II to Type-I, therefore, corresponds to an increase in senescence. With Type-I populations, unlike Type-II, the contribution that mutation accumulation makes to senescence does appear to be sensitive to generation time. It follows that Type-I populations with younger generation times will experience drift at a greater rate than those with older generation times and hence a greater contribution of mutation accumulation to senescence is expected. Considering the transition from a pre-industrial/hunter-gatherer human population to a modern-industrialised population, this also results in a shift towards slower pace and steeper shape (Fig. [1](#Fig1){ref-type="fig"}). However, the drift simulations suggest that, in terms of the rate of the heterozygosity loss, this shift does not correspond to a marked change in the contribution mutation accumulation may make to senescence. Although the pace/shape metrics used to tease apart how fast and how strongly populations senesce indicate an almost linear change from Type-II to pre-industrial/hunter-gatherer to modern populations (Fig. [1b, c](#Fig1){ref-type="fig"}), our results suggest that the contribution of mutation accumulation to senescence is dependent upon population type and, for Type-I populations, this is also dependent upon generation time. Modern human populations may have 'pace' and 'shape' metrics distinct from pre-industrial/hunter-gatherer populations, but both Type-I populations drift at similar rates and are equally sensitive to generation time (Fig. [4](#Fig4){ref-type="fig"}). The relationship *s* \> 1/2*N*~*e*~ describes the conditions required for selection to dominate drift in allele frequency evolution. However, as we have shown, the relative rates of drift between age-structured populations do not always correspond to the relative effective sizes. Figure [5a](#Fig5){ref-type="fig"} shows that when negative selection (*s* = 0.01) is experienced, the rate at which the mutation is lost from the populations still bears the hallmarks of the differential rates of drift between population types. For the older breeding populations (T + 20), the Type-II and Hadza populations have noticeably distinct distributions of heterozygosity values at time point *t* = 1000 (Fig. [5b](#Fig5){ref-type="fig"}). With strong selection (e.g., *s* = 0.1), as predicted, selection dominates the loss of heterozygosity and all three populations evolve at a similar rate and with similar sensitivity to generation time (result not shown). Hence, unless selection is strong relative to 1/2*N*~*e*~, population type will continue to influence the allele frequency distribution of detrimental alleles. Our treatment of life-table, actuarial senescence is possibly naive in that it simply relates to the increased mortality with age, but is nevertheless consistent with a large body of literature drawn upon for this study (e.g., Baudisch [@CR1]; Wrycza et al. [@CR41]; Colchero et al. [@CR7]). More sophisticated analyses of senescence, based upon detailed life history data, have arisen that move away from this traditional life-table approach where, for example, generation time was identified as capturing the 'speed of living' and to be tightly associated with the onset and rate of senescence (Jones et al. [@CR21]). One explicit metric of this is the ratio of fertility rate to age at first reproduction (F/α). Because α is proportional to generation time (*T*), this 'pace of life' metric slows with increasing *T*. This result is consistent with ours, although we consider this in terms of the rate of drift. Metrics such as F/α are valuable in predicting the rate of senescence, but they do not tease apart the genetic underpinnings of senescence. We show that a population's sensitivity to drift and generation time is a function of its 'type', which in turn relates to the relative importance of mutation accumulation as a cause of senescence. Our focus in this paper has been on the contrast between Type-I and Type-II survivorship curves. It remains to be explored how variation in the reproductive distribution (other than just generation time) influences drift. Although the reproductive distribution employed throughout this manuscript corresponds well with modern and pre-industrialised human populations, the chimpanzee reproductive distribution is quite different as it spans their entire adult life (Bronikowski et al. [@CR4]). Fitting a modern reproductive distribution to a Type-II population is likely to be an unrealistic contrivance and in need of further investigation. Nevertheless, the general principle that manipulating survivorship and generation time can influence the relative contribution of mutation accumulation to senescence opens up the possibility of laboratory investigations that could have some bearing on the evolution of human senescence. Data archiving {#Sec5} -------------- The R Script and input data are available from <https://github.com/AndyOverall/DriftAgeStruct>, along with GNU public license details. **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. **Change history** 5/7/2019 This article was originally published under standard License to Publish, but has now been made available under a CC BY 4.0 license. The PDF and HTML versions of the paper have been modified accordingly. We thank David Waxman and David Clancy for insightful comments. We also wish to thank three anonymous reviewers. The authors declare that they have no conflict of interest.

Background {#Sec1} ========== It is widely accepted that there is a vast gap in child mental health and psychosocial support provision in conflict-affected Low and Middle Income Countries (LAMIC), given the consistently higher levels of psychological distress and mental disorders identified in such settings (Belfer [@CR6]; Patel et al. [@CR24]). A recent systematic literature review demonstrated a lack of field-tested or researched intervention models for children in armed conflict (Jordans et al. [@CR20]). Several publications have advocated comprehensive multi-layered care programs (Betancourt [@CR7]; de Jong [@CR11]; Fairbank et al. [@CR12]; IASC [@CR17]; Saltzman et al. [@CR25]). A layered system of care entails provision of complementary supports for sub-populations depending on severity of mental health problems. In practice, however, most reported interventions for children in conflict-affected settings follow a single-intervention approach (Jordans et al. [@CR20]). Multi-layered interventions are necessary because of the wide-ranging impact of war and the limited capacity of the informal and formal health care systems in resource-poor settings. A recent World Health Organization (WHO) rapport states that interventions should not be provided as freestanding activities, but delivered in a variety of packages and through different levels of a health system to provide a way of scaling up interventions in LAMIC (WHO [@CR35]). Moreover, stepped care models have been introduced as an attempt to maximize efficiency of resource allocation in therapies with lower cost interventions implemented first and more costly interventions reserved for those insufficiently helped by the initial intervention (Haaga [@CR14]). Lack of awareness by parents and children of mental health problems and services, lack of understanding, recognition and management by primary health care workers and teachers of mental health problems, and the stigma and discrimination related to mental health have been reported as primary barriers to child and adolescent mental health care in LAMIC (Tyano and Fleishman [@CR29]). Goldberg and Huxley ([@CR13]), in their classic filter-model, present an example of multi-layered pathways to mental health care. It demonstrates how potential patients are selected, through a set of filters, from one level of care to another. Following this combination of pathways and barriers to care, we present a care-utilization model for a community-based care package for children in war-affected areas (Burundi, Sudan, Sri Lanka and Indonesia). This package follows a public health approach including primary-, secondary- and tertiary-prevention interventions (de Jong [@CR11]) and aimed at translating guidelines for complex emergencies (IASC [@CR17]; Weiss et al. [@CR33]) into a replicable delivery framework. Figure [1](#Fig1){ref-type="fig"} depicts utilization of the care package, including filter mechanisms for different levels of care. Level 1 comprises mental health promotion activities for the population at-large. Level 2 consists of peer-resilience group activities (ranging from recreational activities to peer group theme-centred discussions) to prevent healthy, albeit at-risk, populations to develop psychosocial problems. In order to reach as many children as possible in a systematic way the program targets school-going children. As not all children are school going, access constitutes the first filter. Level 3 targets subgroups of children within the school setting that present with elevated levels of psychosocial problems and who are subsequently offered to participate in a 15-session Classroom-Based Intervention (CBI) to reduce psychological distress (Macy et al. [@CR23]). A primary screener was developed and used (see below) for early detection of children between level 2 and 3. Screening presents the second filter. Level 4 represents treatment of sub-groups or children with more severe psychological distress that require more focused individualized care (i.e. counselling or family/parental support) that can be delivered by training para-professionals (Ali et al. [@CR1]). Filter 3 represents the detection, by para-professional CBI facilitators, of children in need of follow-up care due to inability to participate in CBI or sustained problems after termination of CBI, subsequently referred to counselling. Finally, a small group of children will require specialized psychiatric services to treat mental disorders. Mental health facilities in community settings in LAMIC and child and adolescent mental health professionals in specific are sparse (Patel et al. [@CR24]), therefore the limited referral options present the fourth filter in the model. The program was funded by Plan Netherlands, a child-focused development agency concentrating on the promotion of child rights in low-income settings, and took place between September 2004 and August 2009. It was designed as a pilot project to develop a multi-layered psychosocial care response for children in areas of armed conflict, with the aim of sustaining the care package in the four countries and replicating it in other similar settings. For a detailed description on the content of the care delivery framework we refer to previous publications (see: [www.psychosocialcarechildren.org](http://www.psychosocialcarechildren.org)).Fig. 1Utilization and evaluation of care package This paper provides a practice-driven evaluation of a multi-layered psychosocial care package for children in four conflict-affected countries. Evaluation of complex interventions that exist of several components of care present with methodological and feasibility challenges, as different levels of care require different indicators, and consequently tend to be marked by a paucity of high quality research (Campbell et al. [@CR8]). We therefore employ a monitoring and evaluation framework, combining process evaluation methods and outcome methods (Hubbard and Miller [@CR16]), applying the following criteria; (a) perceptions on treatment outcome by client, parents and service providers; (b) client and therapist treatment satisfaction; (c) therapist burden; (d) level of selection and access to care following the levels outlined above, and; (e) care package costs. The latter two criteria cover the entire care package and the first three criteria cover the core interventions (see Fig. [1](#Fig1){ref-type="fig"}). Methods {#Sec2} ======= Participants and Settings {#Sec3} ------------------------- We studied populations in four countries; Burundi, Sudan, Sri Lanka and Indonesia, all active or post-armed conflict settings. The Burundian sample contains children from 3 north-western provinces. The Republic of Burundi has been afflicted by killings and violence along ethnic and regional lines re-erupting in a civil war from 1993 until 2003, when a peace agreement was signed (Amnesty International [@CR2]). The last rebel group, *Palipehutu*-*FNL*, joined the peace process in 2009, but the peace process is still fragile. The Sri Lankan sample consists of school-going children from 3 educational zones in and around Jaffna (Northern Sri Lanka). The origins of the violence in Sri Lanka, between the mainly Buddhist Sinhalese community and the mainly Hindu/Roman Catholic Tamil community, date back to the time of independence from Britain. In 1983, the Liberation Tigers of Tamil Eelam (LTTE) launched an armed struggle for a Tamil Homeland, due to perceived discrimination by the Sinhalese Government (Amnesty International [@CR3]). The Indonesian sample consists of children from five sub-districts of Poso, Central Sulawesi, an area that has seen recurrent violence since 1998 as a result of hostility between Muslim and Protestant populations, though the conflict has its roots in wider economic and political shifts (Aragon [@CR4]). Major hostilities lasted until the Malino accords in 2001, with continued incidences of violence ever since. The Sudanese sample consists of children in 3 *Payams* in Yei County, South Sudan. Since 1955, Sudan has been afflicted for more than 36 years by inter and intra-tribal and ethnical regional conflicts, caused by competition over meagre resources and power-positions. The civil war between the Islamist Central Government (consisting of Northern Sudanese elites) and peripheral areas represented by the Sudan People Liberation Movement/Army (SPLM/A) formally ended in a Comprehensive Peace Agreement in 2005 (Berghof Foundation for Peace Support [@CR26]). All data collection procedures were consistent with the Declaration of Helsinki. All interventionists were from the project countries and where possible hired from targeted communities. Training of the different interventionists were all conducted by local staff and varied in length depending on the service. Community resilience activities were conducted by Community Psychosocial Workers, a group of community volunteers who were typically trained for a period of 1--2 weeks. CBI facilitators received a training course of 3 weeks, which was designed and master-trained by the Center for Trauma Psychology from Boston, USA (further information available upon request). Finally, counsellors were trained following a skills-based training course over a period of 4 months, which was overseen by technical advisors of HealthNet TPO \[MJ, WT\] (Jordans et al. [@CR21]). Average monthly remuneration for the interventionists varied between fees for CBI facilitators (45 € in Indonesia; 60 € in Sri Lanka; 55 € in Burundi, and 145 € in Sudan) and salaries for counselors (e.g. 154 € in Indonesia; 80 € in Sri Lanka; 180 € in Burundi, and 25 € in Sudan). Remunerations were compatible with local standards and organizational practices in each setting. Applied Evaluation Indicators {#Sec4} ----------------------------- ### Selection and Admittance to Different Levels of Care {#Sec5} One of the primary process indicators to evaluate a multi-layered care package is the level of detection and utilization of each level of care. The flow of clients through different levels of treatment displays feasibility and barriers of a care package. Throughout the project (2005 to mid-2008) the number of direct and indirect beneficiaries at each level of the care package (see Fig. [1](#Fig1){ref-type="fig"}) has been continuously monitored. The most structural detection mechanism involved primary screening to distinguish children with elevated psychosocial distress for optimal allocation of children to the group-based CBI (filter between levels 2 and 3 in the model above). The Child Psychosocial Distress Screener (CPDS) is a brief, context-specific and multi-informant primary screener developed for that purpose. A validation study has demonstrated the CPDS to have good concurrent validity (Area Under the Curve \[AUC\] = .81; optimal cut-off score \>8) for detection of caseness for psychosocial care in Burundi (Jordans et al. [@CR19]). The score range for the CPDS is between 0 and 14. As schools were the entry point of the program this is where primary screening took place. The CPDS was finalized by December 2006, therefore data presented below were collected between 2006 and mid-2008. This data present a baseline level of psychosocial distress experienced by children. In addition, we monitored the number of children referred for follow-up care as a result of secondary screening by the project's service providers. ### Treatment-Related Criteria {#Sec6} Treatment-related criteria include; (a) perception of treatment outcome; (b) treatment satisfaction; and (c) therapist burden. Continuous monitoring of services was integrated throughout the care delivery phase. Evaluation of treatment-related indicators was limited to level 3 and 4 interventions because of the level of standardization in implementation of these services. Children, service-providers and parents were asked to complete structured (combined quantitative and qualitative) questionnaires upon completion of the intervention, designed for the purpose of this program. The self-report questionnaires consisted of 4--9 items, with an approximate administration time of 5--7 min. The questionnaires were pilot-tested and adapted in the initial phase of the project. We evaluated perceived treatment outcome (i.e. levels of perceived problem reductions), mean client satisfaction (i.e. level of satisfaction of needs and satisfaction with services) and mean therapist burden (i.e. levels of distress as a result of providing care) for CBI and the counselling interventions. Two satisfaction questions were taken from the Client Satisfaction Questionnaire (Attkisson and Zwick [@CR5]). Quantitative questions were scored on a 5-point response format. All questionnaires were verbally administered. Interviewers read the questions out loud and recorded respondents' responses. This was done to control for the variable literacy aptitude of participants and because of respondents' unfamiliarity with completing standardized questionnaires. Content analyses were conducted on qualitative data. ### Cost Criteria {#Sec7} Mental health care in LAMIC is scarcely subjected to economic evaluation, yet cost analyses are important to assess efficiency, inform policy development regarding allocation of limited financial resources or to stimulate investment (Chisholm et al. [@CR10]). We aimed to analyze the cost of developing and implementing a comprehensive care package in resource poor settings, in order to evaluate viability for future replication. Mean cost per service user was calculated by dividing the total sum of program development and intervention delivery costs (numerator) by the number of beneficiaries included in the care package (denominator). We used annual financial reports for 2005--2008 for the numerator data and the total reached beneficiaries for denominator data. Beneficiaries were defined as individual who received and visited services, regardless of outcome of services. We have made two types of cost calculations. To calculate actual costs of development and implementation of a comprehensive care system all costs, including project management, administrative and service providing staff, transport (including fuel, insurance and maintenance), capacity building of service providers, equipment, office costs, intervention and supervision costs were included. To calculate actual costs of sustaining a developed and downsized care package for children, all above costs except initial capacity building and counselling costs were included. Similarly, we made a distinction in calculating number of beneficiaries; (a) All children who received direct service delivery subsequent to screening (i.e. CBI, resilience groups and counselling), and; (b) Total beneficiaries, including above-mentioned children screened as well as children, parents and teachers reached through group psycho-education sessions. Cost ratio was calculated per country sample over a 3.5-year period. The author(s) declare that there are no known conflicts of interests. All authors read and approved the final manuscript and certify responsibility for the paper. Results {#Sec8} ======= Selection and Admittance to Services {#Sec9} ------------------------------------ Over a period of 3.5 years 19,065 children were enrolled in the Classroom Based Intervention, 24,690 children in the child resilience groups, 1,705 children received counselling services and 571 children have been referred to more specialized services. Additionally, 25,723 parents and teachers have been reached through psycho-education groups. Breakdown of age groups and gender divisions per country and per intervention haven been summarized in Table [1](#Tab1){ref-type="table"}.Table 1Overview of beneficiariesTargetsBurundiIndonesiaSudanSri LankaLevel 1Psycho-education groups^a^ (*N*)12,1514,7956,6382,139Level 2Child Resilience Groups (*N*)13,6901,8587,3751,767Age (%) 7--9NA391718 10--12576469 13--1541913Gender (%) Male485651 Female524449Level 3Classroom based intervention (*N*)8,3353,5944,4722,664Age (%) 7--910361710 10--1252618152 13--15363234Gender (%) Male54546750 Female46463350Level 4Counselling^b^ (*N*)731150548276Age (%) 7--9824418 10--1244643868 13--15/1945125814Gender (%) Female53376647 Male47633453Level 5Referrals (*N*)3591313168*Note*: *NA* not available^a^Only parents and teachers^b^This also includes parental care/family support provided by the counsellors. Total number of children screened is the total of the level 1 + level 2 beneficiaries The above data provide an overview of the total number of children reached by the program. Data on levels of detection and admittance to levels 3--5 can be obtained from the screening data. CPDS data of 2.5 years provide baseline levels of experienced distress across the four countries (see Table [2](#Tab2){ref-type="table"}). Comparable percentages of children were included in the group-based CBI intervention (41.1% in Burundi; 42.4% in Indonesia; 38.1% in Sudan; 42.4% in Sri Lanka). Upon secondary screening by CBI facilitators, the number of children internally referred to follow-up care by a counsellor is 6.6% for Burundi; 3.1% for Indonesia; 8.6% for Sudan and 10.4% for Sri Lanka. Subsequently, the numbers of children referred by counsellors to specialized care ranged between 0.1 and 2.9%.Table 2Total scores and percentages of children indicated for treatmentBurundiIndonesiaSudanSri Lanka*N* = 12,008*N* = 3,770*N* = 8,428*N* = 5,086CPDS Total score *M* (SD)6.77 (2.91)6.74 (2.67)6.01 (4.85)3.86 (2.67)Indicated (%)41.442.438.142.4Referred to counsellors^a^ (%)6.63.18.610.4Referral to mental health professional^b^ (%)2.90.30.11.3*Note*: CPDS = child psychosocial distress screener^a^Children referred after receiving CBI subsequent to screening outcomes; community referrals to counsellors therefore excluded^b^Children referred after receiving counselling Treatment Evaluation Classroom Based Intervention (CBI) {#Sec10} ------------------------------------------------------- Most beneficiaries and service providers of CBI rated the service positively, among service providers 92% in Burundi, 89% in Indonesia, 98% in Sudan and 94% in Sri Lanka reported to be quite or very much satisfied. Among children 94, 97, 95 and 98%, respectively, reported to be quite or very appreciative of the service. Levels of perceived problem reductions were somewhat lower. Among children, 71% in Burundi, 48% in Indonesia, 78% in Sudan and 85% in Sri Lanka reported quite a bit or a lot of problem reductions. Among parents these percentages were 83, 87, 82 and 84%, respectively. The major changes that children (*n* = 11,595) reported are improved emotional wellbeing (i.e. reduced fear, sadness, anger and negative thoughts), as well as positive social and behavioural changes (i.e. increased patience, sociability, self-esteem, concentration, unity, tolerance). Service providers (*n* = 1,595) most commonly reported behavioural changes (i.e. reduced indiscipline, aggression, hyperactivity and increased obedience) and social changes (i.e. increased unity, peer interaction, trust, inclusion and collaboration). Parents and teachers (*n* = 4,660) similarly reported emotional (i.e. reduced fear, nightmares, depressed moods, anger) and behavioural changes (i.e. increased concentration, diligence, independence and reduced aggression). The perceived changes in obedience and respect were most commonly mentioned across the four settings, which respondents often related to improved school functioning (i.e. increased interest in school and study). In addition to these common categories, we found that children and parents/teachers in Sri Lanka and Indonesia commonly mentioned increased skills (i.e. improved drawing, increased creativity, knowledge and awareness of talents) and in Burundi reduced stress complaints. A commonly reported change in Indonesia was further the reduction of tensions and suspiciousness between children of different religions. Across the 4 countries, service providers experienced significant distress while conducting CBI groups (44% in Burundi; 52% in Indonesia; 35% in Sudan; 33% Sri Lanka, scored quite a bit or very much). Treatment Evaluation Counselling {#Sec11} -------------------------------- Most commonly presented intake problems among the children were; fears (23%), behavioral problems (i.e. aggression, theft, violation of rules; 17%), sleeping problems (16%), and learning difficulties (10%) in Burundi; school performance problems/learning difficulties (36%), aggression (32%), concentration problems (26%) and family conflict (22%) in Indonesia; worries and negative thought (68%), learning difficulties (53%), sadness (55%) and fears (46%) in Sri Lanka; social withdrawal and friendship formation difficulties (32%), behavioral problems (i.e. aggression, indiscipline, disrespect; 24%), poor concentration problems (17%), stress complaints (14%), sadness (12%) and low self-esteem (8%) in Sudan. Percentages do not add up to 100% as children could present with multiple intake problems. Average clients' satisfaction of needs after receiving counselling ranged from 3.43 to 4.31 (see Table [3](#Tab3){ref-type="table"}). In other words, 63% in Burundi, 44% in Indonesia, 87% in Sudan and 43% in Sri Lanka reported to be quite a bit or very much satisfied. When comparing with rates of satisfaction with overall service, levels of high satisfaction were reversed for Burundi (32%) and Indonesia (84%), and comparable for Sudan (94%) and Sri Lanka (55%). Clients alike reported high levels of general improvement after counselling (95% in Burundi; 95% in Indonesia; 89% in Sudan; 89% in Sri Lanka). When asked for level of problem reduction, 72% in Burundi, 67% in Indonesia, 61% in Sudan and 30% in Sri Lanka reported to be quite a bit or very much reduced. In terms of reported changes, counselors in Burundi most frequently mentioned decreased fear (11%), decreased stigmatization (10%), reduced nightmares (6%), and reduced substance use (6%); in Indonesia reduced behavioral changes (i.e. aggression, obedience; 34%), anxiety and fear (17%), as well as improved school-performance (12%); in Sudan decreased bad thoughts (14%) and fear (12%), as well as improved sleep (17%) and communication (12%); in Sri Lanka decreased fear (44%), sadness (32%), naughtiness (28%), as well as increased school attendance (40%) and self-esteem (27%). Children's most frequently reported changes were similar, with additional reporting of performance-oriented changes related to improved concentration, problem-solving, self-confidence and overall school functioning. Counselors experienced significant distress as a result of service provision (41% in Burundi; 47% in Indonesia; 5% in Sudan, and 28% in Sri Lanka, report quite a bit or very much distress). In Indonesia, with the smallest sample size (*n* = 106), we observed the highest average counsellor distress score (3.47), and the lowest average perceived client progress score (3.14). The average counsellor distress score per country was lower than that of CBI facilitators. Moreover, it appeared that lower client progress was associated with increased therapist burden for counsellors, as well as for group facilitators (Table [4](#Tab4){ref-type="table"}).Table 3Evaluation data CBIItemCountrySample (*N*)AverageResponse rangeFacilitator distressBurundi5713.185 = Very muchIndonesia5153.434 = Quite a bitSudan2623.033 = A littleSri Lanka2993.052 = Hardly1 = Not at allFacilitator satisfactionBurundi5714.24Indonesia5154.16Sudan2584.60Sri Lanka2974.35Child satisfactionBurundi7,2304.61Indonesia3,3064.77Sudan1,8734.64Sri Lanka2,0344.87Problem reduction according to childrenBurundi7,2303.871 = Not al all reducedIndonesia3,2703.222 = Hardly reducedSudan1,7954.043 = A little reducedSri Lanka2,0344.354 = Quite a bit reduced5 = A lot reducedProblem reduction according to parents & teachersBurundi5023.87Indonesia3,6384.00Sudan2004.18Sri Lanka6094.07Table 4Evaluation data counsellingItemCountrySample (*N*)AverageResponse rangeCounsellor perception of client progressBurundi2703.281 = DeteriorationIndonesia1063.142 = No changeSudan1693.383 = Some improvementSri Lanka1623.564 = Much improvementCounsellor distressBurundi2702.895 = Very muchIndonesia1063.474 = Quite a bitSudan1652.203 = A littleSri Lanka1532.702 = Hardly1 = Not at allChild satisfaction of needsBurundi2563.445 = Almost all needs metIndonesia553.434 = Most needs metSudan1584.313 = Some needs metSri Lanka1013.492 = Only few needs met1 = None needs metChild satisfaction of serviceBurundi2563.205 = Yes, definitelyIndonesia554.254 = Yes, I think soSudan1584.273 = MaybeSri Lanka1013.892 = No, I do not think so1 = No definitely notChild general improvementBurundi2562.951 = DeteriorationIndonesia553.252 = No changeSudan1393.803 = Some improvementSri Lanka1013.554 = Much improvementChild perception of extent of problem reductionBurundi2563.651 = Not at all reducedIndonesia553.722 = Not really reducedSudan1483.713 = A little reducedSri Lanka1013.334 = Quite reduced5 = Much reduced Cost Calculations {#Sec12} ----------------- Cost analyses demonstrated that the development and implementation of the entire care package (combination of 3-tiered intervention structure; i.e. psycho-education, non-therapeutic resilience groups, CBI and counselling, including development, supervision and ongoing capacity building) costs 5.31 € (Burundi), 16.37 € (Indonesia), 6.66 € (Sri Lanka), 13.67 € (Sudan), per reached person (see Table [5](#Tab5){ref-type="table"}). Comparisons between countries were difficult due to national economical differences, with GDP rates (in million of US Dollars) over 2008 varying from 1,163 for Burundi; 514,389 for Indonesia; 40,714 for Sri Lanka; 58,443 for Sudan---which in turn does not account for within-country economical differences (World Bank [@CR36]). To demonstrate the costs of sustaining (as opposed to developing) a downsized intervention package, costs for development (i.e. capacity building) and focused care (i.e. counselling and supervision) were excluded, which resulted in cost per person of 3.46 € (Burundi), 8.47 € (Indonesia), 4.93 € (Sri Lanka), and 8.46 € (Sudan). The same analyses, but including only children who received services and excluding the number of parents and teachers reached through group psycho-education sessions, revealed that the cost per user increases to 5.37 € (Burundi), 15.63 € (Indonesia), 7.62 € (Sri Lanka), 10.78 € (Sudan), when considering running costs only. The above analyses across countries, revealed that mean cost per user ranges from 6.33 € (reduced cost and all beneficiaries), 9.85 € (reduced cost and screened children), 10.50 € (conservative cost and all beneficiaries), to 16.52 € (conservative cost and screened children).Table 5Cost analysesBurundi(USD)Indonesia(USD)Sri Lanka(USD)Sudan(USD)Cost (conservative)^a^181,169171,334125,334135,954Cost (reduced)^b^117,98288,59492,73984,119Screened children^c^21,9705,67012,1747,805All beneficiaries^d^34,12110,46518,8129,944Mean cost per user (conservative cost & all beneficiaries)5.317.0616.3721.776.668.8513.6718.18Mean cost per user (reduced cost & all beneficiaries)3.464.608.4711.264.936.558.4611.25Mean cost per user (conservative cost & screened children)8.2510.9730.2240.1910.3013.6917.3223.03Mean cost per user (reduced cost & screened children)5.377.1415.6320.787.6210.1310.7814.33*Note:* Cost and number of targets are accumulated of entire project period (2005--2008). All costs are in Euros, except for the USD columns. USD rates were calculated by using an average conversion rate for the project period (average: 1 € = 1.33 \$)^a^All program costs included: project management, administrative and service providing staff, transport (including fuel, insurance and maintenance), capacity building of service providers, equipment, office costs, intervention and supervision costs (all interventions)^b^Program costs excluding capacity building, counselling and supervision costs, other costs identical^c^Screened children includes beneficiaries receiving CBI, child resilience groups, counselling^d^All beneficiaries includes screened as well as parents and teachers receiving group psycho-education As can be seen in Table [5](#Tab5){ref-type="table"}, the mean cost per service user ranged considerably between countries (ranging from 5.31 to 16.37 for the entire package, including development). By far the largest contributor to the cost was human resources, accounting for \~58 (Burundi), 59 (Indonesia), 56 (Sri Lanka) and 46% (Sudan). The next largest contributors to cost were transportation (9, 16, 3 and 18%, respectively) and capacity building (9, 12, 10 and 8%, respectively). Cost ratio was dependent on determination of beneficiaries, with reductions of mean cost per user ranging from 35% (Burundi), 46% (Indonesia), 36% (Sri Lanka) to 12% (Sudan) when *including* mental health promotion activities. Discussion {#Sec13} ========== While comprehensive multi-layered approaches are frequently advocated for psychosocial and mental health care in LAMIC (WHO [@CR35]), few examples are presented in the literature, and hardly any evaluation is available (Jordans et al. [@CR20]). An exception is a recent study by Layne and colleagues (2008) who conducted a simultaneous trial into the efficacy of first tier and second tier interventions of a 3-tier public health program in Bosnia. Another example is an evaluation by Wessells and Monteiro ([@CR34]) who used a combined quantitative and qualitative monitoring and evaluation system to demonstrate outcomes of a multi-component psychosocial care program in Angola. This paper aimed to present a practice-driven evaluation of a multi-component care package for children in four conflict-affected settings. Results demonstrate that in low-resource settings with scarce available mental health care, development and implementation of a multi-layered care system is feasible and largely satisfactory, but is dependent on external financial support. Based on an adaptation of the filter model of Goldberg and Huxley ([@CR13]) we assessed the utilization of different levels of care. Relatively large percentages of children were screened for elevated psychosocial distress and subsequently much smaller percentages were referred to individualized or specialized care. The percentages of people indicated for the different levels of care are in line with population projections of distress and disorders in complex emergencies, i.e. 30--50% moderate chronic psychological distress, 10--20% mild to moderate common mental disorders and 2--4% severe mental disorders (Van Ommeren et al. [@CR31]). The percentages for severe mental disorders are lower in our study, which is likely due to under-detection or to unavailability of services to refer to in two settings. With regard to treatment perceptions, we find high levels of client satisfaction for the group-based intervention and moderate-to-high levels for counselling across settings, and, albeit to a somewhat lesser degree, of perceived post-treatment problem reductions. In the literature, client satisfaction has commonly been associated with treatment gains (Holcomb et al. [@CR15]; Veerman and van Yperen [@CR32]). High levels of satisfaction are also found among therapists, yet this seems to come at some cost. Levels of therapist distress are significant. Client satisfaction is more prominent for the group intervention compared to counseling. We hypothesize this is due to the enjoyable, active and creative-expressive nature of the group intervention and the increased severity of problems seen by counselors. In contrast, therapist burden is higher for the group facilitators compared the counselors. While distress is high amongst both groups of service providers, the more thorough training of counselors and the increased status and remuneration are likely buffers for the perceived distress. The presented data on CBI supplement efficacy research we conducted previously (Jordans et al. [@CR18]; Tol et al. [@CR27]) in several ways. The present study demonstrates high rates of perceived problem reductions and satisfaction, suggesting a high level of community acceptance of the intervention. Moreover, the beneficiary-mentioned indicators of post-treatment change are mostly social-behavioural (i.e. diligence, obedience, respect, social inclusion, self-esteem, shyness and interest in school), rather than symptoms of psychopathology. The trials present a more conservative perspective on treatment impact, likely due to scientific rigour of a trial and possible higher social desirable answering tendency in practice-driven evaluations. Economic evaluation of mental health care in LAMIC is scarce, but has been shown feasible for community based mental health care in India and Pakistan (Chisholm et al. [@CR10]). Cost analyses demonstrate that mean cost per service user varies heavily between countries and type of calculations, making between-country comparison difficult. While project expenditures were by no means excessive, the average cost (€10.50 per service user) is relatively high when compared to per capita health expenditures (Burundi \$16; Sudan \$54; Indonesia \$118; Sri Lanka \$163; UNDP [@CR30]). Similarly, the cost per capita for the provision of a scaled-up core mental health care package (mainly pharmaceutical and psychosocial treatment for four mental health conditions contributing largely to the burden of disease) is between \$1.85 and 2.60 for low-income countries (Chisholm et al. [@CR9]; Lancet Mental Health Group [@CR22]). At the same time, straightforward comparison with these figures is not possible as the presented costs in this paper are per reached service user. Costs should be aggregated to total coverage population numbers to get an actual per person per year figure, which was not feasible with the data available to us. Secondly, the presented costs reflect both development and sustainability of a basic psychosocial care package, as opposed to only care delivery costs as is common practice. Third, the following aspects may have had an impact on the average cost ratio: scaling up of the CBI intervention happened only after the termination of a phase focused on research; service delivery frameworks were developed without any existing infrastructure; the program focused on quality of service rather then on reaching the highest number of people. For example conducting clinical supervision and providing counselling increased costs significantly, while not increasing numbers in any considerable way. Nonetheless, the presented costs require external financial aid for most LAMIC and require more attention to establish cost-effective psychosocial and mental health care (Tomlinson et al. [@CR28]). The cost analyses represent basic calculations. Presented data did not allow for more sophisticated cost-effectiveness or cost-benefit analyses of treatments. The results have several implications. First, a combined primary and secondary screening procedure is feasible and facilitates detection and distribution of children over different available levels of care, thereby better matching limited resources with needs. Second, furthermore, high levels of distress among service providers need to be addressed for example by adapting workload, training, provision of care-for-caregivers and addressing logistic difficulties of service providers. Future qualitative assessments (e.g. focus group discussions) should elucidate the perceived causes of distress among the interventionists to inform the process of adaptations. Third, we argue that evaluations of care packages should ideally combine efficacy and cost-effectiveness research into separate treatments within the package with evaluations that emphasize practice driven evaluations (i.e. recipient perspectives, cost; Veerman and van Yperen [@CR32]). A combined approach permits triangulation of trial outcomes with evaluations focusing on beneficiaries' perspectives, cost analyses or other indicators of evaluation. Fourth, there is an obvious need for further cost reductions to increase sustainability. More children need to be reached with the same human resources (cost) to make a care package such as the one presented financially feasible. Conclusion {#Sec14} ========== This study evaluated a multi-layered psychosocial and mental health care delivery framework for children in four conflict-affected countries. It demonstrated access to, and utilization of, different levels of care within resource-poor settings. Practice-driven evaluations demonstrated high levels of satisfaction combined with considerable levels of perceived post-treatment problem reductions and treatment gains. Therapist burden is higher than expected, especially for community workers or for lower perceived client progress. Analyses of cost per service user give rarely reported data on actual cost of development and implementation of community based psychosocial care in LAMIC. The reported project has been made possible through funding from PLAN Netherlands. We would like to thank Drs. Chisholm and van Ommeren for their advice on the cost analyses. **Open Access** This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited. Appendix {#Sec101} ======== See Fig. [2](#Fig2){ref-type="fig"}.Fig. 2Four project countries

Introduction {#s1} ============ Myotonic dystrophy \[Dystrophia Myotonica (DM); Steinert's disease\] is the most common muscular dystrophy in adults. The prevalence rates are high in Europe and countries with European descendants (Thornton, [@B60]; Wenninger et al., [@B69]). In addition to myotonia, a delayed relaxation after forceful muscle contraction, muscle weakness/atrophy and features involving brain, heart, eyes and other organs usually accompany and worsen the course of disease. Although trials applying a variety of strategies are underway, currently no effective therapy is available (Ashizawa and Sarkar, [@B6]; Udd and Krahe, [@B61]; Thornton, [@B60]). There are two types of DM: type 1 (DM1) and type 2 (DM2). Both DM1 and DM2 are microsatellite disorders that carry extended short tandem repeats within non-coding regions of different genes (Fu et al., [@B19]; Liquori et al., [@B34]). The mutations do not cause loss of function of the genes where they are located. Instead, extended (CTG)~n~ repeats in the 3′-UTR of *DMPK* gene (DM1) or (CCTG)~n~ repeats in the intron 1 of *CNBP* gene (DM2) generate pathological effects through RNA toxicity, such as stabilizing CELF1 protein and generation of repeat-associated non-ATG (RAN) proteins (Ranum and Day, [@B48]; Thornton, [@B60]; Gourdon and Meola, [@B22]). One of the main mechanisms is Muscleblind-like (MBNL) loss of function- When the *DMPK* gene is transcribed, the extended RNAs retain in the nuclei and formed "RNA foci," colocalized and sequestered MBNL proteins. MBNLs are RNA-binding proteins (RBPs) that normally regulate splicing of their downstream targets during developmental transition. Once one of the targets *Clcn1* is mis-spliced, myotonia occurs due to loss of CLCN1 protein production (Miller et al., [@B38]; Mankodi et al., [@B35]; Pascual et al., [@B42]; Wheeler et al., [@B70]). It has been shown that RNA foci also exist in neuronal populations, and mis-spliced targets (e.g., GRIN1, MAPT and APP) were found in DM brains (Jiang et al., [@B28]). In fact, the consequence of MBNL loss of function may alter several key features of RNA processing, including polyadenylation and transportation. Therefore, it involves a very intricate underlying mechanism that is in need of further investigation (Gourdon and Meola, [@B22]). Adult DM patients develop various neuropsychological symptoms including excessive daytime sleepiness, cognitive decline, depression, apathy and avoidant personality (Thornton, [@B60]; Minier et al., [@B39]). Similar to DM1, the CNS symptoms in DM2 are common but less severe. In congenital DM (CDM) with several hundreds to more than a thousand of (CTG)~n~ repeats, intellectual disability is as frequent as respiratory distress that patients may face. Autism spectrum disorder (ASD), attention deficit hyperactivity disorder (ADHD) and social difficulties are accompanied in one-third of the patients (Douniol et al., [@B17]; Gourdon and Meola, [@B22]). Brain MRI studies showed diffuse brain atrophy, predominant white matter changes in DM1 brains involving mainly frontal and temporal lobes and milder in DM2 patients (Minnerop et al., [@B41], [@B40]; Wozniak et al., [@B71]). Neuropathological approaches revealed cell loss, presence of neurofibrillary tangles with tauopathy as the major findings (Caillet-Boudin et al., [@B11]). Initial *in vitro* study on stable transfectans of PC12 cell line expressing extended (CTG)~90~ has shown inhibition of differentiation but not short repeats (CTG)~5~ and (CTG)~60~, suggesting the pathological (CTG)~n~ expansion is sufficient to affect terminal differentiation in neuronal cells (Quintero-Mora et al., [@B46]). In terms of *in vivo* models, several mouse models have been created for investigating DM mechanisms (Braz et al., [@B9]), but only limited transgenic lines showed brain phenotypes: one carrying DM human mutation (DM300-328 and DMSXL lines thereafter; Hernández-Hernández et al., [@B24]); the other is a Tamoxifen-inducible EpA960/CaMKII-Cre line which expresses 960 copies of interrupted (CTG) repeats within human *DMPK* 3′-UTR (Wang et al., [@B67]). Our group proved the "MBNL loss of function" hypothesis by creating knockout (KO) mouse lines of three MBNL family members (Mbnl1, Mbnl2 and Mbnl3): While *Mbnl1* KO mice (*Mbnl1*^ΔE3/ΔE3^) recapitulate myotonia and typical DM muscle pathology (Kanadia et al., [@B30]), *Mbnl2* constitutive KO mice (*Mbnl2* KO, *Mbnl2*^ΔE2/ΔE2^) exhibit REM sleep misregulation and cognitive dysfunction (Charizanis et al., [@B12]). The deprivation of Mbnl1/2 and all three members in the skeletal muscle recapitulated symptoms of muscle atrophy and respiratory distress, the end-stage DM1 or CDM features (Lee et al., [@B31]; Thomas et al., [@B59]). Based on these results, MBNL2 is considered the most critical player among the three for DM brain pathology. To investigate if loss of Mbnl1/2 may cause deleterious effects to the brain and to overcome the embryonic lethality of constitutive KO of Mbnl1/2, we chose *Nestin-Cre* mouse, a transgenic line which specifically expresses Cre recombinase in neuronal and glial cell precursors, to conditionally KO Mbnl2 specifically in the nervous system. We found combined loss of Mbnl1 and Mbnl2 in *Nestin-Cre* double KO (DKO, *Mbnl1*^ΔE3/ΔE3^; *Mbnl2*^cond/cond^; *Nestin-Cre*^+/--^) mice showed pronounced missplicing in the hippocampi similar to that of DM patients (e.g., Tau exon 2/3 and exon 10), indicating Mbnl1 may also play a compensatory role in DM brains. Since many of the downstream targets regulated by Mbnl2 were associated with the function of brain wiring (e.g., NMDAR1 or GRIN1, MAPT and DCLK1; Shin et al., [@B54]; Jadhav et al., [@B27]; Perez-Rando et al., [@B43]), we wondered if structural changes of cortical neurons may also be present in our *Mbnl* KO lines. Materials and Methods {#s2} ===================== *Mbnl* KO Mouse Lines {#s2-1} --------------------- Original articles describing the generation of *Mbnl1* KO (*Mbnl1*^ΔE3/ΔE3^), *Mbnl2* neuron-specific conditional KO (*Mbnl2* CKO or 2KO; *Mbnl2*^cond/cond^; *Nestin-Cre*^+/−^) and *Nestin-Cre* DKO (*Mbnl1/2* DKO or DKO; *Mbnl1*^ΔE3/ΔE3^; *Mbnl2*^cond/cond^; Nestin-Cre^+/−^) were published (Kanadia et al., [@B30]; Charizanis et al., [@B12]; Goodwin et al., [@B21]). Following the protocols we developed in the University of Florida, mice were generated by crossing two single KO lines in the animal center of Chang Gung Memorial Hospital, Keelung, Taiwan, one of the AAALAC accredited institutes that is conducting serious medical research. Due to the breeding difficulties and reduced body weight of the DKO mice, we had to delay weaning and the tail clipping/genotyping process until 5--6 weeks after their birth. In addition, the lifespans of the DKO mice were shortened and they could rarely survive longer than 5 months. Therefore, 2--4-month-old mutant mice were collected in this study. Age-matched mice of *Mbnl1*^+/+^; *Mbnl2*^cond/cond^; *Nestin-Cre*^−/−^ were used as controls. The number of mice used and experimental procedures were reviewed and approved by the IACUC committee (IACUC No. 2015121903). Preparation of Brain Tissues {#s2-2} ---------------------------- Mice were anesthetized with sodium pentobarbital (50 mg/kg i.p.) and perfused with phosphate-buffered saline (PBS), followed by 4% paraformaldehyde (for immunohistochemistry and Golgi staining) or 2% glutaraldehyde and 2% paraformaldehyde (for electron microscopy). Brains were collected and half brains were cut into 40 μm (for immunohistochemistry) or 100 μm (for electron microscopy) using a vibrating microtome (VT1000S, Leica Biosystems, Wetzlar, Germany). Paraformaldehyde-fixed remaining brain halves were used for Golgi staining. Immunohistochemistry {#s2-3} -------------------- Coronal sections 40 μm-thick containing the sensorimotor cortex were treated with 0.3% H~2~O~2~ in PBS to block the endogenous peroxidase activity, followed by treating with blocking solution to reduce the non-specific bindings. The sections were then incubated with primary antibodies against NeuN (1:500; Merck Millipore, Darmstadt, Germany), Foxp1 (1:1,000; Abcam, Cambridge, UK), Cux1 (1:1,000; Santa Cruz Biotechnology, Dallas, TX, USA), GAD65/67 (1:2,000; Sigma, St. Louis, MO, USA), Parvalbumin (Pvalb or PV, 1:3,000; Sigma) and Calretinin (Calb2 or CR, 1:3,000; Merck Millipore) at room temperature overnight. Next, sections were incubated with biotinylated secondary antibodies against mouse or rabbit IgG (1:500; The Jackson ImmunoResearch Laboratories, West Grove, PA, USA) for 2 h at room temperature followed by Vectastain \[avidin-biotin complex (ABC) kit, Vector Laboratories, Burlingame, CA, USA\] incubation for 1 h. Last, sections were processed with 2 mg/ml of 3,3′-Diaminobenzidine (DAB) with 0.01% H~2~O~2~ in PBS and mounted with a glycerol-based aqueous mounting medium. Measurement of Cortical Neurons {#s2-4} ------------------------------- To measure the density of immunochemistry-labeled neurons in the cortex, we counted the immunopositive cells within each counting square of 100 μm × 100 μm in the upper, middle and lower regions of the sensorimotor cortex. To estimate the relative distribution of cortical neurons, the thickness of the cortex was equally subdivided into 10 counting bins, starting from the pia surface (top) to the border of white matter (bottom). The width of the counting bins was 50 μm in counting NeuN-, Foxp1- and Cux1-positive neurons and 100 μm in counting GAD65/67-, PV and CR-positive neurons, respectively. In this counting system, bin 1 corresponds to cortical layer I, bins 2--3 to layers II/III, bin 4 to layer IV, bins 5--7 to layer V and bins 8--10 to layer VI (Yu et al., [@B73]). The number of cells was counted and the percentages of each counting bins were calculated. Golgi Staining {#s2-5} -------------- Brain samples were impregnated and processed as previously described (Li et al., [@B32]). In brief, paraformaldehyde-fixed brain halves were kept in the impregnation solution of FD Rapid GolgiStain kit (NeuroTechnologies, Ellicott City, MD, USA) for 3 weeks. After being washed with ddH~2~O, the 100 μm sections were acquired using a vibratome (Leica), followed by incubation with developer and fixer solutions and finally mounted on gelatin-coated slides. The images of Golgi-stained layers II/III pyramidal neurons in the sensorimotor cortex were captured by the StereoInvestigator system (MicroBrightField Bioscience, Williston, VT, USA) using a light microscope (Olympus, Tokyo, Japan). The soma and dendritic structures of collected neurons were reconstructed from stacks of images using the Neurolucida system (MicroBrightField Bioscience). Size-related parameters (dendritic length) and topological parameters (numbers of branching nodes, terminal endings and dendritic segments) were subsequently analyzed. The complexity of dendrites was determined by counting the number of dendritic segments of different orders, intersections between dendritic branches and the concentric rings method (Sholl analysis; Wang Y. C. et al., [@B68]). Any protrusion was defined as spine and the spine density was calculated and shown as numbers per micrometer of dendritic length. Dendritic spines were further classified by their morphology based on the published criteria: mushroom (spine head greater than 0.6 μm in diameter), stubby (short without a head), and thin (long with a head less than 0.6 μm in diameter) and filopodia (long without a head; Sorra and Harris, [@B55]). The length and width of individual spines were also measured. Transmission Electron Microscopy {#s2-6} -------------------------------- The sections of 100 μm thickness containing the sensorimotor cortex were collected from Bregma 0.98 mm to 0.14 mm and post-fixed with 1% aqueous osmium tetroxide for 1 h. The upper portions comprising cortical layers II/III were isolated for further processes. The samples were dehydrated in graded ethanol, washed with propylene oxide, and embedded in epoxy resin (Polysciences Inc., Warrington, PA, USA). Semi- and ultra-thin sections were cut with a diamond knife on a Reichert-Jung Ultracut E ultramicrotome (Leica-Microsystems, Wetzlar, Germany). Ultra-thin sections were collected on copper grids (200 mesh, TAAB, Burks, UK) and stained with lead citrate and uranyl acetate. Digital photographs at a magnification of 10,000× were obtained by a transmission electron microscope (Hitachi H-7100, Tokyo, Japan) at 100 kV. The thickness and width of the postsynaptic densities (PSDs) were measured using ImageJ software (NIH, Bethesda, MD, USA). The glutamatergic synapses were selected and the grayscale values across the synapses were measured. The PSD thickness was defined as the distance between the grayscale values of postsynaptic local maximum and the first local minimum. The PSD width was defined as the length of electron dense band along the postsynatic membrane (Jiang-Xie et al., [@B29]). Statistical Analysis {#s2-7} -------------------- Statistical tests were performed using SPSS software (IBM, Armonk, NY, USA). All data were analyzed by one-way ANOVA followed by Scheffe's *post hoc* analyses and *p* \< 0.05 was defined as the statistical significance. All data were expressed as mean ± SEM and the *n* values of each experiment were noted in the Figure legends. Results {#s3} ======= Mbnl Loss of Function Caused Abnormal Distribution of Cortical Neurons {#s3-1} ---------------------------------------------------------------------- Neuroimaging studies have revealed structural abnormalities in the gray and white matter of the sensorimotor cortex associated with the CNS malfunctions in DM1 patients (Minnerop et al., [@B40]). We first examined the features in the cortex of 2KO and DKO mice, the brain-specific mouse models for DM. We measured the thicknesses of the sensorimotor cortex (gray matter) and the external capsule (EC; white matter) underneath in these mice ([Figure 1A](#F1){ref-type="fig"}). One-way ANOVA showed no significant difference in the thickness of cortex (*F*~(2,12)~ = 0.471, *p* = 0.63; [Figure 1B](#F1){ref-type="fig"}) and the EC (*F*~(2,17)~ = 2.492, *p* = 0.112; [Figure 1C](#F1){ref-type="fig"}) among the control, 2KO and DKO mice. {#F1} Next, we examined the density and distribution of cortical neurons in the sensorimotor cortex of these mice. The density of cortical neurons labeled with a pan-neuronal marker, NeuN, was measured in 100 μm × 100 μm counting frames in the upper (layers II--IV), middle (layer V) and lower (layer VI) cortical regions ([Figures 2A,B](#F2){ref-type="fig"}). We found no significant difference in the densities in the upper (*F*~(2,8)~ = 2.914, *p* = 0.112) and middle (*F*~(2,8)~ = 3.522, *p* = 0.08) regions among three genotypes. In the lower cortical region (*F*~(2,8)~ = 33.225, *p* \< 0.001); the density of NeuN-positive neurons largely increased in DKO mice, compared with control (*p* \< 0.001) and the 2KO (*p* \< 0.001) mice ([Figure 2B](#F2){ref-type="fig"}). For a detailed analysis of cell distribution, we checked NeuN-positive neurons in the cortex by dividing the thickness of cortex into 10 counting bins and measuring the number of cells within each bin ([Figure 2A](#F2){ref-type="fig"}). Compared with control mice, the percentages in bin 3 and 4 (corresponding to layers II/III and IV) decreased (2KO *p* \< 0.05, DKO *p* \< 0.001) but slightly increased in bin 9 (layer VI) in both mutants (*p* \< 0.05; [Figure 2C](#F2){ref-type="fig"}). {#F2} To confirm these findings, we evaluated the expression of Foxp1, a member of the FOX transcription factors and a marker for forebrain pyramidal neurons ([Figures 2D--F](#F2){ref-type="fig"}; Ferland et al., [@B18]). In the upper cortical region (*F*~(2,6)~ = 6.713, *p* \< 0.05); the density of Foxp1-positive neurons largely (\~50%) decreased in the DKO mice compared with controls (*p* \< 0.05; [Figure 2E](#F2){ref-type="fig"}). In the middle (*F*~(2,6)~ = 1.355, *p* = 0.327) and lower (*F*~(2,6)~ = 0.6, *p* = 0.943) cortical regions, the difference among groups was not significant. Distribution analysis revealed a distinct decrease of the Foxp1-positive neurons in bin 4 in the DKOs (*p* \< 0.01), and a subtle reduction in the 2KO mice (*p* \< 0.05). In addition, the percentages in bins 8--10 (layer VI) increased in the DKO mice (*p* \< 0.05; [Figure 2F](#F2){ref-type="fig"}). We also double-checked our findings by choosing Cux1, a homeobox transcription factor normally expressed in upper cortical neurons (Cubelos et al., [@B7511]) for the immunostaining studies ([Figures 2G--I](#F2){ref-type="fig"}). Compared with control mice, the density of Cux1-positive cells again decreased in the upper and increased in the middle and lower regions in the DKOs (*p* \< 0.01, *p* \< 0.05, *p* \< 0.05, respectively; [Figure 2H](#F2){ref-type="fig"}). In terms of distribution, we also found an abnormal reduction in bins 3 and 4 (layers II/III and IV; 2KO, *p* \< 0.001 and DKO, *p* \< 0.001) but an increase in bin 5 (layer V; 2KO, *p* \< 0.05 and DKO, *p* \< 0.01) and in bins 8--10 (layer VI) in both mutant mice (*p* \< 0.05; [Figure 2I](#F2){ref-type="fig"}). These results indicated a failure of Cux1-positive cortical neurons to reach their final destinations in the mutants. Taken together, our results suggested an altered distribution of glutamatergic cortical neurons in both mutants, more prominent in the DKO mice. Density and Distribution of Inhibitory Interneurons in the Cortex {#s3-2} ----------------------------------------------------------------- GABAergic inhibitory interneurons in the cortex play important roles in the processing of neuronal information. Immunostaining of GAD65/67 was performed for labeling GABAergic interneurons ([Figures 3A--C](#F3){ref-type="fig"}), while antibodies against PV ([Figures 3D--F](#F3){ref-type="fig"}) and CR ([Figures 3G--I](#F3){ref-type="fig"}) were used to label two separated subgroups of interneurons (Salaj et al., [@B75011]). We found the density of GAD65/67-positive cells in the upper cortical region was reduced in DKO mice, although the results did not achieve statistical significance (*p* = 0.057; [Figure 3B](#F3){ref-type="fig"}). On the other hand, the distribution of GAD65/67-positive neurons evaluated by 10 counting bins did not significantly change in either 2KO or DKO mice ([Figure 3C](#F3){ref-type="fig"}). Alternatively, using different markers, we found a reduced density of PV-positive interneurons in DKO mice, but this still did not reach significance (*p* = 0.062; [Figure 3E](#F3){ref-type="fig"}). However, the distribution analysis revealed a reduction of the percentage of PV-positive cortical interneurons in bin 4 in both 2KO and DKO mice compared to controls (both *p* \< 0.01; [Figure 3F](#F3){ref-type="fig"}). In addition, although only a trend of reduction was observed in the mutants during the density analysis for CR-positive interneurons ([Figure 3H](#F3){ref-type="fig"}); the percentage of CR-positive cortical interneurons in bin 3 significantly decreased in 2KO and DKO mice compared to the control mice during distribution analysis (both *p* \< 0.05; [Figure 3I](#F3){ref-type="fig"}). Taken together, these results may suggest a milder distributional defect also exists in GABAergic cortical interneurons of both KO mice. {#F3} Reduced Dendritic Complexity in the *Mbnl* KO Mouse Models {#s3-3} ---------------------------------------------------------- To evaluate the cytoarchitecture of glutamatergic pyramidal neurons, we examined the apical (emerged upward from the apex) and basal (emerged downward from base) dendrites in layers II/III pyramidal neurons, collected from the sensorimotor cortex ([Figure 4A](#F4){ref-type="fig"}). Significant differences in dendritic length were noted in both apical (*F*~(2,40)~ = 5.378, *p* \< 0.01) and basal (*F*~(2,33)~ = 17.944, *p* \< 0.001) dendrites. Compared with that of the control group, the length of apical dendrites was reduced in the DKO mice (*p* \< 0.01), while the length of basal dendrites was largely reduced in both mutants (*p* \< 0.001; [Figure 4B](#F4){ref-type="fig"}). Furthermore, the number of nodes, ends and segments decreased in the basal, but rather mildly, in apical dendrites of both mutants ([Figures 4C--E](#F4){ref-type="fig"}). We then measured the complexity of neurons using the aforementioned methods. We found that in both mutants, the dendritic complexity was significantly reduced in the basal dendrites but much less prominent in the apical dendrites ([Figures 4F,G](#F4){ref-type="fig"}). For most of the categories in this analysis, the effects of single KO Mbnl2 and combined KO Mbnl1/2 were very similar, especially in the basal dendrites. These results suggested Mbnl2 loss of function alone was sufficient to disrupt the dendritic morphogenesis in cortical neurons. {#F4} Immature Dendritic Spines Increased in the Absence of Mbnl 1/2 {#s3-4} -------------------------------------------------------------- Dendritic spines are tiny protrusions from dendritic shafts that are critical for receiving excitatory inputs from glutamatergic neurons. Their morphology may change during development or in response to neuronal activity and are directly linked to synaptic function. Neuroscientists developed methods to classify the dendritic spines into stubby, mushroom, thin and filopodia. The different morphology may reflect their maturity: the stubby and mushroom-like spines are mature and well-developed, whereas the thin and filopodia spines are relatively immature and hypogenetic. We then characterized the spine densities and analyzed their morphology under high power magnification with a light microscope ([Figure 5A](#F5){ref-type="fig"}). Significant difference in spine density was noted among three groups (*F*~(2,81)~ = 10.612, *p* \< 0.001); but only significantly decreased in neurons of the DKO mice, compared with those in control (*p* \< 0.001) and 2KO (*p* \< 0.05) mice ([Figure 5B](#F5){ref-type="fig"}). We further compared the frequencies of different spine types among three genotypes. Significant differences were noted in stubby (*F*~(2,72)~ = 14.415, *p* \< 0.001) and thin (*F*~(2,72)~ = 18.82, *p* \< 0.001) spines. The frequency of stubby type (mature) spines was significantly reduced in the DKO group (22.9%) compared with control (35.3%, *p* \< 0.001) and 2KO (31.2%, *p* \< 0.01) groups, respectively. On the other hand, the ratio of thin type (immature) spines was largely increased in the DKO group (41.3%) compared with control (25.44%, *p* \< 0.001) and 2KO (31.5%, *p* \< 0.01) groups ([Figure 5C](#F5){ref-type="fig"}). These results indicated a defect in the maturation of dendritic spines in the DKO mice. We also measured the length and width of each dendritic spine. The spine length was comparable among three groups (*F*~(2,71)~ = 0.254 *p* = 0.776; [Figure 5D](#F5){ref-type="fig"}); while the width of dendritic spines decreased in both 2KO and DKO groups compared with controls (*p* \< 0.001 for both mutants; [Figure 5E](#F5){ref-type="fig"}). These results indicate a shared feature of defective spines and compromised synaptogenesis in these mutants, more severe in the DKO mice. {#F5} Alterations of the Postsynaptic Density (PSD) Morphology in the DKO Mice {#s3-5} ------------------------------------------------------------------------ The PSDs consist of various postsynaptic receptors and scaffold proteins and the features of PSDs are highly linked to synaptic function. We therefore decided to evaluate the ultrastructure of the PSDs of excitatory synapses in the sensorimotor cortex using electron microscopy ([Figure 6](#F6){ref-type="fig"}). The PSDs of excitatory synapses in the sensorimotor cortex were collected from control, 2KO and DKO mice ([Figure 6A](#F6){ref-type="fig"}). The thickness of PSD was defined as the distance between the local maximum and minimum of postsynaptic grayscale values across a glutamatergic synapse, whereas the PSD width was the length of the high electron-dense band along the postsynaptic membrane ([Figure 6B](#F6){ref-type="fig"}). Although the changes in the 2KO mice were subtle and did not reach significant differences, as expected, we found a significant reduction of PSD thickness and widths in the DKO mice compared to control mice (*p* \< 0.05; [Figures 6C,D](#F6){ref-type="fig"}). These results indicate loss of Mbnl would potentially alter the synaptic transmission, particularly in the DKOs. {#F6} Discussion {#s4} ========== MBNL family proteins are RBPs critical for post-transcriptional regulations (Batra et al., [@B7]; Taylor et al., [@B58]). Aside from the RNA toxicity derived from microsatellite repeat expansions including RAN translation and secondary CELF1 upregulation (Wang et al., [@B64], [@B65]; Zu et al., [@B75], [@B74]), MBNL loss of function is essential for causing DM phenotypes(Thomas et al., [@B59]; Chen et al., [@B13]; Ramon-Duaso et al., [@B47]). The 2KO and DKO models have shown their potential and suitability for the investigations of DM brain pathogenesis due to extensive and distinct splicing alterations similar to DM patients (Charizanis et al., [@B12]; Goodwin et al., [@B21]). However, detailed morphological evaluation has been lacking. Therefore, this project was aiming to look for structural evidence at the cellular level in these mouse models. The macroscopic studies on DM1 adult patients were generally normal, without characterizing significant brain atrophy. However, in the longitudinal studies of air encephalography, progressive atrophy did occur in adult DM1 patients (Refsum et al., [@B49]). On the other hand, the severity of brain atrophy was higher in CDM patients and the observations were further confirmed by most series of CT or MRI image studies showing ventricular dilatation and marked brain atrophy (Regev et al., [@B50]). In our results, we were unable to see major changes in gray and white matter thickness. Although a trend of reduced EC thickness in the mutants was observed, it was unable to reach statistical significance. This may be a result of the relatively young mice selected (2--4 months, due to the reduced life expectancy of the DKOs) of our experimental animals or limited numbers of brain checked. Nevertheless, our results did not notably deviate from the previous findings in adult DM patients. Neuronal migration is a critical step during cortical development for neurons to accomplish proper positioning and form brain circuitry. During the early embryonic stage, newborn cortical neurons starting from the ventricular zone move radially alongside with radial glial fibers to the cortical plate and form six cortical layers in an inside-out manner (Pilz et al., [@B45]; Marin et al., [@B37]). Early cortical plate neurons that settle in layer VI become the corticothalamic projection neurons while layer V pyramidal neurons are mainly corticospinal projection neurons. Layers IV, III and II neurons receive thalamic inputs and projections from other cortical regions and layers (Petreanu et al., [@B44]). Migration defects could cause detrimental effects and may potentially be linked to neuropsychiatric disorders such as autism (Reiner et al., [@B51]). Recently, neuroscientists have started to realize that RBPs may potentially regulate migration. For example Nova2, a splicing factor, has shown its function in modulating migration for the upper layer cortical neurons in mice, suggesting the importance of RBPs during this process (Yano et al., [@B72]; Vuong et al., [@B62]). Previous reports on gyral architecture in CDM patients and if migratory defects exist in DM were under debate. However, a report showing multiple cases with poor lamination of cortex and architectural arrangement of neurons was published more than half a century ago (Rosman and Kakulas, [@B52]; Hageman et al., [@B23]). In our study, a considerable amount of cortical neurons, such as Cux1-positive upper cortical neurons, failed to reach their final destination in the mutants, suggesting a defect in cortical neuron migration. Unlike glutamatergic cortical neurons, GABAergic inhibitory interneurons derive from another proliferative region, the ganglionic eminence, in the ventral telencephalon. PV-positive interneurons, which play an important role in the gating of sensory-motor circuits (Sachidhanandam et al., [@B3200]), are mostly derived from the medial ganglionic eminence; whereas CR-positive interneurons are largely derived from the caudal ganglionic eminence (Cauli et al., [@B7500]). During development, inhibitory cortical interneurons first migrate dorsally and then tangentially toward the cortex (Marin, [@B500]). The patterns of PV- and CR-positive cells in the upper cortical regions of our KO mice suggested a relatively mild defect in the distribution of cortical interneurons as well. Our results demonstrated for the first time that loss of function of Mbnl proteins may cause abnormalities in neuronal distribution, which may be caused by migratory alteration. To further confirm if these changes were directly caused by migration deficits rather than a shift in the neuronal lineage, alteration of progenitor proliferation or a result of apoptotic cell death, BrdU pulse-chase experiments would be necessary in the future to directly answer this question. Positioned cortical neurons initiate dendrite and axon identities postnatally for their specific input/output functions after migration. The growth and formation of dendrites is a fundamental process for proper synaptic connections, equally important as axonal maturation. Dendritic structures are constantly changing, under the influence of experience-dependent plasticity (Maravall et al., [@B36]). Alterations of dendritic morphology have also been reported in neuropsychiatric diseases, such as Fragile X syndrome (Irwin et al., [@B26]). In our morphological studies on dendrites, we chose parameters that were functionally relevant. For example, the number of branch orders (bifurcation nodes) may influence signal propagations (Brown et al., [@B10]). Surprisingly, the impact of lacking Mnbl proteins on apical and basal dendrites was different. We observed greater structural changes in the basal dendrites than the apical dendrites in layers II/III pyramidal neurons of both 2KO and DKO mice. Layers II/III basal dendrites receive glutamatergic excitatory inputs from the thalamus as well as layer VI neurons and the branching patterns are sensitive to sensory experience (Bose et al., [@B8]). Our results suggest Mbnl loss of function may alter this feedforward signals in the sensory-motor circuitry (Petreanu et al., [@B44]). Interestingly, the defective basal dendrites in layers II/III cortical neurons were evident and comparable between 2KO and DKO lines, indicating the predominant role of Mbnl2 in determining dendritic branching. Apical and basal dendrites are functionally specified structures. The molecular mechanisms underlying the formation and arborization of apical and basal dendrites are just about to be revealed (Chow et al., [@B14]). For example, TAOK2 and EPAC2 are ASD---related genes. EPAC2 is a guanine nucleotide exchange factor (GEF) for the Ras-like small GTPase Rap, downstream of sema3A-Npn1-PlexinA4 signaling cascade, which is activated by TAOK2 (Srivastava et al., [@B56]; de Anda et al., [@B16]). Both *Epac2* knockdown and *Taok2* KO mice exhibit poor basal dendrite complexity yet relatively normal apical dendrites in layers II/III neurons (Srivastava et al., [@B56]; de Anda et al., [@B16]), similar to our findings. These phenotypes of basal dendrites may be a shared feature in disease showing autism-associated symptoms, including DM1/CDM (Angeard et al., [@B3]). Dendritic spines are micron-scale protrusions from dendritic shafts and the sites where most excitatory synapses are located. These dendritic spines functionally connect nerve fibers, which are fundamental for brain circuits. The structure of dendritic spines is dynamic and influenced by synaptic activity. Neurotrophic factors and extracellular matrix glycoproteins are key players that regulate their formation. They are crucial for learning and memory but may be severely damaged under pathological conditions (Arikkath, [@B5]; Adrian et al., [@B2]). During spinogenesis, the number of spines were quite few in fetal or newborn animals. The immature appendage/filopodia/filopodia-like spines emerged first and were followed by mature spines later postnatally (García-López et al., [@B20]). Compared to the length that is more specifically linked to calcium compartmentation and calcium-dependent learning, the width of the spine has a stronger correlation with the synaptic strength (Arellano et al., [@B4]). Therefore, the primary width alterations suggested the indispensable roles of Mbnl1/2 in maintaining synaptic strength. Impaired hippocampal electrophysiological activity and neuronal circuits have been identified in the *Mbnl2* constitutive KO mice (Charizanis et al., [@B12]; Chen et al., [@B13]; Ramon-Duaso et al., [@B47]). Since dendritic spine morphology is correlated with synaptic function, we applied electron microscopy to evaluate PSDs that were most likely affected by impaired dendritic compositions (Arellano et al., [@B4]). The PSDs are composed of a variety of proteins including glutamatergic receptors and adhesion molecules, located at the tip of dendritic spines and critical for maintaining structural and physiological integrity of the synapses (Sheng and Hoogenraad, [@B53]). As expected, the thickness and width of PSDs were reduced in the DKO mice, which is compatible with the results found in dendritic spine analysis, suggesting defects in synaptic transmission and plasticity may occur in these mutants. Based on previously published data, Mbnl1 and Mbnl2 are highly expressed in the neurons. The distribution of Mbnl1 and Mbnl2 are both nuclear and cytoplasmic (Jiang et al., [@B28]; Charizanis et al., [@B12]; Wang E. T. et al., [@B63]). The distribution of Mbnl proteins may differ in different cell types. For example, the Mbnl1 expression in the cerebellar Purkinje neurons is exclusively cytoplasmic, while in the cortical neurons, cerebellar molecular interneurons and deep cerebellar nuclei neurons, the Mbnl1 expression is both nuclear and cytoplasmic. On the other hand, Mbnl2 is highly expressed in the nucleus of neurons in the frontal cortex (Jiang et al., [@B28]; Daughters et al., [@B15]; Charizanis et al., [@B12]; Wang et al., [@B67]). KO mouse models have demonstrated the functions of Mbnl proteins. The overall splicing changes were very mild in *Mbnl1* KO but readily detectable through RT-PCR, microarray and RNAseq experiments in *Mbnl2* KO. The behavioral evaluation of *Mbnl1* KO was limited due to its constitutive KO and marked myotonia that may affect locomotion. On the other hand, we saw spatial learning deficits as well as eletrophysiological changes in *Mbnl2* KO mice. Therefore, we thought MBNL2 is the key player in DM brains, in contrast to the relative predominant role of MBNL1 in the skeletal muscle. However, this does not mean an exclusive tissue-specific role for individual MBNL. Evidence from *Myo-Cre* DKO that KO of Mbnl1/2 in muscle shows severe muscular atrophy and *Nestin-Cre* DKO that KO of Mbnl1/2 in the nervous system shows complete Tau spliceopathy suggest compensatory roles Mbnl1 and Mbnl2 in different organs (Lee et al., [@B31]; Goodwin et al., [@B21]). In addition to nuclear MBNL that regulate splicing, both cytoplasmic Mbnl1 and Mbnl2 may participate in RNA processing such as RNA localization and facilitate local translation. For example, cytoplasmic Mbnl1 binds to 3′-UTR of downstream gene targets and transports them to the membrane organelles, including the synapses (Adereth et al., [@B1]; Wang E. T. et al., [@B63]). Of note, the nuclear Mbnl1 splicing regulation may also be associated with 3′-UTR binding, since spliced transcripts regulated by Mbnl1 have twofold chances of 3′-UTR binding by Mbnl1, suggesting a collaborative work between nuclear and cytoplasmic Mbnl1. The potential cytoplasmic function of Mbnl1 is intriguing especially in the asymmetrical cells such as neurons, since focal structures far away from the nucleus, such as axon and synapse may be more sensitive to RNA localization. The enriched Mbnl1 binding sites in the synapse suggest its potential role in regulating synaptic function (Wang E. T. et al., [@B63]). It has been known that nuclear Mbnl1 and Mbnl2 autoregulate splicing of themselves, including exon 7 (54 nt) of Mbnl1 and exon 6 (54 nt) of Mbnl2 with sequence of nuclear localizing signals. This translocation from cytoplasm into nucleus is especially important during developmental transition from fetal to adult (Lin et al., [@B33]). In the *Mbnl* KO mice, the splicing shifted these alternative-spliced cassettes towards exon-inclusion, therefore the nuclear portion of Mbnl1 and Mbnl2 was increased (Lee et al., [@B31]). Interestingly, it was reported that an alternative post-translational modification pathway may determine the translocation of Mbnl1 from cytoplasm to nucleus by deubiquitination (Wang et al., [@B66]). In the transgenic mice overexpressing (CTG)~960~ specifically in the brain, they found reduced cytoplasmic Mbnl1 is an early event and this cytoplasmic isoform is necessary for maintaining neurite growth. In addition, loss of function of cytoplasmic Mbnl1 showed negative impact on the synaptic transmission, prior to the nuclear Mbnl2-mediated mis-splicing of downstream targets. Therefore, the mechanism and impact of loss of Mbnl1 or Mbnl2 on the dendrite growth may be different (Wang et al., [@B67], [@B66]). In our *in vivo* system, we chose CKO strategy in order to get DKO mice in the nervous system. Initially, we did not include *Mbnl1* KO due to its subtle brain phenotype as we reported previously (Suenaga et al., [@B57]). Although Mbnl1 plays a predominant role in the cytoplasm and Mbnl2 functions more significantly in the nuclei, our observation on *Nestin-Cre* DKO nevertheless revealed a phenotype that contributed by compound loss of MBNL1/2 in the nervous system. To compare and verify specific consequences of loss of Mbnl1 or Mbnl2 to the dendrite/dendritic spines, further work on morphological analysis will be worth doing by using individual *Mbnl* KO lines. In conclusion, through the morphological studies on *Mbnl* KO models, we once again proved that Mbnl1/2 proteins play a critical role in the developing brain through affecting the distribution patterns of cortical projection neurons and inhibitory interneurons as well as dendritic complexity, spine morphology and PSD ultrastructures. The Mbnl1/2 loss of function caused the infant/immature structural phenotypes in mutant brains that were reminiscent of the fetal isoform retention in adults during splicing analysis (Charizanis et al., [@B12]). These deficits may alter the synaptic transmission and brain circuitry in KO mice and potentially also in DM patients. Our findings provided the first morphological evidence of compromised neurons in *Mbnl* KO brains that will be very helpful in understanding CNS pathogenesis in DM. Data Availability {#s5} ================= All datasets generated for this study are included in the manuscript. Ethics Statement {#s6} ================ Mice were generated in the animal center of Chang Gung Memorial Hospital, Keelung, Taiwan, an AAALAC accredited institute. To maintain the highest standard of animal welfare, the IACUC committee ensured the number of mice used was reduced and the experimental procedures were refined (IACUC No. 2015121903). Author Contributions {#s7} ==================== H-CC and CS conducted the experiments and analyzed the data. K-YL and L-JL designed the experiments, analyzed the data and prepared the manuscript. Conflict of Interest Statement {#s8} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. We appreciate the agreement from Dr. Maurice Swanson, the professor in the Department of Molecular Genetics and Microbiology and the Center for NeuroGenetics, University of Florida, College of Medicine, for providing the *Mbnl1* KO and *Mbnl2* CKO mouse lines. We are also thankful for the technical services provided by the Core Facility for Electron Microscopy in the College of Medicine, National Taiwan University. In addition, we appreciate the English grammar check provided by Hubert Lee. **Funding.** These works were supported by the Ministry of Science and Technology, Taiwan (Grant 105-2314-B-182-069); Chang Gung Medical Research Grant, Keelung, Taiwan (CMRPG2G0292) to K-YL; the Ministry of Education, Taiwan, Grant 107L7837 to L-JL. The Ministry of Science and Technology, Taiwan is a department of the Taiwanese government that oversees national-funded research. The Chang Gung Memorial Hospital, Keelung is a teaching hospital which also fund research grants to physician scientists. The Ministry of Education is a department of the Taiwanese government that is responsible for education and research in general. [^1]: Edited by: Nicolas Heck, Université Pierre et Marie Curie, France [^2]: Reviewed by: Richard Belvindrah, Sorbonne Universités, France; Mario Gomes-Pereira, Institut National de la Santé et de la Recherche Médicale (INSERM), France

INTRODUCTION {#SEC1} ============ DEAD-box RNA helicases remodel RNA secondary structures in an ATP-dependent manner ([@B1]--[@B3]). Members of this helicase family are composed of a helicase core, formed by N- and C-terminal RecA-like domains (RecA_N, RecA_C) that are connected by a short flexible linker. The core contains the conserved motifs involved in RNA binding, ATP binding and hydrolysis and in the coupling of ATP hydrolysis to unwinding of double-stranded RNA ([@B3]--[@B5]). In the absence of ligands, the core is in an open conformation, characterized by a large distance between the RecA domains. RecA_N and RecA_C approach each other upon binding of both ATP and RNA, leading to core closure and promoting RNA unwinding ([@B6]--[@B9]). In addition to the common core, DEAD-box helicases frequently possess flanking regions and ancillary domains that perform diverse functions, including modulation of nucleotide binding and hydrolysis ([@B10],[@B11]), specific or unspecific binding of RNAs ([@B12]--[@B20]), destabilization of the RNA bound to the core ([@B21]) and interaction with other binding partners ([@B22]). Most structural studies have focused on the helicase core or isolated ancillary domains, and the structural information on full-length helicases is scarce. The underlying mechanisms of how ancillary domains affect helicase function are therefore largely unknown. RNA binding domains may anchor helicases on their target RNAs ([@B18]). Yet it is unknown if they act as passive anchors or actively present the bound RNA to the helicase core for unwinding (see ([@B23])). Specifically, it has not been studied if these domains move with respect to the helicase core upon RNA binding, and how RNA binding to the ancillary domain may be communicated to the helicase core. The *Bacillus subtilis* DEAD-box protein YxiN (Figure [1A](#F1){ref-type="fig"}) contains an ancillary RNA recognition motif (RRM) C-terminal to its helicase core that specifically binds to hairpin 92 of the 23S ribosomal RNA (rRNA) ([@B16],[@B17],[@B24]). Binding of the RRM to this hairpin is thought to anchor YxiN on ribosomal RNA during ribosome biogenesis ([@B24],[@B25]). Here, we investigated the mechanism by which the RRM of YxiN affects RNA binding and unwinding, as well as ATP hydrolysis by the helicase core. Using single molecule Förster resonance energy transfer (smFRET), we show that the RRM shifts from a position close to RecA_C in the absence of RNA to a position in proximity of RecA_N when RNA is bound. RNA binding to the RRM alters the conformational space of the helicase core, and stimulates ATP hydrolysis and RNA unwinding. Communication between the RRM and the core is possibly mediated by the short connecting linker that has a high propensity to form α-helical structure. The YxiN RRM is thus not merely a passive anchor but actively modulates the functions of the helicase core. {ref-type="table"}.](gkx014fig1){#F1} MATERIALS AND METHODS {#SEC2} ===================== RNA oligomers {#SEC2-1} ------------- The 32mer, 32merHairpin, 9mer and 15mer, 9merTrap and 14mer RNAs, with or without covalently-linked fluorescein (FAM) or cyanine fluorophores (Cy3 or Cy5), were purchased from Purimex (Grebenstein, Germany) in PAGE-purified form or from Axolabs (Kulmbach, Germany) in HPLC-purified form. The sequences are 5΄-CGAGGUCCCAAGGGUUGGGCUGUUCGCCCAUU-3΄ (32mer), 5΄-UGGGCUGUUCGCCCAU--3΄ (32merHairpin), 5΄-UUGGGACCU-3΄ (9mer), 5΄-CGAGGUCCCAAGGGU-3΄ (15mer), 5΄- AGGUCCCAA-3΄ (9merTrap) and 5΄-GGGCGGGCCCGCCC-3΄ (14mer). The 32mer, 32merHairpin and 14mer RNAs were incubated at 95°C for ∼2 min and cooled to room temperature during ∼1 h to allow hairpin formation or hybridization. RNA concentrations were determined by UV absorption of denaturedRNAs, using the extinction coefficients 298 890 M^−1^ cm^−1^ (32mer), 142 560 M^−1^ cm^−1^ (32merHairpin) and 206 400 M^−1^ cm^-1^ (14mer). The 154mer RNA was produced by *in vitro* transcription using T7 RNA polymerase as previously described ([@B26]). Protein production and purification {#SEC2-2} ----------------------------------- YxiN double-cysteine variants for FRET experiments, S108C/D429C, S108C/N444C, S108C/N464C, S108C/L472C, D262C/N444C or S108C/S229C, were produced as GST fusions and purified as previously described ([@B7],[@B9],[@B26]). A GST fusion of YxiN_A115C/S229C containing an N-terminal biotinylation tag was co-produced with biotin ligase in the presence of biotin (∼ 40 μg ml^−1^) ([@B27]) and purified according to the standard protocol ([@B7],[@B28]). All variants are based on the YxiN variant YxiN_C61A_C267A (YxiN) that lacks solvent-accessible cysteines ([@B7],[@B8],[@B29]), and have similar secondary structure content and ATP-dependent RNA unwinding activity as the wild-type protein ([@B7],[@B28]). The YxiN helicase core (residues 1--368) was produced and purified as previously described ([@B9]). Determination of ATP hydrolysis rates {#SEC2-3} ------------------------------------- Rates of ATP hydrolysis were determined under steady-state conditions using an Utraspec 2100pro absorption spectrometer (Amersham Biosciences) and a pyruvate kinase--lactate dehydrogenase enzyme-coupled assay ([@B30]) that couples ATP hydrolysis to oxidation of NADH, which is monitored by the decrease in absorbance at 340 nm wavelength. Measurements were performed with isolated YxiN core and YxiN_S108C/N444C (0.25 μM) in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl~2~, 2 mM 2-mercaptoethanol in the absence and presence of RNA at 37°C. The reactant concentrations were 5 mM adenosine triphosphate (ATP), 0.2 mM nicotinamide adenine dinucleotide (NADH), 0.4 mM phosphoenolpyruvate, 23 μg/ml lactate dehydrogenase and 36 μg/ml pyruvate kinase. RNA concentrations were 5 or 20 μM 32mer (YxiN_S108C/N444C), 40 μM of 32mer (YxiN_core), 40 μM of 32merHairpin RNA (YxiN_S108C/N444C) or 10 μM of 14-base pair double-stranded RNA. Fluorescence equilibrium titrations {#SEC2-4} ----------------------------------- Steady-state fluorescence anisotropy experiments were conducted on Jobin Yvon (Fluoromax-3 or 4) spectrometers. The dissociation constants (K~d~) for YxiN_core-32mer-ADPNP, YxiN_S108C/N444C-32mer or YxiN_S108C/N444C-32merHairpin complexes were determined by titrating the protein to 50 nM (FAM)-32mer-3΄ or 5΄-32merHairpin-(FAM) RNA in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl~2~, 2 mM 2-mercaptoethanol at 25°C. The determination of the K~d~ value for the YxiN_core/32mer/ADPNP complex was conducted in the presence of 2 mM ADPNP. Fluorescence anisotropy of the fluorescein (FAM) probe was monitored at 520 nm after excitation at 496 nm. K~d~ values were extracted by describing binding curves according to a 1:1 binding model (Equations [1](#M1){ref-type="disp-formula"} and [2](#M2){ref-type="disp-formula"}): $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{eqnarray*} &&r = {r_{{\rm free}}}\ + \nonumber \\ &&\frac{{[YxiN \cdot RNA] \cdot R}}{{[YxiN \cdot RNA] \cdot R + RN{A_{{\rm {\rm tot}}}} - [YxiN \cdot RNA]}}({r_{{\rm bound}}} - {r_{{\rm free}}})\end{eqnarray*}\end{document}$$ with $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation*}\begin{array}{@{}*{1}{l}@{}} {[YxiN \cdot RNA] = \frac{{Yxi{N_{tot}} + {K_d} + RN{A_{tot}}}}{2} - }\\ {\sqrt {{{\left( {\frac{{Yxi{N_{tot}} + {K_d} + RN{A_{tot}}}}{2}} \right)}^2} - Yxi{N_{tot}} \cdot RN{A_{tot}}} } \end{array}\end{equation*}\end{document}$$ r is the measured anisotropy, r~free~ is the anisotropy of the free RNA, r~bound~ the anisotropy of the bound RNA, R is the change in quantum yield upon binding, YxiN~tot~ is the total concentration of YxiN and RNA~tot~ is the total RNA concentration. RNA unwinding experiments {#SEC2-5} ------------------------- The 32/9mer was prepared by mixing 25 μM Cy5-32mer-3΄ with 50 μM 5΄-9mer-Cy3, or by mixing 25 μM FAM-32mer-3΄ with 50 μM 5΄-9mer-Cy3. The 15/9mer was prepared by mixing 25 μM FAM-15mer-3΄ with 50 μM 5΄-9mer-Cy3. The RNAs were incubated at 95°C for ∼2 min and cooled to room temperature during ∼1 h to allow for hybridization. Unlabeled 9merTrap (5 µM), complementary to the Cy3-labeled 9mer and 5 μM YxiN core or YxiN_S108C/N444C were added successively to hybridized RNA containing 0.5 μM Cy5-32mer-3΄ and 1 μM 5΄-9mer-Cy3 in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl~2~, 2 mM 2-mercaptoethanol at 25°C and unwinding was started by the addition of 5 mM ATP. The unwinding reaction was monitored by changes in Cy5 acceptor fluorescence at 666 nm after excitation of the Cy3 donor at 554 nm. To investigate the stimulatory function of the 32merHairpin, RNA unwinding of hybridized RNA containing 0.5 μM FAM-32mer-3΄ and 1 μM 5΄-9mer-Cy3 or 0.5 μM FAM-15mer-3΄ and 1 μM 5΄-9mer-Cy3 by YxiN_N444C in the absence and presence of 5 μM 32merHairpin was monitored as an increase in fluorescein (FAM) fluorescence at 521 nm after excitation at 495 nm. smFRET experiments by confocal microscopy {#SEC2-6} ----------------------------------------- Single molecule FRET experiments of less than 200 pM YxiN (YxiN_C61/A_C261A lacking solvent-accessible cysteines) statistically labeled with AlexaFluor 488 (A488) and AlexaFluor 546 (A546) were conducted on a Microtime 200 confocal microscope (Picoquant, Berlin, Germany) at room temperature in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl~2~, 2 mM 2-mercaptoethanol. smFRET experiments of YxiN/RNA complexes were performed in presence and absence of 5 mM of ADPNP. The RNA concentrations were 1 and 2 μM 32mer, 5 and 10 μM 32merHairpin and 0.4 and 2 μM 154mer RNA. The donor dye was excited at 485 nm. Back-scattered excitation light was rejected by a 505dcXR dichroic mirror, a 100 μm pinhole and a 532rdc beam splitter. Acceptor and donor emissions were filtered by 570LP or 535/40 bandpass filters, respectively, and detected by SPAD detectors. For the construction of corrected FRET histograms, measured intensities were corrected for differences in quantum yield and detection efficiencies of the detectors, direct excitation of the acceptor dye, leakage of the donor emission into the acceptor channel and leakage of acceptor emission into the donor channel as described (see ([@B31]) for details). The quantum yields of the donor labeled YxiN variants were measured relative to fluorescein in 0.1 M NaOH ([@B32],[@B33]). Förster distances were determined as described ([@B31]). FRET histograms were calculated for each of the two possible donor-acceptor-labeled species: one histogram was generated using correction parameters and Förster distances for the DA-configuration (donor attached to Cys1, acceptor to Cys2), and a second for the inverse AD-configuration (acceptor attached to Cys1, donor to Cys2; Table [1](#tbl1){ref-type="table"}). The two mean FRET efficiencies from these histograms were converted to inter-dye distances *r* according to Equation ([3](#M3){ref-type="disp-formula"}): $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation*}r = {R_o} \cdot \sqrt[6]{{\left( {\frac{1}{{{E_{FRET}}}} - 1} \right)}}\end{equation*}\end{document}$$ where *R*~o~ is the Förster radius (see Table [1](#tbl1){ref-type="table"}). Both distances were used as restraints for rigid-body docking. ###### FRET efficiencies, Förster distances and inter-dye distances YxiN alone YxiN/RNA complex ------- ------- ------------ ------------------ ----- ------ ----- ----- S108C D429C 0.42 5.8 6.0 N/A 5.6 N/A D429C S108C 0.22 5.1 6.3 0.53 4.8 4.7 S108C N444C 0.40 5.7 6.1 0.53 5.5 5.4 N444C S108C 0.41 5.1 5.4 0.30 4.7 5.4 S108C N464C 0.26 5.5 6.6 0.52 5.3 5.2 N464C S108C 0.39 5.3 5.7 0.56 5.2 5.3 S108C L472C 0.35 5.4 6.0 0.36 5.1 5.6 L472C S108C 0.46 5.2 5.3 0.37 4.8 5.3 D262C N444C 0.88 5.3 3.8 0.58 4.9 4.6 N444C D262C 0.90 5.2 3.6 0.35 4.9 5.4 E~FRET~: FRET efficiency, R~0~: Förster distance, r: experimental inter-dye distance, calculated from E~FRET~ for different dye orientations. N/A: could not be determined. FRET-restrained structural modeling {#SEC2-7} ----------------------------------- Rigid-body docking iterations of the RRM structure with the homology model of YxiN_core were performed using the FRET-restrained structural modeling software (FPS) ([@B34]). The physical dimensions of the dye and dye linkers were explicitly considered: A488 (radius of dye R~d~ = 5 Å, linker length L~l~ = 20 Å, linker width L~w~ = 4.5 Å) and A546 (R~d~ = 8.1 Å, L~l~ = 20 Å, L~w~ = 4.5 Å). L~w~ for Cys 472 was set to 3.5 Å. A total of 5000 docking iterations were performed starting from a clash tolerance of 6 Å, through 2.0 to 0.5 Å, in accordance with previous studies ([@B34]). At least 20% (1000 structures) of all the conformers with the lowest chi-squared values were analyzed, in accordance with previous studies ([@B34]). smFRET experiments by TIRF microscopy {#SEC2-8} ------------------------------------- TIRF experiments were performed using an Olympus total internal reflection fluorescecne (TIRF) microscope as previously described ([@B27]). Microscope slides and coverslips were cleaned with 10% alconox (or SDS) and KOH, passivated with polyethylene glycol (PEG) doped with biotinylated PEG, and functionalized by coupling of streptavidin. A total of 0.5 μM YxiN RecA_N was added to saturate surfaces and to prevent unspecific binding of YxiN. N-terminal biotinylated YxiN_A115C/S229C in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM MgCl~2~, 2 mM 2-mercaptoethanol was labeled with A555 and A647, immobilized and imaged in the absence and presence of 15 μM 32mer, 1 μM 154mer and/or 5 mM ADPNP or 20 mM ATP in the presence of oxygen scavenger (5 mM protecatechuate/100 nM protecatechuate-3,4-dioxygenase) and triplet state quencher (2 mM Trolox). The time resolution was 150 ms. Trajectories of donor and acceptor intensities were extracted from TIRF movies using an in-house Labview software. The FRET efficiency was calculated from the acceptor I~A~ and donor intensities I~D~, corrected for differences in quantum yield of the dyes and in detection efficiencies (γ, Equation [4](#M4){ref-type="disp-formula"}) ([@B35]). $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation*}{E_{{\rm FRET}}} = \frac{{{I_{\rm A}}}}{{{I_{\rm A}} + \gamma {I_{\rm D}}}}\ \end{equation*}\end{document}$$ Idealized FRET values were then determined by Hidden Markov Modeling of FRET time traces using vbFRET ([@B36]). Transition density plots (TDPs) were calculated a described ([@B37]). Fluctuation density plots (FDPs) for molecules that do not change their FRET state during the course of the experiments were created similarly to transition density plots using own Matlab scripts. One of the FRET states was altered by an insignificant 0.0001 units to attain non-zero values in the exponents. RESULTS {#SEC3} ======= The YxiN RRM is positioned next to RecA_C in the absence of RNA {#SEC3-1} --------------------------------------------------------------- To investigate whether the RRM moves relative to the helicase core in response to RNA binding, we delineated the positions of the RRM of YxiN relative to the helicase core in both the presence and absence of RNA by performing smFRET experiments on freely diffusing donor-acceptor labeled YxiN using confocal microscopy. YxiN variants with cysteines at either position 108 on RecA_N or position 262 on RecA_C, and at positions 429, 444, 464 or 472 on the RRM ([@B28]), were labeled statistically with Alexa488 (donor) or Alexa546 (acceptor; Figure [1A](#F1){ref-type="fig"}). All cysteine variants unwind a model 32/9mer RNA substrate derived from 23S rRNA comparable to wild-type YxiN ([@B28]). smFRET histograms in the absence of RNA, corrected for cross-talk, different quantum yields and detection efficiencies for donor and acceptor and for direct acceptor excitation ([@B31]), generally exhibit a unimodal distribution consistent with a defined position of the RRM relative to the core (Figure [1B](#F1){ref-type="fig"}). The determined FRET efficiencies are in agreement with previous experiments ([@B28]). Two variants, S108C/L472C and S108C/D429C, showed an additional peak at high E~FRET~ (\>0.7) that is most likely due to aggregation. The tendency to aggregate has been noted previously ([@B28]). The high FRET efficiency of 0.7 implies short distances between C108 and C429 or C108 and C472 that are geometrically impossible in combination with the other distance restraints obtained ([@B28]). These species were therefore not included in the subsequent analysis. FRET histograms were calculated for D/A and A/D configurations (donor on RecA_N or RecA_C, acceptor on the RRM or *vice versa*), and the mean FRET efficiencies (E~FRET~) from both analyses were converted into inter-dye distances (Table [1](#tbl1){ref-type="table"}). Both distances were used as a restraint in rigid-body docking of the crystal structure of the free RRM (PDB-ID 2goc) to a homology model of YxiN core, generated by SWISSMODEL with the structure of mjDEAD (PDB-ID 1hv8) ([@B38]) as a template (Figure [1A](#F1){ref-type="fig"}). The docking procedure was performed with the FRET-restrained Positioning Software (FPS) that takes into account the physical dimensions of the fluorophores ([@B34],[@B39],[@B40]). FPS explicitly computes spatial distributions of dye positions, and enables more accurate and precise determination of the RRM position than previous manual docking attempts ([@B28]). Docking iterations were started from 5000 random initial positions of the RRM relative to the core. Convergence was reached in two rounds of positional refinement. Analysis of \>20% of the 1000 structures with lowest reduced chi-squared values revealed a set of conformers that converged on a single position of the RRM, in proximity to RecA_C (Figure [1A](#F1){ref-type="fig"}; [Supplementary Figure S1](#sup1){ref-type="supplementary-material"}). The molecular shape is in agreement with small angle X-ray scattering results ([@B41]), and the structural model is similar to the previously reported one derived from manual docking ([@B28]). Thus, the RRM populates a unique position relative to the core, in proximity to RecA_C, in the absence of RNA. The RRM shifts to a position closer to RecA_N upon RNA binding {#SEC3-2} -------------------------------------------------------------- To evaluate the effect of RNA binding to the RRM on the overall conformation of YxiN, we performed smFRET experiments in the presence of saturating concentrations of 32mer RNA. The 32mer RNA (Figure [2A](#F2){ref-type="fig"}) contains hairpin 92 of the 23S rRNA, as well as a 5΄-single-stranded extension derived from helix 91 ([@B42]), and has been used as a model substrate for YxiN before ([@B7],[@B8],[@B16],[@B17],[@B26],[@B43]). Binding of the 32mer led to an increase in E~FRET~ for S108C/N464C, S108C/N444C and S108C/D429C, and a decrease for D262C/N444C (Figure [2A](#F2){ref-type="fig"}, Table [1](#tbl1){ref-type="table"}). No significant change in E~FRET~ was observed for S108C/L472C. The decrease in E~FRET~ between dyes attached to RecA_C (D262C) and the RRM (N444C) and concomitant increase between dyes attached to RecA_N (S108C) and the RRM (D429C, N464C, N444C) indicates a shift of the RRM away from RecA_C, toward RecA_N. Docking iterations using FPS, with the experimental inter-dye distances (Table [1](#tbl1){ref-type="table"}) as restraints, defined a unique position of the RNA-bound RRM close to RecA_N (Figure [2B](#F2){ref-type="fig"}). RNA binding thus causes a drastic movement of the RRM relative to the core, with an overall translational movement by 53 Å and a rotation of 37°. {ref-type="fig"}.](gkx014fig2){#F2} RNA binding to the RRM drives its positional shift {#SEC3-3} -------------------------------------------------- The 32mer contacts the RRM through the hairpin loop ([@B16],[@B17],[@B42]), and could in principle interact with the helicase core through the single-stranded extension, raising the question whether RNA binding to the RRM is sufficient to trigger RRM movement, or if simultaneous contacts of the RNA with the core are required. YxiN binds the isolated hairpin of the 32mer (32merHairpin; Figure [2A](#F2){ref-type="fig"}) lacking the single-stranded extension with a 40-fold lower affinity (K~d~ = 7.0 μM) than the 32mer (K~d~ = 0.17 μM; Figure [3A](#F3){ref-type="fig"}) ([@B17]). This loss in affinity can be attributed to interactions of the RRM with the single-stranded region ([@B17],[@B42]). The isolated YxiN core does not bind the 32mer (Figure [3A](#F3){ref-type="fig"}), neither in the absence nor in the presence of the non-hydrolyzable ATP analog ADPNP that increases the affinity of YxiN for RNA ([@B7],[@B9]), supporting that the single-stranded region of the 32mer is not bound by the core. The movement of the RRM upon binding to the 32mer thus seems to originate from the interaction of the RRM with the 32merHairpin. To test this hypothesis, we monitored changes in E~FRET~ for YxiN_S108C/N444C in the presence of the 32merHairpin RNA. The 32merHairpin induced the same change in E~FRET~ as the 32mer (Figure [4A](#F4){ref-type="fig"}), suggesting that RNA binding to the RRM provides the energy for RRM movement, without contribution of direct contacts between the RNA and the core. A much larger fragment of the 23S rRNA, the 154mer ([@B7],[@B16],[@B24]), influenced the position of the RRM similarly to the 32mer (Figure [4B](#F4){ref-type="fig"}), indicating that RRM movement is independent of the context in which hairpin 92 is found. {#F3} {#F4} The RRM is required for RNA-induced ATP hydrolysis and RNA unwinding by the YxiN core {#SEC3-4} ------------------------------------------------------------------------------------- We next investigated whether the shift in position of the RRM upon RNA binding affects the functions of the helicase core, and determined rates of ATP hydrolysis in the absence and presence of RNA in an enzyme-coupled assay ([@B30]). YxiN (both full-length and core) showed negligible ATPase activity in the absence of RNA (Figure [3B](#F3){ref-type="fig"}). Upon adding 40 μM of 32mer RNA, the ATPase activity of the isolated core remained negligible, with k~hyd~ = 2.4 (± 0.4)·10^−3^ ATP YxiN^−1^ s^−1^, in agreement with its undetectably low 32mer affinity (Figure [3A](#F3){ref-type="fig"}). The ATPase activity of YxiN increased to k~hyd~ = 1.24 ± 0.04 ATP YxiN^−1^ s^−1^ in the presence of saturating concentrations of 32mer (20 μM), which is in agreement with previous measurements (0.92 s^−1^; ([@B17])), and over 500-fold higher than the ATP hydrolysis rate of the core. Interestingly, the 32merHairpin (40 μM) stimulated the ATPase activity to k~hyd~ = 0.20 ± 0.04 ATP YxiN^−1^ s^−1^, which is only 6-fold lower than the activity in presence of the 32mer, suggesting that RNA binding to the RRM stimulates the ATPase activity of the core even in the absence of direct contacts between the RNA and core. To further corroborate this finding, we investigated the effect of a double-stranded RNA, formed by a self-complementary 14mer, that binds to the YxiN core ([@B9]) but not to the RRM. In the presence of this RNA, YxiN (both full-length and core) showed only background ATPase activity (Figure [3D](#F3){ref-type="fig"}). Thus, RNA that binds directly to the core but not to the RRM does not stimulate the intrinsic ATPase activity, confirming that RNA binding to the RRM is key for activation of the helicase core. We also tested the effect of the RRM on RNA unwinding, using a Cy5-labeled 32mer hybridized to a 9mer-Cy3 (32/9mer) as an unwinding substrate ([@B8],[@B28],[@B44]). The 9 bp duplex of the 32/9mer represents helix 91 of 23S rRNA. While YxiN_S108C/N444C readily unwound the 32/9mer in the presence of ATP, the isolated core exhibits no detectable unwinding activity under these conditions (Figure [3C](#F3){ref-type="fig"}). Altogether, these results suggest that the RRM is essential for efficient RNA binding, which triggers ATP hydrolysis and RNA unwinding by the core. The peptide connecting the RecA_C to the RRM acts as a communication bridge {#SEC3-5} --------------------------------------------------------------------------- RNA binding to the RRM could be communicated to the helicase core in three ways: (i) through the RNA substrate that contacts both the RRM and the core, (ii) through direct interactions between the RRM and the core or (iii) through the linker connecting RecA_C with the RRM. We showed above that the 32merHairpin activates the ATP hydrolysis by the core without contacting it. Furthermore, the structural model for YxiN in the presence of RNA does not provide evidence for direct RRM/RecA contacts. The RRM cannot restore RNA-dependent ATPase activity of the core when supplied in *trans*, which is also pointing to a lack of extensive interactions between the RRM and the core ([@B17]). This leaves the linker (residues 361--400) between RecA_C and the RRM as the only possible communication bridge, although the mechanism of communication remains unclear. Secondary structure predictions by different algorithms that use position-specific scoring matrixes (PSIPRED ([@B45])), combinations of evolutionary information and neural networks (PHD ([@B46])), cascaded multiple classifiers (OUALI ([@B47])) and the combination of sequence homology and structural homology (PSIPRED and SCRATCH ([@B48])), consistently revealed a high propensity for the formation of two α-helices by the peptide (Figure [5](#F5){ref-type="fig"}; [Supplementary Figure S2](#sup1){ref-type="supplementary-material"}). In an extended conformation, these α-helices (∼60 Å end-to-end distance) would be able to span the distance between the C-terminus of RecA_C and the N-terminus of the RRM both in the free and RNA-bound states (44 and 52 Å, respectively). An α-helical conformation would restrict the positional freedom of the free and RNA-bound RRM relative to the core, and could facilitate communication of changes in the RRM position to the core. The linker between RecA_C and the RRM thus most likely acts as a communication bridge between the RRM and the active site for ATP hydrolysis and RNA unwinding. {#F5} The helicase core exhibits RNA-dependent conformational plasticity {#SEC3-6} ------------------------------------------------------------------ The effect of RNA binding to the RRM on the ATPase activity of the core suggested an effect on core conformation. We therefore monitored conformational changes of the core upon RNA binding by smFRET with freely diffusing YxiN_S108C/S229C, labeled in RecA_N and RecA_C, and different types of RNA (Figure [6](#F6){ref-type="fig"}). YxiN_S108C/S229C has previously been employed to monitor RNA- and nucleotide-induced conformational changes of the YxiN core ([@B7]). Binding of the 32mer or the 32merHairpin to YxiN led to a decrease in E~FRET~ from ∼0.60 to ∼0.35, pointing to a wider cleft between the RecA domains than in the ligand-free open state (Figure [6A](#F6){ref-type="fig"} and [B](#F6){ref-type="fig"}). Subsequent addition of ADPNP to the YxiN/RNA complexes caused a further decrease in FRET (E~FRET~ ∼0.25). Both RNAs thus affect the conformation of the core similarly. In contrast, binding of the 154mer RNA alone to YxiN caused little changes of the FRET efficiency, but higher E~FRET~, consistent with core closure, was observed upon subsequent addition of ADPNP (Figure [6C](#F6){ref-type="fig"}), in accordance with previous observations ([@B7]). The conformational space of the helicase core thus depends on the identity of the RNA that is bound to the RRM. {#F6} YxiN species with widened inter-domain clefts represent off-pathway intermediates {#SEC3-7} --------------------------------------------------------------------------------- According to the current model of RNA unwinding by DEAD-box proteins, helicase activity requires cycling of the helicase core between open and closed states ([@B7],[@B31]). The 154mer stimulates the ATPase activity of YxiN and increases the fraction of the helicase core in the closed state ([@B7]). The 32mer or 32merHairpin both stimulate ATPase activity, yet no high FRET states are observed, suggesting that these are most likely populated only transiently. To comprehensively explore the conformational space of the core, we conducted TIRF experiments on N-terminally biotinylated YxiN_A115C/S229C, labeled with Alexa555 (donor) and Alexa647 (acceptor) dyes and immobilized on streptavidin-functionalized microscope cover slips. Donor-acceptor-labeled YxiN_A115C/S229C reports on RNA- and nucleotide-induced conformational changes of the YxiN core ([@B7],[@B8]). TIRF experiments were performed in the absence and presence of 32mer, 154mer, ADPNP or ATP (Figure [7](#F7){ref-type="fig"}). Representative fluorescence and FRET trajectories for YxiN and YxiN/32mer/ATP are shown in [Supplementary Figure S3](#sup1){ref-type="supplementary-material"}. Histograms of the FRET efficiencies for each molecule at the beginning of the experiment ([@B35]) (during the first three seconds) provide an overview about the populated states (Figure [7](#F7){ref-type="fig"}). In the absence of ligands, FRET histograms for YxiN (Figure [7A](#F7){ref-type="fig"}) show a main peak with E~FRET~ ∼0.34, with a shoulder at E~FRET~ ∼0.45, consistent with an open conformation of the helicase core and in agreement with previous results ([@B8]). The FRET efficiencies are independent of bin size ([Supplementary Figure S4](#sup1){ref-type="supplementary-material"}). FRET histograms of YxiN/154mer/ADPNP (Figure [7B](#F7){ref-type="fig"}) show a predominant state with E~FRET~ ∼0.40, corresponding to the initial open state and high FRET states (E~FRET~ ∼0.60 and ∼0.70) indicative of the functional closed state ([@B7]) that are less populated than observed previously for YxiN in solution ([@B8]). In contrast, histograms for YxiN/32mer/ADPNP and YxiN/32mer/ATP complexes (Figure [7C](#F7){ref-type="fig"} and [D](#F7){ref-type="fig"}) showed a predominant low FRET state (E~FRET~ 0.20--0.30, 'wide-open'), smaller populations of medium FRET (E~FRET~ ∼0.50--0.60) and high FRET states (E~FRET~ ≥ 0.60) that most likely reflect the functional closed state. Non-idealized FRET histograms simply show a shift to low FRET ([Supplementary Figure S5](#sup1){ref-type="supplementary-material"}) because of fluorescence fluctuations ([Supplementary Figure S3](#sup1){ref-type="supplementary-material"}), in agreement with confocal data (Figure [6](#F6){ref-type="fig"}). The YxiN/32mer complex in the absence of nucleotides (Figure [7E](#F7){ref-type="fig"}) is predominantly in the open state (E~FRET~ ∼0.35), with a subpopulation at higher FRET (E~FRET~ ∼0.60). The FRET histograms are in qualitative agreement with FRET changes observed in confocal experiments (Figure [6](#F6){ref-type="fig"}), namely population of high FRET states in presence of the 154mer and ADPNP, and shifts to lower FRET with the 32mer and ADPNP. {#F7} To analyze the dynamic behavior of YxiN across the entire time trajectory (∼180 s), we generated FDPs that reflect the molecules that remain in the same FRET state throughout the observation time, and TDPs that measure the interconversion between individual states (Figure [7](#F7){ref-type="fig"}). The largest error in the determination of FRET efficiencies from FDPs or TDPs is 0.02 (half-width of the peak in the FDP of YxiN in absence of ligands; Figure [7A](#F7){ref-type="fig"}). Thus, FRET states with differences in E~FRET~ \> 0.02 are considered different states. For YxiN, 48% of the 184 molecules remained in the same FRET state (∼0.37 or ∼0.45, Figure [7A](#F7){ref-type="fig"}). The other 52% inter-converted between these two states (∼0.33 or ∼0.45). In contrast to free YxiN, all RNA- and nucleotide-complexes showed less conformational fluctuations, with 60--90% of the molecules remaining in the same FRET state throughout the measurement. In the YxiN/154mer/ADPNP complex, 73% of the 80 molecules analyzed remained in one state with E~FRET~ ∼0.40 (open ([@B7],[@B8])) throughout the experiment. In addition, species with E~FRET~ of ∼0.47, ∼0.55, ∼0.60 and ∼0.65 and minor populations of states with E~FRET~ ∼0.33 and ∼0.78 E~FRET~ (closed ([@B7],[@B8])) were also observed (Figure [7B](#F7){ref-type="fig"}). The TDP for the 27% interconverting molecules reveals transitions between E~FRET~ ∼0.70 (closed) and ∼0.42 (open), ∼0.70 (closed) and ∼0.52 and ∼0.42 (open) and ∼0.29 (Figure [7B](#F7){ref-type="fig"}). The state with intermediate E~FRET~ of ∼0.52 exchanges with the functional ∼0.70 state and is likely a low-populated intermediate. States with intermediate FRET efficiencies (0.45--0.65) have been observed previously for YxiN variants with mutations in the interface that is formed between the RecA domains in the closed state ([@B8]), and may correspond to YxiN molecules where the inter-domain cleft is not completely closed. The state with E~FRET~ ∼0.29 that interconverts with the open state represents a low-populated (wide-)open state. For the YxiN/32mer complex, 81% of the 284 molecules remained in the same state, predominantly in the initial open state (E~FRET~ ∼0.40, Figure [7C](#F7){ref-type="fig"}). A smaller population spreads toward lower E~FRET~ (∼0.30), in agreement with the shift to lower FRET efficiencies observed by confocal microscopy. The 19% interconverting molecules mostly switched between the open state (E~FRET~ ∼0.38) and a high E~FRET~ ∼0.68 state (Figure [7C](#F7){ref-type="fig"}) that is not significantly populated in equilibrium and therefore not detected by confocal microscopy. For the YxiN/32mer/ADPNP complex, nearly all molecules (55 of 61, 90%) remain in the same FRET state throughout the experiment. The FDP shows states of E~FRET~ ∼0.30, 0.36 and 0.67 (Figure [7D](#F7){ref-type="fig"}), with the state of E~FRET~ ∼0.30 being much more populated than the rest. Because of the low number of interconverting molecules (n = 6), a TDP plot is not statistically relevant. In the presence of 32mer and ATP instead of the non-hydrolyzable ADPNP, 60% of 194 molecules are non-interconverting. Most are trapped in a state of E~FRET~ ∼0.30, in agreement with the observed shift to lower FRET in the histograms, while some have an E~FRET~ of ∼0.37, (Figure [7E](#F7){ref-type="fig"}). The remaining 40% of the molecules undergo four types of transitions: a predominant inter-conversion of states with E~FRET~ ∼0.38 (open) and ∼0.24 (wide-open), transitions between ∼0.52 (intermediate) and ∼0.33 (open), between ∼0.76 (closed) and ∼0.55 (intermediate) and some transitions between E~FRET~ ∼0.62 and ∼0.44 (open). The largest number of transitions involves the species with E~FRET~ ∼0.24, which does not link with the functional high FRET state, and therefore represents an off-pathway intermediate (wide-open→open→closed; in contrast to a sequence of open→wide-open→closed that would be observed for an on-pathway intermediate). A summary of the observed conformational states and their interconversion is shown in [Supplementary Figure S6](#sup1){ref-type="supplementary-material"}. In summary, binding of the 154mer RNA to YxiN in the presence of nucleotide predominantly favors high FRET (closed) states. In contrast, binding of the 32mer leads to low FRET species with a wider inter-domain cleft that are off-pathway, and to the transient population of high FRET states, which rationalizes the observed stimulation of ATPase activity by these RNAs. RNA binding to the RRM allosterically activates the helicase core {#SEC3-8} ----------------------------------------------------------------- Our smFRET experiments show that RNA binding to the RRM leads to a shift in its position, and to changes in the conformational space that is accessible to the helicase core. In conjunction with ATPase and RNA unwinding experiments, these findings point to an allosteric activation of the helicase core upon RNA binding to the RRM. We tested this hypothesis by performing unwinding experiments with a 15/9mer that corresponds to the 32/9mer lacking the hairpin. An allosteric activation mechanism would predict that addition of the 32merHairpin and its binding to the RRM should stimulate unwinding of the 15/9mer. YxiN unwinds the isolated 15/9mer slowly, with a rate constant of k~unwind~ = 0.003 s^−1^ (Figure [8](#F8){ref-type="fig"}). In the context of the 32/9mer, unwinding is 8-fold faster, with a rate constant of k~unwind~ = 0.024 s^−1^. In presence of the 32merHairpin, unwinding of the 15/9mer is accelerated 2.5-fold (k~unwind~ = 0.007 s^−1^) compared to unwinding of the isolated 15/9mer. Thus, binding of the 32merHairpin, the associated movement of the RRM and the changes in the conformational dynamics of the helicase core indeed lead to an activation of the helicase core, with the RNA bound to the RRM acting as an allosteric activator. {#F8} DISCUSSION {#SEC4} ========== RNA binding to the RRM alters the position of the RRM {#SEC4-1} ----------------------------------------------------- We have shown that the C-terminal RRM of YxiN is located in defined positions relative to the core, and moves from a position close to RecA_C in the absence of RNA to the proximity of RecA_N in the RNA-bound state. The movement is driven by RNA binding to the RRM and does not require interactions of the bound RNA with the core. Nevertheless, RNA binding to the RRM affects the core: the conformation of the core is modulated, and its RNA-dependent ATPase and RNA unwinding activities are stimulated. Communication between the RNA binding site of the RRM and the active site of the core {#SEC4-2} ------------------------------------------------------------------------------------- The current model on the activation of ATP hydrolysis by RNA is based entirely on motifs identified within the core that mediate ATPase activity ([@B3],[@B49],[@B4]). RNA binding domains flanking the core were believed to function as passive anchors that mediate high affinity binding to defined elements in the RNA substrate, and the helicase core can then unwind nearby RNA duplexes ([@B18],[@B50]). Our results now show that binding of the YxiN RRM to the RNA not only anchors the helicase on the RNA, but also modulates the core conformation and its ATPase and RNA unwinding activities, demonstrating that RNA binding at the RRM is communicated to the core. This communication does not require binding of the RNA recognized by the RRM to the core, which excludes an RNA-mediated interaction of core and RRM. In the RNA-bound state, the RRM is located next to the RecA_N. RecA_N contains most of the conserved motifs for ATP binding and hydrolysis, and it would be conceivable that the RNA-bound RRM stimulates RNA-dependent ATPase activity by affecting the conformation of these motifs. However, the structural models for YxiN in the presence of RNA do no point to extensive interactions between core and RRM. In addition, the isolated RRM cannot restore RNA-dependent ATPase activity of the core when supplied in *trans*, which is also pointing to a lack of extensive interactions between the RRM and the core ([@B17]). The peptide connecting the RecA_C to the RRM has a high propensity for α-helix formation, and the end-to-end distance of such an extended α-helical peptide is sufficient to traverse the distance separating the RecA_C and the RRM. The linker therefore most likely acts as a direct communication bridge between RRM and core, and mediates the allosteric activation of core activities. RNA binding to the RRM and core conformation {#SEC4-3} -------------------------------------------- Binding of RNA to the RRM modulates the conformation of the helicase core. Switching of the core between open and closed states has been linked to RNA-dependent ATPase- and ATP-dependent RNA unwinding activities of the core ([@B6]--[@B8],[@B31]). However, instead of observing high FRET species in presence of the 32mer and 32merHairpin, species with lower FRET appeared, representing conformations of the core with increased separation of the two RecA domains. High FRET states, although not populated in equilibrium (i.e. absent in FRET histograms from confocal experiments (Figure [6](#F6){ref-type="fig"}) and in FDPs from TIRF experiments; Figure [7](#F7){ref-type="fig"}), are formed transiently (evident in TDPs; Figure [7](#F7){ref-type="fig"}), in agreement with the observed increase in ATPase activity upon binding of these RNAs. From the lack of interchange between low FRET species and the functional high FRET state in the presence of ATP, we conclude that the low FRET states induced by binding of the 32mer or the 32merHairpin to the RRM are off-pathway intermediates that are only accessed from the open state of the core. In contrast, binding of the 154mer RNA (in the presence of ADPNP) favors population of the high FRET functional state of the core. Thus, the core can adapt to the type of RNA bound by conformational plasticity. For its role in ribosome biogenesis ([@B51]) YxiN needs to distinguish unproductive assembly intermediates from functional, native ribosomes. The different level of activation of the core by the 154mer and the smaller 32mer RNA may reflect the differential recognition of these elements in their distinct structural contexts. The effect of RNA binding to the RRM on the conformational space of the core, the dominant populations and their interchange are tightly linked to the ATPase and unwinding activities of the core. Our results therefore show that the YxiN RRM is not a passive RNA anchor but undergoes an RNA-induced movement that is communicated to the core and modulates core activity. Allosteric regulation of core activities by RNA; mediated through an RNA-induced movement of ancillary domains, may constitute a general regulatory mechanism of DEAD-box proteins. Supplementary Material ====================== ###### Click here for additional data file. This project was funded by the DFG (KL1153/7-1). The authors thank Ulf Harms for assistance with the functionalization of microscope slides, Jessica Guddorf for excellent technical assistance, and Airat Gubaev for helpful discussions. SUPPLEMENTARY DATA {#SEC5} ================== [Supplementary Data](#sup1){ref-type="supplementary-material"} are available at NAR Online. FUNDING {#SEC6} ======= DFG \[KL1153/7-1\]. The open access publication charge for this paper has been waived by Oxford University Press - *NAR* Editorial Board members are entitled to one free paper per year in recognition of their work on behalf of the journal. *Conflict of interest statement*. None declared.

Inflammatory bowel disease (IBD) is a group of chronic, potentially disabling diseases of the gastrointestinal tract, mainly consisting of Crohn's disease and ulcerative colitis. According to epidemiologic studies, the incidence of IBD is rising worldwide, especially in newly industrialized countries, while the mortality of the disease remains low[@b1]. This leaves the world with an ever-growing burden of IBD[@b1]. Apart from innovations in treatment optimization, study of the underlying mechanism of pathogenesis should also be part of the research priority. Hinted by the trend of incidence changes in the west and east, environmental factors including life style are thought to be associated with, or even responsible for the rising incidence of IBD[@b2]. Antibiotic exposure in early life is among the environmental risk factors of IBD. But epidemiologic studies so far showed disparate data, indicating an important yet complicated role of it. A meta-analysis including several epidemiological reports from western countries showed a significant association between early-life antibiotic consumption and newly-onset Crohn's disease, especially among children[@b3]. However, a recent study in Asia indicated an inverse association of antibiotic use and development of IBD[@b4]. Moreover, another survey found antibiotic consumption increased the risk of CD and UC among Caucasians but decreased the risk among Middle Eastern migrants[@b5]. This disparity warrants a need for basic science researches using animal models, which exclude many confounding factors in human real life. Disturbance of gut microbiota has been proposed as the underlying mechanism of antibiotics' effect on immune-mediating diseases, including IBD, asthma and obesity[@b6][@b7]. Recent years have witnessed the thriver of microbiota research as high-throughput sequencing technique substantially extended our knowledge of previous unculturable microorganisms. Plenty of researches have targeted the shifts of gut microbial composition during and after antibiotic treatment, both in human and animals[@b8][@b9]. Through the modification of gut microbiota, antibiotics may cause profound alterations of gut epithelium, immune cells and even intestinal neural system[@b10][@b11]. It might explain why and how antibiotic-caused microbiota changes in early life can impact IBD pathogenesis. Most of the previous researches applied large doses of antibiotics or even cocktails. The modulatory effects of low-dose antibiotics on microbes have been overlooked while they were capable of altering the intestinal microbiota with lasting consequences on hosts[@b12]. Besides, antibiotic residues in foods and water supply have become an environment pollutant[@b13], which causes an extra low-dose exposure to people including children apart from iatrogenic sources[@b14]. Thus, we aimed to answer whether and how early-life exposure to certain subtherapeutic antibiotics would change the susceptibility of IBD with a mouse model of experimental colitis. In the present study, we found low-dose penicillin had an unexpected protective effect against dextran sodium sulfate (DSS) induced colitis. We further examined the perturbation of gut microbiota and immune system by penicillin, and demonstrated the protection was dependent on the eradication of segmented filamentous bacteria (SFB) in the intestine. Results ======= Low-dose penicillin exposure in early life has a protective effect against later DSS colitis in mice ---------------------------------------------------------------------------------------------------- Early life after weaning is considered a critical window for both the development of gut microbiota and the immune system[@b15]. Here we used a mouse DSS-induced colitis model to examine whether perturbed microbiota may alter the susceptibility to IBD. Before DSS treatment, mice were exposed to low-dose penicillin, metronidazole or enrofloxacin respectively in drinking water for 2 weeks, followed by a 1-week wash-out phase ([Fig. 1a](#f1){ref-type="fig"}). After DSS treatment, colitis symptoms occurred among all groups, but penicillin pre-treated mice exhibited a significantly smaller decline of bodyweight, which was most obvious on day 8 ([Fig. 1b](#f1){ref-type="fig"}). The penicillin group also seemed to have a smallest intestinal bleeding score, though it was not statistically significant ([Fig. 1c](#f1){ref-type="fig"}). Histological analysis further confirmed the disparity of tissue damage between groups. Crypt loss and leukocytes infiltration were common in all groups but mice in penicillin group seemed to preserve more crypts ([Fig. 1d,e](#f1){ref-type="fig"}). Metronidazole pre-treated mice, however, displayed more severe submucosal swelling when compared with others. In all, low-dose penicillin exposure in weanling mice seemed to play a protective role against DSS colitis. Metronidazole and enrofloxacin, on the other hand, failed to yield the same effects. Low-dose penicillin treatment suppresses *Il17* expression and ileal Th17 differentiation ----------------------------------------------------------------------------------------- Help T (Th) cells play an important role in orchestrating inflammation in both chemically induced colitis and human IBD. The protective role of penicillin pre-treatment might be due to an immune fingerprint on T cell differentiation. So we next measured the gene expression of key cytokines of different Th cells in terminal ileum tissues, mesenteric lymph nodes (MLNs) and Peyer's patches (PPs). Several cytokines were expressed differently under treatments, but penicillin decreased *Il17a* mRNA level in all three tissues consistently while metronidazole and enrofloxacin didn't ([Fig. 2a](#f2){ref-type="fig"}). *Il17f* expression in terminal ileum was also down-regulated by penicillin ([Supplementary Fig. S1](#S1){ref-type="supplementary-material"}). IL-17 is a group of pro-inflammatory cytokines implicated in host defensive against infection. They are mainly produced by Th17 cells and these cells mostly differentiate in small-intestinal lamina propia (SI-LP)[@b16][@b17]. So we next examined if differentiation of SI-LP Th17 cells was interfered by penicillin. It turned out there were fewer Th17 cells residing in SI-LP after 2-week penicillin treatment ([Supplementary Fig. S2](#S1){ref-type="supplementary-material"}) and the disparity still existed one week after penicillin cessation ([Fig. 2d](#f2){ref-type="fig"}). As CD4+ Th cells and Th1 cells remained rather intact ([Fig. 2b,c and f](#f2){ref-type="fig"}), the lack of Th17 cells is not due to an over-all suppression of T cells but a specific inhibition. Low-dose penicillin imposed transient and small changes to diversity and structure of fecal bacteria community -------------------------------------------------------------------------------------------------------------- In order to understand the gut microbial profile under low-dose antibiotic pressure, we applied 16S rRNA gene analysis of fecal bacteria from mice treated with regular water, low-dose penicillin or metronidazole, the latter two with disparate inflammation phenotypes and *Il17* expression. High-throughput sequencing produced 1842370 valid reads (with 48483 ± 14751 per sample) from 38 samples collected respectively at baseline, 2 weeks after antibiotic treatment and 1 week after treatment cessation ([Fig. 3a](#f3){ref-type="fig"}). The reads were delineated into 688 operational taxonomy units (OTUs) at the similarity cut-off of 97%. The Good's coverage index is 0.998 ± 0.0001, indicating an adequate sequencing depth, which is further proved by rarefaction analysis ([Supplementary Fig. S3](#S1){ref-type="supplementary-material"}). Estimators of alpha-diversity were calculated after rarefying the sequence depth to 19345 reads per sample, in order to avoid bias ([Supplementary Fig. S3](#S1){ref-type="supplementary-material"}). According to Chao1 and Shannon index, both low-dose penicillin and metronidazole decreased bacterial richness in 2 weeks, while penicillin on its own increased bacterial diversity ([Fig. 3b](#f3){ref-type="fig"}), which was similar to a previous study[@b12]. The impact of both antibiotics on alpha-diversity was transient, as the difference became insignificant one week later ([Fig. 3c](#f3){ref-type="fig"}). In order to preserve data from rare species, which may play important roles in gut homeostasis[@b18], we used unrarefied sequences for other analysis in our study. Principle component analysis (PCA) revealed low-dose penicillin caused unique structural changes of fecal microbiota during and after treatment, as was shown by clustering of samples in plots ([Fig. 3d,e](#f3){ref-type="fig"}). However, Venn diagrams showed the majority of OUTs were preserved and shared among groups under low-dose antibiotic treatment ([Fig. 3f--h](#f3){ref-type="fig"}). The fecal microbiota were mainly composed of Bacteroidetes and Firmicutes in all groups, followed by far less abundant Proteobacteria, Candidate division TM7 and Actinobacteria ([Fig. 3i](#f3){ref-type="fig"}). On genus level, low-dose treatment only caused small shifts of microbiota composition ([Fig. 3j](#f3){ref-type="fig"}), compared with obvious dysbiosis induced by large-dose antibiotics in previous reports[@b8][@b9]. Key bacterial alterations under low-dose penicillin treatment ------------------------------------------------------------- In order to clarify which bacteria might be responsible for low-dose penicillin's unique impact on intestinal inflammation, we conducted the LEfSe analysis for microbial biomarker discovery. By testing both the consistency and effect size of difference in bacteria abundance between groups, LEfSe suggests key bacteria at different taxonomy levels[@b19]. According to our experimental results, we combined control with metronidazole group into a super-group (each as a subgroup), as was opposed to penicillin group. According to the results, the key alteration caused by 2-week low-dose penicillin treatment was a decrease of *Enterohadus, Lactobacillus, streptococcus, Candidatus Arthromitus, Allobaculum*, and *Turicibacter*; and an increase in *Bacteroides, staphylococcus, Marvinbryantia, Roseburia, Peptococcus, Oscillibacter, Escherichia-Shigella* and *Stenotrophomonas*. At phylum level, low-dose penicillin treated mice featured a decrease of Firmicutes and an increase of Bacteroidetes and Gammaproteobacteria ([Fig. 4a--c](#f4){ref-type="fig"}). One week after penicillin cessation, the disparity between two groups became less prominent, as was indicated by a smaller number of biomarkers at all levels. The key alterations in penicillin treated mice included a decrease in *Prevotella, Gemella, Ruminococcaceae, Candidatus Arthromitus, Allobaculum, Turicibacter* and *Erysipelotrichaceae*; and an increase in *Gordonibacter, Alistipes, Rikenella, Streptococcus* and *Marvinbryantia* ([Supplementary Fig. S4](#S1){ref-type="supplementary-material"}). Some of the key bacteria are potentially associated with IBD, such as *Prevotella*[@b20][@b21][@b22], *Gemella*[@b20] and *Erysipelotrichaceae*[@b23]. Among them, *Candidatus Arthromitus*, or segmented filamentous bacterium (SFB), has been known for its special immune-stimulating capacity for years[@b24]. Due to the limitation of culturing technique, we conducted PCR to investigate the bacteria's distribution among mice. Most mice upon purchase were SFB-positive, while some were SFB-free. After one day's treatment of penicillin, the load of SFB dropped dramatically ([Fig. 4d](#f4){ref-type="fig"}). We also sought to define the minimum dose of penicillin for SFB eradication and found as little as 0.1 μg/d · g bodyweight of penicillin (equals to 0.6 mg/L penicillin in drinking water) had the same effect with higher dosages ([Fig. 4e](#f4){ref-type="fig"}). Penicillin's suppression of *Il17* expression and Th17 differentiation is SFB dependent --------------------------------------------------------------------------------------- According to previous studies, adhesive mouse-specific SFB plays a crucial role in Th17 differentiation[@b16][@b17][@b25][@b26]. So we sought to clarify if eradication of SFB is responsible for penicillin's effect. SFB's colonization occurs at weaning, and declines in an age-dependent manner[@b27] (see also [Supplementary Fig. S4](#S1){ref-type="supplementary-material"}). At week 9, all mice became free of SFB[@b27]. We then treated 9-week-old mice with low-dose penicillin, and found no difference in SI-LP Th17 count between penicillin and control groups ([Fig. 5a,b](#f5){ref-type="fig"}). In another group of SFB-free weanling mice, penicillin caused a much less and insignificant drop of *Il17a* expression ([Fig. 5c](#f5){ref-type="fig"}). To sum up, the decrease of Th17 only occurred in SFB-positive mice. Moreover, ileal Serum amyloid A (SAA), which is responsible for SFB's immune-stimulating capacity[@b28][@b29], was also suppressed by penicillin ([Supplementary Fig. S5](#S1){ref-type="supplementary-material"}). However, there is still a possibility that penicillin's effect partly relies on the enriched taxa after treatment, apart from the suppressed ones. To test this, we transplanted bacteria from caecal content of antibiotic treated mice, to 3-week SFB-positive weanling mice through gavage. The recipient mice failed to show a trend of *Il17a* suppression ([Fig. 5d](#f5){ref-type="fig"}). This further strengthened the idea that it was not the enriched genera, but the decreased genera, especially SFB, which were responsible for the penicillin-induced IL-17 inhibition. This is in consistency with a recent theory[@b10]. We also induced DSS colitis in a group of SFB-free mice. This time, the weight variance and histology showed similar degrees of inflammation among different groups ([Fig. 5e--g](#f5){ref-type="fig"}). Discussion ========== In the present study, we examined the impact of low-dose antibiotic pre-treatment on intestinal inflammation with a mouse model. Among the tested antibiotics, penicillin yielded an unexpected protective effect against DSS colitis. A 2-week penicillin treatment led to a prominent drop of *Il17a* gene expression in intestinal immune tissues and a decrease of Th17 cells in SI-LP. We also found the effect of low-dose penicillin was associated with the eradication of commensal bacteria SFB, which is highly sensitive to this antibiotic. Penicillin was the first antibiotic isolated and used against bacterial infection. The bacteriocidal effect depends on the inhibition of DD-transpeptidases by β-lactam rings, which causes a disbalance between building and breaking down of bacterial cell walls. Penicillin and other β-lactams are among the top antibiotics consumed worldwide[@b30][@b31][@b32]. A recent meta-analysis pooling risk-factor studies of IBD in western countries concluded that penicillin is the only antibiotic that wasn't associated with IBD onset, while fluoroquinolones and metronidazole were most strongly related with IBD[@b3]. In our study, a 2-week pre-treatment of enrofloxacin or metronidazole failed to aggravate later DSS colitis, while penicillin turned out to be protective. This further indicates the complexity of the association between antibiotics and IBD pathogenesis. Some basic science researches have revealed penicillin's extra-anti-infective effects on hosts through the modulation of gut microbiota. Zhang *et al*. reported that penicillin therapy could inhibit neutrophil aging and further alleviate inflammation-related tissue damage of sickle cell disease or endotoxin-induced septic shock[@b33]. Cox *et al*. first discovered that low-dose penicillin yielded a lasting growth promotion effect in mice when given in early life[@b12]. Our results enhance the knowledge of low-dose penicillin, showing only a 2-week short-period intervention could leave animals in a less inflammation-favoring state. Gut microbiota have long been recognized as key factors in IBD pathogenesis. Many IBD-associating genes and pathways have roles in microbial defense and intestinal immune homeostasis[@b34]. A number of environmental risk factors of IBD can also be triggers of gut microbiota alterations[@b35]. There are several specific microorganisms considered as possible causative agents of the disease, such as *Mycobacterium avium* subspecies paratuberculosis (MAP), and *Escherichia coli*[@b36]. With the help of low antibiotic dose, which didn't cause severe dysbiosis and suppressed only a few genera, we were able to define a group of key bacteria and highlight the importance of SFB. SFB, or *Candidatus Arthromitus* as the genus name, is a commensal bacterium mainly inhabits the small intestine, forming attachment sites on epithelia cells[@b37]. SFB genome sequencing revealed a lack of ability of biosynthesis, indicating that its life style is highly dependent on host cells[@b38]. SFB had been unculturable, until recently, a team developed an SFB-host cell co-culturing system[@b26]. Previous studies with electron microscopes found prominent morphology changes of SFB within 3--5 hours after the hosts were given a large dose of penicillin, inferring its sensitivity to penicillin[@b39]. We think the high demand for cell wall construction for numerous intracellular septa during a rapid life cycle is the reason why SFB could be eradicated with such a low dose of penicillin in our experiments[@b26]. Low-dose metronidazole, while decreasing SFB load, was not able to achieve a complete wipe-out. This makes penicillin a specific and ideal tool for SFB manipulation. The importance of SFB lies on its unique ability to elicit physiological inflammation. When adhering to epithelial cells, SFB is able to promote Th17 differentiation, IgA secretion and antimicrobial peptide production[@b26]. Though commonly detected in various young animals as a commensal[@b27], SFB is reported to exacerbate autoimmune encephalitis and arthritis in mouse model[@b40][@b41]. In the case of intestinal inflammation, SFB could induce colitis in SCID mice transferred with CD4^+^CD45RB^high^ T cells, when co-colonizing with a cocktail of other bacteria[@b42]. SFB was also found highly coated with IgA, which turned out as a marker of colitogenic microbes[@b23]. Moreover, though not usually found in grown-up animals or adult human, there are case reports providing evidence that SFB could be detected at certain inflammatory sites both in UC and CD patients[@b43][@b44]. Our finding reinforces the notion that SFB, though not able to cause intestinal inflammation on its own, could be part of the pathogenesis[@b42]. SFB's pro-inflammatory potential was mainly attributed to its ability to inducing IL-17-producing cells, including Th17. Similar to a previous report[@b12], we found down-regulation of Th17 cells in SI-LP by low-dose penicillin. Th17 cells are an important part of the immune system. Researchers have discovered an enrichment of balancing selection in genes related to IL-17 production[@b45]. This implies a need for balance between microbial defense and inflammation during evolution[@b45]. Under normal condition, Th17 cell can promote neutrophil function and antibacterial peptide production, but they may also trigger autoimmune diseases including IBD, rheumatoid arthritis, psoriasis and multiple sclerosis[@b46]. Several genes in IL-17/IL-23 pathway have been observed as IBD risk loci, including IL-23R, IL-22, RORC, IL-21, Stat3 and AHR[@b34][@b47]. SFB is recently reported to induce Th17 differentiation through activation of an IL-23R/IL-22 circuit[@b28]. Thus, SFB and IL-17/Th17 are closely associated with IBD pathogenesis. However, whether the alleviation of DSS colitis was caused by Th17 inhibition is still to be elucidated. IL-17, though usually viewed as a pro-inflammatory cytokine, plays a rather complicated role in human IBD as well as experimental colitis. While the dual blockade of IL-17A and IL-17F attenuated intestinal inflammation[@b48]. Th17-deficient mice presented even more severe DSS colitis than wild type ones[@b49]. While antibodies targeting IL-17A proved ineffective in treating CD patients[@b50], some small-molecule inhibitors have shown efficacy in clinical trials[@b51][@b52]. Overall, there is a complex and delicate balance between the protective and pathogenic roles of IL-17 and Th17 cells. Targeting IL-17 during a more acute phase of intestinal inflammation, or even as a prophylactic therapy ahead of disease flare, could be a more rational approach[@b53]. Here in the present study, why did the inhibition of IL-17 yield anti-inflammatory results? We put forward several explanations. First of all, penicillin achieved a dual suppression of *Il17a* and *Il17f* expression, the latter as a more pro-inflammatory cytokine. Secondly, penicillin-treated mice were not deprived of all Th17 cells, which may preserve more protective capacity in compare with the case of IL-17 knock-out mice. Thirdly, the pre-treatment had put mice in a low-Th17 state before DSS colitis, mimicking a prophylactic therapy. DSS colitis is featured with massive neutrophil infiltration in intestinal tissue. Th17 cells and neutrophils can reciprocally recruit each other in inflamed sites, forming a vicious cycle[@b54]. When Th17 cell differentiation is inhibited by penicillin, the vicious cycle can be less strong. Overall, we suggest that the down regulation of IL-17 and Th17 could be responsible for penicillin's protective effects through eradication of SFB. There are several limitations in our research. We failed to conduct further experiments with other bacteria that also had biological importance according to LEfSe analysis, especially Turicibacter and Allobaculum, of which behavior was highly similar to SFB under penicillin pressure. Secondly, we used conventional mice to achieve a more real-life results with a more diverse microbiome, yet at the possible risk of inter-study variations. Further researches with germ-free or gnotobiotic mice together with new culturing methods for key bacteria are needed for better control of the composition of gut microbiota. Thirdly, we applied DSS colitis model in the study, while chemically induced inflammation cannot fully recapitulate human IBD. However, there is another report in which the authors directly examined the colitogenic effect of SFB-containing bacterial cocktails with a T-cell transferred colitis model[@b42]. Interestingly, they found that SFB together with a cocktail of bacteria were effective to trigger intestinal inflammation, which is very supportive of our conclusion. Genetically modified mice should be used in future to elucidate the compound effects of SFB and IBD risk genes especially in IL-17/IL-23 pathway. In conclusion, through a highly specific manipulation of sensitive bacteria, a short-term treatment of low-dose penicillin has shown its potential to interfere with development of intestinal immune system in early life. It may partly answer the epidemiologic question why different antibiotic exposure imposed distinct risk on IBD onset. Our study may further indicate the possibility of achieving a favorable immune state among a certain group of patients with IBD, or other autoimmune diseases, by fine-tuning the gut microbiota. Methods ======= Animals and antibiotic treatment -------------------------------- Three-week-old C57BL/6 mice were purchased from SLAC laboratory (Shanghai, China) and maintained under conventional barrier conditions with a 12-hour light/dark cycle in Shanghai Jiao-Tong University, School of agriculture and biology, or the experimental animal centre of Tongji University. All cages and water were autoclaved prior to use, with bedding and chow irradiated. Low dose antibiotics were supplied in drinking water *ad libitum* at the concentration of 6.67 mg/L, which equals to 1 μg/g bodyweight according to previous reports[@b12][@b55]. Antibiotic intervention began on the next day of animal receipt and lasted for 2 weeks. In order to maintain the concentration of antibiotics against degradation, penicillin and metronidazole were renewed every day, with enrofloxacin every three days[@b56][@b57][@b58]. All animal experimentations were approved by the scientific research ethical committee of Renji Hospital affiliated to Shanghai Jiao-Tong University. Procedures involving animals and their care were conducted following NIH guide for the care and use of laboratory animals (eighth edition). Induction of colitis and determination of clinical scores --------------------------------------------------------- Acute colitis was induced with 2.5% (w/v) DSS (molecular weight 36000--50000, MP Biomedicals, US) dissolved in drinking water for 7 days. As described in a protocol, newly-made DSS solution was given on day 0, day 2 and day 4, followed by regular water on day 7 until termination of the experiment on day 8--10[@b59]. DSS solution was always given 1 week after antibiotic cessation, in order to exclude the direct effect of antibiotic residue in animals. Body weight was determined every day, as well as stool consistency and rectal bleeding state. Occult blood was assessed with a urine-fecal occult blood test kit. An established scoring system for intestinal bleeding was used as an index of disease activity[@b59]. Hemotoxylin and eoxin (HE) staining and histologic analysis ----------------------------------------------------------- Following euthanasia on day 9, 0.5 cm of distal colons were dissected, washed in saline and fixed in 4% paraformalin for 24 h at 4 °C. The tissues were then embedded in paraffin and sectioned to 4 μm thickness. HE staining was performed by an autostaining machine following standard protocols, while images acquired by an Olympus BX43 microscope. Based on a previously described scoring system, each tissue was assigned with four scores for epithelial damage, inflammatory infiltration of mucosa, submucosa and muscularis/serosa. The scores were multiplied with an index according to the extent of tissue damage[@b60]. Microbial community analysis by 16S rRNA gene sequencing using Illumina technology ---------------------------------------------------------------------------------- Mouse feces were collected into autoclaved eppendorf tubes and stored at −80 °C after snap frozen in liquid nitrogen. Fecal genomic DNA was extracted with OMGA-soil DNA kit following the manufacturer's instruction. Hypervariant region V3-V4 of bacterial 16S rRNA gene was amplified with the primer 338F (5′-barcode-ACTCCTACGGGAGGCAGCA-3′) and 806R (5′-ACTCCTACGGGAGGCAGCA-3′) by PCR. PCR reactions were carried out in 20 μL mixtures containing 10 ng of template DNA, 0.8 μL of each primers (5 μM), 0.4 μL of FastPfu polymerase (TransGen, Beijing), 2 μL of 2.5 mM dNTPs, and 4 μL of 5\* FastPfu buffer. Reaction condition was 3 mins at 95 °C, and then 27 cycles of 30 s at 95 °C, 30 s at 55 °C and 45 s at 72 °C, and a final extension phase of 10 mins at 72 °C. Products were purified with AxyPrep DNA Gel Extraction Kit (Axygen Bioscience, CA, USA) and quantified with Quantifluor-ST (Promega, US). Purified amplicons were sequenced on an Illumina MiSeq platform at Majorbio Bio-pharm Technology Co. Ltd according to the standard protocols. Processing of sequencing data ----------------------------- Raw fastq files were quality-filtered using QIIME[@b61] (version 1.17). Reads which could not be assembled were rejected. Operational taxonomic units (OTUs) were clustered by UPARSE[@b62](version 7.1 <http://drive5.com/uparse/>) with 97% similarity and chimeric sequences were identified and removed using UCHIME. The taxonomy of each sequence was analyzed by RDP Classifier against Silva (SSU115) 16S rRNA database with 70% confidence threshold[@b63][@b64]. Rarefaction analysis was performed by Mothur and alpha-diversity indexes were compared using rarefied data. Principle component analysis (PCA) plot and Venn diagram were implemented by R programming language. Significant changes in relative abundance of microbial taxa were detected by linear descriminant analysis effect size (LEfSe) method[@b19]. Segmented Filamentous Bacteria (SFB) quantification by real-time PCR -------------------------------------------------------------------- Fecal genomic DNA was extracted by Tiangen Fecal Genomic DNA Extraction Kit under the manufacturer's instruction. Quantitative PCR was carried out by Applied Biosystems 7500 with SYBR Premix Ex Taq quantitative PCR Kit using the universal bacteria primers (5′-ACTCCTACGGGAGGCAGCAGT-3′ and 5′-ATTACCGCGGCTGCTGCG-3′)[@b17], mouse-specific SFB primers (5′-TGAGCGGAGATATATGGAGC-3′ and 5′-CATGCAACTATATAGCTATATGCGG-3′) and rat-specific SFB primers (5′-TGAAGCGGAGGTAGATGGA-3′and 5′-GCAACTATATAGCTGTATGCGG-3′)[@b29]. PCR reactions were performed as indicated by Takara protocol. The results were expressed as fold changes of relative abundance variation of SFB. Each mouse was tested for SFB upon arrival. Tissue RNA extraction and quantification of gene expression with real-time PCR ------------------------------------------------------------------------------ Tissues were stored in RNAlater (Qiagen) at −20 °C upon harvesting. Total RNA was isolated using TRIzol (Invitrogen, CA, USA) according to the manufacturer\'s instruction. Quantification real-time PCR was performed with SYBR Premix Ex Taq quantitative PCR Kit in a StepOnePlus apparatus. Results were analyzed with ΔΔCT method and normalized to the gene *Gapdh*. Primers used in the experiment are included in [Supplementary Table S1](#S1){ref-type="supplementary-material"}. Small intestinal lamina propria lymphocyte isolation and flow cytometry ----------------------------------------------------------------------- Mouse ilea were collected at sacrifice. Peyer\'s patches were carefully excised. Ilea were then cut longitudinally and washed with saline to fully remove intestinal content. Before further processing, tissues were stored at 4 °C in complete 1640 containing 5% fetal bovine serum (FBS; Gemini bio, US), 200 U/ml penicillin, 200 μg/ml streptomycin, 10 μg/ml amphotericin B and 100 μg/ml gentamycin. Mucus and epithelium were removed by two 20 mins washes at 37 °C in calcium-free PBS containing 5 mM dithiothreitol (DTT), 5 mM EDTA, 5% FBS, 200 U/ml penicillin, 200 μg/ml streptomycin, 10 μg/ml amphotericin B and 100 μg/ml gentamycin. The remaining tissues were excised into fine pieces and washed in PBS, and then digested in complete 1640 containing 1% collagenase (Worthington Biochemical, US), 1% DNase (Sangon, Shanghai, China) and 2% FBS, at 37 °C and 150 rpm, for 20 mins. After a 7 s vortex for adequate isolation, cells were eluted by passing through a 70-micron strainer, washed with PBS and resuspended in 40% Percoll (GE Healthcare, US) in complete 1640. Cell suspension were laid onto 80% Percoll and centrifuged at 750 g for 15 mins at 4 °C without braking. Lamina propria mononuclear cells (LPMCs) were collected at the interface and used for further experiment. For flow cytometry analysis, LPMCs were stimulated with 50 ng/ml phorbol-12-myristate 13-acetate (PMA; Sigma, US) and 750 ng/ml ionomycin (Sigma) for 4 hrs at 37 °C in the presence of monensin (BD biosciences, US). Following stimulation, cells were permeablized by Perm/Wash buffer set (BD biosciences) under the manufacturer\'s instruction and stained with anti-CD4-FITC (BD biosciences), anti-TCRβ-PE (BD biosciences), anti-IL-17A-PE (BD biosciences) and anti-IFNγ-APC (ebioscience). Cells were acquired on a Fortessa flow cytometry and analyzed with Flowjo 7.6.1 software. Manipulation of gut microbiome by gavage ---------------------------------------- Cecal content from donor mice were taken after sacrifice, suspended in 10 volumes (w/v) 20% glycerol/PBS solution and stored at −80 °C. Upon use, frozen suspension was thawed and centrifuged at 4 °C for 8 mins (8000 g). The pellet was resuspended with the same volume of PBS. 100 μL of the suspension was given to each recipient by oral gavage using 22 ga \* 25 mm plastic feeding tubes. Gavage was performed every week for 3 weeks, in order to maintain a stable microbial composition. No adverse effect was observed during experiments. Statistical analysis -------------------- Data are presented as mean ± SEM. Shapiro-wilk test and Levene\'s test were used for determination of normality and homogeneity of variance, respectively. According to distribution of data, comparisons between different groups were carried out with one-tailed unpaired *t*-test (between 2 groups), one-way ANOVA corrected for multiple comparison with SNK tests (among more than 2 groups), or Wilcoxon rank sum test (for data unsuitable for *t*-test or ANOVA). Differences were considered statistically significant at *P* \< 0.05. Analysis was performed with SPSS 19 software. Additional Information ====================== **How to cite this article:** Jin, S. *et al*. Low-dose penicillin exposure in early life decreases Th17 and the susceptibility to DSS colitis in mice through gut microbiota modification. *Sci. Rep.* **7**, 43662; doi: 10.1038/srep43662 (2017). **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary Material {#S1} ====================== ###### Supplementary Material This work was supported by the grants from the National Science Foundation of China (No. 81470820, 81170362, No. 81600435 and No. 81370508), and the National Key Technology Research and Development Program of China (No. 2012BAI06B03). We also thank Prof. Yitong Lu and Prof. Tao Sun from Shanghai Jiao-Tong University School of Agriculture and Biology for their kind help to provide animal housing facility. The authors declare no competing financial interests. **Author Contributions** S.J., D.Z., Z.R. and Q.Z. conceived the experiment. S.J., C.C. and D.S. conducted the experiments and. S.J. wrote the manuscript. J.S., A.X. and Y.Q. helped with the statistical analysis and revised the manuscript. All authors reviewed the manuscript. {#f1} {#f2} {#f3} {#f4} {ref-type="fig"} (**f**) Histologic score of mice from e based on hemotoxylin and eoxin (HE) staining. Mice were sacrificed on day 8. (**g**) HE staining of the distal colon. Each panel is representative of tissue analyzed in (**f**). C: control, P: penicillin, E: enrofloxacin, M: metronidazole. ^ns^: not significant, one-way ANOVA (**c**--**f**), Student\'s *t*-test (**b**).](srep43662-f5){#f5}

Cancer is a major cause of mortality worldwide.[@b1] Lung cancer is the leading cause of cancer-related deaths, accounting for 20% of all cancer deaths.[@b2] Lung cancer is divided into small-cell lung cancer and non-small cell lung cancer (NSCLC), and the latter accounts for more than 80% of all lung cancer types, including adenocarcinoma (AC), squamous cell carcinoma (SCC) and large-cell carcinoma.[@b3] The overall 5-year survival rate of lung cancer has merely improved from 12% to 16% over the recent three decades, as observed in 80% of patients diagnosed with metastatic disease and more than half of patients with distant metastases.[@b4] Early detection and improved monitoring of lung cancer are urgently needed. Sputum cytology and examination of blood biomarkers, including circulating DNA and RNA, exosomal microRNA, circulating tumor cells, and antigens, are potential approaches for the early detection of lung cancer.[@b5] However, although some biomarkers, such as carcinoembryonic antigen (CEA) and cytokeratin 19 fragment antigen 21-1 (CYFRA 21-1), have been used to screen for lung cancer, the specificity and sensitivity remain unsatisfactory.[@b6],[@b7] Therefore, novel tumor biomarkers and tests need to be developed.[@b8] In a previous study, we used NSCLC and normal lung tissues to construct libraries of differentially expressed genes, and found that endoplasmic reticulum-Golgi intermediate compartment protein 3 (ERGIC3) is strongly overexpressed in NSCLC, and this overexpression promotes proliferation and migration of NSCLC cells.[@b9] ERGIC3 (Erv46 and ERp43) is located in the *cis* face of the Golgi apparatus and vesicular tubular structures between the transitional endoplasmic reticulum (ER) and *cis*-Golgi.[@b10] ERGIC3 significantly affects cell growth and causes ER stress-induced cell death, and is involved in the invasion and metastasis in hepatocellular carcinomas (HCC).[@b11],[@b12] Considering the high sensitivity and specificity of ERGIC3 expression in NSCLC, and the roles of ERGIC3 in cancer development and progression, in embarking on this study we believed that ERGIC3 may be a potential biomarker of NSCLC.[@b13] In this study, we developed a new murine monoclonal antibody (mAb) against ERGIC3, and systematically investigated ERGIC3 expressions in a broad variety of normal human tissues and various types of human tumors. The mechanisms of ERGIC3 regulation were also studied. Materials and Methods ===================== Cell lines and tissue samples ----------------------------- Lung cancer cell lines, including A549 (AC), 801-D (large cell carcinoma), EPLC-32M1 (SCC), NCI-H292 (mucoepidermoid carcinoma), and XLA-07 (AC),[@b14] were cultured with RPMI 1640 containing 10% FBS. 16HBE (an immortalized human bronchial epithelial cell line) was maintained in DMEM containing 10% FBS. Normal adult human tissues were obtained from surgical specimens or following autopsies performed \<10 h after death. Only specimens maintaining good histological preservation were used. Among these, 23 lung tissue samples obtained from patients with bullous lung disease and inflammatory pseudotumors through surgical operation were used as "normal lung tissues." A total of 110 resected tumors, including 15 various types, were studied. The tumor types were identified based on microscopic examination according to the international classification.[@b15] All tissue samples were fixed in 10% buffered formalin and embedded in paraffin for immunohistochemistry. Seven cases of fresh NSCLC and adjacent nonmalignant lung tissues were used for isolation of RNA and proteins. The study was approved by the local research ethics committee. Preparation of mAb ------------------ For generation of mAb to ERGIC3, a peptide of ERGIC3 (Gene ID: 51614) coupled to KLH via an N-terminal cysteine was synthesized by GL Biochem, Shanghai, China (Fig.[1a](#fig01){ref-type="fig"}). Eight-week-old BALB/c mice were immunized subcutaneously with the ERGIC3 peptide emulsified using complete Freund's adjuvant (Sigma-Aldrich, St. Louis, MO, USA). A booster injection was given intraperitoneally with the peptide emulsified using incomplete Freund's adjuvant (Sigma-Aldrich) 3 days before the animals were killed. Titers of anti-serums to ERGIC3 were assayed by solid-phase ELISA. Western blot analysis was used to evaluate whether the anti-serum could recognize the native ERGIC3 protein. When titers of ERGIC3-specific antibodies reached 1:10 000, the immunized mice were killed and the spleen was removed for cell fusion.[@b16] All experiments were carried out according to the regulations for animal experimentation and approved by the appropriate authorities. {#fig01} The spleen cells were fused with SP2/0 myeloma using 50% polyethylene glycol solution, and then cultivated in hypoxanthine, aminopterin and thymidine medium. The candidate hybridomas that secreted the antibody with high titer and desired specificity were screened and identified by ELISA. To ensure that the mAb was secreted by the progeny of a single cell, sub-cloning was performed by limiting dilution. ELISA assay ----------- Antigens were coated on microplates. The plates were blocked with 2% BSA, and then incubated with antibodies. Subsequently, peroxidase-labeled anti-mouse antibody was added, followed by color reaction developed with o-phenylenediamine dihydrochloride. After the reaction was stopped with 2.5 M sulfuric acid, the optical density (OD) was determined using a microplate reader (BioRad, Hercules, CA, USA) at 490 nm. Western blot ------------ Total protein was prepared from cultured cells and tissues through protein lysis. The lysate was centrifuged. Supernatants were boiled in sodium dodecyl sulfate-polyacrylamide gel reducing buffer, separated by electrophoresis, and then transferred to polyvinylidene difluoride transfer membrane. After blocking with 2% BSA, the membrane was incubated with antibodies, and then treated with peroxidase-conjugated anti-mouse antibody, followed by the addition of chemiluminescence reagents (Thermo Fisher Scientific, Waltham, MA, USA). GAPDH was used as an internal control. Immunofluorescence staining --------------------------- Cells were seeded on coverslips. The slides were fixed with ice-cold acetone and the permeations were performed using digitonin for 10 min. The cells were blocked with 2% BSA and incubated with antibodies. Subsequently, the slides were incubated with FITC-conjugated goat anti-mouse antibody. The nuclei were counterstained with 4,6-diamidino-2-phenylindole. Finally, the slides were analyzed under confocal laser scanning microscopy. Immunohistochemistry -------------------- The paraffin sections were deparaffinized, and the antigen retrieval was performed using 10 mM sodium citrate buffer through pressure cooker processing for 5 min. Endogenous peroxidase activity was eliminated with 3% H~2~O~2~. The sections were blocked using 2% BSA, incubated with antibodies, and then treated with peroxidase-labeled goat anti-mouse antibody. The color development was performed with 3,3′-diaminobenzidine. Hematoxylin was used for counterstaining. Negative controls were performed using SP2/0 cell supernatant instead of the antibody. Quantitative real-time polymerase chain reaction ------------------------------------------------ Total RNA was isolated from the cultured cells and tissues using TRIzol (Invitrogen, Carlsbad, CA, USA). The cDNA was synthesized with M-MLV Reverse Transcriptase (Promega, Madison, WI, USA) using RNA samples as the template. The quantitative RT-PCR (qRT-PCR) assay was performed for ERGIC3 and internal control (β-actin) as described previously.[@b9] All samples were run in triplicate to ensure quantitative accuracy, and the threshold cycle numbers (*C*~t~) were averaged. Upregulation or downregulation of target genes was determined using the method.[@b17] For examination of miRNA, total RNA were used in 5× miScript HiSpec Buffer (QIAGEN, Suzhou, China) according to the manufacturer's protocol. The cDNA was prepared using the miScript II RT Kit (QIAGEN). The qRT-PCR assay was performed using a target-specific primer and the miScript SYBR Green PCR Kit (QIAGEN), which contains the miScript Universal Primer (reverse primer) and QuantiTect SYBR Green PCR Master Mix. The U6 was used as an internal control. The *C*~t~ of target miRNA were determined as described above. All primers used in the study are listed in Supplementary [Table S1](#sd1){ref-type="supplementary-material"}. Bioinformatics prediction ------------------------- The miRNA that may target ERGIC3 were predicted through bioinformatics analysis using algorithms, RNAhybrid (<http://bibiserv.techfak.uni-bielefeld.de/rnahybrid>) and miRecords (<http://mirecords.biolead.org/prediction_query.php>). MiRNA expression profiling -------------------------- MiRNA expression profiling was determined in cultured cells through miRNA sequencing using Hiseq 2000 platform by BGI Tech (Shenzhen, China: <http://bgitechsolutions.com>). Differentially expressed miRNA were found through comparing miRNA expression profiling of NSCLC cells (A549, 801D, EPLC-32M1) with that of 16HBE. Transient transfection of miRNA ------------------------------- Cultured cells were seeded in 12-well plates at a density of 30--50%, and then transfected, respectively, with 150 nM has-miRNA mimics or inhibitors, as well as the same amount of mimic or inhibitor controls (RiboBio, Guangzhou, China) using lipofectamine 2000 (Invitrogen) according to the instruction manual. Luciferase reporter assay ------------------------- The predicted target site of miR-203a was amplified from the genomic DNA of EPLC-32M1 cells (primers in Suppl. [Table S1](#sd1){ref-type="supplementary-material"}) and was inserted into the *KpnI* and *HindIII* restriction sites of the pGL-3 basic vector (Promega). The cells were co-transfected with the pGL-3 luciferase construct and Renilla luciferase plasmid (pRL-TK) along with the miR-203a mimic or its negative control using Lipofectamine 2000. The pRL-TK plasmid was used as an internal control. After transfection, the luciferase activities were measured using Dual-Luciferase Reporter System (Promega) according to the manufacturer's instructions. Cell viability assay -------------------- Cell viability and proliferation was analyzed by using WST-1 assay. At 48, 72 and 96 h after the treatments, the WST-1 reagent (Roche Molecular Biochemicals, Rotkreuz, Switzerland) was added and incubated for 2--3 h at 37°C. The absorbance of converted dye was measured at 490 nm by microplate reader (BioRad). Statistical analysis -------------------- Data of mRNA and protein levels, as well as cellular proliferation were analyzed using the paired *t*-test. All of the values were evaluated using IBM SPSS 19 (SPSS, Chicago, IL, USA). Differences were considered significant if *P* \< 0.05. Results ======= A new monoclonal antibody to ERGIC3 was developed ------------------------------------------------- After cell fusion was performed, six strongly positive clones were obtained (Suppl. [Fig. S1a](#sd1){ref-type="supplementary-material"}). A monoclonal hybridoma was established from 06-C4 after three rounds of sub-cloning (Suppl. [Fig. S1b](#sd1){ref-type="supplementary-material"}). The mAb secreted by the monoclonal hybridoma was named "6-C4." The isotype of mAb 6-C4 was IgG2b (κ light chain). 6-C4 reacted with the ERGIC3 peptide and the native protein extracted from NSCLC cells, but not with BSA, and plasma, saliva and urine samples from three normal adults, as revealed by ELISA (Fig.[1b](#fig01){ref-type="fig"}). A clear single band was detected at approximately 50 kD using 6-C4 by western blot analysis with the native protein, similar to the preliminary result using the sera of immunized mice (Fig.[1c](#fig01){ref-type="fig"} and Suppl. Fig. S1c). The immunofluorescence staining of 6-C4 was localized around the Golgi apparatus and the ER (Fig.[1d](#fig01){ref-type="fig"}). NSCLC and HCC tissues were strongly stained by immunohistochemistry using 6-C4 (Suppl. [Fig. S1d](#sd1){ref-type="supplementary-material"}). These observations are consistent with our previous finding in which anti-ERGIC3 serum (Abcam, Cambridge, UK) was used, and a previous study.[@b9],[@b12] These results indicated that 6-C4 specifically recognizes ERGIC3. ERGIC3 expression was determined in normal adult human tissues -------------------------------------------------------------- Studies have not been previously conducted on ERGIC3 expression in a broad variety of normal human tissues. Therefore, we examined the expressions of ERGIC3 in various normal human tissues using 6-C4; the results are shown in Table[1](#tbl1){ref-type="table"} and Figure[2](#fig02){ref-type="fig"}. Most normal tissues were not stained with 6-C4. However, the cytoplasm of some epithelial cells was positively stained. By contrast, all non-malignant lung tissues were negative for 6-C4 staining. ###### Immunohistochemical analysis of ERGIC3 in normal human tissues by using mAb 6-C4 Tissue ERGIC3 staining ----------------- ----------------- Cerebral cortex 0/3 Cerebellum 0/1 Heart 0/4 Lung 0/23 Liver 10/11 Gallbladder 0/2 Pancreas 5/6 Esophagus 0/3 Stomach 5/5 Intestine 2/2 Colon 3/3 Kidney 6/6 Urinary bladder 0/3 Testis 0/3 Prostate 1/5 Breast 3/4 Ovary 0/3 Uterus 0/2 Thyroid gland 0/5 Spleen 0/3 Thymus 0/4 Muscle 0/3 Skin 0/3 {#fig02} ERGIC3 expression was determined in various tumor tissues --------------------------------------------------------- Immunohistochemical results of ERGIC3 in 15 types of human tumors using 6-C4 are shown in Table[2](#tbl2){ref-type="table"} and Figure[3](#fig03){ref-type="fig"}. ERGIC3 was strongly expressed in all carcinomas originating from the epithelial cells, but all sarcomas were negative for 6-C4. ###### Immunohistochemical analysis of ERGIC3 in various tumor tissues by using mAb 6-C4 Cancer ERGIC3 staining ------------------------------------ ----------------- Non-small cell lung cancer (NSCLC) 21/22 Pancreatic carcinoma 4/4 Hepatocellular carcinoma (HCC) 4/4 Gastric carcinoma 4/4 Esophagus carcinoma 10/10 Colorectal carcinoma 5/5 Renal cell carcinoma 9/9 Bladder carcinoma 11/12 Mammary carcinoma 5/6 Cervical carcinoma 9/10 Prostate cancer 9/11 Thyroid carcinoma 5/5 Osteosarcoma 0/3 Chondrosarcoma 0/2 Fibrosarcoma 0/3 {#fig03} Using western blot with 6-C4, all cultured NSCLC cells expressed higher levels of ERGLC3 than 16HBE. (Fig.[4a,b](#fig04){ref-type="fig"}). This expression pattern of ERGIC3 protein is consistent with that of ERGIC3 mRNA (Fig.[4d](#fig04){ref-type="fig"}). {#fig04} ERGIC3-related miRNA were identified in non-small cell lung cancer cells ------------------------------------------------------------------------ Many cancer-related genes are regulated by miRNA.[@b18] By bioinformatics analysis we predicted that 398 miRNA may bind to ERGIC3 mRNA. Moreover, we found 87 consensus differentially expressed miRNA by comparing the miRNA profiles of NSCLC cells (A549, 801D and EPLC-32M1) with those of 16HBE. Integrating the candidate miRNA predicted by bioinformatics analysis and the differentially expressed miRNA detected by the miRNA expression profiling, we found that miR-140-3p and miR-203a may target ERGIC3 and that they were differentially expressed in NSCLC cells. Therefore, the two miRNA were selected for further research. MiR-490-3p participates in ERGIC3 regulation in HCC;[@b12] as such, miR-490-3p was also investigated. Through qRT-PCR, it was found that MiR-140-3p expression was higher in NSCLC cells than in 16HBE (Suppl. [Fig. S2a](#sd2){ref-type="supplementary-material"}); by contrast, miR-203a expression was lowed in NSCLC cells than in 16HBE (Fig.[4c](#fig04){ref-type="fig"}). The qRT-PCR results of miR-140-3p and miR-203a are also consistent with those of miRNA expression profiling. No significant difference in miR-490-3p expression was observed between NSCLC cells and 16HBE (Suppl. Fig. S2c). Interestingly, miR-203a expression levels were negatively correlated with ERGIC3 expression (Fig.[4c,d](#fig04){ref-type="fig"}). ERGIC3 levels were negatively correlated with miR-203a in non-small cell lung cancer ------------------------------------------------------------------------------------ MiR-203a was downexpressed in NSCLC tissues compared with adjacent non-malignant tissues (Fig.[4f](#fig04){ref-type="fig"}); by contrast, ERGIC3 was upregulated in NSCLC tissues (Fig.[4e](#fig04){ref-type="fig"}). ERGIC3 and miR-203a expression were negatively correlated in patients' tissues, similar to that in cultured cells. ERGIC3 expression was regulated by miR-203a in non-small cell lung cancer cells ------------------------------------------------------------------------------- MiRNA can downregulated gene expression through either translation inhibition or mRNA degradation via binding to mRNA.[@b18] In bioinformatic analysis, miR-203a can bind to ERGIC3, and the sequence alignment of the miR-203a binding site is conservative among primates, even in mammals (Fig.[5a](#fig05){ref-type="fig"}). A luciferase reporter assay conducted using ERGIC3 construct that contained the potential binding site of miR-203a showed that the luminescence intensities were significantly reduced by the miR-203a mimic (Fig.[5a](#fig05){ref-type="fig"}). Furthermore, the miR-203a mimic treatment increased miR-203a expression (Fig.[5b](#fig05){ref-type="fig"}) and decreased ERGIC3 expression (Fig.[5c--e](#fig05){ref-type="fig"}). However, ERGIC3 expression was increased by the miR-203a inhibitor (Fig.[5f--h](#fig05){ref-type="fig"}). Unfortunately, ERGIC3 expressions were not significantly changed by miR-140-3p and miR-490-3p mimic treatments (Suppl. Fig. S2b,d). {#fig05} MiR-203a expression affected cellular morphology and proliferation ------------------------------------------------------------------ Considering the novel discovery of miR-203a in NSCLC, we further investigated the possible functions of miR-203a. XLA-07 (with low endogenous miR-203a level) and 16HBE (with high endogenous miR-203a level) were treated with the miR-203a mimic and inhibitor, respectively. The MiR-203a mimic caused an evident morphological change (from the spindle-like appearance to the round shape, Fig.[6a](#fig06){ref-type="fig"}) and significantly reduced the cell proliferation in XLA-07 (Fig.[6c](#fig06){ref-type="fig"}). Notably, changes induced by the miR-203a mimic were similar to the alterations caused by ERGIC3 RNA interference in NSCLC cells.[@b9] However, significant alterations in cellular morphology and proliferation were not observed in 16HBE after miR-203a inhibitor treatment was administered (Fig.[6a,b](#fig06){ref-type="fig"}). {#fig06} Discussion ========== Cancer is a disease caused by the accumulation of genetic and epigenetic alterations. As such, expressions and functions of cancer-related genes should be investigated to understand carcinogenesis and to identify cancer biomarkers. In a previous study, ERGIC3 was found to be a novel lung cancer-related gene. In addition, overexpressed ERGIC3 promotes cellular proliferation and migration.[@b9] ERGIC3 is coupled with Erv41p to form an integral membrane protein complex as components of COPII vesicles, which play an important role in protein transportation via the secretory pathway.[@b19]--[@b21] Abnormal ERGIC3 expression can affect cell growth and ER stress-induced cell death, and contribute to epithelial to mesenchymal transition.[@b11],[@b12] Thus, ERGIC3 may be a potential cancer biomarker. In this study, a new mAb (6-C4) against ERGIC3 was developed. The proposed antibody is well suited for immunohistochemistry, immunoblotting and ELISA. To our knowledge, 6-C4 is the first murine mAb towards ERGIC3. 6-C4 could be used to investigate the expression and functions of ERGIC3. We systematically studied expressions of ERGIC3 in normal human tissues and tumors by immunohistochemistry using 6-C4. The staining was not observed in the most normal human tissues. However, 6-C4 reacted with some epithelial cells, such as hepatocytes, gastrointestinal epithelium, ducts and acini of pancreas, proximal and distal tubules of kidney, and mammary epithelial cells. The expression of ERGIC3 in these epithelial cells may be closely related with its functions. ERGIC3 participates to form vesicles of the protein transport and secretion.[@b20],[@b21] These epithelial cells stained by the ERGIC3 mAb are very active in protein synthesis and secretion. We found that almost all carcinomas that originated from the epithelial cells were positively stained with 6-C4; by contrast, all sarcomas were negative for 6-C4. Therefore, we suggested that ERGIC3 may be a biomarker to distinguish between carcinomas and sarcomas in histopathological diagnosis. However, a larger number of cases should be investigated to draw final conclusions. We also observed that NSCLC cells were strongly stained with 6-C4; however, all of the cells (including bronchial epithelial cells and alveolar cells) in non-malignant lung tissues were negative for 6-C4. These findings were similar to our previous observation in which anti-ERGIC3 serum purchased from Abcam was used.[@b9] These results showed that ERGIC3 has higher specificity and sensitivity for NSCLC than CEA and CYFRA 21-1 in immunostaining.[@b6],[@b7] Hence, ERGIC3 may be used to distinguish NSCLC from benign lesions of the lung. More importantly, ERGIC3 mAb might be applied in the sputum test to detect early-stage lung cancer. Early diagnosis and treatment have a decisive influence on lung cancer prognosis. Thus, new methods should be developed to detect early-stage lung cancer. This study is continuing in our laboratory. ERGIC3 overexpression has been observed in NSCLC. However, molecular mechanisms underlying ERGIC3 alteration remain elusive. MiRNA are a family of small non-coding RNA, approximately 21 to 25 nucleotides in length,[@b22],[@b23] that target mRNA at complementary sites in the 3′-UTR and coding domain sequence to regulate gene expression at a post-transcriptional level.[@b24] MiR-203a (also named miR-203), which was first assigned as a skin-related miRNA,[@b25],[@b26] is involved in cancer development and progression by targeting different genes.[@b27]--[@b30] In this study, reduced miR-203a expression was one of the mechanisms found to underlie ERGIC3 overexpression in NSCLC. Our finding was supported by the following evidence: (i) miR-203a can bind to ERGIC3 mRNA by bioinformatic prediction; (ii) a luciferase reporter assay verified that ERGIC3 was targeted by miR-203a directly; (iii) ERGIC3 expressions were negatively correlated with miR-203a in cultured cells and tissues obtained from patients with NSCLC; (iv) ERGIC3 was downregulated at mRNA and protein levels by the increased miR-203a expression in NSCLC cells; (v) ERGIC3 was upregulated at mRNA and protein levels by the reduced miR-203a expression in human bronchial epithelial cells; and (vi) the forced expression of miR-203a led to changes in cellular morphology and proliferation of cultured NSCLC cells, which were similar to the effects induced by the decreased ERGIC3 expression.[@b9] MiR-203a has been regarded as a tumor suppressor miRNA that participates in cancer development and progression.[@b27]--[@b30] Here, miR-203a regulated the cellular proliferation and the expression of functional gene in NSCLC, suggesting that miR-203a might be a tumor suppressor miRNA in lung cancer. Surprisingly, significant alterations in cellular morphology and proliferation were not observed in immortalized human bronchial epithelial cells after miR-203a inhibitor treatment. We cannot explain this phenomenon. We speculated that cancer cells showed higher sensitivity to the change of miR-203a than normal cells, because normal cells have better regulation and balanced systems than cancer cells. Taken together, miR-203a might be a reasonable biomarker and therapeutic target for lung cancer. However, further research is necessary. In a previous study, miR-490-3p increased ERGIC3 mRNA and protein levels by binding to the 3′-UTR of ERGIC3 in HCC cells; this interaction is in contrast to most miRNA--mRNA interactions.[@b12] However, we did not find that miR-490-3p significantly affected ERGIC3 expression in NSCLC cells. The mechanisms of ERGIC3 regulation may differ in various cancers. In conclusion, a new murine mAb against ERGIC3 (6-C4) was developed, and the avid antibody is well suited for immunohistochemistry, immunoblotting and ELISA. 6-C4 strongly stained NSCLC cells but did not react with normal lung tissues. Thus, this mAb might be used in the histopathological diagnosis and cytopathological testing to detect early-stage lung cancer. Furthermore, miR-203a downregulation induced ERGIC3 overexpression in NSCLC cells. This work was supported by the Natural Science Foundation of China (81272617) and the 973 Program (2011CB510104). Disclosure Statement ==================== The authors have no conflict of interest to declare. Supporting Information ====================== Additional supporting information may be found in the online version of this article: **Fig. S1.** Establishment and identification of monoclonal antibodies (mAb) to ERGIC3. ELISA analysis demonstrated positive hybridomas of anti-ERGIC3 mAbs after cell fusion was completed (a) and the positive rate of approximately 95% after rounds of sub-cloning was performed by limiting dilution (b). Sera of mice immunized with the ERGIC3 peptide that recognized the protein extracted from cultured cells at approximately 50 kD in immunoblot (c). 6-C4 strongly stained non-small cell lung cancer (NSCLC) and hepatocellular carcinoma (HCC) tissues in immunohistochemistry, and SP2/0 supernatant was used as a negative control (d). **Fig. S2.** Effects of miR-140-3p and miR-490-3p on the ERGIC3 expression in non-small cell lung cancer (NSCLC) cells. (a) miR-140-3p was highly expressed in cultured NSCLC cells. (b) miR-140-3p mimic treatment did not significantly affect ERGIC3 expression in 16HBE. (c) miR-490-3p was expressed at varying degrees in NSCLC cells. (d) miR-490-3p mimic treatment likely reduce the ERGIC3 expression in A549; however, no significant difference was observed between miR-490-3p mimic treatment and negative control (Student's *t*-test, *P *\> 0.05). ncRNA, negative control miRNA. **Table S1.** Primers used in this study. [^1]: **Funding Information** Natural Science Foundation of China (81272617) and the 973 Program (2011CB510104).

1. Introduction {#sec1-nanomaterials-07-00226} =============== Graphene is a monolayer of hexagonal-oriented carbon atoms with outstanding electronic properties \[[@B1-nanomaterials-07-00226]\]. Its high electrical conductivity combined with its virtually massless character makes graphene interesting for electrodes in the field of acoustic devices operated at radio frequency (RF), such as surface and bulk acoustic wave resonators. Aluminum nitride (AlN) is a typically used material for such RF resonator devices \[[@B2-nanomaterials-07-00226],[@B3-nanomaterials-07-00226]\]. Several groups as \[[@B4-nanomaterials-07-00226],[@B5-nanomaterials-07-00226]\] could show that graphene reveals its potential as a conductive material to minimize the influence on the performance of resonator devices. For an unfailing use as an active electrode, graphene needs to exhibit low sheet resistances since the conductivity of the electrode material strongly affects the quality factor of a resonator \[[@B6-nanomaterials-07-00226],[@B7-nanomaterials-07-00226]\]. These findings require high-quality large-area graphene on insulating substrates. Chemical vapor deposition (CVD) is the most promising technique for the synthesis of large-area graphene layers with a low sheet concentration of structural defects. Since the synthesis occurs on catalytic substrates \[[@B8-nanomaterials-07-00226]\], such as copper (Cu) and nickel (Ni), graphene needs to be transferred on substrates used for applications in electronic devices. In order to ensure a large-area continuous graphene sheet on the target substrate the transfer procedure within the graphene process, technology must be stable, avoiding the formation of broken bonds and cracks which decrease the electric performance of graphene. The standard transfer technique is wet transfer on silicon dioxide (SiO~2~) substrates \[[@B9-nanomaterials-07-00226],[@B10-nanomaterials-07-00226]\]. Sheet resistances *R~S~* down to 1.1 kΩ/sq are achieved reproducibly \[[@B11-nanomaterials-07-00226]\]. Transferred graphene with *R~S~* = 450 Ω/sq is commercially available from certain suppliers \[[@B12-nanomaterials-07-00226]\], which proves the mature process technology und usability of CVD graphene. In the field of RF microelectromechanical systems (RF MEMS), used e.g., in wireless communication, graphene is able to reveal both its conductive and two-dimensional character as an alternative electrode material in full strength. In this case, the transfer substrates are of piezoelectric nature, such as aluminum nitride (AlN). Graphene transfer on AlN is already adopted by many groups \[[@B4-nanomaterials-07-00226],[@B13-nanomaterials-07-00226],[@B14-nanomaterials-07-00226]\], but a thorough examination and experimental optimization of the transfer method has not yet been shown. However, this is essential in order to improve the structural as well as the electric quality of graphene on the transfer substrate. In this publication, we show an optimized wet transfer technique on AlN. As a two-dimensional conductive material, graphene can play an important role as an alternative electrode material regarding the elimination of viscous losses due to its virtually massless character. Using the wet transfer technique \[[@B15-nanomaterials-07-00226]\], the desiccation of graphene on the target substrate is the most critical process step. The interplay between baking after the process and the substrate's wettability strongly influences this desiccation procedure. In our work we found that hydrophobic surfaces tend to be responsible for additional strain-induced cracks in transferred graphene sheets. Therefore, the target surface has a great impact on the success of the wet transfer. Commonly used silicon dioxide (SiO~2~) as a target substrate is highly hydrophilic, therefore, no other surface treatment than standard substrate cleaning is necessary for a successful transfer \[[@B10-nanomaterials-07-00226]\]. However, the RF topic requires AlN, which is a relatively hydrophobic surface substrate, due to its piezoelectric character for RF resonator devices. The emerging cracks in the graphene monolayer result in very high sheet resistances and strongly decrease the performance of graphene as a conducting electrode material. Therefore, a surface plasma pretreatment process was developed to modify the surface energy of AlN. The increased wettability enables a more controlled and smooth desiccation and, concurrently, lower graphene sheet resistance. Our obtained results ensure an unchanged graphene layer quality during both synthesis and transfer and are therefore fundamental for ongoing research activities in the field of acoustic RF devices \[[@B16-nanomaterials-07-00226]\] using graphene as an active electrode on AlN. Depending on the plasma there is both a physical and a chemical effect on the surface properties which influences both the transfer and the performance of the fabricated device. The surface roughness of the substrate surface strongly influences the adhesion between graphene and substrate. The adhesion, represented by a van der Waals interaction, depends on the surface roughness due to a changing mean distance between graphene and the substrate, as Gao et al. showed \[[@B17-nanomaterials-07-00226]\]. Additionally, the chemical effect acts like a "cleaning" of the surface \[[@B18-nanomaterials-07-00226]\]. Our investigations show that using the chemical cleaning effect of plasma pretreatments to optimize the transfer process is beneficial. As a result, we are able to achieve large-area graphene transferred on AlN with *R~S~* ≈ 350 Ω/sq as a best value, which is strongly comparable to commercially available graphene on SiO~2~. 2. Results {#sec2-nanomaterials-07-00226} ========== 2.1. Wettability Investigations on AlN/Si Substrates {#sec2dot1-nanomaterials-07-00226} ---------------------------------------------------- Graphene transferred on an untreated AlN/Si substrate is shown in [Figure 1](#nanomaterials-07-00226-f001){ref-type="fig"}. Optical microscope images ([Figure 1](#nanomaterials-07-00226-f001){ref-type="fig"}b,c) indicate the cracks and gaps in the transferred graphene layer which stem from the desiccation process. The characteristic Raman 2D peak clearly reveals large graphene-free areas ([Figure 1](#nanomaterials-07-00226-f001){ref-type="fig"}d). In contrast, graphene on SiO~2~ does not show cracks after transfer under the same transfer conditions (a). In order to investigate surface effects, SiO~2~ on Si and AlN on Si were characterized via atomic force microscopy (AFM). [Figure 2](#nanomaterials-07-00226-f002){ref-type="fig"} clearly shows that the SiO~2~ surface (a) is explicitly smoother than the AlN surface (b). The calculated root mean square (RMS) as a surface roughness parameter averages between 0.9 ± 0.43 nm and 3.49 ± 0.88 nm for SiO~2~ and AlN, respectively ([Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"}i). The much lower roughness for SiO~2~ might be a first indication that the graphene transfer is less stable on the rougher AlN surface. Since the wettability of the surface depends both on surface roughness and chemical conditioning, contact angle (CA) measurements were performed. According to the sessile drop method \[[@B19-nanomaterials-07-00226]\] very small water droplets of 10 µL volume were used for the measurements, so that additional mass effects influencing the droplet shape can be neglected. Therefore, the contact angle can be obtained via a circular fit of the optical droplet image ([Figure 2](#nanomaterials-07-00226-f002){ref-type="fig"}c). For untreated SiO~2~ and AlN on Si the contact angle is 28° and 48°, respectively. Both AFM and CA measurement clearly reveal the different surface properties resulting in an unsatisfying graphene transfer on AlN. In a next step, AlN/Si samples were plasma treated in different ways before the transfer. The samples were directly exposed to H~2~, N~2~, O~2~ and Ar plasmas. [Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"} shows the related AFM images obtained for each AlN sample. In [Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"}, (a) shows an image of untreated AlN, (b) illustrates AlN after an H~2~ plasma, and (c), AlN after an Ar plasma. [Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"}e--g show a zoom-in to the corresponding samples. Ar plasma treatment significantly changes the topography of AlN whereas there is almost no change for H~2~ plasma pretreatment. The effect of N~2~ and O~2~ (not shown) is fairly similar as the one for Ar. Comparing the corresponding roughness parameters we obtained RMS values for N~2~, O~2~ and Ar (1.1--1.2 nm), which are close to the one for SiO~2~ (0.9 nm) ([Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"}i). For the H~2~ plasma the RMS value merely changes to 3.2 nm. Corresponding CA measurements show a strong change in surface properties for the differently treated AlN samples. The obtained CA ([Figure 4](#nanomaterials-07-00226-f004){ref-type="fig"}a) for measurements 2 h after pretreatment closely fit the RMS values. CA tremendously decreases to approx. 18° for N~2~, O~2~ and Ar plasma pretreatments. In general, the contact angle between a fluid and a solid is described by the Young equation \[[@B20-nanomaterials-07-00226]\]:$$\gamma_{SA} - \gamma_{SL} - \gamma_{LA} \times \cos\theta = 0$$ with the surface tension $\gamma$ and the corresponding indices S (solid), A (air) and L (liquid) for the phases. $\theta$ is called the Young or chemical angle which describes the contact angle for a perfectly flat surface. According to the Wenzel model which describes homogenous wetting regimes, surface roughness changes the contact angle via $$\cos\theta^{*} = r \times \cos\theta$$ where $\theta^{*}$ is the corrected contact angle including the roughness parameter *r*, which is the ratio between the real and the projected solid surface area \[[@B21-nanomaterials-07-00226]\]. The Wenzel model establishes the connection of surface roughness on surface wettability. Higher roughness leads to lower wettability. Heavy atoms like N~2~, O~2~ and Ar physically damage the AlN structure and smoothen its surface, whereas H~2~ has almost no major physical impact on AlN due to its low atomic weight. The surface damage is clearly seen in [Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"}g. The typically pebble-like structure for sputtered AlN is destroyed, although the surface is smoother than without any plasma treatment. Apart from this physical smoothing which influences the substrates wettability, there is another important effect responsible for the surface properties. Plasma treatment also leads to a change in the chemical composition of the surface. For all plasmas there is a cleaning effect of the surface. Contamination atoms like carbon and other carbonic composites, which tend to be hydrophobic, are both physically and chemically removed. These composites can readsorb on the AlN surface after the plasma pretreatment \[[@B22-nanomaterials-07-00226]\], which indicates a time-dependent recovery effect of the AlN substrate's wettability. In addition to this plasma cleaning effect, Sönmez et al. \[[@B18-nanomaterials-07-00226]\] explore another impact of different plasma gases on stainless steel. In case of an O~2~ plasma, there is an oxidation effect of the surface which in turn becomes more hydrophilic due to a smoothening. A certain amount of residual O~2~ within the reactor chamber might enable oxidizing reactions which results also in a smoothening of the AlN surface. A nanolayer of aluminum oxide allotropes directly increases the wettability. Depending on the fabricated application one needs to keep in mind that oxidation might cause undesired side effects on the device performance (see [Section 2.2](#sec2dot2-nanomaterials-07-00226){ref-type="sec"}). Dehydration effects, e.g., during the wet transfer, might decrease the wettability again. Physical damage in combination with N~2~ or H~2~ plasma might also form nitrate composites, such as Al(NO~3~)~3~ and NH~4~NO~3~, due to the aforementioned, non-negligible amount of residual oxygen in the plasma chamber. Despite their negligible effect in physical surface smoothening, H~2~ atoms might create labile hydroxide compounds with the assistance of background oxygen, such as Al(OH)~3~. These chemical reactions can also activate the AlN surface due to the availability of aluminum atoms in the substrate material. The creation of nitrate components as intermediate products of chemical reactions increase, as well as oxides, the surface energy i.e., wettability \[[@B22-nanomaterials-07-00226]\]. Nitrate components are unstable and tend to degrade very soon and also dehydration takes place right after the plasma treatment. On the contrary an oxidized layer on the AlN surface can endure. All of these described effects can strongly influence the total surface wettability and might explain the recovery effect below. The "cleaning" phenomenon of the removal of surface contaminations can be seen in [Figure 4](#nanomaterials-07-00226-f004){ref-type="fig"}b. Time-dependent CA measurements obviously show a recovery effect. It is clearly visible that there is a strong reduction in *θ*\* for all applied plasmas within the first minutes after plasma exposure. This drop in CA is rapidly recovered within one hour after pretreatment. The recovery effect is quickest for H~2~ plasma. There is a specific saturation CA value for each of the plasma pretreatments, which is time-independent and averages between 44° for H~2~ and 18° for N~2~, O~2~ and Ar (dashed lines). Hence these maximum values of $\theta^{*}$ for *t* \> 2 h for different plasmas represent the ones shown in [Figure 4](#nanomaterials-07-00226-f004){ref-type="fig"}a. There exists an explicitly noticeable hydrophobic recovery with increasing time. By repeating the plasma pretreatment, the time-dependent change of the contact angles can be obtained at every measurement cycle. The difference in recovery might be explained with the above mentioned chemical reactions. For Ar, N~2~, and O~2~ plasma several oxidation reactions superpose the cleaning effects, whereas for H~2~ plasma a stable oxidation apparently does not take place. For this purpose a profound analysis of the chemical compounds on the AlN surface after plasma treatment needs to be carried out, this is not within the scope of this publication. In a first summary we can conclude that plasma pretreatments impact the surface both physically and chemically, whereas a chemically activated surface shows an aging effect. Transfer experiments indicate that already a physical impact on the AlN surface resulting in as little as *θ*\* \< 20° improves the quality of transferred graphene. The Raman D/G ratio decreases below 0.2 and the crack density can be lowered significantly. As shown later on this influences directly the graphene sheet resistance. 2.2. Wettability Investigations on BAW-SMR Substrates {#sec2dot2-nanomaterials-07-00226} ----------------------------------------------------- Since we are interested in graphene as a top electrode for resonator devices, we adapted the transfer process including the plasma treatment on solidly mounted bulk acoustic wave AlN resonator structures (BAW-SMR, described in detail in [Section 4.1](#sec4dot1-nanomaterials-07-00226){ref-type="sec"}) B1 and B2, which were grown on a Bragg reflector structure and different bottom electrodes. In this case, an increase in wettability without a change in surface roughness and stable oxide allotropes is preferable, to avoid a potentially different performance characteristic of the resonator devices due to a changed AlN surface roughness. This is important regarding the performance characteristics of actual devices, such as BAW resonators, where a structural change of the surface can influence future device functionality. For this reason we adjusted the plasma treatment in a further experimental step by reducing power and ambient pressure during the plasma process to lower the direct plasma exposure of the AlN substrates. In this case the change in wettability occurs only because of the chemical conditioning of the surface and not as a consequence of physical damage of the AlN surface. Monolayer graphene was therefore transferred on AlN grown on a material stack working as a solidly mounted resonator device (BAW-SMR). [Figure 5](#nanomaterials-07-00226-f005){ref-type="fig"}a,b show RMS obtained from AFM images and CA values, respectively, for both resonator structures. SiO~2~ and AlN on Si are shown as a reference. Although the two samples differ in their surface roughness (RMS = 4.4 ± 0.7 nm (1.8 ± 0.2 nm) for B1 (B2)) the CA is remarkably high for B2 (63 ± 6°). When an Ar plasma is applied on both samples RMS values do not change (AFM image in [Figure 3](#nanomaterials-07-00226-f003){ref-type="fig"}d,h), but CA strongly decreases to 19--22° ([Figure 5](#nanomaterials-07-00226-f005){ref-type="fig"}c). The comparably large error bars result from a relatively fast recovery effect (CA measurements are performed within 30 min after plasma treatment). Hence both a physical damage and stable oxidation effects can be ruled out. These experimental results demonstrate that wettability manipulations of the AlN surface are possible without physical surface smoothening which might deteriorate other device properties. For this, the transfer step needs to occur directly after the plasma pretreatment. 2.3. Strain and Doping Investigations via Spatially-Resolved Raman Spectroscopy {#sec2dot3-nanomaterials-07-00226} ------------------------------------------------------------------------------- Additionally, the strain distribution across the transferred graphene layer was investigated to further understand the effect of plasma pretreatment. Using spatial Raman spectroscopy analysis of the 2D and G peak positions, ω~2*D*~ and ω*~G~*, respectively, one is able to draw conclusions about strain and charge carrier density due to the strain and doping sensitivity of the Raman G and 2D modes \[[@B23-nanomaterials-07-00226]\]. In [Figure 6](#nanomaterials-07-00226-f006){ref-type="fig"}, ω~2*D*~ versus ω*~G~* is plotted for graphene on differently pretreated AlN substrates. The red (blue) line shows different levels of hole doping (strain). The slopes are obtained from \[[@B23-nanomaterials-07-00226]\], as $$\left( \frac{\mathsf{\Delta}\omega_{2D}}{\mathsf{\Delta}\omega_{G}} \right)_{\varepsilon}~ = 2.2 \pm 0.2,$$ $$\left( \frac{\mathsf{\Delta}\omega_{2D}}{\mathsf{\Delta}\omega_{G}} \right)_{n}^{hole} = 0.70 \pm 0.05.$$ According to the authors, electron doping would show a nonlinear slope and in this case Equation (4) would not hold. Hall measurements of graphene on AlN showed hole doping, which allows the use of the aforementioned equations. Das et al. \[[@B24-nanomaterials-07-00226],[@B25-nanomaterials-07-00226]\] revealed a ω~2*D*~ dependency on hole doping as $\mathsf{\Delta}\omega_{2D}/\mathsf{\Delta}n = - 1.04~$ cm, *n* in terms of 10^12^ cm^−2^. Yoon et al. \[[@B26-nanomaterials-07-00226]\] calculate the dependence of strain on a shift of ω*~G~* as $\mathsf{\Delta}\omega_{G}/\mathsf{\Delta}\varepsilon = - 23.5~$ cm^−1^/%. These two relations lead to a direct identification of strain and doping of graphene on the differently pretreated AlN substrates. In [Figure 6](#nanomaterials-07-00226-f006){ref-type="fig"}, a Raman line scan of 121 data points is plotted (a--d). ω~2*D*~ versus ω*~G~* reveals a significant change for plasma pretreated AlN on Si. Strain and doping levels are labelled in [Figure 6](#nanomaterials-07-00226-f006){ref-type="fig"}g and hold for each of the graphs. A rather linear dependency is visible for both O~2~ and Ar plasmas, whereas there is no dependency identifiable for untreated AlN and AlN pretreated with an H~2~ plasma. Graphene on Ar (O~2~) plasma pretreated AlN shows dominantly tensile strain between 0.1--0.3% (0.1--0.2%); the doping is significantly lower for the Ar plasma (*n* = 2 × 10^11^--1 × 10^13^ cm^−1^) compared to the O~2~ plasma (*n* = 2 × 10^13^--4 × 10^13^ cm^−1^). The same result is obtained for N~2~ (not shown). We can conclude that variations of strain and doping are strongly reduced applying an N~2~, O~2~ or Ar plasma instead of an H~2~ or no plasma pretreatment at all. To obtain a greater spatial distribution map measurements were done, shown in (e--g) for untreated and Ar plasma treated AlN substrate (AlN/Si and B1). Due to observable D peak intensities for graphene on untreated AlN (e) information about actual doping concentrations cannot be extracted due to overlapping effects in the Raman shifts. However, a clear reduction of strain variation is apparent for the Ar plasma for both AlN/Si and B1 (f,g). The same results hold for B2 (not shown). This indicates that the transfer-induced strain results not from surface roughness itself but from wettability effects on the graphene sheet during transfer. For the transfer of graphene on B1 we added exemplary AuCl~3~ during the Cu etching, which results in an additional doping of the transferred layer. This is directly visible in a shift of the 2D/G ratio (g). AuCl~3~ might play an important role regarding the decrease of sheet resistance due to the doping effect. This will be investigated further. 3. Discussion {#sec3-nanomaterials-07-00226} ============= In order to quantify the enhanced graphene quality due to the applied plasma treatments the samples were investigated via optical microscopy and Raman spectroscopy. [Figure 7](#nanomaterials-07-00226-f007){ref-type="fig"} shows an obvious improvement of the graphene wet transfer. Cracks and large-area gaps are avoided and the graphene sheet is transferred completely (microscopic images). Raman spectroscopy shows a clear quality enhancement. The defect-related D peak (1350 cm^−1^) vanishes almost completely, whereas the 2D/G intensity ratio increases up to 2.6. The 2D peak lies at 2772 cm^−1^. Both indicate the existence of mostly monolayer graphene on AlN \[[@B27-nanomaterials-07-00226]\]. This result directly corresponds with the reduction of strain variation after Ar plasma treatment ([Section 2.3](#sec2dot3-nanomaterials-07-00226){ref-type="sec"}). Non-uniform in-plane strain in graphene raises the probability of creating cracks and wrinkles in the layer. The obtained quality improvement is directly proven via four-point measurements. Due to plasma pretreatment, the sheet resistance *R~S~*, shown in [Figure 8](#nanomaterials-07-00226-f008){ref-type="fig"}, drops almost one order of magnitude for all applied plasmas (from 22 kΩ/sq to \<5 kΩ/sq). The high error bars, especially for the untreated AlN samples, result from the probe distance of 1 mm, which makes resistance measurements strongly dependent on the created gaps. According to Raman measurements, small graphene domains might have slightly lower sheet resistances. The most reliable results are achieved for Ar plasma pretreatment. Here, *R~S~* constantly reaches values below 1 kΩ/sq. Best values were obtained as *R~S~* = 350 Ω/sq. Sheet resistance measurements were homogenous on an area of 40 × 40 mm^2^. So far, we can exclude the effect of induced additional strain due to doping, as shown in [Figure 6](#nanomaterials-07-00226-f006){ref-type="fig"}g. Therefore, a further reduction of *R~S~* seems to be promising. Additional experiments on this topic still need to be carried out. As a conclusion, our experiments clearly reveal that the most delicate process step of the wet transfer technique is the graphene's desiccation on the target substrate. [Figure 9](#nanomaterials-07-00226-f009){ref-type="fig"}a shows the wettability concept on different substrates. Due to their high wettability, hydrophilic surfaces (such as SiO~2~ (1)) ensure a continuous desiccation via capillary forces in between the substrate and the graphene/PMMA (Poly (methyl methacrylate)) whereas hydrophobic surfaces (such as AlN (2)) form water droplets instead. These droplets either remain in between the substrate and the graphene layer (1) or they break the graphene layer for sublimation, creating cracks and discontinuities (2). As shown experimentally in this work, additional strain is applied on the transferred graphene sheets, which critically influences the sheet resistance *R~S~* of the graphene. The underlying effect of hydrophobic surfaces is the concept of surface and interfacial energies, which we identified indirectly via CA measurements. Hydrophobic surfaces feature a low surface energy and hydrophilic surfaces feature a high surface energy \[[@B28-nanomaterials-07-00226]\]. Because of the high surface energy of water due to its polar character, the surface energy of AlN must be increased. In our experiments, this was done by plasma treatment. Thinking of a further optimization of graphene as an electrode material, a stacking of several graphene sheets transferred on top of each other might be reasonable to further decrease *R~S~*. Problems regarding the transfer of graphene onto graphene will occur due to its extremely hydrophobic behavior. CA measurements revealed *θ*\* to be very close to 90° ([Figure 9](#nanomaterials-07-00226-f009){ref-type="fig"}b). This means that the topic of surface wettability will continue to arise with respect to the fabrication of high-quality large-area graphene multilayers. Furthermore, an area-increase of transferred graphene towards wafer-size makes the desiccation process even more challenging. 4. Materials and Methods {#sec4-nanomaterials-07-00226} ======================== 4.1. Materials {#sec4dot1-nanomaterials-07-00226} -------------- In this work, the synthesis of a graphene monolayer was achieved via chemical vapor deposition (CVD), a mature and commonly applied technique of growing graphene \[[@B29-nanomaterials-07-00226]\], on copper (Cu) foil (25 µm thickness, 99.8% purity, Alfa Aesar, Haverhill, MA, USA) used as a catalytic metal substrate in a cold wall low-pressure reactor (BlackMagic Pro, Aixtron, Herzogenrath, Germany) using methane and hydrogen as precursor and carrier gas, respectively (flow ratio CH~4~/H~2~ = 25%, 1600 Pa, 980 °C, 30 min deposition time). Primarily, AlN substrates were investigated which are fabricated via sputtering (2 µm AlN on 500 µm silicon (AlN/Si, 4″ wafer size)). The AlN wafers were cut into 20 × 20 mm^2^ pieces for 10 × 15 mm^2^ graphene transfers. Solidly mounted BAW resonators (BAW-SMR) with sputtered AlN as a piezoelectric layer (3″ wafer size, 1.7 µm AlN on TiAlCuW (named B1) and 1.5 µm AlN on MoAlN (B2)) were used for 40 × 40 mm^2^ graphene transfers ([Table 1](#nanomaterials-07-00226-t001){ref-type="table"}). Both AlN/Si and AlN-SMR wafers were fabricated by our project partner within the Graphene Flagship and provided to us for transfer experiments. 4.2. Methods {#sec4dot2-nanomaterials-07-00226} ------------ The graphene on Cu foil was routinely investigated by a Raman spectroscopy line scan of 121 data points (InVia Raman microscope, 532 nm excitation laser, grating 1800 lines/mm, Renishaw, Gloucestershire, UK), since it is an approved, easily applicable, non-destructive characterization method \[[@B30-nanomaterials-07-00226]\]. The representative 2D/G peak intensity ratio of reproducibly above 2.5 and the FWHM (Full Width at a half Maximum) of the 2D peak of \<30 cm^−1^ clearly identifies this as a monolayer structure \[[@B27-nanomaterials-07-00226],[@B31-nanomaterials-07-00226]\]. An almost imperceptible defect peak (D/G ratio \< 0.15) supports the conclusion that our synthesized material is a low-defect monolayer of graphene. After CVD synthesis, the graphene monolayers were transferred on differently pretreated AlN substrates using the wet transfer technique ([Figure 10](#nanomaterials-07-00226-f010){ref-type="fig"}), described in detail in \[[@B15-nanomaterials-07-00226]\], in a slightly aligned procedure. Using a PMMA (Poly (methyl methacrylate)) coating as a protection layer of the graphene/Cu stack the copper foil was etched off with an aqueous ammonium persulfate solution ((NH~4~)~2~S~2~O~8~, 0.5 mol/L). This turned out to result in a cleaner interface between AlN and graphene after transfer due to lower formation of precipitates and chemical byproducts than commonly used etching solvents like iron chloride (FeCl~3~) and iron nitrate (Fe(NO~3~)~3~). Wu et al. \[[@B32-nanomaterials-07-00226]\] suggest (NH~4~)~2~S~2~O~8~ as an etching solvent since it avoids additionally induced Fe compounds during etching, which reduce the quality of the grown graphene. The PMMA/graphene stack was then deposited onto different AlN target substrates. Before the graphene transfer, the AlN substrates were exposed to different plasmas (H~2~, N~2~, O~2~ and Ar, 10 min) in order to vary the surface conditions regarding surface energy and roughness. The SiO~2~ substrates received no plasma treatment and were used as reference samples. For comparison purposes, graphene transferred on untreated AlN was also investigated. Plasma pretreatments of AlN samples were done in a direct current (DC) plasma reactor (Leybold, 100--300 W, 9 × 10^−3^--5 × 10^−2^ mbar, 10 min exposure time). The wettability of the AlN sample was characterized via contact angle measurements using the sessile drop method \[[@B33-nanomaterials-07-00226]\] in a self-construction experimental setup. Here, a precise test droplet of deionized (DI) water (10 µL) was dispensed on the AlN surface. The obtained video image of the droplet was fitted (ImageJ, image processing software, contact angle plugin) to extract the contact angle. The AlN surface roughness was investigated with atomic force microscopy (AFM; Nanowizard III, JPK instruments, Berlin, Germany). In addition, the sheet resistance of the transferred graphene was measured with the four-point method (Loresta-GX MCP-T700, 1 mm probe distance, Mitsubishi Chemical Analytech, Yamato, Japan). 5. Conclusions {#sec5-nanomaterials-07-00226} ============== The wet transfer of CVD graphene monolayers onto AlN was investigated. We identified a strong dependence of the wettability of AlN surfaces on different plasma pretreatments due to both a physical and a chemical effect on the AlN surface. AFM images and contact angle measurements reveal a change of surface properties resulting in spatially lower strain variation and a tremendous reduction of layer cracks and gaps for N~2~, O~2~ and Ar plasma pretreatments. Focusing on the chemical "cleaning" of the surface leads to a temporal but strong increase of wettability, which enables a defect-free graphene transfer without changing the AlN's surface morphology. Sheet resistance measurements show a direct effect of this optimization regarding sheet conductivity. *R~S~* can reproducibly be reduced below 1 kΩ/sq for transferred graphene with the size of 40 × 40 mm^2^ and an Ar plasma pretreatment. With this new knowledge, the wet transfer method for the assembly of multilayer graphene sheets for further sheet resistance reduction should be investigated further. A sensitive adjustment of the surface properties of graphene regarding its strong hydrophobic behaviour is essential to obtain even lower sheet resistances. Furthermore, the effect of substrate surface pretreatment is an interesting and inevitable topic for the transfer of high quality graphene onto other substrate materials. Our work clearly shows a strong improvement in the quality of graphene and therefore emphasizes the promising future of graphene as a virtually massless and ultrathin electrode material for high-frequency devices. The authors would like to thank Michaela Fritz, Benedikt Tritschler and Andres Rösch for the technical and exploratory assistance and to acknowledge the funding from the European Union within the framework of Graphene Flagship. Marius Knapp, René Hoffmann and Oliver Ambacher conceived and designed the experiments; Marius Knapp performed the experiments and analyzed the data; René Hoffmann, Volker Cimalla and Oliver Ambacher contributed reagents/materials/analysis tools; Marius Knapp wrote the paper. The authors declare no conflicts of interest. In this manuscript the following abbreviations are used: RFRadio frequencyMEMSMicroelectromechanical systemCVDChemical vapor depositionAlNAluminum nitridePMMAPoly (methyl methacrylate)BAWBulk acoustic waveSMRSolidly mounted resonatorAFMAtomic force microscopyCAContact angle {#nanomaterials-07-00226-f001} {#nanomaterials-07-00226-f002} {ref-type="sec"}, reveals that the typically columnar structure of the AlN surface is preserved, which is shown exemplary in (**d**,**h**). (**i**) Comparison of RMS surface roughness parameter for SiO~2~ (black) and differently plasma treated AlN/Si surfaces.](nanomaterials-07-00226-g003){#nanomaterials-07-00226-f003} {#nanomaterials-07-00226-f004} {#nanomaterials-07-00226-f005} {#nanomaterials-07-00226-f006} {#nanomaterials-07-00226-f007} {#nanomaterials-07-00226-f008} {#nanomaterials-07-00226-f009} {#nanomaterials-07-00226-f010} nanomaterials-07-00226-t001_Table 1 ###### Overview about used AlN samples for the graphene wet transfer. Substrate Substrate Size Graphene Layer Size ----------------- ---------------- --------------------- AlN/Si 20 × 20 mm^2^ 10 × 15 mm^2^ B1 (AlN/TiAlCu) 3″ wafer 40 × 40 mm^2^ B2 (AlN/MoAlN) 3″ wafer 40 × 40 mm^2^

 Executive Editors ================= *Senior Editors*: K.R.Fox, *Southampton, UK*; foxnar\@soton.ac.uk B. Stoddard, *Seattle, WA, USA*; stoddnar\@fhcrc.org M.Madan Babu, *Cambridge, UK*D.Corey, *Dallas, TX, USA*M.Wegner, *Erlangen, Germany*<narlmb@mrc-lmb.cam.ac.uk><david.corey@utsouthwestern.edu><m.wegner@biochem.uni-erlangen.de>J.M.Bujnicki, *Warsaw, Poland*W.Dynan, *Atlanta, GA, USA*E.Westhof, *Strasbourg, France*<iamb@genesilico.pl><wsd.nar@gmail.com><nar@ibmc.u-strasbg.fr>M.Churchill, *Aurora, CO, USA*A.D.Sharrocks, *Manchester, UK*J.A.Wise, *Cleveland, OH, USA*<nar@uchsc.edu><nar@manchester.ac.uk><jaw17@po.cwru.edu> *Methods Editors*: G.Sczakiel, *Lübeck, Germany*; <sczakiel@imm.uni-luebeck.de> A.R.Kimmel, *Bethesda, MD, USA*; <ark.nar@gmail.com> *Database Issue Editor*: M.Galperin, *Bethesda, MD, USA*; <nardatabase@gmail.com> *Web Server Issue Editor*: G.Benson, *Boston, MA, USA*; <gbenson@bu.edu> Editorial Board =============== F.Allain, *Zurich, Switzerland* C.Bailly, *Lille, France* A. Bateman, *Hinxton, UK* P.B.Becker, *Munich, Germany* M.Belfort, *Albany, NY, USA* B.Berkhout, *Amsterdam, The Netherlands* E.H.Bresnick, *Madison, WI, USA* J.F.Caceres, *Edinburgh, UK* J.S.Carroll, *Cambridge, UK* X.Cheng, *Atlanta, GA, USA* J.J.Collins, *Boston, MA, USA* J.Dinman, *College Park, MD, USA* G.Divita, *Montpellier, France* S.Eddy, *Ashburn, VA, USA* M.Egli, *Nashville, TN, USA* D.R.Engelke, *Ann Arbor, MI, USA* A.Engelman, *Boston, MA, USA* M.J.Gait, *Cambridge, UK* H.U.Gö ringer, *Darmstadt, Germany* B.Gottgens, *Cambridge, UK* T.Grange, *Paris, France* S.Gri?ths-Jones, *Manchester, UK* W.Gruissem, *Zurich, Switzerland* J.G.Hacia, *Los Angeles, CA, USA* P.Hsieh, *Bethesda, MD, USA* P.A.Jeggo, *Brighton, UK* A.Khvorova, *Lafayette, CO, USA* R.Kierzek, *Poznan, Poland* H.Klein, *New York, NY, USA* D.Klostermeier, *Muenster, Germany* E.T.Kool, *Stanford, CA, USA* E.Koonin, *Bethesda, MD, USA* A.N.Lane, *Louisville, KY, USA* B.Lehner, *Barcelona, Spain* B.Lenhard, *London, UK* D.M.J.Lilley, *Dundee, UK* S.Linn, Berkeley, *CA, USA* L.J.Maher, *Rochester, MN, USA* D.Mann, *Newcastle-upon-Tyne, UK* A.Maxwell, *Norwich, UK* S.Mueller, *Greifswald, Germany* S.Neidle, *London, UK* R.Parker, *Tucson, AZ, USA* M.R.Parthun, *Columbus, OH, USA* R.Pillai, *Grenoble, France* M.Porteus, *Stanford, CA, USA* N.Proudfoot, *Oxford, UK* R.J.Roberts, *Ipswich, MA, USA* D.B.Roth, *Philadelphia, PA, USA* T.D.Schneider, *Frederick, MD, SA* B.Schwer, *New York, NY, USA* B.Séraphin, *Gif sur Yvette, France* P.M.Sharp, *Edinburgh, UK* S.K.Silverman, *Urbana, IL, USA* H.Siomi, *Tokyo, Japan* C.Smolke, *Stanford, CA, USA* I.Stancheva, *Edinburgh, UK* A.Stasiak, *Lausanne, Switzerland* N.Sugimoto, *Kobe, Japan* J.Vogel, *Würzburg, Germany* M.C.Williams, *Boston, MA, USA* S.H.Wilson, *Research Triangle Park, NC, USA* R.D.Wood, *Smithville, TX, USA* S.A.Woodson, *Baltimore, MD, USA* S.Yokoyama, *Tokyo, Japan* Founding Editors ================ R.T.Walker, *Birmingham, UK* D.Söll, *New Haven, CT, USA* A.S.Jones, *Birmingham, UK* Editorial and Production ======================== *Editorial ManagerProduction Editor*Martine Bernardes-SilvaOliver Barham<nar@soton.ac.uk><narese@oup.com>

Introduction {#Sec1} ============ Bioorthogonal chemistry has attracted much attention and played critical important roles to dissect complex intracellular processes and perform specific chemical reactions in living systems^[@CR1]--[@CR3]^. Being a relatively new sector of chemical biology, bioorthogonal chemistry has made significant progress in imaging, regulation, and therapeutic applications^[@CR4]--[@CR6]^. Different kinds of bioorthogonal reactions have been gratifyingly realized in living systems^[@CR7],[@CR8]^. Transition metal catalysts (TMCs) become outstanding candidates for catalyzing different bioorthogonal reactions^[@CR9]--[@CR12]^. Compared with natural enzyme systems, TMCs can rapidly catalyze chemical transformations that cannot be realized by natural enzymes for specific biological applications^[@CR13]--[@CR15]^. However, in terms of accuracy and intelligence, current catalytic behaviors of TMCs are undeniably far less than natural biological processes. In fact, life is a highly ordered system and all the activities of proteins, genes, and other signal molecules are precisely controlled in timing and location. Molecules arranged in a random fashion cannot form a living system^[@CR16],[@CR17]^. That means, reactions in living systems are intelligently controlled on demand. Therefore, for practical applications, it is highly desirable and demanding to precisely determine when, where, and to what extent a bioorthogonal reaction is started and ended. Unfortunately, most of previously reported bioorthogonal catalysts do not provide the capability of precisely controlling the catalytic processes according to extrinsic stimulus. To this end, people developed different extrinsic stimulus to control bioorthogonal reactions by liberating reactive groups through mild chemistry or light^[@CR18]--[@CR20]^. Though it has made significant advances in responsive bioorthogonal reactions, these strategies rely on design of precursors or photoremovable protecting groups (PPGs), limiting the reaction types in Cu-click or photo-click reactions^[@CR21]--[@CR24]^. Moreover, the extra chemical reactions would increase the complexity and indeterminacy inevitably. The leave of PPGs usually needs high extra energy along with relatively slow reaction rates, and the PPGs have irreversibility and instability under certain conditions. Recently, cucurbit\[7\]uril has been used to block the access to the catalytic site of Ru or Pd catalysts on the surface of Au nanoparticles, and the catalyst can be re-activated by extra addition of 1-adamantylamine (ADA)^[@CR25]^. This is an innovative design for TMC-mediated bioorthogonal chemistry. However, extra addition may cause potential interference in cellular processes and is still difficult to realize reversible control and temporal-spatial resolution. Light is an ideal external trigger for manipulating chemical reactions^[@CR26]^. It is highly selective and also harmless when applied properly. Several excellent examples have successfully applied light in living systems with temporal-spatial resolution^[@CR27]--[@CR32]^. Meggers and co-workers reported that the organometallic complex \[Cp\*Ru(η^6^-pyrene)\]PF~6~ (Cp\* = pentamethylcyclopentadienyl) could catalyze the conversion of *N*-allylcarbamates into their amines in the presence of thiols, water, and air under UV light^[@CR33]^. However, the homogeneous organometallic catalyst \[Cp\*Ru(η^6^-pyrene)\]PF~6~ was difficult in synthesis, separation, and purification. Moreover, the application of the homogeneous organometallic catalysts in living cells was challenging due to the issues of biocompatibility, water solubility, stability, and rapid efflux of catalysts from living cells^[@CR10],[@CR25]^. Currently, although many TMCs have been used to catalyze multiple bioorthogonal reactions, light-controllable heterogeneous TMCs have not been reported. Herein, we designed Pd nanoparticles-imbedded macroporous silica nanoparticles (SP), then modified the nanocatalyst with an azobenzene switch, abbreviated as ASP (Fig. [1](#Fig1){ref-type="fig"})^[@CR34]^. The resulting catalyst can be reversibly functionalized by cyclodextrin (CD) through host--guest interactions and named as CASP. When the access to catalytic site of CASP is blocked by CD, the catalytic activity will be inhibited. By utilizing the isomerization of Azo, the CD blocker can be released from the catalysts resulting in the recovery of catalytic activity under UV illumination or returned to the original inactive status again under visible light^[@CR35]^. Since the energy band between isomerization of Azo is much lower than the bond energy of PPGs^[@CR36]^, lower dose of extra light is needed. The efficacy of our designed heterogeneous catalyst was verified in cuvettes and living cells through two applications: (1) the gated generation of fluorophore through deallylation of non-fluorescent precursors;^[@CR9],[@CR37]^ (2) the gated synthesis of a mitochondria-specific probe in situ by Suzuki--Miyaura cross-coupling reaction and light-controllable activation of a prodrug in living cells^[@CR10]^. Herein, there are three major advantages of our designed catalysts: (1) the synthetic method of the heterogeneous TMCs is versatile and easy to be applied for constructing different kinds of catalysts, such as Pt, Ru, Rh, and so on. (2) The reversible interaction between Azo and CD endows the bioorthogonal catalyst with well reversibility, providing a great promise for broadening potential application of bioorthogonal catalysts in biosystem. (3) Low-dosed light as the external stimulus is an accepted green regulation method and possesses excellent controllability in space and time.Fig. 1Illustration of design of photo-responsive bioorthogonal catalysts in cells. **a** Macroporous silica (DMSN), Pd-embedded silica nanoparticles (SP), Azo-modified complex catalyst (ASP), and CD-capped complex catalyst (CASP) used in the study. **b** The photo-isomerization process between *trans*-azobenzene and *cis*-azobenzene can cause the dissociation and recombination of CD. **c** Intracellular bioorthogonal catalysis with CASP In this study, we demonstrate that the designed heterogeneous catalyst could effectively mediate the bioorthogonal reactions in situ through light. Such light-gated control of bioorthogonal catalysis in living cells has not been demonstrated to date although this issue is important for activating catalytic process at target locations and artificially changing catalytic activity to maintain homeostasis for long-term therapeutics. Results {#Sec2} ======= Design and construction of the complex nanocatalyst {#Sec3} --------------------------------------------------- Encouraged by the vast possibilities and applications of palladium chemistry, the preparation of a bio-friendly light-controlled Pd^0^-based heterogeneous catalyst was undertaken. The detailed synthesis process was provided in supporting information. Briefly, macroporous silica nanoparticles (DMSN) were used as the scaffold to stabilize the transition metal (Pd^0^)^[@CR38]^. The average particle diameter and central-radial pore size of DMSN were 80 nm and 12.3 nm, respectively, as indicated by scanning electron microscopy (SEM), transmission electron microcopy (TEM) images (Supplementary Fig. [1a, b](#MOESM1){ref-type="media"}), and nitrogen adsorption measurements (Supplementary Fig. [1c, d](#MOESM1){ref-type="media"}). Next, ultrafine palladium nanoparticles of diameter around 1.3 nm were synthesized by in situ reduction and attached to the pores' inner surfaces, as clearly shown in the SEM-EDX (Supplementary Fig. [2b](#MOESM1){ref-type="media"}), TEM, and elemental mapping images (Supplementary Fig. [3](#MOESM1){ref-type="media"}), powder X-ray diffraction (XRD) analysis (Supplementary Fig. [4](#MOESM1){ref-type="media"}) and diameter distribution analysis (Supplementary Fig. [5](#MOESM1){ref-type="media"}). For further modification, the SP was successfully functionalized with azobenzene group (ASP) and then combined with CD (CASP). The variation of C1s, N1s XPS (Supplementary Fig. [6](#MOESM1){ref-type="media"}), and the changes of Fourier transformation infrared spectroscopy (FTIR) spectrum (Supplementary Fig. [7](#MOESM1){ref-type="media"}) confirmed the conjugation of the azobenzene groups and CD. Besides, the amount of Pd loaded in DMSN was estimated by inductively coupled plasma mass spectrometry (Supplementary Table [1](#MOESM1){ref-type="media"}). Catalytic efficacy of SP nanoparticles in vial {#Sec4} ---------------------------------------------- As a prerequisite, we first validated the catalytic efficacy of SP in vitro by catalyzing the allylcarbamate cleavage of *N*-allyloxycarbonyl coumarin (N-alloc-cumarin), **2** (Fig. [2a](#Fig2){ref-type="fig"}). Comparing with the standard fluorescence spectrum of compounds **1** and **2** (Supplementary Fig. [8](#MOESM1){ref-type="media"}), the increased fluorescence intensity at 450 nm after incubation with SP (Fig. [2b](#Fig2){ref-type="fig"}) indicated the catalytic capability of SP. After 24 h, bright blue fluorescence was observed in the tube with SP. As expected, no reaction occurred in the tube with control DMSN particles (Supplementary Fig. [9](#MOESM1){ref-type="media"}). For further studying the catalytic efficiency and kinetic process between substrates and products, we recorded the changes of fluorescence spectrum from 0 to 9 h (Supplementary Fig. [10](#MOESM1){ref-type="media"}), and analyzed the supernatants of the reactions with high-performance liquid chromatography (HPLC) (Supplementary Fig. [11](#MOESM1){ref-type="media"}). The data showed clearly that the catalytic process was almost finished in about 9 h. The relating statistics of concentration of reagent and product and translation ratio were shown in Supplementary Fig. [12](#MOESM1){ref-type="media"}. The relationship between natural logarithm of reactant concentration and reaction time (ln\[c\]-t) indicated that the allylcarbamate cleavage of N-alloc-cumarin was belonged to the first-order reaction under our experimental condition. More importantly, the translation ratio of reagent was more than 95% at 9 h with our catalysts, while the valve \<1% with reagent alone (Supplementary Fig. [12c](#MOESM1){ref-type="media"}). This indicated that the confinement effect of DMSN, excellent catalytic performance of Pd^0^ and large specific surface area of the ultra-small Pd nanopaticles synergistically improved the catalytic activity. Besides, the different reaction conditions had no obvious disturbance on the catalytic activity of SP (Supplementary Fig. [13](#MOESM1){ref-type="media"} and Supplementary Table [2](#MOESM1){ref-type="media"}), showing the stability of this catalyst.Fig. 2Catalytic efficacy of SP and CASP nanoparticles in vial. **a** SP and CASP catalyzed allylcarbamate cleavage of N-alloc-cumarin, **2** to generate a fluorescent compound, **1**, in vial. **b** The fluorescent spectra of alloc-coumarin alone at 0 h (blue) or 20 h (red) and reacted with SP for 20 h (black). **c** Fluorescence was increased when the system added SP, while CASP showed no significant change during 0--40 min. **d** After irradiation with UV light for 10 min, the catalytic activity of CASP was restored and became similar to SP, while UV light had no effects on SP. The attached graph reflected the reaction rates of SP and CASP before and after the UV light. Data were presented as mean ± s.d. (*n* = 3). **e** The fluorescent gel experiments: the polyacrylamide gel (PAMG) was prepared in the presence of CASP and N-alloc-coumarin (**f**). Under ambient conditions, the PAMG showed minimal fluorescence (**g**), when irradiated the gels with UV light under a mask that featured lower-left cycle (**h**), upper-right triangle (**i**), right rectangle (**j**), and middle rectangle (**k**) Light-mediated reversible catalysis process {#Sec5} ------------------------------------------- We next studied the light-gated bioorthogonal catalysis. For the light-gated process, the Azo modified on DMSN was one key factor in our design. The quantity of Azo modified on DMSN was estimated about 50 μg mg^−1^ (Supplementary Fig. [14](#MOESM1){ref-type="media"}). More importantly, the Azo modified on DMSN still retained its photo-responsive property (Supplementary Fig. [15](#MOESM1){ref-type="media"}). After treated with β-CD, the catalytic activity of CASP was inhibited because almost no fluorescent change was observed at 450 nm (Supplementary Fig. [16a](#MOESM1){ref-type="media"}). As expected, the catalyst could be activated by light: after irradiation by UV light, the catalytic activity was restored again (Supplementary Fig. [16a](#MOESM1){ref-type="media"}). Besides, the catalyst could be reset to its inactive state reversibly by visible light (Supplementary Fig. [16b](#MOESM1){ref-type="media"}). Kinetic studies further supported the light-activated catalysis (Fig. [2c, d](#Fig2){ref-type="fig"}). When the system was irradiated by UV light for 10 min (Fig. [2d](#Fig2){ref-type="fig"}), the fluorescence was increased suggesting the activation of CASP. Control studies of SP showed its catalytic activity was unaffected by light indicating the critical role of reversible interactions between Azo and CD. LC-MS observed the fluorescent reaction product in the solution with CASP treated by light (Supplementary Fig. [17](#MOESM1){ref-type="media"}). Therefore, we could control the chemical reaction to initiate or terminate by light at proper time through this system. Moreover, the spatial resolution of our light-mediated process was also confirmed with a fluorescent gel experiment (Fig. [2e](#Fig2){ref-type="fig"})^[@CR39]^. When re-activating the catalysts in thin polymer gels through irradiation locally under a mask, we might be able to create a corresponding fluorescent pattern. The polyacrylamide gel (PAMG) was prepared in the presence of CASP and compound, **2**, (Fig. [2f](#Fig2){ref-type="fig"}). Under ambient conditions, the PAMG showed minimal fluorescence (Fig. [2g](#Fig2){ref-type="fig"}). To demonstrate our supposition, we irradiated the gels with UV light under a mask that featured lower-left cycle (Fig. [2h](#Fig2){ref-type="fig"}), upper-right triangle (Fig. [2i](#Fig2){ref-type="fig"}), right rectangle (Fig. [2j](#Fig2){ref-type="fig"}), and middle rectangle (Fig. [2k](#Fig2){ref-type="fig"}). The fluorescence in the exposed regions turned on, which indicated fluorescent precursors were converted to fluorescent product. Besides, the heterogeneous nanocatalysts were recyclable with high catalytic activity (Supplementary Table [3](#MOESM1){ref-type="media"}). These results suggested that the light-gated transition metal catalysts had good spatial and temporal resolution. The light activation of CASP in living cells {#Sec6} -------------------------------------------- Having confirmed the catalytic activity of nanocatalysts in solution, we next studied their intracellular performance in human cervical cancer cells (HeLa cells) as the model system (Fig. [1c](#Fig1){ref-type="fig"}). Before that, the cellular uptake of the nanocatalysts was quantified by tracking \[^106^Pd\] by ICP-MS, both SP and CASP could be effectively endocytosed by HeLa cells, respectively (Supplementary Fig. [18](#MOESM1){ref-type="media"}). In addition, according to the cytotoxicity experiments (Supplementary Fig. [19a](#MOESM1){ref-type="media"}), SP and CASP were biocompatible at the concentrations used in our study. Besides, the low-dose UV light did not cause cytotoxicity avoiding extra light damage (Supplementary Fig. [19b](#MOESM1){ref-type="media"}). Then, the catalytic activity of the complex catalysts was investigated in living cells. Transformation of rhodamine 110 derivative (**4**) into rhodamine 110 (**3**) was chosen as the model reaction^[@CR25]^. The nanocatalysts were incubated with HeLa cells first, and then the old medium was replaced by the fresh one containing substrate, **4**. After 24 h, the cells were washed with phosphate buffer (PBS) for three times. The resulting fluorescent compound, **3**, was retained within the cell once the allylcarbamate cleavage took place (Fig. [1c](#Fig1){ref-type="fig"}). As shown in the flow cytometry analysis (Fig. [3a](#Fig3){ref-type="fig"}), a significant fluorescence increase was observed for the cells treated with SP in comparison with the controls. The results indicated that the nanocatalyst could still catalyze the reaction in cells. However, CASP showed little catalytic activity under this condition, indicating that CASP was stable in cells.Fig. 3Triggered cleavage of alloc-Rhodamine 110 (**4**) in living cells. **a** The flow cytometry analysis of control (cells alone), DMSN, SP, and CASP. **b** Treated the CASP with UV light, the cells produced an increased fluorescence in the flow cytometry. **c**--**f** Fluorescence images of HeLa cells. A green fluorescence was observed with SP (**d**) and CASP + UV light (**f**), while no fluorescence was obtained with substrate only (**c**) and CASP (**e**) (scale bar = 50 µm). Data were presented as mean ± s.d. (*n* = 3) We next investigated the intracellular light-gated process of CASP. When treated the catalyst with UV light, CASP restored its catalytic activity with an obvious fluorescence increase (Fig. [3b](#Fig3){ref-type="fig"}). The increased catalytic efficiency was echoed in fluorescence imaging (Fig. [3c--f](#Fig3){ref-type="fig"}). We synchronously monitored the intracellular catalytic reactions at first 1 h, the relatively steady reaction rate suggested the stable catalytic activity of CASP in living cells after treated by UV light (Supplementary Fig. [20](#MOESM1){ref-type="media"}) and the recycle experiment indicated that in short term our system could realize the reversible activation of TMCs in cells more than five cycles (Supplementary Fig. [21](#MOESM1){ref-type="media"}). The fluorescence images showed similar results (Supplementary Fig. [22](#MOESM1){ref-type="media"}). What's more, after the activation of the catalyst by UV light, if we treated the system with visible light again, the catalyst showed no activity once more (Supplementary Fig. [22c](#MOESM1){ref-type="media"}). Certainly, the UV and Vis light had no obvious effect on the catalytic activity of SP showing the activity switch was due to the interaction between Azo and CD (Supplementary Fig. [23](#MOESM1){ref-type="media"}). Also, the LC-MS analysis of the final product (331.10) of the bioorthogonal reaction catalyzed by CASP ensured that the catalytic reaction was carried out correctly (Supplementary Fig. [24](#MOESM1){ref-type="media"}). Light-gated synthesis of a mitochondria-specific probe {#Sec7} ------------------------------------------------------ In addition to the cleavage reaction, we explored the possibility of catalyzing other types of reactions. Among them, cross-coupling reactions play an important role in chemical synthesis of molecules^[@CR40]^. The cross-coupling reaction under mild conditions especially attracts attention^[@CR41]^. To explore the formation of the C--C cross-coupled product in cells, we decide to synthesize a fluorescent dye based on the palladium-mediated synthesis via Suzuki--Miyaura cross-coupling reaction (Fig. [4a](#Fig4){ref-type="fig"})^[@CR10]^. The non-fluorescent compounds, **8** and **11**, can produce fluorescent target product **12** through the couple of triflate and phenylboronate groups. Meanwhile, since the product **12** has a triphenylphosphonium moiety, it can act as a mitochondria-specific fluorescent probe^[@CR11],[@CR42]^. The light-gated synthesis of Suzuki--Miyaura probe **12** was initiated by incubation of the nanocatalysts with HeLa cells. After washing with PBS, the two reagents **8** and **11** were added. Then, we treated the cells with UV light. After another 24 h, the cells were labeled with a common mitochondrial red dye (MitoTracker Red CMXRos Ex/Em ≈ 579/599 nm) and Hoechst 33258 (nuclei stain) as controls. Cellular fluorescence was investigated by fluorescence images and flow cytometry. As shown in Fig. [4b--i](#Fig4){ref-type="fig"}, upon activation by UV light, the fluorescence images of synthesized dye (Fig. [4c](#Fig4){ref-type="fig"}) showed an obviously increased fluorescence co-localized with MitoTracker (Fig. [4d](#Fig4){ref-type="fig"}) when compared with the controls (Supplementary Fig. [25](#MOESM1){ref-type="media"}). The flow cytometry showed similar results indicating the light-gated catalytic reaction occurred (Supplementary Fig. [26](#MOESM1){ref-type="media"}). Besides, the two-channel flow cytometry analysis verified the independence between probe **12** and MitoTracker, eliminating mutual interference (Supplementary Fig. [27](#MOESM1){ref-type="media"}). These results demonstrated that the Suzuki--Miyaura coupling reaction could be also activated by light as expected. The fluorescent spectra of the reaction in solution and LC-MS of the products provided further supports (Supplementary Figs. [28](#MOESM1){ref-type="media"}--[30](#MOESM1){ref-type="media"}). All results proved that light-gated Suzuki--Miyaura reaction could be realized successfully in living cells.Fig. 4Light-gated Suzuki--Miyaura coupling reaction. **a** Scheme of the light activation^[@CR10]^ of CASP catalyzed cross-coupling reaction generates the mitochondria-localized fluorescent product **12**. **b-i** Collection of the fluorescence images of the CASP catalyzing the Suzuki--Miyaura coupling reaction with the UV light activation: **b** bright-field image of HeLa cells, **c** emission of the compound **12** (green channel), **d** the mitochondrion labeled with MitoTracker^®^ Red CMXRos (red channel), **e** the cell nucleus labeled with Hoechst 33258 (blue channel), **f** merging with **c** and **d**, **g** merging with **c** and **e**, **h** merging with **d** and **e**, **i** merging with **c**, **d**, and **e** (scale bar = 10 µm). The low fluorescence intensity of green channel indicated the low catalytic activity of CASP as we designed Intercellular light activation of prodrug by the catalyst {#Sec8} --------------------------------------------------------- Furthermore, we also carried out the intracellular conversion of prodrug 5-fluoro-1-propargyl-uracil (Pro-5FU) into 5-fluorouracil (5FU) by using our system (Supplementary Fig. [31](#MOESM1){ref-type="media"}). As a preliminary experiment, the toxicity profiles of Pro-5FU and 5FU were studied at various concentrations (Supplementary Fig. [31b](#MOESM1){ref-type="media"}). Compared with Pro-5FU, 5FU showed obvious toxicity with increasing the concentration. For the experiment of light-gated activation of Pro-5FU, the relative cell viability studies were carried out in the presence of the catalysts. As shown in Supplementary Fig. [31c](#MOESM1){ref-type="media"}, the cells that were incubated with CASP + UV light showed increased toxicity at high concentration of Pro-5FU, while CASP alone retained high cell viability. These results indicated that the toxicity was due to light-gated conversion of Pro-5FU into 5FU, but not the catalyst itself. Therefore, our system could be used in therapeutic applications such as the activation of multiple prodrugs on site of action. Discussion {#Sec9} ========== We have fabricated a light-gated heterogeneous transition metal catalyst, which is capable to artificially control the bioorthogonal reactions in living cells. The ultrafine Pd nanoparticles deposited in DMSN motif have excellent catalytic activity in Pd-mediated bioorthogonal reactions. The light-gated host--guest interactions between azobenzene headgroup and β-CD play a critical and important role in reversibly regulation of the catalytic activity. This work not only provides new perspective to customize different heterogeneous catalysts with novel functions, but also has a huge potential in activation of prodrugs and synthesis of active molecules for precision therapy. Selective activation of bioorthogonal catalysts on demand can minimize the interference of organisms and realize multiple chemical reactions in situ. Our design is versatile and adaptable to a variety of chemical reactions in living cells for precision imaging and therapy. Methods {#Sec10} ======= Reduction of palladium in DMSN motif {#Sec11} ------------------------------------ For synthesis of SP nanoparticles, \[PdCl~4~\]^2−^ ions were absorbed onto the inner pore surfaces by coordinating and electrostatic interaction with amino groups. After in suit reduction, \[PdCl~4~\]^2−^ ions were converted to Pd^0^ leading to growing of palladium nanoparticles. The resulting nanoparticles DMSN (0.1 g) were dispersed in distilled water (10 mL) along with sonication for 30 min, followed by addition of H~2~PdCl~4~ (0.1 mL, 1 M) solution in 2 mL distilled water, after 20 min, a freshly prepared NaBH~4~ (36 mg in 4 mL cold water) was added into the above aqueous solution under vigorous stirring. After that, the resulting suspension was stirred for another 3 h. Finally, the mixture was centrifuged at 11,000 × *g* for 10 min to separate the SP. Then, SP was washed by water three times and dried under vacuum. The fluorescent gel experiments {#Sec12} ------------------------------- The polyacrylamide gel (PAMG) was prepared via radical-copolymerization of acrylamine in the presence of CASP and N-alloc-coumarin. Aliquots of 750 mg acrylamide (AAm) and 25 mg Bis-acrylamide (MBA) were dissolved in 2.5 mL distilled water as solution **1**. The CASP (100 µg mL^−1^) and 10 µM N-alloc-cumarin (40 mM in dimethyl sulfoxide (DMSO)) were mixed in 2.45 mL distilled water as solution **2**. We mixed solution **1** and solution **2** to obtain a uniformly dispersed solution and added 40 µL 10% ammonium persulphate (APS) and 10 µL tetramethylethylenediamine (TEMED) to the mixture. Then, the mixture was transferred to the mold and reacted for 2 h at room temperature. The prepared PAMG was cut into square (1 cm × 1 cm) and irradiated with UV light for 10 min under the different masks that featured lower-left cycle, upper-right triangle, right rectangle, and middle rectangle. After 20 h, the fluorescent images were recorded by a camera on the UV transilluminators. Light-gated catalytic reactions in living cells {#Sec13} ----------------------------------------------- HeLa cells were cultured with Dulbecco's modified Eagle medium (DMEM) containing fetal bovine serum (10%) at 37 °C under a humidified atmosphere with 5% CO~2~. For flow cytometry analysis, HeLa cells were seeded at around 60,000 cells per well in 6-wells plates for 24 h prior the experiment. Thereafter, the nanocatalysts SP or CASP (40 µg mL^−1^) were incubated with pre-seeded HeLa cells, respectively. After 24 h incubation, old media was removed and cells were washed three times with PBS to eliminate extracellular nanoparticles. Then, the cells were treated with 0.12 W cm^−2^ UV light for 10 min. Protected rhodamine 110, **4** (10 mM in DMSO) was added to the culture at final concentration of 20 µM and incubated for 24 h. After that, cells were rinsed with PBS twice, harvested with trypsin, and resuspended in PBS buffer. The intracellular presence of fluorescent compound, **3**, was analyzed by flow cytometry using FITC-like band pass emission filters (530/30). Suzuki--Miyaura coupling reaction in living cells {#Sec14} ------------------------------------------------- The cells were incubated on sterilized cover slips in a 24-wells plate at 37 °C. Old media was removed and replaced by fresh media containing SP or CASP (40 µg mL^−1^) and incubated for another 24 h. Excess of extracellular nanoparticles was removed by washing with PBS buffer thrice. Compounds **8** and **11** (20 mM in DMSO) were diluted in fresh media to give a final concentration of 20 µM and incubated at 37 °C for 24 h. After that, the cells were washed with PBS twice, subsequently, mitochondria were stained by 20 min incubation with 50 nM of MitoTracker^®^ Red CMXRos at 37 °C. The nuclei were stained by incubation with a 10 µg mL^−1^ solution of Hoechst 33258 in media for 15 min. The fluorescence images were collected. Data availability {#Sec15} ----------------- All data are available from the authors on reasonable request. Electronic supplementary material ================================= {#Sec16} Supplementary Information(PDF 3974 kb) **Electronic supplementary material** **Supplementary Information** accompanies this paper at 10.1038/s41467-018-03617-x. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This work was supported by 973 Project (2012CB720602), NSFC (21210002, 21431007, 21533008), and Key Program of Frontier of Sciences, CAS QYZDJ-SSW- SLH052. J.R. and X.Q. designed the research; F.W., Y.Z., and Z.D. performed the research; F.W., Y.Z., and Z.D. analyzed the data; F.W., J.R., and X.Q. wrote the paper. Competing interests {#FPar1} =================== The authors declare no competing interests.

Introduction {#Sec1} ============ Developed by Kennedy and Eberhart \[[@CR1], [@CR2]\] particle swarm optimization (PS0) is a stochastic optimization method modeled on social behavior and intelligence of animal such as flocks of birds and fish schooling. Similar to other evolutionary methods, it is based on the population. The mechanism of the PSO method relies on particles following their best personal particle and globally the best particle in the swarm towards the most promising areas of the search space. Because of its easy implementation and high convergence rate, it is widely used in solving various optimization problems, including energetic \[[@CR3]\], mechanics \[[@CR4]\], scheduling problem \[[@CR5]\], antenna design \[[@CR6], [@CR7]\], control systems \[[@CR8]\], image classification \[[@CR9]\] and many others. However likewise other evolutionary algorithms, PSO encounters some troubles including stagnation in local optima, excessive loss of diversity and premature convergence \[[@CR10]\]. A variety of different variants of PSO have been introduced to counteract these disadvantages and enhance the efficiency of PSO. Among them, the following improvements can be distinguished:Adjustment of basic coefficients. According to Shi and Eberhart \[[@CR11]\], a key to the improvement of the PSO performance is inertia weight, which should be linearly decreased from 0.9 to 0.4. Clerc \[[@CR12]\] recommended to use fixed factors, and indicates that inertia weight of 0.729 with fixed acceleration coefficients of 1.494 can enhance convergence speed. Five years later Trelea \[[@CR13]\] proved that PSO with inertia weight of 0.6 and constant acceleration coefficients of 1.7 allowed to get faster convergence than that achieved by Eberhart \[[@CR11]\] and Clerc \[[@CR12]\]. The PSO method with nonlinear factors were proposed by Borowska \[[@CR14], [@CR15]\]. Furthermore, the efficiency of changing factors was examined by Ratnawera et al. \[[@CR16]\]. The cited authors concluded that time-varying acceleration coefficients (TVAC) helped to control local and global searching process more efficiently.Modification of the update equations. To improve searching process the researches propose to use a new update equation \[[@CR17], [@CR18]\] or add a new component to existing velocity equation \[[@CR19]\]. Another approach is to introduce, for ineffective particles, a repair procedure \[[@CR10]\] with other velocity updating equations that helps more precisely determine swarm motion and stimulate particles when their efficiency decreases.Topology structure. According to Kennedy \[[@CR20]\] topology structure affects the way information exchange and the swarm diversity. Many different topological structures have been proposed including: square, four clusters, ring, pyramid and the von Neumann topology \[[@CR20]--[@CR23]\]. Another approach is a multi-swarm structure recommended by Liang and Suganthan \[[@CR24]\] and Chen et al. \[[@CR25]\]. In contrast, Gong et al. \[[@CR22]\] have introduced a two-cascading-layer structure. In turn, Wang et al. \[[@CR26]\] developed PSO based on multiple layers.Learning strategy. It is used to improve performance of algorithm by breading high quality exemplars from which other swarm particles can acquire knowledge and learn to search space. A multi-swarm PSO based on dynamic learning strategy has been presented by Ye et al.\[[@CR27]\]. Likewise, Liang et al.\[[@CR28]\] has proposed a comprehensive learning strategy (CLPSO) according to which, particle velocity is updated based on historical best information of all other particles. To greater improve the performance and adaptability of CLPSO, Lin et al. \[[@CR29]\] recommend to use an adaptive comprehensive learning strategy with dynamically adjusting learning probability level according to the performance of the particles during the optimization process. Another approach is based on social learning PSO as described by Cheng et al. \[[@CR30]\].Hybrid methods combine beneficial features of two or more approaches. They are used to strength PSO efficiency and achieve faster convergence as well as better accuracy of the resultant solution. Holden et al. \[[@CR31]\] have proposed to join PSO with an ant colony optimization method. Li et al.\[[@CR32]\] have combined PSO with jumping mechanism of SA (simulated annealing). A modified version based on PSO and SA has been developed by Shieh et al. \[[@CR33]\]. In turn, PSO with chaos has been presented by Tian and Shi\[[@CR34]\], whereas Chen et al. \[[@CR35]\] have proposed learning PSO based on biogeography. Furthermore, a hybrid approach based on improved PSO, cuckoo search and clustering method has been developed by Bouer and Hatamlou \[[@CR36]\]. In order to enhance the PSO performance, Gong et al. \[[@CR22]\] have merged the latter two categories and proposed genetic learning particle swarm optimization (GL-PSO). In GL-PSO, except PSO and genetic operators, a two layer structure have been applied in which the former is used to generate exemplars whereas the latter to update particles through the PSO algorithm. The GL-PSO method improves the performance of PSO by constructing superior exemplars from which individuals of the population learn to move in the search space. Unfortunately, this approach is not free from disadvantages. In fact, the algorithm can achieve high convergence rate but in case of complex problems, due to global topology, the particle diversity quickly decreases and, as a result, impairs the exploration capability. In order to enhance the diversity and adaptability of GL-PSO as well as to improve its performance in solving complex optimization problems, in this paper, a new modified genetic learning method, referred to as GL-PSOIF, has been demonstrated. The proposed GL-PSOIF method is based on GL-PSO in which two modifications have been introduced. Specifically, instead of global topology, an interlaced ring topology has been introduced. The second modification relies on introducing a flexible local search operator. The task of the interlaced ring topology is to increase the population diversity and improve effectiveness of the method by generating better quality exemplars. In turn, a flexible local search operator has been introduced to enrich searching and improve the exploration and the exploitation ability. To evaluate the impact of the proposed modifications on performance of the proposed method, the interlaced ring topology has been first integrated with GL-PSO only (referred to as GL-PSOI) and then together with a flexible local search operator (referred to as GL-PSOIF). Both methods were tested on a set of benchmark problems and a CEC2014 test suite \[[@CR38]\]. The results were compared with five different variants of PSO, including the genetic learning particle swarm optimization (GL-PSO) \[[@CR22]\], the comprehensive particle swarm optimizer (CLPSO) \[[@CR28]\], the standard particle swarm optimization (PSO), the global genetic learning particle swarm optimization (GGL-PSOD) \[[@CR23]\], and the heterogeneous comprehensive learning particle swarm optimization (HCLPSO) \[[@CR39]\]. The PSO Method {#Sec2} ============== The PSO method was inspired by the social behavior of flocks of organisms (bird flocking, fish schooling, bees swarm) living in their natural environment \[[@CR2], [@CR3]\]. Likewise other evolutionary method, PSO is based on a population. Individuals of the population are called particles, and the population itself is called a swarm. In the PSO, the optimisation process is achieved by migration the particles towards the most promising area of the search space. Assuming that migration occurs in the *D*-dimensional search space, we can imagine particle swarm as a set of points each of which possess knowledge about: its actual position described by the position vector *x*~*j*~ = (*x*~*j1*~*, x*~*j2*~, *...,*  *x*~*jD*~), its current speed of movement described by velocity vector *v*~j~ = (*v*~*j1*~*, v*~*j2*~, *...,* *v*~*jD*~), its best position encountered by itself described by *pbest*~*j*~ = (*pbest*~*j1*~*, pbest*~*j2*~, *...,* *pbest*~*jD*~), and the best position encountered in all swarm described as *gbest *= (*gbest*~*1*~*, gbest*~*2*~, *...,* *gbest*~*D*~). In the first iteration, the position vector value and the velocity vector value are randomly generated. In subsequent iterations, values of the vectors are updated based on the knowledge and acquired experience of the particles. Changing of the particles velocity is achieved based on the Eq. ([1](#Equ1){ref-type=""}).$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ v_{j} (l + 1) = w \cdot v_{j} (l) + c_{1} \cdot r_{1} (pbest_{j} - x_{j} (l)) + c_{2} \cdot r_{2} (gbest - x_{j} (l)) $$\end{document}$$ Changing the particle position is realized by adding its actual velocity to its previous position ([2](#Equ2){ref-type=""})$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ x_{j} (l + 1) = x_{j} (l) + \cdot v_{j} (l + 1) $$\end{document}$$where: *w* - inertia weight, *pbest*~*j*~ -the best *j* particle position., *gbest* - the best position. in a swarm, *r*~1~*, r*~2~ .-random numbers generated from (0, 1), *c*~1~*, c*~2~- acceleration coefficients. Genetic Learning Particle Swarm Optimization {#Sec3} ============================================ In contrast to PSO, the GL-PSO algorithm possess a two-cascading-layer structure. One layer is used to generate exemplars, the other to update particles position and velocity through the PSO algorithm. To generate exemplars, three operators (crossover, mutation and selection) of the GA algorithm \[[@CR37]\] are applied. Exemplars *e*~*j*~ are selected from offspring. To generate offspring *o*~*j*~ for each dimension of particle *j*, a crossover operator is applied according to the formula:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ o_{j} = \left\{ {\begin{array}{*{20}l} {r \cdot pbest_{j,} + \left( {1 - r} \right) \cdot gbest, if f\left( {pbest_{j} } \right) < f(pbest_{k} ) } \hfill \\ {pbest_{k} \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \;otherwise} \hfill \\ \end{array} } \right. $$\end{document}$$where *k* is random selected particle, *r* --random number from (0, 1). Next, for each dimension, a random number $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ r\left[ {0,1} \right] $$\end{document}$ is generated and then if *r* \< *p*~m,~ (where *p*~m~ probability mutation) the offspring is mutated. Then the offspring undergoes the selection operation according to the formula:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ e_{j} \leftarrow \left\{ {\begin{array}{*{20}l} {o_{j} , \quad \quad if f\left( {o_{j} } \right) < f(e_{j} )} \hfill \\ {e_{j} , \quad \quad otherwise} \hfill \\ \end{array} } \right. $$\end{document}$$ The particle velocity is updated based on the following equation:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ v_{j} (l + 1) = w \cdot v_{j} (l) + c \cdot r_{{}} (e_{j} - x_{j} (l)) $$\end{document}$$where *e*~*j*~ is the exemplar of the *j* particle. The Proposed Method {#Sec4} =================== In order to improve the performance of global genetic learning particle swarm optimization (GL-PSO), in this article two modifications have been proposed: interlaced ring topology and flexible local search operator. Interlaced Ring Topology {#Sec5} ------------------------ One of the main reason for inability to obtain and pursue satisfactory performance of the GL-PSO is the lack or weakennes ability to maintain diversity of the population (swarm). This leads to a loss of balance between exploration and exploitation and consequently to premature convergence and unsatisfactory results. To avoid this, it is necessary to develop tools that could help increase adaptability of the algorithm, which, in turn, should give satisfactory results. Lin et al. \[[@CR22]\] have introduced ring and a global learning component with linearly adjusted control parameters to enhance a GL-PSO diversity. This improves the adaptability of the method but is not sufficient. Hence, the problem remains open and other solutions should be sought. To improve the adaptability of the GL-PSO, in this paper, instead of global learning, the interlaced ring topology has been proposed. This approach uses two neighbour particles, like in the ring topology, but in every next iteration (except the first one), the order of the particles is changed as follows. The particle collection is divided into two parts (sets) and particles of the second part take up spaces between the particles of the first part alternately (one particle from the first set, another particle from the second set, and next one from the first set, another from the second set etc.) according to Eq. [6](#Equ6){ref-type=""} and [7](#Equ7){ref-type=""}.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ n_{j} \, = \,\begin{array}{*{20}c} {\frac{j + 1}{2}} & {for\, odd \,j} \\ \end{array} $$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ n_{j} \, = \,\begin{array}{*{20}c} {\frac{j + N}{2}} & {for \,even \,j} \\ \end{array} $$\end{document}$$where *n*~j~ is the position of the particle to be moved to the *j* place in the ring, *j *= *1... N*, *N* is a swarm size (for example *n*~2~ = 5 means that the second position in the ring is occupied by a particle from 5th place in the swarm). Then, the position of exemplars are generated according to Eqs. [8](#Equ8){ref-type=""}, [9](#Equ9){ref-type=""} and [10](#Equ10){ref-type=""}.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ o_{j} = r \cdot pbest_{{n_{j1} }} + ( 1 - r) \cdot pbest_{{n_{j2} }} $$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ n_{j1} = \left\{ {\begin{array}{*{20}l} {N, \quad \quad j = 1} \hfill \\ {j - 1, \quad j > 1} \hfill \\ \end{array} } \right. $$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ n_{j2} = \left\{ {\begin{array}{*{20}l} {1, \quad \quad \;\,j = N} \hfill \\ {j + 1, \quad j\, < \,N} \hfill \\ \end{array} } \right. $$\end{document}$$where according to the ring topology *n*~*j*1~ and *n*~*j*2~ are the indexes of the adjacent particles from the left and right side of the particle *j*, respectively. Flexible Local Search Operator {#Sec6} ------------------------------ To improve the searching behavior of PSO and improve the exploitation capacity of the swarm, a flexible local search operator is introduced. The particle positions are updated according to the formula:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ x_{j}^{k + 1} = \left\{ {\begin{array}{*{20}l} {pbest_{j} \cdot (1 + N(0,1)), \quad \quad otherwise} \hfill \\ {x_{j}^{k} + v_{j}^{k + 1} , \quad \quad \quad \quad \quad \quad \;\,if\, p < s} \hfill \\ \end{array} } \right. $$\end{document}$$where *p* is a a randomly selected number in the range \[0,1\], *s* is a real number linearly increasing from 0.6 to 0.8. This means that each particle has a 40 to 20% possibility to perform search in the vicinity of its personal best position. This means that, according to \[[@CR16]\], in the early stage of the optimization process, the exploration is enhanced, and the local exploitation in the latter stage is facilitated. Test Results {#Sec7} ============ In order to investigate the efficiency of the proposed modifications, the GL-PSOI (in which only the interlaced ring topology was adopted) and GL-PSOIF (with interlaced ring topology and flexible local search operator) were evaluated, separately. Both strategies were tested on a set of classical benchmark problems, and on the CEC2014 test suite. Twelve of them (6 selected benchmark function and 6 CEC2014 functions) are described in this article and depicted in Tables [1](#Tab1){ref-type="table"} and [2](#Tab2){ref-type="table"}.Table 1.Optimization test functions. Table 2.Selected CEC2014 test suite.Functions NameRange*F(x\*)*F7Rotated Bent Cigar Function\[−100,100\]^*n*^100F8Shifted and Rotated Rosenbrock's Function\[−100,100\]^*n*^400F9Shifted and Rotated Ackley's Function\[−100,100\]^*n*^500F10Shifted Rastrigin's Function\[−100,100\]^*n*^800F11Shifted and Rotated Rastrigin's Function\[−100,100\]^*n*^900 The results of the tests were compared with performances of CLPSO, HCLPSO, PSO, GL-PSO and GGL-PSOD. The parameter settings of this algorithms are listed in Table [3](#Tab3){ref-type="table"}.Table 3.Parameters settings.AlgorithmParameter settingsCLPSO*w *= 0.9-0.4, *c *= 1.496HCLPSO*w *= 0.99-0.2, *c*~*1*~= 2.5-0.5, c~2~ = 0.5-2.5, c = 3-1.5PSO*w *= 0.9-0.4, *c*~1~ = 2.0, c~2~ = 2.0GL-PSO*w *= 0.7298, *c *= 1.49618, *p*~m~ = 0.01, s~g~ = 7GL-PSOD*w *= 0.7298, *c *= 1.49618, *p*~m~ = 0.01, s~g~ = 7 Both in the GL-PSOI and GL-PSOIF, the inertia weight *w* = 0.6 \[[@CR13]\]. The acceleration coefficients used in the computations were equal *c*~1~ = *c*~2~ = 1.7. In case of the set of benchmark functions, the population consisted of 20 particles, the dimension of the search space was 30, the maximum number of function evaluations was 300000. The search range depends on the function used as shown in Table [1](#Tab1){ref-type="table"}. For each problem, the simulations were run 30 times. For CEC2014 functions, the population consisted of 50 particles, the dimension of the search space was *D *= 30, and the maximum number of function evaluations was *D* × 10^4^. The search range was \[-100,100\]^n^. For CEC2014 functions, the algorithms were run 31 times independently. The exemplary results of the tests are summarized in Tables [4](#Tab4){ref-type="table"} and [5](#Tab5){ref-type="table"}.Table 4.The comparison test results of the PSO algorithms on the benchmark functions.FunctionsCriteriaCLPSOHCLPSOGL-PSOPSOGGL-PSODGL-PSOIGL-PSOIFF1Mean0.00E+00(=)0.00E+00(=)0.00E+00(=)3.48E−25(+)0.00E+00(=)0.00E+000.00E+00Std0.00E+000.00E+000.00E+002.08E−240.00E+000.00E+000.00E+00F2Mean6.88E+01(+)5.57E+00(+)2.43E−20(+)2.71E−11(+)6.74E−20(+)**3.15E**−**22**4.52E−21Std3.24E+014.03E+003.16E−204.29E−114.82E−202.67E−213.84E−20F3Mean2.34E+01(+)2.16E+00(+)6.48E−01(+)4.16E+01(+)6.53E−01(+)**5.02E**−**01**5.16E−01Std1.58E+014.24E+002.54E−013.92E+016.07E−015.48E−012.58E−01F4Mean1.02E−11(+)6.32E−12(+)7.14E−14(+)3.89E+01(+)4.32E−14(+)6.44E−15**3.50E**−**16**Std3.21E−128.40E−123.62E−149.22E+005.36E−145.37E−143.68E−15F5Mean2.05E−14(+)1.41E−12(+)7.86E−15(+)3.59E−13(+)6.29E−15(+)5.85E−15**5.32E**−**16**Std3.41E−154.07E−133.92E−157.91E−142.23E−152.73E−151.98E−15F6Mean1.82E−32(+)1.65E−32(+)1.73E−31(+)3.47E−02(+)2.11E−31(+)1.62E−32**1.57E**−**32**Std5.56E−485.56E−481.94E−325.89E−023.73E−325.04E−364.86E−34 Table 5.The comparison test results of the PSO algorithms on the CEC2014 test suite.FunctionsCriteriaCLPSOHCLPSOGL-PSOPSOGGL-PSODGL-PSOIGL-PSOIFF7Mean**3.24E**+**02(-)**4.15E +02(-)5.96E+02(+)8.09E+02(+)7.12E+02(+)4.58E+024.41E+02Std4.85E+026.73E+023.63E+023.34E+027.29E+026.73E+021.18E+02F8Mean6.93E+01(+)3.82E+01(-)**2.76E**+**01**(-)1.62E+02(+)6.27E+01(+)5.75E+014.64E+01Std3.15E+013.36E+016.59E+015.16E+013.49E+015.18E+012.37E+01F9Mean2.08E+01(=)**2.00E**+**01**(=)2.05E+01(=)2.32E+01(+)**2.00E**+**01**(=)**2.00E**+**012.00E**+**01**Std5.37E−026.24E−033.42E−028.89E−023.27E−022.83E−022.12E−02F10Mean4.07E−02(+)2.38E−01(+)1.95E−10(+)2.66E+01(+)2.43E−12(+)2.35E−13**1.57E**−**13**Std2.19E−025.40E−017.23E−118.19E+007.68E−136.48E−131.88E−13F11Mean4.20E+01(+)4.43E+01(+)5.84E+01(+)7.81E+01(+)3.57E+01(+)2.97E+01**2.35E**+**01**Std7.17E+001.26E+012.13E+012.69E+011.49E+011.56E+011.06E+01 The exemplary charts showing the mean fitness selected functions in the following iterations for GL-PSO, GGL-PSO, CLPSO, HCLPSO, PSO, GL-PSOI and GL-PSOIF algorithms, are depicted in Figs. [1](#Fig1){ref-type="fig"}, [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}.Fig. 1.Convergence performance for *f*~2~ function. Fig. 2.Convergence performance for *f*~4~ function. Fig. 3.Convergence performance for *f*~6~ function. The results of the tests confirmed that both GL-PSOI and GL-PSOIF are more effective and can achieve superior performance over the remaining tested methods. In case of unimodal functions, the GL-PSOI with interlaced ring topology obtained superior results over the ones for GL-PSOIF. For multimodal functions superior results were achieved by GL-PSOIF. In case f2 function, GL-PSO achieved worse results than GL-PSOI and GL-PSOIF but better than those obtained by the CLPSO, HCLPSO and PSO. For f3 function, GL-PSOI achieved the best result. The performance of GL-PSO was worse than that obtained by GL-PSOI but superior then performance of GL-PSOIF. For unimodal f7 function the best results were obtained by CLPSO. The outcomes achieved by GL-PSOI and GL-PSOIF were worse than results obtained by CLPSO but better than the results achieved by the remaining tested methods. For multimodal functions, the results show that (almost in all cases) GL-PSOIF exhibit the best performance. The convergence curves presented in Figs. [1](#Fig1){ref-type="fig"}, [2](#Fig2){ref-type="fig"}, and [3](#Fig3){ref-type="fig"} indicate that both GL-PSOI and GL-PSOIF converge slower in the early stage of the optimization process than most of the compared methods. At this stage, each algorithm, except PSO, is faster. Then both algorithms accelerate and converge faster than the others. In case of the unimodal f2 function, both algorithms initially revealed slower convergence, which was followed by a further rapid acceleration after about 5x10^4^ iterations showing superiority over the rest evaluated methods. For the unimodal f2 function, GL-PSOIF performed a bit slower than GL-PSOI, which could be due to the introduction of flexible search operator, which did not improved the GL-PSOIF run. In case multimodal functions (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}), GL-PSOIF converges slowly (other methods are faster) but after about 1.3x10^5^ iterations accelerates and after 2x10^5^ iterations becomes the fastest. Statistical Test {#Sec8} ================ In order to evaluate the differences between algorithms, a statistical t-test was used. A confidence level of 0, 05 was selected for all statistical comparisons. Tables [4](#Tab4){ref-type="table"} and [5](#Tab5){ref-type="table"} shows the results of the t-test performed on the test functions. The signature '+' indicates that GL-PSOIF is significantly better than the other algorithms, '−' worse to the other algorithms, and '=' equal to the other algorithms. The rows in Table [6](#Tab6){ref-type="table"} named '+', '−' and '=' mean the number of times that the GL-PSOIF is better than, worse than or equal to the other algorithms. The results of the t-test indicate that proposed algorithm is significantly better than other methods with 95% confidence level in a statistically meaningful way.Table 6.The comparison test results of the PSO algorithms.SignatureCLPSOHCLPSOGL-PSOPSOGGL-PSOD+878118−12101=22202 Conclusion {#Sec9} ========== In this study, a new genetic learning particle swarm optimization with interlaced ring topology and flexible local search operator (GL-PSOIF) has been proposed. To assess the impact of introduced modifications on performance of the evaluated method, first the interlaced ring topology was integrated with GL-PSO only (referred to as GL-PSOI) and then with the flexible local search operator (GL-PSOIF). The efficiency of the new strategy was tested on a set of benchmark problems and the CEC2014 test suite. The results were compared with five different variants of PSO, including GL-PSO, GGL-PSOD, PSO, CLPSO and HCLPSO. The results of the experimental trials indicated that the genetic learning particle swarm optimization with interlaced ring topology is effective for unimodal function. In case of the multimodal function, GL-PSOIF showed superior performance over the remaining tested methods.

1. Introduction {#sec1} =============== Many herbs have been shown to possess therapeutic potential towards several medical conditions, and hence, they are of a substantial medicinal value. For centuries, people around the globe have been using numerous medicinal herbs to alleviate the signs and symptoms of various disorders [@bib1]. Herbal medicine, also known as botanical medicine, phytomedicine, phytotherapy, herbology, and herbalism, is a form of therapy that uses plants or plant extracts to prevent or treat different diseases and to boost the overall health status [@bib1]. Herbal medicine is one of the oldest, if not the oldest, and probably remains to be of growing popularity. It is really intriguing that despite the great advancement in the fields of conventional medicine and drug discovery, the use of herbal formulations is still extremely widespread throughout the world, indicative of peoples\' perception of the safety and therapeutic efficacy of such medicinal herbs. Although herbal medicine is more prevalent in Asia, Africa, and to a lesser extent in Europe, the use of medicinal herbs has witnessed a significant, gradual increase in North America [@bib1], [@bib2]. It is most likely the gentle, nourishing, efficacious, synergistic, cost-effective, and safe properties of medicinal herbs that make them an attractive option for many people as therapeutic agents [@bib1], [@bib2]. In fact, the discovery of the vast majority, if not all, conventional drugs is based on the chemical, physiological, and therapeutic actions of the bioactive constituents of many medicinal herbs. The recent advancement in pharma and medicine, manifested by the development of biotechnologies, and mass production of highly specific, chemically-synthesized drugs, has certainly revolutionized the therapeutic approach to health care and disease management worldwide. However, herbal medicine continues to be a primary ideology in many populations today and a very common practice in different parts of the world. Herbs and spices are known to be major taste enhancers in most cuisines, primarily used as a source of flavor and aroma. Besides their use as food additives, a wide range of herbs and spices have been used to prevent or treat medical conditions including cancer. Over the past few decades, research investigating dietary factors and their effects on various medical conditions has been constantly growing. A large number of studies focused on these naturally-occurring products and reported a plethora of anti-cancer properties manifested through various molecular mechanisms. For instance, an ethanolic extract of *Piper nigrum* (black pepper) has been shown to induce DNA damage and reduce cell viability in MCF-7 human cancer cells [@bib3]. Treatment with the ethanolic extract of *P. nigrum* inhibited cell proliferation by 57% and elevated ROS levels by 65%. Moreover, the same extract increased Bax and p53 levels, both of which are key proteins in regulating the cell cycle arrest. Another study used flow cytometric analysis to describe the anti-cancer effects exerted by *Fagonia cretica*, a tea herb and a food additive [@bib4]. Indeed, the *F. cretica* extract caused a dose-dependent arrest of the cell cycle at G0/G1 phase and enhanced the rate of apoptosis in MCF-7 and MDA-MB-231 human cancer cells. Another example of a widely used active food constituent is sesamin, a major lignin in sesame seeds. Siao and colleagues showed that sasamin plays a strong preventive role against cancer by modulating apoptotic signaling pathways and restricting angiogenesis [@bib5]. Hence, various herbs and food additives are becoming widely used for the treatment and/or the prevention of acute and chronic conditions ranging from mild allergies to more serious diseases including cancer. Yet, despite the intensive research efforts devoted to the identification of herbs with therapeutic properties, the exact molecular pathways and cellular mechanisms by which these herbs induce their therapeutic effects are not fully understood. *Nigella sativa* is an annual flowering plant that is grown almost all over the world but is native to South and Southwest Asia and commonly found in Northern Africa, the Middle East, and Southern Europe [@bib6], [@bib7]. *N. sativa* is also known as nigella, blackseed, black cumin, black caraway, Roman coriander, fennel flower, nutmeg flower, "kalonji" (in India), "Kalo jeera" (in Bangladesh), "Hak Jung Chou" (in China), and "habbat al-barakah" (in the Middle East). *N. sativa* belongs to the botanical family *Ranunculaceae* [@bib8], [@bib9]. The mature plant grows to 20--90 cm of height with finely divided leaves, and white, pale blue, or pale purple delicate flowers containing 5--10 petals [@bib10]. The follicles within the fruits contain many small (2.0--3.5mm × 1.0--2.0 mm) angular, black seeds with whitish interior [@bib10]. Aside from its use as a food flavoring additive, *N. sativa* seeds oil and extracts have been used since ancient times to treat several diseases and medical conditions. *N. sativa* plant extracts have been commonly used in various traditional systems of medicine like Ayurveda, Siddha, Unani, Arabic, Islamic, etc. Several *N. sativa* crude extracts have been popularly used in traditional medicine as appetite stimulants, bronchodilators, liver tonics, and analgesics as well as to treat various conditions like diabetes, asthma, hypertension, cardiovascular disease, liver and kidney diseases, digestive problems, diarrhea, skin disorders, microbial infections, cancer, etc. [@bib7], [@bib8], [@bib9], [@bib10], [@bib11], [@bib12], [@bib13], [@bib14], [@bib15], [@bib16], [@bib17], [@bib18], [@bib19] Such uses of *N. sativa* extracts in traditional medicine have been validated by well-designed experiments showing that such extracts possess cardio-protective, anti-microbial, anti-histaminic, anti-diabetic, antihypertensive, anti-hyperlipidemic, anti-diarrheal, hepato-protective, renal protective, gastro-protective, spasmolytic, immunomodulatory, anti-inflammatory, anti-oxidant, and anti-cancer properties [@bib7], [@bib8], [@bib9], [@bib10], [@bib11], [@bib12], [@bib13], [@bib14], [@bib15], [@bib16], [@bib17], [@bib18], [@bib19]. Hence, traditional medicine uses that are validated by experimental evidence strongly suggest that *N. sativa* extracts can be of potent therapeutic efficacy in the prevention and treatment of various infectious and non-infectious diseases. In this review, the *in vitro* and *in vivo* anti-cancer properties of *N. sativa* extracts are discussed. Special emphasis is given to the molecular and cellular mechanisms that mediate the anti-proliferative, pro-apoptotic, and anti-oxidant effects of *N. sativa*. Recent advances in the establishment of an effective therapeutic potential of *N. sativa* extracts, leading to suppressed tumor initiation and progression, are also underscored. 2. Anti-proliferative and pro-apoptotic effects of *N. sativa* {#sec2} ============================================================== The potent anti-cancer potential of *N. sativa* is well established through *in vitro* and *in vivo* studies using different cell lines and animal models. Driven by traditional medical practices in Sri Lanka, a decoction (hot-water extract) comprised of *N. sativa* (seeds), *Hemidesmus indicus* (roots), and *Smilax glabra* (rhizome), a polyherbal mixture used to treat different types of cancer, has been shown to ameliorate diethylnitrosamine-induced hepatocarcinogenesis in male Wistar rats at a dose of 4--6 g/kg/day after 10 weeks of oral feeding [@bib20]. The researchers of the aforementioned study indicated that the potential anti-cancer effects of the extracts of the individual plants in the decoction were not examined because only the decoction is traditionally used in cancer chemotherapy [@bib20]. Subsequent studies suggest that flow cytometric analysis conducted using Annexin V and propidium iodide staining demonstrated that HepG2 cells were in the late stage of apoptosis and/or necrosis 24 h post treatment with the polyherbal mixture [@bib21]. Consistently, oral administration (6 g/kg/day) of the polyherbal mixture of *N. sativa*, *H. indicus*, and *S. glabra* led to a long-term protection against diethylnitrosamine-induced hepatocellular adenoma in Wistar rats [@bib22], [@bib23]. In fact, a great deal of literature underscores many *in vitro* and *in vivo* effects of pure *N. sativa* extracts. In an early *in vivo* study, topical application of *N. sativa* extract (100 mg/kg) inhibited the two-stage initiation/promotion of skin carcinogenesis and delayed the onset of skin papilloma in mice challenged with 7,12-dimethylbenzanthracene/croton oil [@bib24]. The same study revealed that intraperitoneal administration of *N. sativa* extract significantly reduced methylcholanthrene (MCA)-induced soft tissue sarcomas in albino mice by about 70% following 30 days of subcutaneous administration of MCA [@bib25]. Aqueous and ethanolic extracts of *N. sativa* seeds, both separately and in combination, were shown to exert potent anti-proliferative effects on MCF-7 human breast cancer cells in presence and absence of H~2~O~2~, which seems to play a synergistic role [@bib26]. In another study, Salim and Fukushima examined the effects of *N. sativa* oil on the development of colon tumors in a murine model of 1,2-dimethylhydrazine (DMH)-induced colon cancer [@bib27]. Fourteen weeks post DMH challenge, Fischer 344 rats that were treated with *N. sativa* oil at the initiation and post-initiation stages of colon carcinogenesis displayed significantly reduced DMH-induced aberrant crypt foci (ACF), which are putative pre-neoplastic lesions for colon cancer [@bib27]. Immunohistochemical analysis revealed that *N. sativa* oil exerted potent anti-proliferative activity in the colonic ACF in rats that were treated with *N. sativa* oil at both the initiation and post-initiation stages of DMH challenge [@bib27]. Similarly, using 7,12-di-methylbenz(a)anthracene (DMBA)-induced mammary carcinoma model, female Sprague--Dawley rats that were injected with DMBA were subsequently orally treated with *N. sativa* oil (4 g/kg/day) starting 2 weeks before or at the time of DMBA injection, and the experiment lasted for 3 months [@bib28]. The frequency of mammary papillary, comedo, and cribriform carcinoma was reduced in rats treated with *N. sativa* oil at the time of DMBA injection, and this frequency was more potently recused in rats that were pre-treated with *N. sativa* oil for 2 weeks before DMBA challenge [@bib28]. The reduced frequency of mammary carcinoma was associated with reduced serum levels of tumorigenicity markers (total sialic acid (TSA) and lipid-bound sialic acid (LSA)), serum levels of endocrine derangement markers (prolactin, estradiol, and progesterone) and levels of apoptotic markers (serum tumor necrosis factor α (TNFα), tissue caspase-3 activity, and DNA fragmentation) [@bib28]. Using the essential oil, an ethanolic extract, and a butanol extract of *N. sativa* and different cell lines (P815, IC01, Vero cells, and BSR cells), Ait Mbarek and colleagues demonstrated that the potency of the *in vitro* anti-cancer activity of *N. sativa* depends, at least partially, on the tumor cell type [@bib29]. In the aforementioned study, the anti-cancer activity of *N. sativa* essential oil was also evaluated *in vivo*. Injection of 30--50 μl (28.5--47.5 mg/mouse) *N. sativa* essential oil into the tumor site of a DBA2/P815 (H2d) mouse model led to a significant suppression of solid tumor development (more than 10-fold decrease in tumor size) and resulted in a significantly delayed mortality of P815 mastocytoma tumor-bearing mice [@bib29]. Recently, the administration of *N. sativa* ethanolic extract treatment was shown to improve the histopathological changes in the malignant liver tissue which were caused by diethylnitrosamine (DENA) treatment, without causing any direct cytotoxic effect [@bib30]. In a similar study, the effects of a methanolic extract of *N. sativa* on the modulation of glyco-regulatory enzymes in an albino rat model of hepatocellular carcinoma were investigated [@bib31]. Hepatocellular carcinoma was induced in albino rats by intraperitoneal injection of DENA and carbon tetrachloride (CCl~4~), leading to a significant increase in the serum level of α-fetoprotein (AFP), the relative liver weight, and the activities of hexokinase, glyceraldehyde phosphate dehydrogenase, and G6P dehydrogenase in both the serum and liver homogenate of treated rats. Oral administration of a methanolic extract of *N. sativa* (1 g/kg/day) for 2 weeks prior to induction of hepatocellular carcinogenesis improved the histopathological changes associated with DENA and CCl~4~ treatment, bringing the physiological and biochemical parameters indicated above back to normal levels [@bib31]. Very recently, an *in vitro* study demonstrated that an aqueous extract of *N. sativa* (0.1--1.0% concentration) caused a significant decrease in cell proliferation and varying morphological changes including cell shrinkage and membrane damage in HepG2 cells, accompanied by DNA damage and cell death [@bib32] ([Fig. 1](#fig1){ref-type="fig"}). 3. Anti-oxidant and cytotoxic effects of *N. sativa* {#sec3} ==================================================== Among the first reports pointing to the potential anti-cancer properties of *N. sativa*, Swamy and Tan demonstrated that an aqueous extract and an ethyl acetate chromatographic fraction of *N. sativa* seeds (50 μg/ml) caused significant cytotoxic effects against various types of cancer cell lines (HepG2, MOLT4, and LL/2), but not against normal, non-cancerous human umbilical cord endothelial cells [@bib33]. Aside from their anti-proliferative effects, both aqueous as well as ethanolic extracts of *N. sativa* seeds were found to induce significant cytotoxic effects on MCF-7 cells in presence and absence of H~2~O~2~ [@bib26]. However, the ethanolic extract of *N. sativa* exerted more potency against MCF-7 cells compared to the aqueous extract (LC~50~ values in presence of H~2~O~2~ were 377 μM and 725 μM, respectively). Also, the aforementioned anti-cancer polyherbal mixture, which is comprised of *N. sativa* (seeds), *H. indicus* (roots), and *S. glabra* (rhizome), was shown to exert cytotoxic effects in human hepatoma HepG2 cell line at 5--50 mg/ml concentration [@bib21]. In fact, the three individual plant extracts exerted cytotoxic efficacy in the order *N. sativa* \> *H. indicus* \> *S. glabra* [@bib21]. Such anti-cancer effects were confirmed by Samarakoon and colleagues who demonstrated that both the aqueous and ethanolic extracts of the polyherbal mixture of *N. sativa*, *H. indicus*, and *S. glabra* caused strong dose-dependent cytotoxicity in HepG2 cells [@bib22]. However, most of these studies do not yield insightful results since *N. sativa* extracts were used in combination with *H. indicus* and *S. glabra* extracts, making it challenging to draw plausible conclusions regarding the anti-cancer activity of *N. sativa* itself. Nonetheless, several studies examined the effects of *N. sativa* and its extracts on various cell lines. An *in vitro* cytotoxic study showed that a crude methanolic extract of *N. sativa* caused about 50% cytotoxicity in Ehrlich ascites carcinoma (EAC), Dalton\'s lymphoma ascites (DLA), and Sarcoma-180 cells (S-180 cells) [@bib24]. Another *in vivo* study demonstrated that 6-month oral administration of *N. sativa* seeds (0.2 g/rat/day) provided protective effects against methylnitrosourea-induced oxidative stress and colon carcinogenesis in Sprague Dawely rats due to reduced expression of malondialdehyde (MDA), a biomarker of lipid peroxidation, and nitric oxide (NO) [@bib34]. Zaoui and colleagues examined the possible biochemical and histopathological effects of *N. sativa* fixed oil in *Iops of a* mice and Wistar-Kyoto rats [@bib35]. Acute toxicity of *N. sativa* fixed oil was assessed in mice that received a single oral or intraperitoneal dose, and the LD~50~ values were determined to be 28.8 ml/kg and 2.06 ml/kg, respectively [@bib35]. The chronic toxicity of *N. sativa* fixed oil was assessed in rats receiving a daily oral dose of 2 ml/kg for a period of 12 weeks. It was demonstrated that chronic treatment with *N. sativa* fixed oil did not affect the level or catalytic activity of key hepatic enzymes including aspartate-aminotransferase, alanine-aminotransferase, and gamma-glutamyltransferase, nor it had any marked histopathological effects in the heart, liver, kidney, and pancreatic tissues [@bib35]. The very low toxicity of the chronic treatment with *N. sativa* fixed oil was evidenced by biochemical stability and high LD~50~ values, suggesting that the indicated doses are sub-toxic and do not raise major safety concerns. Moreover, Islam and colleagues demonstrated that *N. sativa* oil exerts cytotoxic effects against a panel of four human cancer cell lines (SCL, SCL-6, SCL-37′6, and NUGC-4) and 3T6 fibroblast mouse cell line with LC50 values 155.02 ± 10.4, 185.77 ± 2.9, 120.40 ± 20.5, 384.53 ± 12.1, and 286.83 ± 23.3 μg/ml, respectively, with no significant cytotoxic effects on normal cells [@bib36]. In another study, Ali assessed the ability of *N. sativa* oil to ameliorate the nephrotoxicity associated with gentamicin, an antibiotic, in rats [@bib37]. Intramuscular injection of gentamicin was associated with proximal tubular damage, histopathological and biochemical signs of nephrotoxicity, elevated levels of creatinine and urea, as well as decreased level of glutathione (GSH) and total anti-oxidant status [@bib37]. Such effects were abrogated by oral administration of *N. sativa* oil (1--2 ml/kg/day) for 10 days, without any detectable overall toxicity [@bib37]. Similarly, oral *N. sativa* treatment (4 g/kg/day) of rats with DMBA-induced mammary carcinoma for a period of 3 months resulted in reduced tissue levels of oxidative stress markers (NO and lipid peroxides) [@bib28]. Intragastric administration of *N. sativa* oil for 12 days in male albino rats potently reduced the hepatic and overall toxic effects associated with intraperitoneal administration of cyclophosphamide, an anti-cancer drug that causes a high degree of lipid peroxidation and reactive oxygen species (ROS) over-production [@bib38]. Similarly, oral administration of *N. sativa* oil (90 mg/kg/day) in albino rats for 30 and 60 days significantly ameliorated, in a time-dependent manner, the toxic effects and pathological tissue damage in the spleen and thymus resulting from treatment with chloramphenicol, a potent antibiotic [@bib39]. These findings suggest that *N. sativa* oil co-treatment could potentially reduce the toxicity-related side effects that accompany the bactericidal and anti-cancer chemotherapy. In a recent study, the hepatotoxic effects of *N. sativa* were evaluated in Spargue Dawley rats by measuring the catalytic activity of key liver enzymes (ALT and AST) and by histopathological assessment of liver tissue [@bib40]. Rats were fed diet supplemented with 0.01--1 g/kg/day of *N. sativa* seeds powder for 28 days. It was demonstrated that *N. sativa* powder supplementation did not lead to a significant change in the catalytic activity of ALT and AST, histopathological abnormalities, inflammation, or necrosis in the liver tissue even at the highest dose of 1 g/kg/day [@bib40]. This study showed that 0.01--1 g/kg/day doses of *N. sativa* seeds powder caused no marked toxic effects on liver function in rats and they are considered safe. Very recently, Hadi and colleagues performed a clinical trial to assess the anti-oxidant effects of *N. sativa* oil in patients with rheumatoid arthritis (RA) [@bib41]. It was revealed that a daily dose of 1 g *N. sativa* oil for 8 weeks significantly reduced the serum levels of MDA and NO, suggesting that *N. sativa* can potentially be employed in the treatment of RA due to its ability to suppress RA-associated oxidative stress responses [@bib41] ([Fig. 1](#fig1){ref-type="fig"}). 4. Anti-mutagenic effects of *N. sativa* {#sec4} ======================================== A few studies have examined the potential of *N. sativa* to exert anti-mutagenic activity against *N*-methyl-*N*′-nitro-*N*-nitrosoguanidine (MNNG), a directly acting mutagen. Although an aqueous extract of *N. sativa* had no cytoprotective nor anti-mutagenic activity against MMNG in primary rat hepatocytes [@bib42], an ethanolic extract of *N. sativa* exerted an inhibitory effect against MNNG mutagenicity due to significantly reduced chromosomal aberrations in primary rat hepatocytes [@bib43]. The anti-mutagenic activity of the ethanolic extract of *N. sativa* was observed in MNNG-challenged primary rat hepatocytes that were pre-treated, co-treated, or post-treated with the extract, without inducing direct apoptosis [@bib43]. Such anti-mutagenic effects against MNNG were attributed to possible induction of detoxifying enzymes that degrade MNNG, chemical interaction with or absorption of MNNG (or its electrophilic degradation products), enhanced fidelity of DNA replication, and/or improved DNA repair [@bib43]. Such factors that prevent or reduce chromosomal aberrations [@bib43]. Although some findings provide evidence of a potent anti-mutagenic activity of *N. sativa*, research is still in its early stages of establishing a direct link between the specific ingredients in *N. sativa* extracts and the anti-mutagenic activity of *N. sativa* ([Fig. 1](#fig1){ref-type="fig"}). 5. Anti-metastatic effects of *N. sativa* {#sec5} ========================================= Awad investigated the effect of *N. sativa* oil on HT1080 human fibrosarcoma cell lines with regard to their fibrinolytic potential, a hallmark of malignant tumors [@bib44]. *N. sativa* oil (25--200 μg oil/ml) caused a significant dose-dependent down-regulation of key fibrinolytic products including tissue-type plasminogen activator (t-PA), urokinase-type plasminogen activator (u-PA), and plasminogen activator inhibitor type 1 (PAI-1), both in sub-confluent and confluent cell cultures [@bib44]. This study highlights the ability of *N. sativa* to hinder local tumor invasion and metastasis. Ait Mbarek and colleagues reported similar findings, whereby injection of 30--50 μl (28.5--47.5 mg/mouse) *N. sativa* essential oil into the tumor site of a DBA2/P815 (H2d) mouse model resulted in inhibition of liver metastasis even after 30 days of treatment [@bib29] ([Fig. 1](#fig1){ref-type="fig"}). 6. Effects of *N. sativa* on natural killer (NK) cytotoxic activity {#sec6} =================================================================== Enhancement of NK cytotoxic activity has been proposed by several research groups to serve as a mechanism underlying the anti-cancer effects of *N. sativa* [@bib45], [@bib46], [@bib47], [@bib48], [@bib49], [@bib50]. In an *in vivo* study performed on healthy volunteers, El-Kadi and colleagues showed that ingestion of *N. sativa* oil for 4 weeks enhanced the ratio of helper to suppressor T cells and significantly improved NK cytotoxic function [@bib46]. In agreement, an *in vivo* study performed in mice revealed that 1-week oral administration of an aqueous extract of *N. sativa* caused a significant increase in the number of splenic NK cells and a significant enhancement of splenic NK cytotoxic activity against YAC-1 tumor cells [@bib48]. These *in vivo* findings were supported by *in vitro* studies. Abuharfeil and colleagues demonstrated that a fresh aqueous extract of *N. sativa* (50 and 100 μg/ml) led to a significant increase in splenic NK cytotoxic activity against YAC-1 tumor cells (% cytotoxicity 43.7 ± 3.6 and 62.7 ± 5.6 at 200:1 E:T ratio, 45.7 ± 5.7 and 44.6 ± 6.2 at 100:1 E:T ratio, and 13.6 ± 2.7 and 18.3 ± 3.1 at 50:1 E:T ratio, respectively) [@bib47]. Indeed, the fresh aqueous extract of *N. sativa* appeared to be more potent in inducing NK cytotoxic activity compared to the old dried aqueous extract or the ethanolic extract [@bib47]. A similar study from our laboratory provided further *in vitro* experimental evidence indicating that an aqueous extract of *N. sativa* (50--100 μg/ml) significantly enhanced killing of YAC-1 tumor cells due to augmented NK cytotoxic activity leading to 14% (3 folds) and 23% (4.5 folds) cytotoxicity 200:1 E:T ratio at 50 μg/ml and 100 μg/ml concentrations, respectively [@bib50]. Importantly, enhanced killing of YAC-1 tumor cells is due to the ability of *N. sativa* extract to improve NK cytotoxic activity rather than inducing an immediate cytotoxic effect. This is evidenced by the findings that *N. sativa* extract had no significant, direct cytotoxic effect against YAC-1 tumor cells in absence of NK cells [@bib50]. We have reported similar observations in which aqueous extracts (100 μg/ml) of black pepper (*P. nigrum*) and cardamom (*Elettaria cardamomum*) caused a significant increase (35% and 45% cytotoxicity, respectively) in the NK cytotoxic activity against YAC-1 tumor cells [@bib51]. Therefore, it seems that boosting the cytotoxic potential of NK cells against tumor cells is at least one mechanism exploited by several plant extracts to exert their tumoricidal action. Interestingly, Abuharfeil and colleagues assessed NK cytotoxic activity in the presence of the aqueous extract of *N. sativa* using splenocytes obtained from BALB/c mice [@bib47], whereas in our study the NK cytotoxic activity was assessed using splenocytes obtained from C57/BL6 mice [@bib50]. Although the enhancement of NK cytotoxic activity caused by *N. sativa* does not seem to be strain-specific, more studies are required to confirm this argument using splenocytes from a wide range of mice strains and even different animal models. Along the same lines, an aqueous extract of *N. sativa* (10--500 μg/ml) was shown to significantly enhance the cytotoxic activity (26.6--67.7% cytotoxicity) of NK cells isolated from human blood against K562 tumor cells *in vitro* [@bib49]. The improved cytotoxic potential of NK cells was primarily due to the ability of *N. sativa* extract to significantly enhance the production of interferon γ (IFNγ) and TNFα, immunostimulatory cytokines with potent tumoricidal activity, from NK cells [@bib49]. Moreover, treatment of NK cells with *N. sativa* extract led to a significant increase in the release, and hence activity, of granzyme A and N-acetyl-β-[d]{.smallcaps}-glucosaminidase, key proteolytic enzymes involved in target cell killing [@bib49]. These findings suggest that augmentation of NK cytotoxic activity against tumor cells serves as an effective immunomodulatory mechanism that may explain, at least partially, the reported *in vitro* and *in vivo* anti-cancer effects of *N. sativa*. An early *in vivo* study, however, demonstrated that intraperitoneal injection of *N. sativa* oil (100 μg/100 ml/mouse) for 7 days caused a significant decrease in the number of splenic NK cells in non-infected and CMV-infected BALB/c mice [@bib52]. Interestingly, although *N. sativa* oil treatment had no effect on NK cytotoxic activity in non-infected mice, it caused a significant suppression of NK cytotoxic activity in CMV-infected mice [@bib52]. The same study revealed that *in vitro* treatment of splenic NK cells isolated from non-infected mice with *N. sativa* oil (100 μg/ml) significantly decreased their cytotoxic activity against YAC-1 tumor cells [@bib52]. These findings are inconsistent with those reported by El-Kadi and his colleagues [@bib46] regarding the effects of *N. sativa* oil on NK cytotoxic effects against cancer cells. These inconsistent findings are most likely due to different experimental conditions including the dose of *N. sativa* extract or oil, cell type, species, incubation time, and method of detection. It is worth mentioning that *N. sativa* oil may exert toxic effects against NK cells, which could be another factor influencing the outcome of the reported experiments. Future studies, with a carefully-designed experimental approach that addresses the raised possible experimental variables, are required to shed more light on the potential modulatory effects of *N. sativa* oil on NK cytotoxic activity. Although some of the signaling molecules involved in mediating the immunostimulatory effects of *N. sativa* extracts in NK cells have been identified, the exact signaling pathways and molecular targets implicated in these pathways are largely unknown. Future *in vitro* and *in vivo* studies should focus on elucidating the targeted receptors and intracellular/extracellular factors involved in the signal transduction pathways that are modulated in NK cells by *N. sativa* extracts. Furthermore, we suggest that the stimulatory potential of *N. sativa* toward NK cytotoxic activity be further confirmed by *in vitro* and *in vivo* studies using a wide range of primary and transformed NK cells against numerous primary tumors and cancer cell lines ([Fig. 1](#fig1){ref-type="fig"}). 7. Signaling pathways underlying the anti-cancer effects of *N. sativa* {#sec7} ======================================================================= Several *in vitro* and *in vivo* studies were conducted in an attempt to elucidate the molecular and cellular mechanisms underlying the anti-cancer activity of *N. sativa*. The key mechanisms underlying the documented anti-cancer effects of *N. sativa* have been largely attributed to their ability to modulate the activity of key enzymes [@bib31], [@bib43], [@bib44], [@bib49], [@bib53], [@bib54], [@bib55], [@bib56], [@bib57], [@bib58]. suppress inflammation [@bib8], [@bib50], [@bib53], [@bib55], [@bib59], [@bib60], [@bib61], [@bib62], [@bib63], [@bib64], [@bib65], [@bib66], [@bib67], [@bib68], [@bib69], [@bib70], [@bib71], [@bib72], [@bib73], [@bib74], [@bib75], [@bib76], [@bib77], [@bib78], [@bib79], [@bib80], [@bib81], [@bib82], and induce apoptosis in tumor cells [@bib21], [@bib41], [@bib54], [@bib55], [@bib72], [@bib83], [@bib84], [@bib85], [@bib86], [@bib87], [@bib88], [@bib89], [@bib90], [@bib91], [@bib92], [@bib93], [@bib94], [@bib95], [@bib96], [@bib97], [@bib98], [@bib99], [@bib100], [@bib101]. One mechanism that is implicated in tumorigenesis involves the inducible nitric oxide synthase (iNOS) pathway. NO, which is synthesized by iNOS or other nitric oxide synthase (NOS) isoforms during physiological reactions including inflammation, is an endogenous radical implicated in predisposition to tumor development. In a recent study, Fathy and Nikaido investigated the effect of an ethanolic extract of *N. sativa* on modulating the iNOS pathway in rats with DENA-induced hepatocarcinogenesis [@bib30]. Oral administration of *N. sativa* ethanolic extract (250 mg/kg/day) for 5 days led to a significant reduction in the serum levels of AFP, NO, interleukin-6 (IL-6), and TNFα, factors whose production was significantly increased after treatment with DENA [@bib30]. Very recently, Alhamzi and colleagues demonstrated that a methanolic extract of *N. sativa* seeds (50--100 μl/ml) induced apoptosis in MCF-7 cells in a time- and dose- dependent manner, as judged by TUNEL assay [@bib97]. The methanolic extract of *N. sativa* led to a significant time- and dose-dependent increase in the expression of apoptotic factors including caspase-3, caspase-8, caspase-9, and p53 in MCF-7 cells, indicating that *N. sativa* manifests its anti-cancer activity by targeting the p53 and caspase signaling pathways [@bib97]. A brief summary about the reported *in vitro* and *in vivo* anti-cancer activities of *N. sativa* is given in [Table 1](#tbl1){ref-type="table"}. 8. Anti-cancer effects of *N. sativa* phytoconstituents {#sec8} ======================================================= Many of the anti-cancer activities of *N. sativa* have been attributed to its major active constituent, thymoquinone (TQ). TQ has been shown to exert anti-proliferative, pro-apoptotic, anti-oxidant, anti-oxidant, anti-mutagenic, anti-angiogenic, and anti-metastatic effects against cancer cells [@bib6], [@bib12], [@bib14], [@bib16], [@bib17], [@bib18], [@bib19], [@bib38], [@bib53], [@bib55], [@bib63], [@bib64], [@bib66], [@bib70], [@bib71], [@bib77], [@bib82], [@bib83], [@bib84], [@bib85], [@bib86], [@bib87], [@bib88], [@bib89], [@bib90], [@bib91], [@bib92], [@bib93], [@bib94], [@bib95], [@bib96], [@bib97], [@bib98], [@bib99], [@bib100]. TQ seems to mediate its anti-cancer effects by targeting a number of cellular pathways involving p53, NF-κB, PPARγ, STAT3, MAPK, and PI3K/AKT transducing signals [@bib67], [@bib69], [@bib72], [@bib83], [@bib84], [@bib85], [@bib86], [@bib87], [@bib88], [@bib89], [@bib90], [@bib91], [@bib92], [@bib93], [@bib94], [@bib95], [@bib96], [@bib97], [@bib98], [@bib99], [@bib100]. Besides TQ, other phytoconstituents of *N. sativa* have also been shown to contribute to the anti-cancer potential of *N. sativa* extracts. α-hederin is a pentacyclic triterpene saponin found in *N. sativa* seeds that exerts effective anti-cancer effects, both *in vitro* and *in vivo* [@bib54], [@bib102], [@bib103], [@bib104], [@bib105], [@bib106], [@bib107]. Moreover, thymol, thymohydroquinone, dithymoquinone, nigellimine-N-oxide, nigellicine, nigellidine, and carvacrol are phytoconstituents of *N. sativa* that have been demonstrated to play anti-cancer and cytotoxic functions [@bib13], [@bib108], [@bib109], [@bib110], [@bib111], [@bib112], [@bib113], [@bib114], [@bib115], [@bib116], [@bib117]. Yet, the exact molecular mechanisms underlying the anti-cancer effects of these phytoconstituents are not fully known, and future studies are needed to elucidate the detailed mechanisms of action that mediate the anti-cancer effects of *N. sativa* phytoconstituents. 9. Conclusions {#sec9} ============== *N. sativa* is among the most commonly used herb in the history of mankind. *N. sativa* is considered by many to be a "miracle" herb due to its effective therapeutic potential to alleviate signs and symptoms of many diseases including cancer. The anti-cancer properties of *N. sativa* have been mainly attributed to its ability to exert potent anti-proliferative, pro-apoptotic, anti-oxidant, anti-mutagenic, and anti-metastatic roles. The protective effects of *N. sativa* against tumor initiation and progression have also been attributed, at least in part, to their ability to suppress inflammation and exert immune-boosting effects. Enhancement of NK cytotoxic activity against cancer cells and regulation of signaling pathways, such as iNOS, p53, and caspases, mediate the potential of *N. sativa* to subdue tumorigenesis and cancer. *In vitro* and *in vivo* experimental findings suggest that *N. sativa* extracts can potentially be employed in the development of effective therapeutic agents that can be employed in the regulation of various stages of tumorigenesis and treatment of many types of cancer. Further studies are definitely needed to shed more light on the molecular and cellular mechanisms underlying the anti-cancer effects of *N. sativa*. Such research endeavors will hopefully elucidate the exact signaling pathways implicated in the suppressive role that *N. sativa* extracts play in tumorigenesis and cancer. Moreover, although the preclinical, experimental evidence suggesting potent anti-cancer effects of various *N. sativa* extracts is compelling, preventive and clinical studies that directly point to the anti-cancer potential of *N. sativa* extracts are still lacking. Future studies should focus on establishing a direct link between the reported anti-cancer effects of *N. sativa* extracts and cancer prevention/treatment in preclinical and clinical settings. Sources of support {#sec10} ================== None. Conflicts of interest {#sec11} ===================== There are no conflicts of interest. Peer review under responsibility of Transdisciplinary University, Bangalore. {#fig1} ###### A brief summary of the reported *in vitro* and *in vivo* anti-cancer activities of *N. sativa*. Table 1 Activity *N. sativa* ---------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Anti-proliferative and pro-apoptotic effects •Stimulation of anti-proliferative effects on MCF-7 cells [@bib26].•Reduction in frequency of mammary papillary, comedo, and cribriform carcinoma in DMBA-induced carcinoma model [@bib28].•Reduction in serum levels of total sialic acid (TSA), lipid-bound sialic acid (LSA), prolactin, estradiol, progesterone, serum TNFα, tissue caspase-3 activity, and DNA fragmentation [@bib28].•Amelioration of diethylnitrosamine-induced hepatocarcinogenesis [@bib20].•Induction of late-stage apoptosis and/or necrosis as well as inhibition of both DNA synthesis and cell proliferation in HepG2 cells [@bib21], [@bib32].•Protection against diethylnitrosamine-induced hepatocellular adenoma [@bib23].•Reduction of serum AFP levels, relative liver weight, and activities of hexokinase, glyceraldehyde phosphate dehydrogenase, and G6P dehydrogenase [@bib31].•Inhibition of the two-stage initiation/promotion of skin carcinogenesis and delays the onset of skin papilloma [@bib25].•Reduction of methylcholanthrene (MCA)-induced soft tissue sarcomas [@bib25].•Reduction in formation of pre-neoplastic lesions for colon cancer [@bib27].•Delay in mortality of P815 mastocytoma bearing cells [@bib29]. Anti-oxidant and cytotoxic effects •Induction of cytotoxic effects against HepG2, MOLT4and LL/2 cells but no effects on normal cells [@bib33].•Induction of cytotoxic effects against MCF-7 cells [@bib26].•Induction of cytotoxic effects against EAC, DLA, S-180 cells [@bib24].•Induction of cytotoxic effects against SCL, SCL-6, SCL-37′6, NUGC-4 cells [@bib36].•Reduction in expression of MDA and NO [@bib34], [@bib41].•Reduction in lipid peroxides and NO levels [@bib28].•No effect on not affect the level or catalytic activity of aspartate-aminotransferase, alanine-aminotransferase, and gamma-glutamyltransferase [@bib35].•Amelioration of nephrotoxicity through reduction in creatinine and urea as well as elevation of GSH levels [@bib37].•Amelioration of anti-cancer drug-induced hepatic cytotoxicity [@bib38].•Amelioration of antibiotic-induced cytotoxicity in the thymus and spleen [@bib39]. Anti-mutagenic effects •Inhibition against MNNG mutagenicity [@bib43].•Enhancement of DNA replication and reduction in chromosomal aberrations [@bib43]. Anti-metastatic effects •Down regulation of t-PA, u-PA, and PAI-1 [@bib44].•Inhibition of liver metastasis [@bib29]. Effects on NK cytotoxic activity •Enhancement of helper to suppressor T cell ratio and improvement of NK cytotoxic activity [@bib46].•Improvement and increase in cell numbers [@bib48].•Enhanced killing of YAC-1 tumor cells [@bib50].•Increase in production of IFNγ and TNFα [@bib49].•Increase in production and activity of granzyme A and N-acetyl-β-[d]{.smallcaps}-glucosaminidase [@bib49].•Suppression of NK cytotoxic activity in CMV-infected mice [@bib52].

INTRODUCTION {#s1} ============ T-cell acute lymphoblastic leukemia (T-ALL), a neoplasm of lymphoblasts committed to the T-cell lineage involving the bone marrow and blood, is observed in ∼15% and 25% of childhood and adult ALL, respectively ([@MCS003533PETC20]; [@MCS003533PETC1]). In general, childhood T-ALL is a clinically aggressive neoplasm that requires intensive chemotherapy regimens and is associated with a higher risk of induction failure and early relapse when compared to childhood B-ALL ([@MCS003533PETC1]). Although recurrent genetic abnormalities in many hematologic neoplasms add diagnostic, prognostic, and therapy-related value, the clinical significance of the known recurrent genetic abnormalities observed in T-ALL are mostly unclear or controversial ([@MCS003533PETC1]; [@MCS003533PETC19]). Interestingly, the detection of *ABL1* gene rearrangements, which are often associated with *BCR/ABL1*-positive B-ALL and *BCR-ABL1*-like (Ph-like) B-ALL ([@MCS003533PETC14]; [@MCS003533PETC18]), have also been described in T-ALL---namely, *NUP214*/*ABL1*---and may be amenable to tyrosine kinase inhibitor (TKI) therapy ([@MCS003533PETC9]; [@MCS003533PETC2]; [@MCS003533PETC15]; [@MCS003533PETC6]; [@MCS003533PETC10]; [@MCS003533PETC11]; [@MCS003533PETC4]; [@MCS003533PETC5]; [@MCS003533PETC8]; [@MCS003533PETC16]; [@MCS003533PETC3]; [@MCS003533PETC13]). Considering the limited treatment options and overall unfavorable prognosis of T-ALL, detecting genetic abnormalities that may respond to targeted therapy is critical. Herein, we describe the detection of a cryptic *NUP214*/*ABL1* gene fusion in a newly diagnosed case of pediatric T-ALL by mate-pair sequencing (MPseq), a unique next-generation sequencing (NGS)-based technology that can detect both chromosomal structural and copy number abnormalities with significantly increased resolution and precision when compared to conventional chromosome and fluorescence in situ hybridization (FISH) methodologies ([@MCS003533PETC12]; [@MCS003533PETC17]). RESULTS {#s2} ======= Clinical Presentation {#s2a} --------------------- The patient is a 4-yr-old male with a prior diagnosis of autism spectrum disorder who presented with a 2-wk history of extreme fatigue and significant bruising. On exam he had pallor, splenomegaly, and hepatomegaly, with the splenic tip palpated just below the umbilicus, and the hepatic border palpated at the costal edge. In addition, he had several subcentimeter cervical and inguinal lymph nodes. Petechiae were noted overlying the anterior upper legs and beneath both eyelids, and large bruises were noted over the legs bilaterally and over the midthoracic vertebrae. Initial laboratory results revealed leukocytosis, anemia, and thrombocytopenia. His presenting pertinent labs were as follows: white blood cell count 480 K/µL, hemoglobin 4.8 g/dL, platelets 23 K/µL, blasts 83%, absolute blast count 399 K/µL, absolute neutrophil count 0 K/µL, lactate dehydrogenase 8574 U/L, uric acid 6.6 mg/dL, potassium 4.0 mEq/L, phosphorus 4.2 mg/dL, ionized calcium 1.45 mg/dL, INR 1.7, and PT 20.3. A chest X ray obtained at presentation revealed no mediastinal mass. The patient was admitted to the pediatric intensive care unit for management of anticipated tumor lysis syndrome given the significant leukocytosis. Peripheral blood was sent for morphologic evaluation and flow cytometry studies. Normocytic normochromic anemia with mild anisopoikilocytosis was observed with absence of rouleaux and nucleated red blood cells. Marked leukocytosis with predominantly lymphoblasts and rare mature lymphocytes were observed with virtually absent neutrophils, monocytes, and platelets ([Fig. 1](#MCS003533PETF1){ref-type="fig"}). Flow cytometry of peripheral blood showed blasts represented 95% of total leukocytes and expressed the following antigens: CD1a (partial dim), CD2, cCD3, double positive for CD4 and CD8, CD5, CD7, CD10, CD34 (partial dim), CD56, and TdT ([Fig. 2](#MCS003533PETF2){ref-type="fig"}). Blasts were negative for CD19, CD22, and MPO. Based on morphologic and flow cytometry results, a diagnosis of T-ALL was rendered. Cerebral spinal fluid was sampled and found to be negative for malignant cells, classifying the patient as CNS1. Of note, he did not receive corticosteroids within 4 wk prior to his diagnostic lumbar puncture. In addition, he did not have testicular disease appreciated on exam. A bone marrow biopsy and aspirate were obtained for morphologic assessment, chromosome studies, and FISH studies. The bone marrow aspirate and biopsy were hypercellular (95%--100%) and predominantly composed of lymphoblasts (90%--95% of marrow cellularity). Myeloid and erythroid lineages and megakaryocytes were all markedly decreased. {#MCS003533PETF1} {#MCS003533PETF2} Genomic Analyses {#s2b} ---------------- All genomic studies were performed on the submitted diagnostic bone marrow aspirate specimen. Conventional chromosome studies indicated 20 normal metaphases (46,XY) from an unstimulated culture ([Fig. 3](#MCS003533PETF3){ref-type="fig"}A), whereas T-ALL panel FISH studies revealed both complete and partial *CDKN2A* deletions in the majority of nuclei (96.5% of 200 interphase cells) ([Fig. 3](#MCS003533PETF3){ref-type="fig"}B) and what appeared to be a partial *ABL1* deletion (89.4% of 500 interphase cells; observed using a *BCR/ABL1* D-FISH probe set) ([Fig. 3](#MCS003533PETF3){ref-type="fig"}C). Importantly, the remaining T-ALL and TKI responsive FISH probes were negative (see Methods for a complete list of FISH probes), including the *ABL1* BAP probe set ([Fig. 3](#MCS003533PETF3){ref-type="fig"}D). {#MCS003533PETF3} To further characterize the abnormal clone, MPseq was performed and revealed complex rearrangements involving both Chromosome 9 homologs and a 10.1-Mb duplication on 10q ([Fig. 4](#MCS003533PETF4){ref-type="fig"}A). A 3.8-Mb deletion of 9p (Chr 9:20,866,257-24,692,658) was observed on one Chromosome 9 homolog, resulting in loss of multiple genes including *CDKN2A* and *CDKN2B* ([Fig. 4](#MCS003533PETF4){ref-type="fig"}B). In addition, a pericentric inversion of the other Chromosome 9 homolog was observed that resulted in deletions at the breakpoint sites (9p21.3 and 9q21.12), including an ∼583-kb deletion (Chr 9:21,805,717-22,389,662) that also spanned *CDKNA* and *CDKN2B* ([Fig. 4](#MCS003533PETF4){ref-type="fig"}B). These results were consistent with homozygous *CDKN2A* deletions observed by FISH analysis. Lastly, a "cryptic" t(9;[9]{.ul})(q34.13;q34.12) was observed with breakpoints located in intron 31 of *NUP214* (NM_001318324.1) and intron 1 of *ABL1* (NM_007313.2) and is predicted to create an in-frame chimeric gene consisting of exons 1--31 of *NUP214* and exons 2--11 of *ABL1* ([Fig. 5](#MCS003533PETF5){ref-type="fig"}). Sanger sequencing confirmed the translocation breakpoints indicated by MPseq ([Table 1](#MCS003533PETTB1){ref-type="table"}). {#MCS003533PETF4} {#MCS003533PETF5} ###### Mate-pair sequencing (MPseq) and Sanger sequencing results for the *NUP214/ABL1* gene fusion ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Gene 1 Gene 2 Position 1 Position 2 Transcript 1 Intron number 1 Transcript 2 Intron number 2 Fusion junction sequence^a^ Frame-shift class -------- ---------- ------------------- ------------------- -------------- ----------------- ---------------- ----------------- --------------------------------- ------------------- *ABL1* *NUP214* Chr 9:130,835,718 Chr 9:131,215,806 NM_007313.2 1 NM_001318324.1 31 **CAGTTCTCTTATATTCTGTCTCTC**\ In-frame **TTTCTTTCTCTCTGTG**cgaatttttt\ CCCAAAAGTTGAAGCCTTT\ AATATTTCTCACCTCTTACAT ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ^a^Bold, uppercase sequence maps to gene 1. Sequence in lowercase is found at the breakpoint but does not map to either chromosome. Nonbolded uppercase sequence maps to gene 2. Treatment Outcomes {#s2c} ------------------ The patient was initially treated with methylprednisolone for 72 h for gentle leukoreduction followed by systemic chemotherapy as per the Children\'s Oncology Group (COG) AALL1231 trial protocol. As the family declined to enroll in the trial, the patient was treated off study as per the standard therapy arm. Induction chemotherapy consisted of the standard four-drug regimen of vincristine, dexamethasone, daunorubicin, and pegaspargase, and the patient also received standard CNS prophylaxis with intrathecal cytarabine and methotrexate. Peripheral blasts were not cleared from circulation until day 14 of induction therapy. During induction, he experienced multiple complications, including steroid-induced hypertension requiring several antihypertensives, vincristine-induced syndrome of inappropriate antidiuretic hormone secretion, sepsis due to methicillin-sensitive *Staphylococcus aureus*, and a large saddle embolus. Despite these complications, therapy was not withheld. At the end of induction, a bone marrow study revealed measurable minimal residual disease (MRD) of 0.006%. Per the AALL1231 protocol, patients with who MRD \< 0.01% at the end of induction that are CNS1 with no testicular involvement and who did not receive steroid pretreatment prior to the diagnostic lumbar puncture are stratified into the standard risk group. Consolidation therapy consisted of cyclophosphamide, cytarabine, mercaptopurine, vincristine, and pegaspargase. The patient tolerated consolidation therapy well and achieved remission as evidenced by an end of consolidation MRD of 0%. Since obtaining remission, this patient has completed the interim maintenance and delayed intensification phases of therapy as per the nonrandomized, standard risk arm of the COG AALL1231 protocol. The patient is currently in his second phase of maintenance therapy and is doing well clinically with no signs of relapse. DISCUSSION {#s3} ========== Following conventional chromosome and T-ALL and TKI panel FISH studies that yielded little prognostic and treatment-related information, MPseq was performed to further elucidate what appeared to represent a partial *ABL1* deletion by FISH analysis using the *BCR*/*ABL1* D-FISH probe set. Rather than confirming a partial or complete deletion as suggested by the *ABL1* D-FISH probe footprint, a *NUP214*/*ABL1* gene fusion resulting from a "cryptic" t(9;[9]{.ul})(q34.13;q34.12) was revealed by MPseq and subsequently confirmed by Sanger sequencing. This discrepant result was resolved after the translocation breakpoints by MPseq were compared to the *ABL1* D-FISH footprint. Although a segment of the disrupted *ABL1* D-FISH probe on der(9) resulted in a diminished (dim) *ABL1* signal, the intact *ABL1* D-FISH probe on der([9]{.ul}) was enhanced via juxtaposition of the translocated portion of the *ABL1* D-FISH probe from the der(9) ([Fig. 6](#MCS003533PETF6){ref-type="fig"}). ![A focused view of the *ABL1* and *NUP214* gene regions on the derivative copies of Chromosomes 9. Horizontal dashed red lines indicate the breakpoints on each derivative Chromosome 9 and disruption of each FISH probe footprint. The solid red vertical bars indicate the *ABL1* FISH probe footprint from the *BCR/ABL1* D-FISH probe set (*ABL1* D-FISH). The close proximity of *ABL1* and *NUP214* on each Chromosome 9 resulted in a diminished *ABL1* signal on the der(9) homolog while enhancing the *ABL1* FISH probe signal intensity on the der([9]{.ul}) homolog. The striped vertical green bars indicate the 5′*ABL1* FISH probe footprint from the *ABL1* BAP (5′*ABL1* BAP) and the striped vertical red bars indicate the 3′*ABL1* FISH probe footprint from the *ABL1* BAP (3′*ABL1* BAP). Although the *ABL1* gene was rearranged on the der(9) homolog, the 3′*ABL1* BAP spanning the *NUP214* rearrangement from the der([9]{.ul}) homolog created a fusion signal on the der(9), thus "masking" the *ABL1* rearrangement.](MCS003533Pet_F6){#MCS003533PETF6} Similarly, because the *ABL1* BAP generated two apparently normal, intact fusion signals, the *ABL1* BAP FISH footprints were compared to the *NUP214*/*ABL1* breakpoints. Although the *ABL1* gene on the der(9) was disrupted because of the translocation event and should be associated with a disruption of the *ABL1* BAP, the MPseq results indicate a fusion signal was ultimately "re-created" on the der(9) because of retention of the 5′*ABL1* BAP from the der(9) with juxtaposition of the translocated portion of the 3′*ABL1* BAP from the der([9]{.ul}), thus "masking" the *ABL1* rearrangement ([Fig. 6](#MCS003533PETF6){ref-type="fig"}). The initial break on the der([9]{.ul}) disrupts the *NUP214* gene rather than the *ABL1* gene and is not associated with a true disruption of the BAP *ABL1* probe ([Fig. 6](#MCS003533PETF6){ref-type="fig"}). However, the der([9]{.ul}) is ultimately the location of the *NUP214/ABL1* fusion that is generated by the "cryptic" t(9;[9]{.ul}) event. Importantly, amplification of the *NUP214/ABL1* fusion gene appears to be required for oncogenicity and may be a late event occurring in only a subpopulation of cells ([@MCS003533PETC10]). Although amplification of the *NUP214/ABL1* fusion gene was not observed in our case, follow-up testing to detect clonal evolution would be of clinical importance. In addition to the *NUP214*/*ABL1* rearrangement, MPseq identified heterozygous and homozygous 9p deletions, a pericentric inv(9) and a partial gain of 10q. Taken together, of the multiple structural and numerical abnormalities identified by MPseq, FISH only identified complete and partial *CDKN2A* deletions, whereas chromosome studies were normal. MPseq is a unique NGS-based technology that provides significantly higher resolution and precision when compared to chromosome and FISH methodologies, often characterizing breakpoints to the exact intron or exon. Library preparation involves circularizing long DNA fragments (∼2--5 kb), followed by paired end sequencing of the mate-pair fragments. Rather than direct sequencing, long insert sequences are inferred from short paired end reads, thus allowing for the robust detection of structural abnormalities without extensive sequencing. Importantly, MPseq leverages inferred sequence in between reads to achieve "bridged coverage"; thus the amount of sequencing necessary to gather evidence for structural abnormalities and the resulting cost are less compared to traditional paired end sequencing across the entire genome. Utilization of bridge coverage also allows for a breakpoint resolution smaller than fragment size, typically within a 1-kb window (or smaller, depending on the coverage of the study). However, MPseq also has the potential to capture "split reads," wherein a single sequence read, rather than a set of paired reads, contains sequences from both sides of the rearrangement. These split reads identify the breakpoint down to the base pair level, essentially providing the maximum level of resolution without requiring follow-up Sanger sequencing. The ability to detect both structural and copy number abnormalities with high resolution and precision significantly exceeds the detection capabilities of FISH, as we have demonstrated in this case. Last, the ability to detect structural rearrangements of clinical significance will continue to grow as disease-specific and potentially targetable abnormalities are discovered. For example, the list of kinase genes that respond to TKI therapy in Ph-like B-ALL continues to grow ([@MCS003533PETC14]; [@MCS003533PETC18]), and it will become infeasible or cost-effective to continue the development of FISH probes or other methodologies targeting single-gene regions for application to the rapidly growing list of gene rearrangements. In summary, we have demonstrated the clinical utility of MPseq in a newly diagnosed case of pediatric T-ALL with a cryptic *NUP214*/*ABL1* fusion. The detection of the *NUP214*/*ABL1* gene fusion by MPseq, in addition to other structural and copy number abnormalities, were undetected by chromosome studies and two standard, clinical FISH probe sets with *ABL1* coverage, thus illustrating the power of this novel NGS-based technology. Although this patient has not yet been treated with TKI therapy, the ability of MPseq to reveal a genetic abnormality of potential clinical significance that was missed by standard cytogenetic evaluation has been demonstrated, and the *NUP214/ABL1* fusion could provide an alternative treatment target if this patient experiences a T-ALL relapse. METHODS {#s4} ======= Conventional Chromosome Analysis {#s4a} -------------------------------- Cells from the bone marrow aspirate specimen were cultured, harvested, and banded utilizing standard cytogenetic techniques according to specimen-specific protocols. Twenty metaphases were analyzed by two qualified clinical cytogenetic technologists and interpreted by a board-certified (American Board of Medical Genetics and Genomics \[ABMGG\]) clinical cytogeneticist. Fluorescence In Situ Hybridization (FISH) {#s4b} ----------------------------------------- The T-ALL and TKI-responsive FISH panels were performed on interphase nuclei including locus-specific (*CDKN2A*/D9Z1 \[Abbott Molecular\], *TP53*/D17Z1 \[Abbott Molecular\]), break-apart (*KMT2A*(*MLL*) \[Abbott Molecular\], *TRB* \[laboratory-developed test, LDT\], *TRD* \[LDT\], *TAL1/STIL* \[LDT\], *ABL1* \[LDT\], *PDGFRβ* \[LDT\], *JAK2* \[LDT\], *ETV6* \[LDT\], *NUP98* \[LDT\]), and D-FISH (*BCR/ABL1* to detect *ABL1* amplification \[Abbott Molecular\], *TLX3/BCL11B* \[LDT\], *MLLT10/PICALM* \[LDT\]) probe sets. The bone marrow aspirate specimen was subjected to standard FISH pretreatment, hybridization, and fluorescence microscopy according to specimen-specific protocols. FISH analysis was performed by two qualified clinical cytogenetic technologists and interpreted by a board-certified (ABMGG) clinical cytogeneticist. Mate-Pair Sequencing (MPseq) {#s4c} ---------------------------- DNA was extracted from a fixed cell pellet and 1 µg of DNA was utilized for mate-pair sequencing library preparation and was processed using the Illumina Nextera Mate Pair library kit (Illumina). Library preparation consisted of tagmentation to simultaneously shear and biotinylate the genomic DNA, strand displacement to fill any gaps left by the tagmentation step, and overnight circularization (16--20 h) to produce stabile 2- to 5-kb DNA fragments. AMPure purification (Beckman Coulter) was performed after the tagmentation and strand displacement steps (0.56× and 0.4×, respectively) to ensure only the longest fragments are selected to complete library preparation. After overnight circularization, noncircularized DNA was digested with exonuclease prior to mechanical shearing of the circularized fragments with a Covaris LE220 System (Covaris). The resulting biotinylated DNA fragments were bound to Dynabeads M-280 Streptavidin (Thermo Fisher Scientific) and subsequently processed through end repair, A-tailing, ligation of 7-bp Illumina adapters (a component of the TruSeq DNA library prep kit), and PCR using the PCR Primer Cocktail (Illumina) and KAPA HiFi HotStart Ready Mix PCR Kit (KAPA Biosystems). A 0.67× AMPure purification was performed to complete library preparation. MPseq libraries were multiplexed at two samples per lane to be sequenced on the Illumina HiSeq 2500 in rapid run mode. On both ends of each mate-pair fragment, 101 base pairs were sequenced to a bridged coverage of 43× and a base coverage of approximately 6×. Data were aligned to the reference genome (GRCh38) using BIMAv3, and abnormalities were identified and visualized using SVAtools, an in-house developed bioinformatics tool. Additional information regarding MPseq technology and bioinformatics tools have been previously described ([@MCS003533PETC7]; [@MCS003533PETC12]; [@MCS003533PETC17]). Sanger Sequencing {#s4d} ----------------- Reference DNA sequences spanning the minimal 5′ and maximal 3′ positions of MPseq approximate breakpoints were used for primer design using Primer3Plus. End-point PCR was performed on patient DNA with a 50% 2× Paq5000 Hotstart PCR Master Mix (Agilent) using touchdown PCR. Results were visualized on a 2% agarose gel in a UV light box, and amplicon sizes were estimated. Selected amplicons were purified using Exo-SAP-IT PCR Product Cleanup Reagent (Thermo Fisher Scientific) and Sanger sequencing was performed on a 3730*xl* DNA Analyzer (Thermo Fisher Scientific). The resulting sequences were analyzed using Sequencher DNA Sequence Analysis Software (Gene Codes Corporation) and mapped to the GRCh38 genome using the BLAT function in the UCSC genome browser to determine precise breakpoints in this rearrangement. ADDITIONAL INFORMATION {#s5} ====================== Data Deposition and Access {#s5a} -------------------------- This variant was submitted to ClinVar (<http://www.ncbi.nlm.nih.gov/clinvar>) and can be found under accession number SCV000882548. Ethics Statement {#s5b} ---------------- The results described in this manuscript were derived from clinical testing rather than research testing. Therefore, consent was not required for testing, and because patient identifiers were removed for the purposes of this manuscript, consent (oral or written) was not obtained. Mayo Clinic does not require IRB approval for the publication of single case reports. Author Contributions {#s5c} -------------------- J.F.P. drafted and edited the manuscript. B.A.P., S.A.S., S.J.K., M.R.W., R.P.K., P.T.G., and N.L.H. analyzed the data and reviewed and edited the manuscript. J.B.S., S.H.J., G.V., and L.B.B. reviewed and edited the manuscript. M.M. and M.G.B. collected clinical and pathologic data and reviewed and edited the manuscript. R.A.R. participated in manuscript writing and reviewed and edited the manuscript. Competing Interest Statement {#s5d} ---------------------------- The authors have declared no competing interest. Funding {#s5e} ------- Funding for this project was provided via internal Mayo Clinic funds; no external/National Institutes of Health funding was used.

INTRODUCTION {#S0001} ============ Anthracosis (*anthrac*- meaning coal, carbon + -*osis* meaning condition) is defined in Bioline as, "the asymptomatic, milder type of pneumoconiosis as caused by the accumulation of carbon in the lungs due to repeated exposure to air pollution or inhalation of smoke or coal dust particles" ([@CIT0001]). Anthracosis may be seen as a superficial black discoloration (simple anthracosis) ([Figure 1A](#F0001){ref-type="fig"}) or scattered foci of black spots, which retract mucosa inward due to the effect of adjacent anthracotic lymphadenopathy ([Figure 1B](#F0001){ref-type="fig"}). Anthracosis is an ancient disease discovered in mummies ([@CIT0002]--[@CIT0004]). The early scientific reports of this disease were mainly from Western countries and the term "anthracosis" was coined by Pearson in 1813 ([@CIT0005]). Pearson and others believed that anthracosis was a complication of coal worker pneumoconiosis ([@CIT0006]). However, the interest of Western countries in this disease declined as the frequency of anthracosis declined in their countries. The second wave of anthracosis in the literature started in Asia, as it is still a problem in this continent. Most of these studies showed that pneumoconiosis and exposure to coal were not the most frequent risk factors and thus researchers excluded pneumoconiosis patients from the category of anthracosis ([@CIT0007], [@CIT0008]). Chung et al. ([@CIT0007]) introduced BAF as a unique clinical syndrome. It is the severe form of disease, which distorts and narrows the bronchial lumen ([Figure 1C](#F0001){ref-type="fig"}). Later, some new terms were introduced such as anthracostenosis ([@CIT0009]) or anthracotic bronchitis ([@CIT0010]) used to describe extensive deposition of carbon in the main bronchial walls; which in the majority of cases is accompanied by severe submucosal edema, bronchial stenosis, protruded mucosal folds and lung collapse ([@CIT0010]) ([Figure 1C](#F0001){ref-type="fig"}). {#F0001} Prevalence {#S20002} ---------- Prevalence of anthracosis in the general population has been roughly estimated because the exact diagnosis of anthracosis requires bronchoscopy, which is impossible for the general population due to ethical considerations. Available data from large series of patients who underwent bronchoscopy for other reasons have shown the frequency of simple anthracosis to be 3.4 - 21% ([@CIT0011], [@CIT0012]). This rate was 0.1-22.5 for BAF ([@CIT0013], [@CIT0014]) (cumulated means 5.7%) ([Table 1](#T0001){ref-type="table"}). Frequency of BAF is lower in Western countries and Wynn et al. reported seven BAF cases among 7000 bronchoscopies ([@CIT0015]). Reports from other continents such as North America or Africa were also scanty ([@CIT0014], [@CIT0016]) and reports of anthracosis in children are very rare ([@CIT0017]). As mentioned above, anthracosis was previously prevalent in coal workers, but new reports are now mostly reporting this disease in farmers (40%) ([@CIT0011])([Table 1](#T0001){ref-type="table"}) and rural dwellers (55-66%) ([@CIT0008], [@CIT0010], [@CIT0018]). The number of affected females in some large series has been equal to males ([@CIT0008], [@CIT0010]), but accumulation of data show that BAF in females is slightly more prevalent than males ([Table 1](#T0001){ref-type="table"}). Moreover, almost all studies have shown that anthracosis subjects are elderly ([Table 1](#T0001){ref-type="table"}). In a meta-analysis, the mean age of patients affected with anthracosis was 63±3.8 years, which shows that these patients are significantly older than non-anthracosis subjects (52±6.4 years) (t= 3.43, P = 0.02) ([@CIT0019]). ###### Patient demographics and clinical manifestation of lung anthracosis Studies Anthracosis Number Frequency Mean age or median (range) \% Non-smokers Female/male ratio Biomass exposure Occupation Female Occupation Male Dyspnea Cough ---------------------------------------- -------------------- --------------------------------------- ---------------------------- ----------------- ------------------- ------------------ ------------------- ----------------- --------------- ---------------- **Chung et al 1998**^[@CIT0007]^ 28 3% 64 (42-86) 86% 2.5/1 0 Not miner Not miner 60% 71% **Amoli 1998**^[@CIT0026]^ 10 1.1% 62.5 (46-72) 100% 1/0 100% Housewives \-\-- 90% 70% **Amoli 2009**^[@CIT0046]^ 102 NA 60 (29-77) 73% 1.4/1 Most Housewives Workers Most Most **Kim et al2009**^[@CIT0008]^ 333 11% 72.3 (47-90) 78.1% 2.1/1 100% Housewives Farmers 38.4% 30% **Ghanei et al 2011**^[@CIT0035]^ 71 7.7% 68.2±10.7 NA 1.2/1 27% Not miner Not miner 72% 80% **Wynn et al 2008**^[@CIT0015]^ 7 0.1%[†](#TF0001){ref-type="table-fn"} 72.7±6.4 (67-82) 42% 2/5 NA Worker Miner 100% 85% **Najafizadeh et al 2003**^[@CIT0010]^ 47 16% 70 (51-82) 89.6% 1/1 29.3% NA NA NA 93% **Hwang et al2010**^[@CIT0014]^ 10 16.7% 69±9.8 100% 4/1 0% NA NA NA NA **Towhidi et al 2003**^[@CIT0058]^ 96 8% 71 (52-90) 72% 2.2/1 0 NA NA 51% 100% **No et al 2003**^[@CIT0075]^ 166 12% 72.4 (56-91) 77% 6.2/1 25% NA NA 50% 48% **Na et al 2002**^[@CIT0037]^ 30 3.4% 71 (53-88) 66% 5/1 NA Not miner Not miner 36% 50% **Park et al2008**^[@CIT0048]^ 49 NA 76 (56-90) 76% 3.7/1 NA Not miner Not miner NA NA **Kim et al 2004**^[@CIT0070]^ 37 3% 67.5±8.2 (21-97) 75% 5.1/1 83% NA NA 93% 40% **Kim et al 2000**^[@CIT0018]^ 54 NA 67 (33-78) 85% 2.3/1 NA Not miner Two miner NA NA **Mirsadraee et al 2005**^[@CIT0012]^ 63 11.7% 60±16 81% 1.1/1 NA Housewives Farmers 71% 100% **Jang et al 2007**^[@CIT0073]^ 54 NA 75 (50-99) 80% 1.8/1 NA NA NA 57% 44% **Mirsadraee et al**^[@CIT0033]^ 70 NA 69.3 ± 9.4 79% 1.5/1 40% Housewives Farmers 81% 68% **Törün et al 2007**^[@CIT0027]^ 27 NA 66.8 (53-77) 100% 12.5/1 100% Housewives Farmers 63% 100% **Sigari et al 2009**^[@CIT0011]^ 778 5.4% 63 (25-80) NA 1/1 NA Housewives Farmers 35.7% 83.6% **Hemmati et al 2008**^[@CIT0072]^ 34 16% 61.8 NA 1.2/1 52% NA NA 47% 73% **Najafizadeh et al 2008**^[@CIT0039]^ 87 NA 69 ± 13 (36-84) 88.5% 1.9/1 29.9% NA NA NA NA **Rezaei Talab 2007**^[@CIT0013]^ 225 22.5% 65 ± 12.5 NA 1.3/1 7% Housewives Farmers NA NA **Mirsadraee et al 2011**^[@CIT0038]^ 54 NA 70.5 ± 10.4 82% 1/1 39% housewives Farmers 66% 74% **Cumulated frequency (Average)** 2457 5.7% (0.1-22.5) 67.9 ±10.1 (21-99) 77.6% (42-100%) 1.7/1 61.5% (7-100%) 64.2% (35-93) 98.5% (30-100) This frequency was omitted from cumulated frequency because of non-homogeneity with other studies. Proposed etiological factors {#S20003} ---------------------------- ### Dust The exact reason of anthracosis and the origin of anthracotic nodule have yet to be discovered. Dust exposure, especially coal dust in anthracosis subjects was reported during 1960-1980 in Europe ([@CIT0005], [@CIT0006]). Later on, Wynn et al. ([@CIT0015]) reported seven subjects who were exposed to coal and tile dust in their work place. Recent studies in Europe have shown deposition of mica and silica crystals in calcified mediastinal lymph nodes of anthracotic patients using transmission electron microscopy ([@CIT0020], [@CIT0021]). Hwang et al. ([@CIT0014]) studied 10 BAF subjects from Asia and the Indian subcontinent. They reported that aluminum silicates and quartz were present in 37-70% and 2-7% of subjects, respectively. In a study in Japan, elemental constitution of lobectomized lung tissue obtained by autopsy was analyzed by a wavelength-dispersive X-ray fluorescence spectrometer ([@CIT0022]); the most important non-carbonaceous fraction of intrapulmonary particulate pollutant was silicon and aluminum, especially in farmers compared to other occupational categories. Electron microscopic evaluations of the lung tissue from mummies also showed silica, aluminum, and iron deposits ([@CIT0003]). Although the effect of coal on BAF subjects was again reported in Eastern countries ([@CIT0023]); heavy exposure to dust such as in miners, stone breakers and well diggers' was reported in only 3% of BAF subjects ([@CIT0011]). Therefore, the pathogenesis of these crystals causing anthracosis should be different from that of routine air pollution and occupational exposure, and according to a recent widespread investigation ([@CIT0024]) and in mummies ([@CIT0004]), concurrent exposure to carbon smoke from the combustion of fuels and inorganic compounds that contain limestone and alumina-silicates is the most possible cause for anthracosis. ### Smoke Biomass smoke has been mostly reported as a risk factor for anthracosis in Asian countries such as Korea ([@CIT0008]), India ([@CIT0025]), Iran ([@CIT0026]) and Turkey ([@CIT0027]). Some reports from Africa ([@CIT0028]) and Latin America ([@CIT0029], [@CIT0030]) have indicated chronic respiratory disease induced by biomass smoke exposure. The hypothesis about biomass as a causative factor for anthracosis is due to the resemblance of anthracotic pigments to carbon particles and the fact that most of the anthracotic subjects (77.6%) are non-smokers ([Table 1](#T0001){ref-type="table"}). Biomass is produced by burning of wood, leaves or dung (manure) of farm animals for heating, cooking or baking ([@CIT0008]). Depending on the geographical and socioeconomic status, exposure to biomass was reported in 7-100% of BAF subjects, but the cumulated incidence rate was 61.5% ([Table 1](#T0001){ref-type="table"}). The mean duration of biomass exposure was reported to be 36 years ([@CIT0012]--[@CIT0060]) ([@CIT0008]) and duration (years) of smoke exposure showed significant association with anthracosis (OR: 1.05, 95%CI: 1.01-1.09) ([@CIT0031]). Amoli ([@CIT0026]) described using rustic traditional wood ovens requiring the baker to put his/head into the oven to place or remove the bread. In other studies, indoor exposure to wood smoke similar to bread baking in traditional ovens increased the risk of BAF by 4.3 to 4.8 folds especially in women ([@CIT0032], [@CIT0033]). Furthermore, some investigators have attempted to explain the pathophysiology of anthracosis, including bronchial narrowing, lymph node enlargement with or without calcification and susceptibility to TB and malignancy on the basis of biomass smoke inhalation ([@CIT0008]). However, there were some anthracotic subjects who had used other fuels such as kerosene or gas; these cases cannot be explained by this hypothesis. Cigarette smoking is not a risk factor for anthracosis and its frequency was significantly lower in anthracotic subjects compared to COPD subjects who underwent bronchoscopy ([Table 1](#T0001){ref-type="table"}). Moreover, anthracotic plaque and bronchial deformity were not observed in typical COPD subjects (who had a history of cigarette smoking); therefore, in case of detecting anthracosis during bronchoscopy, it should not be considered as a variant of COPD. Increased motility of bronchial cilia was postulated to be the cause of low prevalence of anthracosis in cigarette smokers ([@CIT0024]). ### Tuberculosis Three decades ago, presence of TB in BAF was shown by Chung et al ([@CIT0007]). Thereafter, several studies have reported the association of anthracosis with TB ([Table 2](#T0002){ref-type="table"}). In a meta-analysis, which reviewed all studies on the association of TB and anthracosis, it was shown that the cumulated incidence of TB in anthracosis and BAF subjects who underwent routine bronchoscopy was 16.6% (95% CI = 8.5-31) and 32.3% (95% CI = 21-57) respectively ([@CIT0019]). The risk of TB increased in anthracosis with a cumulated odds ratio of 3.16 (95% CI= 2.49-6.85), which was significantly higher than that in the control group ([Figure 2](#F0002){ref-type="fig"}) ([@CIT0019]). History of close contact with subjects suffering from TB and PPD skin test more than 10 mm was reported to be useful for diagnosing TB in BAF subjects suspicious for TB ([@CIT0034]). It is noteworthy that 38% of BAF subjects who suffered from associated TB reported close contact with TB patients ([@CIT0034]). PPD reaction more than 10 mm was reported in 63% of TB-proven BAF subjects, while this finding was shown in 34% of BAF non-TB subjects ([@CIT0035]). Serological markers for the activity of TB, such as IL-2 sRα, IFN-γ and TBGL antibody, were evaluated in BAF cases, but they were not useful for evaluation of TB activity in patients with anthracofibrosis ([@CIT0036]). ![Forest plot for evaluating the association and risk of tuberculosis in anthracosis and BAF subjects ([@CIT0019]).](Tanaffos-13-001-g002){#F0002} ###### Summary of diagnostic procedures in subjects with anthracosis of the lungs Studies BAF Number CT findings Bronchoscopy Associated disease --------------------------------------------------------------- ------------ --------------- -------------- -------------------- ------- ------- ----- ------- -------- ------ **Chung et al 1998**^[@CIT0007]^ 28 42% (28%) 85% 100% 75% 75% 64% 78% 60% 0 **Kim et al 2009**^[@CIT0008]^ 333 66% (44%) 63% 72% 90% 80% 62% 69% 34% 3.6% **Park et al 2008**^[@CIT0048]^ 49 91% (NA) 95% NA 98% 84% 40% 65% 20% NA **Wynn 2008**^[@CIT0015]^ 7 28% (28%) 28% 100% 71% 57% 57% 86% 0 0 **Kim et al 2000**^[@CIT0018]^ 54 94% (57%) 84% 94% 26% 63% 15% 52% 40% 0 **Hwang et al**^[@CIT0014]^[\*](#TF0002){ref-type="table-fn"} 10 70% (NA) 60% 90% 10% 80% 40% 30% 100% 0 **Najafizadeh et al 2003**^[@CIT0010]^ 47 NA (61%) 74% 62% 80% NA 63% NA 27.7% 3% **Lee et al 2002**^[@CIT0065]^ 43 89% (38%) 84% 76% NA NA NA NA 29% 7% **Towhidi et al 2003**^[@CIT0058]^ 96 NA NA 68% 82% 41% 31% 79% 30% 0 **No et al 2003**^[@CIT0075]^ 166 69% (45%) 55% 59% 90% 75% 60% 70% 21.7% 6.6% **Jang et al 2007**^[@CIT0073]^ 54 65% (53%) 68% 75% NA NA NA NA 27% 4% **Na et al 2002**^[@CIT0037]^ 30 58% (26%) 67% 96% 70% 76% NA 50% 37% 0 **Razi et al2007**^[@CIT0076]^ 51 5.9% (NA) NA 47% 47% 33% 23% 60% 25.4.% 0 **Törün et al 2007**^[@CIT0027]^ 27 29.6% (22.2%) NA 74% 92% 81% 74% 100% 25.9% 0 **Hemmati et al 2008**^[@CIT0072]^ 34 79% (NA) NA 64% 91% 52% NA NA 44% 0 **Najafizadeh et al 2008**^[@CIT0039]^ 87 NA 36% 29.8% 35% 29.8% 34% 48.2% 26% NA **Mirsadraee et al**^[@CIT0038]^ 70 88% (88%) 48% 97% 80% 35% 43% 60% 33% 0 **Cumulated frequency** 1186 55% (36%) 56.2% 50% 79.3% 60% 45% 64.7% 32.3% 3.4% BAF was collected from TB proven subjects. Parenchymal lesion included atelectasis, mass or alveolar type infiltration Two studies used polymerase chain reaction for evaluation of TB and the frequency of TB superimposed on BAF was reported to be 37% and 34% ([@CIT0037], [@CIT0038]), which were slightly higher than the result of traditional methods such as acid fast bacilli testing or culture (31%) ([@CIT0038]). Evaluation of space-oligonucleotide (spoligotyping) of *M. tuberculosis* by PCR showed that *M. tuberculosis* in anthracosis is not a special subtype (in comparison to the common subtype of their community) ([@CIT0039]). Other non-TB mycobacterial infections such as *Mycobacterium kansasii* were also reported to be associated with BAF ([@CIT0040]). The reason for TB susceptibility has with wood smoke on the immune system have been postulated ([@CIT0008], [@CIT0024]). ### Bronchogenic carcinoma Malignant lesions associated with anthracosis have been sparsely reported ([@CIT0041]--[@CIT0043]), but accumulation of data from studies reported the frequency of malignancy revealed lung cancer in 0-7% of subjects (cumulated frequency = 3.4%) ([Table 2](#T0002){ref-type="table"}). The reason for the variation in frequency of lung cancer associated with anthracosis has yet to be understood, but in one study by Ohshima ([@CIT0022]), subjects with lung cancer had a high level of iron, calcium, copper, lead, chromium and nickel in their lung tissue and lower levels of silicon and aluminum as the main mineral intrapulmonary particulate pollutant of routine anthracosis subjects ([@CIT0022]). Therefore, we can conclude than there is no epidemiological or etiological association between anthracosis and lung cancer. Pathology {#S20008} --------- Anthracosis involvement mainly starts from the respiratory bronchioles ([@CIT0044], [@CIT0045]). Histopathology of the lung tissue has shown carbon-like particles inside the cytoplasm of the macrophages in the bronchial wall ([@CIT0046]) ([Figure 3](#F0003){ref-type="fig"}) and free particles in the mediastinal lymph nodes ([@CIT0047]). Submucosal fibrosis may also be seen in the bronchial wall ([@CIT0037]) and the epithelial lining is usually intact ([@CIT0026]). In a study, lobectomy was done on two BAF subjects and fibrosis of the bronchi and reactive hyperplasia with anthracotic pigmentation were the major histopathological findings ([@CIT0048]). Perforation of anthracotic lymph nodes into the bronchial lumen may be the mechanism that produces anthracosis with retracted mucosa ([Figure 1B](#F0001){ref-type="fig"}) ([@CIT0018]). Bronchial cytology showed macrophage-containing anthracotic nodules in 71% of subjects ([@CIT0012]). Associated pathology such as TB or cancer is usually distinct from BAF histopathology ([@CIT0037], [@CIT0049]). These findings are completely distinct from the histopathological findings of COPD as the most important clinical differential diagnosis of BAF. {#F0003} Animal studies {#S20009} -------------- Anthracosis was discovered in the lung tissue of wild and domestic animals ([@CIT0047], [@CIT0050]). A study in cattle showed anthracosis in 3.8% of lung and lymph node tissues ([@CIT0047]). In some experimental studies, investigators were successful in inducing anthracosis in animal models ([@CIT0051]); moreover, the frequency of anthracosis in rats that were exposed to exhaust smoke was evaluated as well ([@CIT0052]). Anthracosis in other organs {#S20010} --------------------------- Anthracosis has also been reported in the liver, spleen ([@CIT0053]) and esophagus ([@CIT0054], [@CIT0055]). Anthracosis in the esophagus is important as it mimics the picture of malignant melanoma ([@CIT0056]). An interesting case of sinusitis was reported, where the subject suffered from anthracosis of the sinuses ([@CIT0057]). Clinical manifestations {#S20011} ----------------------- BAF usually presents with pictures very similar to COPD with the exception of a history of cigarette smoking. Cough and dyspnea are the most frequent symptoms of BAF and anthracosis in most reports ([Table 1](#T0001){ref-type="table"}). Association of TB does not change the chronic symptoms, but it may cause new onset weight loss or fever (25%) ([@CIT0058]) which should be differentiated from pneumonia, which is reported to be superimposed on 30% of BAF subjects ([@CIT0008]). Physical examination of the lungs usually shows wheezing ([@CIT0007]) and less frequently rales or decreased breath sounds ([@CIT0010]). Some anthracosis subjects had normal physical examinations, but the frequency has not been mentioned in the literature. Anthracosis may present with complications of enlarged mediastinal lymph nodes such as vocal cord paralysis ([@CIT0049]) or broncholithiasis ([@CIT0059]). Many of these cases were also reported in association with tuberculosis ([@CIT0049], [@CIT0060]). Pulmonary function tests {#S20012} ------------------------ Preliminary studies on BAF subjects showed a mild obstructive pattern (mean FEV1 83.9%±22.9% predicted) and class I and II GOLD classification ([@CIT0008]) ([Table 3](#T0003){ref-type="table"}). Generally, BAF subjects could be classified under obstructive lung disease, but some of them have shown normal or restrictive patterns ([Table 3](#T0003){ref-type="table"}). In a study on BAF subjects, post bronchodilator response was not significant and DLCO and DLCO/VA were mainly within the normal range ([@CIT0061]). Another study showed that the resistance of the airways had a significant correlation with the number of bronchial stenoses ([@CIT0062]). Therefore, we conclude that obstruction should be fixed in the bronchi and the respiratory unit should be intact. Statistical analysis did not show a correlation between the severity of clinical findings and spirometry ([@CIT0061]). ###### Review of results of spirometry in anthracosis subjects Studies BAF Number Classification of spirometry Mean of predicted value ---------------------------------- ------------ ------------------------------ ------- ------- ------------------------- ----------- ----------- ----------- -------- ---------- **Jung et al 2005**^[@CIT0062]^ 113 49.6% 8.8% 41.5% 94.3±24.7 84.1±25.2 69±13 41.1±21.4 119±39 94±13.3 **Mirsadraee et al**^[@CIT0059]^ 40 95% 5% 0 75.8±19.5 57.3±18.4 60.6±13.3 25.7±14 144±80 104±29.1 **Jang et al 2007**^[@CIT0073]^ 21 62% 5% 33% NA NA NA NA NA NA **No et al 2003**^[@CIT0075]^ 113 47.8% 12.4% 39.8% NA **Lee et al 2002**^[@CIT0065]^ 43 50% 0 50% NA **Kim et al 2009**^[@CIT0008]^ 151 NA NA NA 90.8±22 83.9±22.2 68.8±11.4 NA NA NA **Amoli 2009**^[@CIT0046]^ 39 62% 31% 7% **Ghanei 2011**^[@CIT0035]^ 71 46% 36.5% 17.5% **Cumulated frequency** **(Average)** 440 59% 14% 27% Included mixed pattern These findings were across the findings of Tanaka et al. ([@CIT0063]), who evaluated the peripheral bronchi with an ultrafine bronchoscope. They showed that anthracosis started from the small bronchi and then spread to the larger bronchi. For this reason, we conclude that when BAF is found in the proximal bronchi, all other distal bronchi should assume to be occluded by anthracosis. Therefore, anthracosis and BAF should be considered in the list of differential diagnoses of obstructive lung disease. Radiological findings {#S20013} --------------------- ### Chest X-ray Chest X-ray (CXR) was reported to be normal in only 7% of subjects ([@CIT0033]). The most frequent abnormalities reported in CXR were non-homogeneous pulmonary infiltrate ([Figure 4](#F0004){ref-type="fig"}), subsegmental atelectasis and mass lesions (40% and 16%, respectively in BAF subjects) ([@CIT0033]). CXR in some cases showed resolution of abnormalities when the anthracosis subjects were proven to suffer from TB and were treated with anti-TB medications ([@CIT0007], [@CIT0037], [@CIT0064]). {#F0004} Computed tomography {#S20015} ------------------- Computed tomography (CT) was more sensitive for anthracosis and showed more specific radiological findings ([Figure 5](#F0005){ref-type="fig"}). The earliest reports have shown mediastinal or hilar lymphadenopathy in 94% of cases, 57% of them were calcified (high attenuation); followed by bronchial narrowing with or without atelectasis in 94% ([@CIT0018]). A study evaluated the most important CT findings, and showed a significantly higher frequency and chance of lymph node high attenuation (calcification) (80%, odds ratio = 22.9, Cl 95% = 7.31-75), bronchial wall high attenuation (calcification) (62%, odds ratio = 6, Cl 95% = 2.07-18.3), bronchial stenosis (48%, odds ratio = 2.91, Cl 95% = 0.96-8.99), atelectasis (20%, odds ratio = 4.75, 95% Cl = 0.8-386.8) and mass lesion (14%) in BAF compared to non-anthracotic subjects ([@CIT0033]) ([Table 2](#T0002){ref-type="table"}). Bronchial stenosis was usually smooth without endobronchial nodules (80%), and although not frequent, distal atelectasis was not seen in some cases ([@CIT0018]). Bronchial wall thickening was also reported in 20% of BAF subjects ([@CIT0018]). Involvement might be unilateral or bilateral, but the right middle lobe, followed by the upper lobes were frequently reported as the most commonly involved lobes ([@CIT0007], [@CIT0018], [@CIT0033], [@CIT0065]). {#F0005} CT images were unremarkable in 17% of simple anthracosis subjects and 6% of anthracofibrosis subjects ([@CIT0033]). Pleural disease was observed in a quarter of BAF subjects ([@CIT0008]). As a routine practice, anthracosis may be erroneously diagnosed as TB, lung cancer, atelectasis or pneumonia ([@CIT0016], [@CIT0048]). Park et al. used CT scans for differentiating BAF from endobronchial TB ([@CIT0048]). The results of their study showed that BAF subjects tend to show bilateral smooth bronchial stenosis and peribronchial lymphadenopathy. This is in contrast to subjects with endobronchial TB who tend to have limited ipsilateral irregular bronchial stenosis, especially in the lobar bronchus and it can extend to contiguous bronchus and trachea. Choe et al, ([@CIT0066]) also showed that necrotic lymph nodes, multiple poorly defined small nodules, including branching opacities (including tree in bud) and consolidation with internal low density were in favor of TB. MRI was used to differentiate BAF from lung cancer in case of a mass lesion ([@CIT0067]), which showed that BAF had low density in the T2 weighted image, and differentiated BAF from lung cancer. But some reports showed positive FDG-PET results in BAF, which make the use of FDG-PET scan for differentiating between malignancy and BAF difficult ([@CIT0068], [@CIT0069]). ### Bronchoscopic findings Bronchoscopy is the gold standard for diagnosing anthracosis. As mentioned above, anthracosis may be detected in different images as simple flat anthracosis ([Figure 1A](#F0001){ref-type="fig"}), deep seated retracted anthracosis (originating from an anthracotic lymph node besides the bronchus) ([Figure 1B](#F0001){ref-type="fig"}), and protruded black discoloration of mucosa with or without narrowing of bronchi (BAF) ([Figure 1C](#F0001){ref-type="fig"}). In addition to black lesions, bronchial swelling with infiltration, erythema ([Figure 1C](#F0001){ref-type="fig"}) and thickening that may cause obliteration of bronchi may be seen ([@CIT0046]). Anthracosis can be localized or disseminated, unilateral or bilateral ([@CIT0018]) and the most frequent place of involvement is the right middle lobe (RML), followed by the upper lobe bronchi ([Table 2](#T0002){ref-type="table"}), especially at the bifurcation or inlet of the lobar or segmental bronchi ([@CIT0008], [@CIT0012]). Tracheal involvement is rare and was detected in 3.8% of BAF subjects ([@CIT0011]). Also, bronchial washing may show free black particles ([@CIT0045]) and biopsy is usually difficult as the mucosa has a hard consistency ([@CIT0007]). Bleeding during biopsy is frequent (9%) ([@CIT0008], [@CIT0037]), but it usually causes no further complications ([@CIT0007], [@CIT0008], [@CIT0037]). Detecting endobronchial TB associated with BAF with bronchoscopy is difficult; Kim et al. ([@CIT0070]) described edematous-hyperemic mucosa and ulceration as a useful picture for diagnosing TB associated with BAF. Endobronchial ultrasound has not been studied extensively in anthracotic subjects, but Mirsadraee and Farshchi ([@CIT0071]) reported a picture from typical anthracosis that showed a scattered nodular hypoechoic pattern in the subepithelial area of the bronchus or lymph node adjacent to the bronchial mucosa ([Figure 6](#F0006){ref-type="fig"}). {#F0006} Diagnosis {#S20017} --------- Anthracosis should be considered in the list of differential diagnoses of diseases such as COPD, TB (without anthracosis), lung cancer, fungal infection (such as mucormycosis and actinomycosis) and amyloidosis. A history of long-standing dyspnea and/or cough in an elderly non-smoker man or woman exacerbated in winter is suggestive for anthracosis. Wheezing during lung auscultation is in favor of BAF. Obstructive lung disease in spirometry with lymph node or bronchial calcification (high attenuation), especially in subjects who also show mass lesion (or atelectasis) strongly makes the diagnosis of anthracosis more likely. Kim et al, ([@CIT0018]) also described smooth bronchial narrowing with enlarged calcified lymph nodes as a useful marker for differentiating BAF from lung cancer. However, in all suggested cases, bronchoscopy as the gold standard of diagnosis should be performed and bronchoscopy specimens should be sent for TB evaluation. In case of a mass lesion in radiological findings, open lung biopsy, transthoracic lung biopsy or advanced bronchoscopic techniques may be necessary to rule out TB ([@CIT0072]) or malignancy ([@CIT0043], [@CIT0044]). The most important differential diagnosis of anthracosis (BAF) is COPD. [Table 4](#T0004){ref-type="table"} shows the most important differences between anthracosis (BAF) and COPD. According to this evidence, anthracosis should not be considered as a variant of COPD (with exposure to biomass instead of cigarette smoke). Progressive massive fibrosis should also be considered as a differential diagnosis, but history of work related exposure to dust and more diffuse involvement of the lung could differentiate it from pneumoconiosis. ###### Important differences between bronchial anthracofibrosis and COPD Bronchial anthracofibrosis COPD^[@CIT0077]^ -------------------------------- --------------------------------------------- ---------------------------------------------------------------------------------- **Age** Over 67 years Over 55 years **Smoking** 22.4% 80% **Female/ Male ratio** 1.7/1 1/1 **Biomass exposure** 61.5% 3.7%^[@CIT0078]^ **Histopathological findings** Macrophage containing black pigments Submucosal gland hypertrophy, smooth muscle hyperplasia and alveolar destruction **Clinical findings** Cough, dyspnea and wheeze Cough, dyspnea and wheeze **PFT** Obstructive and restrictive and Normal DLCO Obstructive, Low DLCO in emphysema **Radiological findings** Infiltration Normal or hyper-inflation, low attenuation with no visible wall **Mass lesion** Yes No **Lymphadenopathy** Yes No **Bronchoscopy** Black discoloration+ obstruction Normal Treatment {#S20018} --------- No established treatment for anthracosis has been reported thus far. Empirical treatment including bronchodilators (short or long acting), corticosteroids (inhalation or systemic) ([@CIT0073]) and antibiotics have been used. Systemic corticosteroids showed temporary relief in 60% (9/15) of non-TB BAF subjects ([@CIT0073]). In case of confirmed TB associated with anthracosis, anti-TB medications can improve the general condition and sometimes, the radiological manifestations of patients ([@CIT0007], [@CIT0033], [@CIT0064]). However, anthracosis alone can usually be controlled with conservative management, although in some cases of severe localized bronchial obstruction of large airways, bronchial stents were successfully used ([@CIT0074]). Clinical course, quality of life and follow up {#S20019} ---------------------------------------------- Follow up of radiographic mass lesions has shown a slow progress of the lesion not similar to the spread pattern of malignant lesions. In these subjects, open lung biopsy may be indicated. Kim et al. ([@CIT0008]) described the clinical course of 280 BAF subjects and showed that subjects who suffered only from BAF or a combination of BAF and malignancy had significantly lower survival rates than subjects with BAF and TB, acute exacerbation of airway disease or pneumonia. Causes of death in this study (18/288) were malignancy ([@CIT0006]), acute infection ([@CIT0005]), cardiac disease ([@CIT0003]), trauma ([@CIT0002]), acute exacerbation ([@CIT0001]) and hemoptysis ([@CIT0001]). Na et al. ([@CIT0037]) treated BAF subjects with anti-TB treatment and showed improvement and resolution of lesions in 66.6% (10/15) of subjects. The author wishes to thank Mrs. Zahra Mercedes Gonzalez for precise English editing of this article.

Introduction {#sec1-1} ============ "Intellectuals solve problems; geniuses prevent them!!" Caries occur five times more frequently in occlusal fissures, than on smooth surfaces.\[[@ref1]\] Sealant not only prevents tooth decay before it starts, but also halts the progress. This effect relies on the sealant\'s ability to fill the fissure and not detach. Retention is provided by resin tags that form an effective mechanical interlock between resin material and enamel surface. Viscosity of sealant enhances penetration, thus helping in retention.\[[@ref2]\] However it still remains to be evaluated as to how exactly do the factors, such as, resin tag length, microleakage and viscosity affect the clinical efficacy of these sealants. This study aims at evaluating the relationship between sealant viscosity, resin tag length and prevention of microleakage. Materials and Methods {#sec1-2} ===================== Three pit and fissure sealants were used in this study. The following is the list of the materials and armamentarium used in this study. Test Materials: \[[Figure 1](#F1){ref-type="fig"}\] {#sec2-1} --------------------------------------------------- {#F1} Guardian seal Kerr orange CA 92867Helioseal^®^ Assortment, Ivoclar vivadent AG, FL -9494 Schaan/ LiechtenstienEmbrace wetbond Pulpdent Corporation Watertown NA 02471-0780 USA Equipments used {#sec2-2} --------------- Stereomicroscope (Leica Wild M3Z, Germany)Scanning electron microscope (IISc Bangalore)Brooks field viscometer (DV-E viscometer Brookfield, Analytical technologies, Bangalore) The collected teeth were cleaned off the calculus, plaque and debris with an ultrasonic scaler and were stored in thymol solution. Pretreatment of the occlusal surfaces was done by cleaning the teeth with pumice slurry. The teeth were then randomly divided into three groups of 10 samples each: Group E:Embrace wetbond sealant.Group H:Helioseal sealant.Group G:Guardian seal sealant. Application of sealant {#sec2-3} ---------------------- ### Procedure for Helioseal and Guardian seal fissure sealants {#sec3-1} Occlusal surfaces of the teeth were etched with 37% phosphoric acid (Enamel preparator, Ivoclar) for 30 seconds and rinsed with water. The teeth were then dried with a mild oil-free air stream to achieve a characteristic frosty white, chalky appearance of enamel.\[[@ref3]\] Two coats of bonding agent (G bond, GC) were applied (using applicator on the etched tooth surface) and light cured for 20 seconds using visible light curing unit (470 nm to 420 nm wavelength). Sealants were then applied and cured according to the manufacturer\'s instructions. ### Procedure for Embrace wet bond fissure sealant {#sec3-2} Occlusal surfaces of the teeth were etched with 37% phosphoric acid (Enamel preparatory, Ivoclar) for 30 seconds and rinsed with water. With Embrace Wetbond, the typical dull, frosted appearance of the etched surface is not desired. Rather, the surface should be lightly dried and very slightly moist with a glossy appearance. To accomplish this, a cotton pellet was used to remove the excess moisture.\[[@ref4]\] No bonding agent was applied. The sealant Embrace wet bond was applied as per manufacturer\'s instructions \[[Figure 2](#F2){ref-type="fig"}\], followed by light curing for 40 seconds.\[[@ref5]\] The prepared teeth were stored in double de-ionized distilled water. {#F2} Procedure for microleakage testing {#sec2-4} ---------------------------------- The teeth were coated with double layer of nail varnish to prevent the leakage of the dye. The occlusal surface was excluded.\[[@ref6]\] A different colour nail varnish was used for each group for easy distinguishing. Red - Helioseal sealantBlue - Guardian sealantOrange - Embrace wetbond sealant All the groups were then immersed inverted, (such a way that unpainted surface was in contact with the dye) in 1% methylene blue for 24 hours. \[[Figure 3](#F3){ref-type="fig"}\] The samples were removed and gently brushed to remove excess dye.\[[@ref6]\] All the three groups were then subjected to thermocycling at a temperature range of 5-55°C for 500 cycles, with a dwell time of 30 seconds.\[[@ref6]\] The procedure used to determine marginal leakage was similar to the one described by Theodoridou -- Pahini and Tolidis.\[[@ref7]\] Each tooth was sectioned longitudinally in a mesio distal direction through the center of the sealant with a diamond wheel measuring 0.02 mm in thickness. The root portion of the teeth was then sectioned and removed. One half of every tooth section was utilized for microleakage testing and the other half for evaluating the length of resin tags. {#F3} The sections for microleakage evaluation were cleaned and examined under stereomicroscope for the dye penetration and were scored by 2 independent examiners as follows.\[[@ref7]\] \[[Figure 3](#F3){ref-type="fig"}\] 0= no dye penetration.1= dye penetration down the mesial or distal wall.2= dye penetration down the mesial and distal walls.3= dye penetration underneath the sealant and down the mesial or distal wall.4= dye penetration all around the sealant. Procedure for scanning electron microscopy and resin tag measurement {#sec2-5} -------------------------------------------------------------------- The other half of the tooth section was prepared for evaluation under scanning electron microscopy. The tooth sections were polished using a carbide stone. The polished sections were then decalcified using 37% phosphoric acid for 15 seconds to etch away any enamel mineral component not protected by sealants and then rinsed and stored in distilled water. The tooth sections were dried thoroughly under the heat lamp, and then mounted on brass rings using a non conductor tape made of carbon. This was then applied to the sections, in the areas that did not need scanning. These mountings were then placed inside an ion spluttering device for 30 minutes using vacuum evaporation at 200 -300 A°.\[[@ref8]\] The gold spluttered sections were then placed inside the scanning electron microscope of 20 kV capacity and photographs of the sections were obtained. The resin tag lengths were then measured \[[Figure 4](#F4){ref-type="fig"}\]. The average of each photograph was calculated. {#F4} Preparation of the samples for viscosity measurement {#sec2-6} ---------------------------------------------------- The viscosity was checked by diluting the sealant with methyl methacrylate monomer liquid.\[[@ref9]\] The viscosity of the monomer liquid was evaluated first. The liquid was placed in the sample holder of the Brooke\'s field viscometer and was calibrated at 60 revolutions per minute, 100% torque. Further, s-18 spindle was used and the reading was noted. 1 ml of the sealant was drawn out of its container and poured into a 10 ml test tube to which the methyl methacrylate liquid was added and mixed till the total liquid amounted to 10 ml (since the minimum amount needed to evaluate the viscosity was 10 ml). Each sample was individually placed in the Brook\'s field viscosity meter at room temperature and the viscosity was measured in cP (centipoise). Care was taken to see that the light exposure of the sealant was minimal. The obtained measurements were then subjected to statistical analysis. Results {#sec1-3} ======= The recorded values were represented as mean ± standard deviation and range values, and were statistically analyzed using one way analysis of variance (ANOVA) for multiple group comparison, post hoc tukey\'s test for group wise comparison, the spearman\'s rank correlation coefficient for correlation analysis and the Mann Whitney test for multiple group comparison of the microleakage scores. [Table 1](#T1){ref-type="table"} and [Graph 1](#F5){ref-type="fig"} show the comparison of values obtained for microleakage between group E, group H and group G. The mean value of microleakage for group E is 0.4, group H is 1.0 and group G is 1.6. No statistically significant differences were found between the group E v/s group H and between group H v/s groups G. Statistically significant difference (*p* \< 0.05) was seen between group E v/s group G. Group E showed the least microleakage followed by group H and G. ###### Mean microleakage scores and inter group comparison of microleakage scores of group E, group H, group G  {#F5} [Table 2](#T2){ref-type="table"} and [Graph 2](#F6){ref-type="fig"} show the comparison of values obtained for resin tag length between group E, group H and group G. The mean resin tag length for Group E is 10.14 ± 4.84; for group H is 9.65 ± 4.28 and for group G is 5.86 ± 1.85. The mean difference of group E v/s group H is 0.49, group G v/s group H is 3.79 and group E v/s group G is 4.28. No statistically significant differences were found between group E v/s group H and group H v/s group G. Statistically significant difference (*p* \< 0.05) was seen between group E v/s G. The mean resin tag length was highest for group E. ###### Mean difference of the resin tag lengths and inter group comparison of the resin tag length between group E, group H and group G  {#F6} [Table 3](#T3){ref-type="table"} shows the relationship between the microleakage and resin tag length in group E, group H and group G. Statistically significant differences were seen in all the three groups. Group E showed a *p*-value of *P*\< -0.85, group H showed a *P*-value of *P*\< - 0.94 and group G showed a *P*-value of *P*\< -0.96. The relationship between microleakage and resin tag length was inversely proportional. As the resin tag length increased, the microleakage decreased. ###### Relationship between resin tag length and microleakage of group E, group H and group G (correlation analysis)  [Table 4](#T4){ref-type="table"} and [Graph 3](#F7){ref-type="fig"} show the compiled data showing the mean difference and range of resin tag length, microleakage scores and viscosity of group E, group H and group G. The viscosity of the group E was the lowest at 0.96 cP, group H was 1.59 cP and the group G was 1.92 cP. ###### Compiled data showing the mean difference and range of resin tag length, microleakage scores and viscosity of group E, group H and group G  {#F7} Discussion {#sec1-4} ========== The occlusal surface, especially the pits and fissures of posterior teeth have been recognized for their high caries susceptibility over many years.\[[@ref10]\] It is undoubtedly more vulnerable due to the unique morphology of the pits and fissures. Occlusal pits and fissures vary in shape, but are generally narrow (\~ 0.1 mm wide) and tortuous, and are considered to be an ideal site for the retention of bacterial and food remnants. This is because the morphology renders the mechanical means of debridement inaccessible\[[@ref11]\] as an average tooth brush bristle (0.2 mm) is too large to penetrate most of the fissures.\[[@ref12]\] Other factors such as lack of salivary access to the fissures, the close proximity of fissure base to the dentino-enamel junction and remnants of debris and pellicle in the fissures increase caries susceptibility of fissures by many folds. Therefore, to prevent initiation of caries in these fissures, the concept of pit and fissure sealants evolved.\[[@ref13]\] The cariostatic properties of sealants are attributed to the physical obstruction of the pit and fissures. This prevents colonization of the pits and fissures with new bacteria and also prevents the penetration of fermentable carbohydrates to any bacteria remaining in the pits and fissures.\[[@ref14]\] Various studies show strong correlation between the sealant and absence of caries.\[[@ref15]\] *Requisites of an efficient sealant include:* A viscosity allowing penetration into deep and narrow fissures even in maxillary teeth, adequate working time, rapid cure, good and prolonged adhesion to enamel, low sorption and solubility, resistance to wear, minimum irritation to tissues, and cariostatic action.\[[@ref9]\] The third molars extracted for therapeutic purpose, which were free of caries, developmental defects, enamel micro-fractures and discoloration were included in this study, as previous studies\[[@ref16][@ref17]\] have revealed that any preexisting alteration of surface morphology of the tooth directly influenced the caries progression. As compared to the formaldehyde, thymol has no effect on the protein structure and neither does it alter the enamel structure. Hence in this study, 0.01% thymol solution was used as a storage solution and disinfectant.\[[@ref3]\] Some studies show that pumice prophylaxis does not completely and consistently remove the pellicle and debris, especially in the depth of the fissure.\[[@ref18]\] Studies by Blackwood JA *et al*. showed that between enameloplasty, air abrasion and pumice prophylaxis, the least microleakage was seen with the conventional pumice prophylaxis.\[[@ref19]\] In the present study, despite of all the controversies, conventional pumice method was used for cleaning prior to etching. Timing of the sealant placement is critical. The teeth that have newly erupted are the ones that are most susceptible to caries and hence need protection from pit and fissures. But isolating them is the most difficult. Until now the only moisture tolerant sealants were glass ionomers.\[[@ref20]\] Their mechanism of adhesion is ionic bonding, not micromechanical retention to an acid etched enamel surface. Paradi and co workers reported low sealant retention rates with glass ionomer cement (GIC).\[[@ref21]\] Recently, resin-based sealant technology has introduced a moisture-tolerant chemistry. While traditional sealants are hydrophobic, Embrace^™^ Wetbond^™^ is hydrophilic. On light-curing, this sealant has physical properties similar to other commercially available sealants.\[[@ref5]\] Embrace Wetbond incorporates di-, tri- and multifunctional acrylate monomers into an advanced acid-integrating chemistry that is activated by moisture. When placed in the presence of moisture, the sealant spreads over the enamel surface. Because of its unique chemistry, Embrace Wetbond is miscible with water and flows into moisture-containing etched enamel and combines with it.\[[@ref5]\] Additionally, a longitudinal study done *in vivo* with embrace wetbond has shown a retention rate of 90%. Only 32 teeth out of the 334 teeth in the study required replacement of the sealant.\[[@ref22]\] Early in the development of sealants, it was recognized that the addition of fluoride to a sealant could have the potential benefits of additional caries protection. The fluoride release achieved high levels initially, with salivary fluoride concentration being high within 30 minutes after sealant placement; however, returned to baseline levels within 1-2 days after the sealant placement.\[[@ref23]\] Evaluation of the combined use of fluoride and dental sealants has showed retention of 92% after 4 years. This suggests that pit and fissure sealants confer additional caries preventive benefits beyond those of fluoride therapy alone.\[[@ref24]\] A sealant is effective in preventing caries, only when it is retained in the fissures. Hence the retention becomes a major factor, influencing the efficacy of the sealant. The retention in resin based pit and fissure sealant is through micromechanical interlocking between the resin and the enamel. Mechanical retention of sealant is the direct result of resin penetration into the porous enamel forming tags. This occurs by capillary action. Monomer in the material polymerizes and the material becomes interlocked with the enamel surface.\[[@ref25]\] In the present study the mean length of tags obtained was in the range of 10 *μ*m to 5 *μ*m.\[[@ref19]\] Other studies by Gomez *et al*.\[[@ref26]\] and Karina Tonini *et al*.\[[@ref27]\] have also concurred with the measurements obtained. It was seen in our study that the pit and fissure sealant with the longer resin tags had lesser microleakage. Thus, it can be postulated that, the longer resin tags formed, lesser will be microleakage. Other than surface tension, the viscosity of the sealant also influences the penetration of sealants.\[[@ref13]\] Viscosity is the resistance of a liquid to flow. This resistance of the fluid to flow is controlled by internal frictional forces within the liquid. A highly viscous fluid flows slowly. The viscosity is measured in units of MPa (Mega pascal) per second or cP (centipoise). Higher viscosity may cause poorer adaptation and incomplete penetration to the bottom of the pit and fissures resulting in decreased retention. With low viscosity sealants, there is a greater potential of the sealant to flow, spread more rapidly over the surface and penetrate. At times the size of the filler particles may be larger than the porosities of enamel. Faster penetration rates are found with larger holes, less dense liquids and those with high surface tension.\[[@ref28]\] On the contrary, the study by DM Barnes, P kihn, A Elsabach and JA von Fraunhofer has shown that the viscosity and flow properties of the fissure sealants do not affect their sealing ability.\[[@ref29]\] Hence, we tried to analyze the nature of relationship between viscosity and the other factors of pit and fissure sealants. The viscosity of sealants can be assessed in centipoise units using ultrasonic vibratile viscometer or by conventional capillary tube method.\[[@ref29][@ref30]\] The ultrasonic vibratile viscometer is a more accurate method as compared to the capillary tube method, as it gives a digital read out. The viscometer rotates the spindle in the liquid to overcome the viscous resistance to the induced movement, and thus a reading is obtained.\[[@ref9]\] In this study, the Brookfield viscometer was used. All Brookfield laboratory viscometers are accurate within the range of +/- 1%, and have a reproducibility within the range of +/- 0.2%, thus allowing the test results to be duplicated anywhere in the world when the same model instrument is used. In the present study, the viscometer utilized required a minimum volume of 10 ml sample. To achieve this, 1ml of each sealant was first diluted in 9 ml of the diluent, methlymethacrylate.\[[@ref9]\] Hence, the relative viscosities of the three sealants were obtained. In the present study, the sealant with fluoride i.e. Guardian seal showed highest viscosity. It has been shown that sealant containing fluoride tends to be thicker than those without fluoride.\[[@ref29]\] The least viscous was the Embrace wetbond. Urethane monomer based sealant may confer more elasticity and adhesiveness to the resin than does Bisphenol A-Glycidyl Methacrylate (Bis-GMA).\[[@ref31]\] In this study, two of the sealants Helioseal and Guardian seal contained Bis-GMA, where as the Embrace wet bond contained di, tri and multiacrylate monomers. That may be the reason why the viscosity of Embrace wetbond was lower as compared to the other groups.\[[@ref5]\] Another factor thought to be central to the clinical success of sealants is microleakage.\[[@ref32]\] Microleakage is defined as the passage of bacteria and their toxins between restorative margins and the tooth preparation walls. Clinically microleakage is important when one considers that pulpal inflammation is more likely caused by bacteria than by chemical toxicity of restorative material. Microleakage may support the caries process beneath the sealant, hence the ability of the sealant to adequately seal the pits and fissures and prevent microleakage is important. If restoration is hermatically sealed, then the bacteria cannot survive.\[[@ref32][@ref33]\] Microleakage assessment may be qualitative or quantitative with different systems, including both simple and computer based methods. Dye penetration has been used in several studies, to assess the presence of marginal leakage around the sealant--enamel surface.\[[@ref34][@ref35]\] In the present study, a qualitative technique of dye penetration was used. Penetration of dye can indicate the lack of a perfect seal. The presence of a dye in the under penetration zone of the etched enamel can indicate a susceptible microleakage pathway. Clinically, this may imply that the remaining etched area could be a factor for development of caries by microleakage, if the sealant was partially or completely lost. All the sealants used in this study showed some degree of microleakage and these factors agreed with previous studies.\[[@ref7][@ref31][@ref36]\] The least microleakage was seen with the Embrace wetbond sealant, followed by Helioseal and Guardian seal. Complete penetration of sealant into complex fissure systems is difficult due to the phenomenon of closed end capillaries or isolated capillaries. Some lateral fissures arising from the main fissures also fail to be filled with sealant.\[[@ref34]\] Hence, further research on the effect of fissure morphology on sealant microleakage and penetration are necessary. Stereomicroscope is the gold standard in microleakage studies and hence was used here. The SEM was used to measure the length of the resin tags as the SEM can produce very high-resolution images of a sample surface, and reveal details about less than 1 to 5 nm in size. Due to the very narrow electron beam, SEM micrographs have a large depth of field yielding a characteristic three-dimensional appearance useful for understanding the surface structure of a sample.\[[@ref37]\] It should be noted that the results of the present study are valid for *in vitro* conditions. Depending on the environment, all pit and fissure sealants may act differently due to other variables like type of fissures, preparation of fissures, enamel etching and conditioning, application of bonding agent and contamination of prepared surfaces of fissures. Appropriate method of application of sealants and viscosity of the sealant, are also a factors influencing the microleakage, and if a proper application method is followed, it can increase the length of resin tag and thus improve the efficiency of the sealant in preventing caries. Conclusion {#sec1-5} ========== This study clearly indicated that, there does exists a relationship between the resin tag length and microleakage, and the longer the resin tags formed, the lesser is the microleakage, and the better will the cariostatic action of the pit and fissure sealant be. The viscosity too plays a role in the formation of these all important resin tags. The lower viscosity sealants are better. Also, with the newly developed hydrophilic sealant Embrace wetbond, it is now possibly to go ahead and seal the newly erupted teeth that were previously left unprotected due to moisture control problems. The authors would like to acknowledge Dr. Sadashiva Shetty, Principal of Bapuji Dental College; for his support in making this study a success, the Indian institute of sciences for their gracious support in the utilization of the scanning electron microscope; and, last but definitely not the least, Dr.Thimmashetty, Head of Department, department of pharmaceutics, Bapuji Pharmacy College, for his help. **Source of Support:** Nil. **Conflict of Interest:** None declared.

(J Am Heart Assoc. 2018;7:e009172 DOI: 10.1161/JAHA.118.009172.) Clinical PerspectiveWhat Is New?Visceral adipose tissue was not associated with incident clinical atherosclerotic cardiovascular disease (ASCVD) events in older men, without or with adjustment for age and other ASCVD clinical risk factors, even among those without any major pre‐existing comorbid medical conditions, overweight men, or men with normal body mass index.Android‐to‐gynoid fat mass ratio (measured using dual energy x‐ray absorptiometry) was also not associated with incident clinical ASCVD events in older men.What Are the Clinical Implications?Dual energy x‐ray absorptiometry measures of central adipose tissue depots are not clinically useful for ASCVD risk stratification in older men.Weight loss among older men to reduce central obesity may not reduce their risks of incident ASCVDS.Since weight loss can result in loss of muscle mass and functional capabilities, the risks and benefits of weight for older men need to be carefully considered on a case‐by‐case basis. {#jah33368-sec-0008} Previous studies have linked central obesity (around the abdomen rather than appendicular) with metabolic syndrome, dyslipidemia, and a higher risk of incident atherosclerotic cardiovascular disease (ASCVD).[1](#jah33368-bib-0001){ref-type="ref"} Central adipose tissue is composed of both subcutaneous adipose tissue (SAT) and visceral adipose tissue (VAT) compartments. VAT may be particularly detrimental[2](#jah33368-bib-0002){ref-type="ref"}, [3](#jah33368-bib-0003){ref-type="ref"}, [4](#jah33368-bib-0004){ref-type="ref"}, [5](#jah33368-bib-0005){ref-type="ref"} because it is infiltrated with macrophages and other inflammatory cells associated with higher levels of serum inflammatory markers,[6](#jah33368-bib-0006){ref-type="ref"} and is also associated with metabolic syndrome and markers of insulin resistance (including raised fasting insulin concentrations).[7](#jah33368-bib-0007){ref-type="ref"}, [8](#jah33368-bib-0008){ref-type="ref"}, [9](#jah33368-bib-0009){ref-type="ref"} We have previously shown that higher levels of VAT, measured by either computed tomography (CT) or dual energy X‐ray absorptiometry (DXA), are associated with a detrimental ASCVD risk factor profile among older men including lower high‐density lipoprotein cholesterol, higher triglycerides, and greater insulin resistance.[10](#jah33368-bib-0010){ref-type="ref"} In spite of this, it is unclear whether VAT or other measures of central adiposity predict ASCVD events in the older population, given that adipose tissue may be a marker of vitality and a low burden of comorbid illness in this age group.[11](#jah33368-bib-0011){ref-type="ref"} Although obesity (body mass index \[BMI\] ≥30 kg/m^2^) is associated with increased mortality among adult men and women age 64 years or younger, this association is much weaker or absent among those age 65 and older,[12](#jah33368-bib-0012){ref-type="ref"}, [13](#jah33368-bib-0013){ref-type="ref"} a phenomenon referred to as the "obesity paradox."[11](#jah33368-bib-0011){ref-type="ref"} Some have postulated that this paradox is caused by increasing comorbid illness burden in older adults being associated with both weight loss and incident ASCVD. Only 1 other study has examined the association of VAT with any type of incident ASCVD events in adults older than age 70 years; this study reported that higher VAT was associated with a higher risk of myocardial infarction in older women but not older men. Other ASCVD outcomes were not assessed.[3](#jah33368-bib-0003){ref-type="ref"} Therefore, our primary study aim was to estimate the association of VAT with incident ASCVD events in older men. Other measures of central adipose tissue (eg, greater waist circumference) have been associated with increased risks of all‐cause and ASCVD mortality, but greater gluteofemoral adipose tissue (operationally defined as hip circumference) may be associated with decreased risks.[14](#jah33368-bib-0014){ref-type="ref"}, [15](#jah33368-bib-0015){ref-type="ref"} Therefore, our second aim was to also estimate the association of another measure of central obesity, the android--gynoid fat mass ratio, with incident ASCVD events. Given the hypothesis that the obesity paradox is related to a much higher incidence of multimorbidity in those over age 70 years, a third aim was to estimate the association of VAT with incident ASCVD events in the subset without key common comorbid medical conditions, on the supposition that these men biologically may resemble those who are chronologically younger. Finally, data from the National Health and Nutritional Examination Survey suggest that central obesity may be particularly associated with ASCVD mortality in adult men without obesity as defined by BMI.[16](#jah33368-bib-0016){ref-type="ref"}, [17](#jah33368-bib-0017){ref-type="ref"} Hence, we also performed all of the above analyses for the subsets of overweight men (BMI 25.0--29.9 kg/m^2^), and men with normal BMI (18.5--24.9 kg/m^2^). Methods {#jah33368-sec-0009} ======= MrOS (Osteoporotic Fractures in Men) data, analytic methods, and study materials have been made available to other researchers for purposes of reproducing the results or replicating the procedures in this study.[18](#jah33368-bib-0018){ref-type="ref"} Between 2000 and 2002, the MrOS study enrolled 5994 community‐dwelling ambulatory men age 65 years and older at 6 geographic sites in the United States (Birmingham, AL; Minneapolis, MN; Palo Alto, CA; Pittsburgh, PA; Portland, OR; and San Diego, CA), as described in previous publications,[19](#jah33368-bib-0019){ref-type="ref"}, [20](#jah33368-bib-0020){ref-type="ref"} after Institutional Review Board approval at each of the 6 study sites. All participants signed informed consent documents. Between December 2003 and March 2005, 3135 of these men were also enrolled in the Outcomes of Sleep Disorders in Older Men (MrOS Sleep) ancillary study. Men with the following characteristics were excluded: (1) use of oxygen therapy in the past 3 months; (2) a history of an open tracheotomy; or (3) sleeping with a mouthpiece, continuous positive airway pressure, or bilevel positive airway pressure for snoring or sleep apnea in the prior 3 months. Of these men, 2899 had valid VAT measurements on a DXA body composition study at the MrOS Sleep baseline visit, adjudication of incident ASCVD events, and complete covariate data (Figure [1](#jah33368-fig-0001){ref-type="fig"}). Over half of the 236 excluded men did not have a valid VAT measurement on DXA and/or had missing data with respect to an incident ASCVD event. There was no difference in other baseline characteristics between the 2899 men included in and the 236 men excluded from these analyses (Table [S1](#jah33368-sup-0001){ref-type="supplementary-material"}). {#jah33368-fig-0001} Ascertainment of Incident ASCVD Events {#jah33368-sec-0010} -------------------------------------- We used the composite outcome of ASCVD events as defined by the American Heart Association--American College of Cardiology guidelines (coronary heart disease death, myocardial infarction, fatal stroke, or nonfatal stroke). This excludes events that have poor reliability (such as angina) and those that are influenced by clinical practice preferences (such as revascularization procedures).[21](#jah33368-bib-0021){ref-type="ref"} Participants were contacted by postcard and/or telephone every 4 months after the baseline MrOS Sleep substudy visit until February 28, 2015 (mean \[SD\] follow‐up 7.9 \[3.4\] years), and queried regarding possible ASCVD events; 99% of these follow‐up contacts were successfully completed among active surviving participants. Relevant medical records and supporting documentation from any potential incident clinical events identified by phone or postcard contact were obtained by the clinical center and forwarded to the data coordinating center. For fatal events, the death certificate and hospital records from the time of death (if available) were collected. For fatal events that did not occur in hospital, a proxy interview with next of kin and hospital records from the most recent hospitalization in the past 12 months were obtained. These documents were used to determine the underlying cause of death. Documentation for potential nonfatal and fatal ASCVD events were reviewed and adjudicated by a centrally trained board‐certified cardiologist using a prespecified adjudication protocol that had been successfully used for both prior randomized trials and epidemiological studies of ASCVD. Only events confirmed by the adjudicator are included for analysis. Measurement of Regional Fat Mass Depots on DXA {#jah33368-sec-0011} ---------------------------------------------- At the MrOS Sleep baseline visit, DXA whole body composition studies were performed using Hologic QDR4500A densitometers. A whole body phantom was circulated among the 6 study sites to cross‐calibrate regional fat and lean tissue mass measurements. The variability across study enrollment sites was within acceptable limits, and cross‐calibration correction factors were not required. To adjust for study site differences, statistical models include indicator variables for the individual scanners. The whole body DXA scans were re‐analyzed centrally in 2016 with Hologic APEX software version 5.5 to obtain VAT measurements, using a standard algorithm.[22](#jah33368-bib-0022){ref-type="ref"} The software calculated the total adipose tissue within a 5‐cm transverse slice on the 2‐dimensional projection of the abdominal‐pelvic region, the inferior border of which is placed at the top of the iliac crests (at about the L4 vertebral level). The lateral and medial edges of the abdominal wall musculature are identified (Figure [2](#jah33368-fig-0002){ref-type="fig"}); all of the adipose tissue in the areas outside of the lateral edges is SAT. VAT is contained in the visceral cavity inside the medial edges of the abdominal wall musculature, but this area also includes SAT anterior and posterior to the abdominal wall musculature superimposed on the 2‐dimensional projection. The amount of SAT in the medial VAT area can be estimated from the SAT lateral to the abdominal wall musculature, and the estimated visceral fat area (cm^2^) is then calculated as total adipose tissue overlying and within the visceral cavity minus the SAT overlying this area. {#jah33368-fig-0002} The android region of interest was situated above the line between the right and left iliac crests and below the horizontal line 20% of the distance between the inferior boundary and the chin. Fat mass measurements in the android region included both VAT and SAT. The upper boundary of the gynoid region was a line drawn a distance 1.5 times the height of the android region below the iliac crests, and the gynoid region lower boundary was a line 2 times the height of the android region below the upper boundary.[23](#jah33368-bib-0023){ref-type="ref"} Other Covariates {#jah33368-sec-0012} ---------------- Height was measured with a Harpenden stadiometer, weight recorded with a balance beam or electronic scale, and BMI calculated as weight (kg) divided by height squared (m^2^) at the MrOS Sleep baseline visit. Participants self‐reported if they were a current or a past smoker, if they were currently taking aspirin or a statin medication, whether or not they were using medications for hypertension, and whether or not they had been diagnosed previously by a physician with a myocardial infarction, stroke, atrial fibrillation, congestive heart failure, chronic obstructive pulmonary disease, diabetes mellitus, Parkinson\'s disease, rheumatoid arthritis, liver disease, or renal disease. Systolic blood pressure was measured in the right arm twice with the participant sitting, and averaged. Oxidized low‐density lipoproteins were measured on Beckman Coulter Biomek NXp (Beckman Coulter, Inc, Fullerton, CA) using a direct sandwich enzyme immunoassay (Oxidized LDL ELISA; Mercodia AB, Uppsala, Sweden). Statistical Analyses {#jah33368-sec-0013} -------------------- Baseline characteristics of participants were compared in 2 groups, stratified by incident ASCVD outcome (yes versus no) using χ^2^ statistics for categorical variables, Student *t* test for normally distributed continuous variables, and Mann--Whitney *U* test for continuous variables with a non‐normal distribution. Separate Cox proportional hazards models were used to estimate the hazard ratios (per SD increase) of BMI (model 1), VAT (model 2), android‐to‐gynoid fat mass ratio (model 3), with incident ASCVD events, adjusted for age, education, race, systolic blood pressure, smoking (ever versus never), oxidized LDL level, treatment for hypertension, statin use, aspirin use, and presence of diabetes mellitus, censoring for ASCVD event, mortality for any cause, and loss to follow‐up. To test whether the associations of our primary adipose tissue varied by age, additional models were run including the appropriate interaction term (BMI\*age, VAT\*age, or \[android/gynoid\]\*age). Schoenfeld residuals were used to test that the proportional hazards assumption was not violated. Gluteofemoral adipose tissue may be protective against incident ASCVD and mortality, and some have therefore recommended that prediction models include measures of central and gluteofemoral adipose tissue as separate covariates in order to capture their independent effects, adjusted for each other, on outcome events.[14](#jah33368-bib-0014){ref-type="ref"}, [15](#jah33368-bib-0015){ref-type="ref"} Therefore, we also estimated the separate associations of android and gynoid fat mass, adjusted for each other (model 4). Given that weight loss is associated with and can even begin shortly before the onset of a variety of illnesses in older people,[24](#jah33368-bib-0024){ref-type="ref"} the "obesity paradox" may be a function of comorbid conditions that become more prevalent with age, rather than age per se. Hence, we repeated these analyses in men without major self‐reported physician diagnosed pre‐existing comorbid illness (prior myocardial infarction, stroke, atrial fibrillation, congestive heart failure, diabetes mellitus, chronic obstructive pulmonary disease, Parkinson\'s disease, rheumatoid arthritis, liver disease, or renal disease). For all models, postdiagnostic tests were done to ensure that the proportional hazards assumption for the adipose tissue predictor variables was not violated. Results {#jah33368-sec-0014} ======= Over a mean follow‐up period of 7.9 (SD 3.4) years, 424 men (14.6%) experienced 1 or more incident ASCVD events; 102 (3.5%) died of coronary heart disease, 223 (7.7%) had an acute myocardial infarction, 133 (4.6%) had a nonfatal stroke, and 39 (1.3%) had a fatal stroke. Men who had an incident ASCVD event were older, had a higher systolic blood pressure, and were more likely to self‐report that they were using medication for hypertension at baseline (Table [1](#jah33368-tbl-0001){ref-type="table"}). None of the crude, unadjusted associations of baseline VAT, android fat mass, gynoid fat mass, and android--gynoid fat mass ratio with incident ASCVD events were statistically significant (Table [1](#jah33368-tbl-0001){ref-type="table"}). ###### Patient Characteristics Stratified by ASCVD Incident Event Status Characteristic All (N=2899) ≥1 ASCVD Event (N=424) No ASCVD Event (N=2475) *P* Value -------------------------------------------- ------------------ ------------------------ ------------------------- ----------- VAT area (cm^2^), mean (SD) 182.2 (69.18) 186.9 (71.30) 181.4 (68.79) 0.13 Android fat mass (g), mean (SD) 2253.5 (913.14) 2292.9 (914.57) 2246.7 (912.92) 0.24 Gynoid fat mass (g), mean (SD) 3151.5 (1056.87) 3155.6 (1092.10) 3150.8 (1050.94) 0.83 Android/gynoid fat mass ratio, mean (SD) 0.714 (0.19) 0.728 (0.19) 0.711 (0.18) 0.05 Age (y) at sleep visit, mean (SD) 76.3 (5.53) 78.1 (5.64) 76.0 (5.45) \<0.0001 Educational status, N (%) High school or less 611 (21.1) 104 (24.5) 507 (20.5) 0.06 Some college 649 (22.4) 106 (25.0) 543 (21.9) College 538 (18.6) 78 (18.4) 460 (18.6) Some graduate school 320 (11.0) 40 (9.4) 280 (11.3) Graduate school 781 (26.9) 96 (22.6) 685 (27.7) Race, N (%) White 2616 (90.2) 387 (91.3) 2229 (90.1) 0.44 Other 283 (9.8) 37 (8.7) 246 (9.9) Cigarette smoking status, N (%) Never 1151 (39.7) 163 (38.4) 988 (39.9) 0.57 Ever 1748 (60.3) 261 (61.6) 1487 (60.1) BMI at sleep visit, N (%) \<18.5 kg/m^2^ 9 (0.3) 1 (0.2) 8 (0.3) 0.83 18.5 to 24.9 kg/m^2^ 874 (30.1) 122 (28.8) 752 (30.4) 25 to 29.9 kg/m^2^ 1427 (49.2) 209 (49.3) 1218 (49.2) ≥30 kg/m^2^ 589 (20.3) 92 (21.7) 497 (20.1) Systolic blood pressure (mm Hg), mean (SD) 126.9 (16.18) 131.0 (18.38) 126.2 (15.67) \<0.0001 oxLDL (U/L), mean (SD) 44.1 (12.25) 43.8 (12.00) 44.1 (12.30) 0.65 Diabetes mellitus, N (%) No 2522 (87.0) 362 (85.4) 2160 (87.3) 0.28 Yes 377 (13.0) 62 (14.6) 315 (12.7) Current blood pressure medication, N (%) No 1249 (43.1) 138 (32.5) 1111 (44.9) \<0.0001 Yes 1650 (56.9) 286 (67.5) 1364 (55.1) Current statin medication, N (%) No 1698 (58.6) 253 (59.7) 1445 (58.4) 0.62 Yes 1201 (41.4) 171 (40.3) 1030 (41.6) Aspirin, N (%) No 1208 (41.7) 174 (41.0) 1034 (41.8) 0.78 Yes 1691 (58.3) 250 (59.0) 1441 (58.2) ASCVD indicates atherosclerotic cardiovascular disease; BMI, body mass index; OxLDL, oxidized low‐density lipoprotein; VAT, visceral adipose tissue. There was no association between VAT, android‐to‐gynoid fat mass ratio, or BMI with incident ASCVD events among the overall cohort, adjusted for age, study enrollment site, race, education, systolic blood pressure, current hypertension medication use, current aspirin use, current statin use, oxidized low‐density lipoproteins, diabetes mellitus, and smoking status (Table [2](#jah33368-tbl-0002){ref-type="table"}, Table [S2](#jah33368-sup-0001){ref-type="supplementary-material"}). Interaction terms between age and BMI category, age and VAT, and age and android--gynoid ratio were all not significant (data not shown). Increasing age, higher systolic blood pressure, and current medication for hypertension were each independently associated with a higher risk of ASCVD events. There was no association between all other covariates (including oxidized low‐density lipoproteins) and incident ASCVD. ###### Multivariable Adjusted Associations[a](#jah33368-note-0004){ref-type="fn"} (HR \[95% CI\]) of DXA‐VAT and Android--Gynoid Fat Mass Ratios Both With Major ASCVD Events Characteristic Model 1 (N=2899) Model 2 (N=2899) Model 3 (N=2899) Model 4 (N=2899) ------------------------------------------------- ------------------- ------------------- ------------------- ------------------- BMI at sleep visit \<18.5 kg/m^2^ 1.05 (0.15, 7.49) 18.5 to 24.9 kg/m^2^ Reference 25 to 29.9 kg/m^2^ 1.04 (0.83, 1.31) ≥30 kg/m^2^ 1.13 (0.85, 1.51) VAT area (cm^2^) (per SD increase) 1.02 (0.92, 1.13) Android/gynoid fat mass ratio (per SD increase) 1.05 (0.95, 1.17) Android fat mass (per SD) 1.08 (0.91, 1.28) Gynoid fat mass (per SD) 0.94 (0.80, 1.12) ASCVD indicates atherosclerotic cardiovascular disease; BMI, body mass index; CI, confidence interval; DXA, dual energy x‐ray absorptiometry; HR, hazard ratio; LDL, low‐density lipoprotein; VAT, visceral adipose tissue. Adjusted for age, education, race, systolic blood pressure, current use of hypertension medication, oxidized LDL, smoking status, diabetes mellitus, use of statin medication, use of aspirin, and study enrollment site; parameter coefficients with CIs are shown in Table [S2](#jah33368-sup-0001){ref-type="supplementary-material"}. Within subsets of men defined by BMI category or the subset without any self‐reported major comorbid illness at baseline (Table [3](#jah33368-tbl-0003){ref-type="table"}), neither VAT nor android--gynoid fat mass ratio were associated with incident ASCVD events. Sensitivity analyses showed that android fat mass and gynoid fat mass, adjusted for each other, were also not associated with incident ASCVD events among all men (Table [2](#jah33368-tbl-0002){ref-type="table"}), overweight men (Table [3](#jah33368-tbl-0003){ref-type="table"}), or men with no pre‐existing comorbid illness (Table [3](#jah33368-tbl-0003){ref-type="table"}). Among men with normal BMI, each SD increase of gynoid fat mass, adjusted for android fat mass and all other covariates, was associated with a 37% reduction of incident ASCVD events (hazard ratio 0.63, 95% confidence interval, 0.41--0.97). ###### Multivariable‐Adjusted Associations (HR, 95% CI)[a](#jah33368-note-0006){ref-type="fn"} of Regional Fat Depots With Incident ASCVD Events in Key Subsets of Men Predictor(s) (per SD Increase) Subset -------------------------------- ------------------- ------------------- ------------------- VAT 0.90 (0.67, 1.21) 1.02 (0.83, 1.24) 1.09 (0.92, 1.28) Android--gynoid fat mass ratio 1.07 (0.89, 1.28) 1.03 (0.89, 1.19) 1.14 (0.98, 1.33) Android fat mass 1.17 (0.80, 1.73) 1.04 (0.81, 1.34) 1.25 (0.96, 1.65) Gynoid fat mass 0.63 (0.41, 0.97) 1.00 (0.78, 1.29) 0.83 (0.63, 1.08) ASCVD indicates atherosclerotic cardiovascular disease; BMI, body mass index; CI, confidence interval; HR, hazard ratio; LDL, low‐density lipoprotein; VAT, visceral adipose tissue. Adjusted for age, education, race, systolic blood pressure, current use of hypertension medication, oxidized LDL, smoking status, diabetes mellitus, use of statin medication, use of aspirin, and study enrollment site. Absence of myocardial infarction, stroke, atrial fibrillation, congestive heart failure, chronic obstructive pulmonary disease, diabetes mellitus, Parkinson\'s disease, rheumatoid arthritis, liver disease, or renal disease. Discussion {#jah33368-sec-0015} ========== In this large cohort of older men, VAT and android--gynoid fat mass ratio had no significant bivariate (unadjusted) or multivariable‐adjusted association with incident ASCVD events. This is consistent with the "obesity paradox" hypothesis that with advancing age adipose tissue depots may be associated with higher vitality reflected in lower burden of comorbid illness. However, we also did not find any association of VAT or android--gynoid fat mass ratio within the subset of older men without any pre‐existing major comorbid conditions. With advancing age, adipose tissue may also be associated with better preservation of muscle mass. Whether this, or other positive nutritional factors associated with higher levels of adipose tissue in adults over age 70 years, negate the negative effects of VAT are unknown. Some have hypothesized that central adipose tissue may be particularly deleterious among those with normal BMI ("normal weight obesity"). However, neither VAT nor android--gynoid fat mass ratio were associated with incident ASCVD in this subset of older men. Considered as separate predictors, gynoid fat mass adjusted for android fat mass may be protective against ASCVD events among older men with normal BMI. However, further studies would be required to confirm this and explicate the reason(s) for this association. Our findings are consistent with those from the Health ABC study where VAT predicted incident myocardial infarction among older women but not older men.[3](#jah33368-bib-0003){ref-type="ref"} That prior study had a relatively small number (71) of outcomes events; our study had a much larger number (424) of outcome events, and suggests that VAT is also not associated with a broader range of ASCVD outcomes in older men. While our study was not designed to examine the effects of interventions to reduce central obesity in these older men, our findings suggest that it is possible that weight loss to reduce central obesity among older obese individuals may not reduce their risks of incident ASCVD. On the other hand, weight loss results in loss of both fat and lean mass, and can reduce muscular strength and increase functional impairment even among obese individuals[25](#jah33368-bib-0025){ref-type="ref"}; these changes may be particularly detrimental among very old adults who already have experienced significant age‐related loss of muscle mass,[26](#jah33368-bib-0026){ref-type="ref"} resulting in further functional decline[27](#jah33368-bib-0027){ref-type="ref"} and incident frailty.[28](#jah33368-bib-0028){ref-type="ref"} Therefore, the risks versus benefit of recommendations to lose weight need to be carefully considered in older men on a case‐by‐case basis. At the least, intentional weight loss in obese very old individuals should be accompanied by appropriate exercise and nutritional interventions to reduce or prevent the loss of lean mass and muscle strength and to preserve functional status.[29](#jah33368-bib-0029){ref-type="ref"}, [30](#jah33368-bib-0030){ref-type="ref"} It is uncertain why associations of central fat depots, including VAT, with incident ASCVD may weaken with age and may be different for older men versus older women. VAT is associated with metabolic risk factors such as high‐density lipoprotein cholesterol in both sexes, albeit somewhat more strongly in women.[31](#jah33368-bib-0031){ref-type="ref"} Moreover, higher levels of VAT on DXA were associated with lower high‐density lipoprotein cholesterol, higher triglycerides, greater insulin resistance, and lower adiponectin concentrations in this cohort of older men.[10](#jah33368-bib-0010){ref-type="ref"} However, the associations between VAT and inflammatory cytokines such as tumor necrosis factor alpha and interleukin‐6 are weak in older men.[10](#jah33368-bib-0010){ref-type="ref"} Whether or not the associations of inflammatory cytokines and adipocytokines with central adiposity weaken with advancing age, and whether or not changes of these associations with age varies by sex, is not known. It is also possible that the excess ASCVD risk conferred by central obesity simply becomes less important relative to other risk factors, known and unknown, associated with advancing age. Consistent with this, development and validation of the pooled ASCVD risk calculator of the American Heart Association and American College of Cardiology revealed that the associations of high‐density lipoprotein and total cholesterol with incident ASCVD weaken with advancing age in whites and blacks of both sexes.[21](#jah33368-bib-0021){ref-type="ref"} Similarly, the association of systolic blood pressure with incident ASCVD weakens with advancing age among black women.[21](#jah33368-bib-0021){ref-type="ref"} While we did not see that the association of VAT or android‐to‐gynoid fat mass ratio varied by age in these analyses, our cohort did not include men younger than age 65 years, and a larger age range may be necessary for this to be clearly seen. It is also possible that men who survive to be at least age 70 years have VAT that is less injurious to the arterial vasculature compared with younger men. VAT is not homogeneous. For example, there is individual variability in regard to its density (as measured in Hounsfield units on CT), and less dense VAT is more strongly associated with cardiometabolic risk factors.[31](#jah33368-bib-0031){ref-type="ref"} Whether or not this or other characteristics of central adipose tissue depots are different for middle‐age compared with older men, and/or vary by sex is unknown. Further investigations of the associations of VAT characteristics with age and sex are needed. Finally, it may be that the extent of atherosclerotic vascular damage may be more strongly associated with the duration of exposure to a high level of VAT, rather than a 1‐time value. VAT tends to increase with increasing age[32](#jah33368-bib-0032){ref-type="ref"}; hence, many men with a high level of VAT at the baseline MrOS sleep visit may not have had much VAT at earlier stages of their lives. If there is a tendency for middle‐aged men with lower levels of VAT to "catch up" to their peers who survive to advanced ages, that could weaken the association of a 1‐time VAT measurement at older ages with incident ASCVD events. There are several important strengths of our study. Our study was conducted in a large cohort of older men with comprehensively assessed measures of cardiovascular disease risk factors and obesity. While enrollment in MrOS was limited to community‐dwelling men, characteristics of the MrOS study population resemble those of the older men enrolled in the population‐based National Health and Nutritional Examination Survey.[33](#jah33368-bib-0033){ref-type="ref"} Incident ASCVD events were carefully adjudicated using a validated protocol. Our study had a large number (424) of outcome events and was well powered to detect small‐to‐modest associations of our main central adipose tissue measures with incident ASCVD events. There are also limitations of this study. The MrOS cohort enrolled only older men and therefore is not generalizable to women or younger men. Furthermore, 90% of the present study cohort was white, and this might also limit generalizability to other racial or ethnic groups. We cannot rule out the possibility that measures of VAT using CT or magnetic resonance imaging are associated with incident ASCVD in older men. However, we think that is unlikely since DXA‐VAT is highly correlated with both CT[34](#jah33368-bib-0034){ref-type="ref"}, [35](#jah33368-bib-0035){ref-type="ref"}, [36](#jah33368-bib-0036){ref-type="ref"}, [37](#jah33368-bib-0037){ref-type="ref"} and magnetic resonance imaging[34](#jah33368-bib-0034){ref-type="ref"}, [38](#jah33368-bib-0038){ref-type="ref"} measures of VAT, and since DXA‐VAT is as strongly associated with cardiometabolic ASCVD risk factors as CT‐VAT in older men.[10](#jah33368-bib-0010){ref-type="ref"} Finally, we ascertained prior diagnoses of comorbid conditions at baseline by self‐report, and did not have medical records available to confirm self‐reported diagnoses. In conclusion, central obesity, measured as VAT or android--gynoid fat mass ratio, was not associated with incident ASCVD events among this cohort of older community‐dwelling men. Sources of Funding {#jah33368-sec-0016} ================== The Osteoporotic Fractures in Men (MrOS) Study is supported by the National Institute on Aging (NIA), the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS), the National Center for Advancing Translational Sciences (NCATS), and NIH Roadmap for Medical Research (grant numbers: U01 AG027810, U01 AG042124, U01 AG042139, U01 AG042140, U01 AG042143, U01 AG042145, U01 AG042168, U01 AR066160, and UL1 TR000128). This study was also supported by an unrestricted grant from Hologic, Inc. Disclosures {#jah33368-sec-0017} =========== Taylor, PhD, MPH reports grants from NIH outside the submitted work. Schwartz, PhD received an unrestricted grant from Hologic, Inc, to measure VAT on all MrOS participant whole body DXA scans. Lewis, MD, MSPH reports grants from NIH, during the conduct of the study and grants from Novo Nordisk outside the submitted work. The remaining authors have no disclosures to report. Supporting information ====================== ###### **Table S1.** Comparison of Baseline Characteristics of Men Included in the Analyses vs Men Excluded From the Analyses **Table S2.** Multivariable Adjusted Associations of All Covariates (HR \[95% CI\]) With Major ASCVD Events, Adjusted for Each Other ###### Click here for additional data file.

Anterior cruciate ligament (ACL) reconstruction has evolved constantly since it was implemented in clinical practice more than 3 decades ago.^[@bibr8-2325967119832594]^ There have been many shifts in surgical trends, with the current focus mainly on anatomic reconstruction.^[@bibr29-2325967119832594]^ Intraoperative and postoperative imaging is often used to assist and assess graft placement.^[@bibr13-2325967119832594],[@bibr32-2325967119832594],[@bibr44-2325967119832594]^ Various imaging modalities and measurement methods are used to improve the reproducible assessment of postoperative graft placement.^[@bibr31-2325967119832594],[@bibr41-2325967119832594],[@bibr47-2325967119832594]^ There is no consensus on which modality to use in clinical practice. The matter is further complicated by the fact that different measurement methods are used on the various radiological modalities for the assessment of femoral graft placement.^[@bibr10-2325967119832594],[@bibr24-2325967119832594],[@bibr34-2325967119832594]^ The Bernard and Hertel grid is commonly used on radiographs and 3-dimensional computed tomography (3D-CT), with accurate measurements of graft placement in the high-low direction only possible on 3D-CT.^[@bibr31-2325967119832594]^ There is less controversy regarding tibial graft placement, as measurements are performed in a similar manner in all modalities.^[@bibr4-2325967119832594],[@bibr31-2325967119832594],[@bibr42-2325967119832594]^ The angle measurements of graft inclination or obliquity are mostly performed on magnetic resonance imaging (MRI).^[@bibr3-2325967119832594],[@bibr4-2325967119832594],[@bibr27-2325967119832594],[@bibr38-2325967119832594],[@bibr46-2325967119832594]^ Studies have examined the normal anatomic locations of ACL insertions, which are used to judge postoperative placement as either "in" or "out" of the anatomic range.^[@bibr1-2325967119832594],[@bibr9-2325967119832594],[@bibr11-2325967119832594],[@bibr19-2325967119832594],[@bibr25-2325967119832594],[@bibr34-2325967119832594],[@bibr35-2325967119832594]^ Several studies have also examined the normal angles of the native ACL.^[@bibr3-2325967119832594],[@bibr5-2325967119832594],[@bibr36-2325967119832594]^ In clinical practice, it is common to perform MRI before primary surgery and revision to assess the soft tissues to detect/identify concomitant meniscal injuries and other ligament injuries. However, some also suggest the use of CT and 3D-CT to aid surgery planning before revision.^[@bibr15-2325967119832594],[@bibr37-2325967119832594]^ In the clinical setting, it is not common to measure graft angles on CT or grids on MRI, and the methods are not used interchangeably. Recent studies show that there is little difference between angle measurements performed on MRI and CT and grid measurements performed on CT and 3D-MRI.^[@bibr10-2325967119832594],[@bibr14-2325967119832594]^ However, it is not known if both methods are equally adept in identifying the nonanatomic placement of grafts. The purpose of this study was to compare the angle and grid measurements of identifying anatomic versus nonanatomic tunnel placement on CT performed in patients undergoing ACL reconstruction. We hypothesized that the rate of nonanatomic placement would be higher in patients undergoing revision surgery compared with primary ACL reconstruction and aimed to assess the ability of the measurement methods to follow this hypothesis. Methods {#section1-2325967119832594} ======= The study was approved by a regional ethics board, and informed consent was waived. From January 2011 to the end of December 2017, all patients who were evaluated for revision ACL reconstruction and who underwent preoperative CT at a single institution were initially included retrospectively. Revision surgery was planned in patients who had an unsatisfactory function in the knee joint; examples of underlying causes were nonanatomic placement of graft tunnels, impingement due to grafts that were too long, or stretching of grafts during a new injury. Potential patients were identified through a manual search in our PACS system. From January 2011 to December 2015, we also included 100 postoperative CT scans obtained after primary reconstruction with either a hamstring graft or bone--patellar tendon--bone (BPTB) graft consecutively as separate groups. Postoperative CT was performed within 1 to 3 days after ACL reconstruction with the knee in full extension. Exclusion criteria were cases with multiple ligament reconstructions, known graft ruptures, or cases with previous revision. Graft ruptures were excluded, as graft angle measurements were not possible to perform without visible fibers on CT. In all cases, sex, age, and knee laterality were recorded. In addition, the type of surgical technique (anteromedial portal or transtibial) and type of graft used (hamstring or BPTB) were recorded. In the revision group, months since primary ACL reconstruction were also recorded. All CT examinations were performed in our institution on either a 64- or 512-detector CT machine (GE Healthcare). All images were acquired at a tube voltage of 100 kV and tube current of 80 mA with a 0.625-mm slice thickness and reconstructed in 3 planes in soft kernel and bone algorithms with 2 mm--thick slabs. In all cases, measurements of femoral tunnel placement were performed according to Bernard and Hertel, tibial tunnel placement was assessed according to Stäubli and Rauschning,^[@bibr42-2325967119832594]^ and ACL graft angles were measured in the coronal and sagittal planes. All measurements were performed and recorded by an experienced radiologist (\>15 years; A.P.P.) ([Figure 1](#fig1-2325967119832594){ref-type="fig"}). Normal ranges for the grid measurements were defined according to the literature: femoral deep-shallow, 24% to 37%; femoral high-low, 28% to 43%; and tibial anterior-posterior, 39% to 46%.^[@bibr33-2325967119832594]^ ![(A) Femoral tunnel measurement according to Bernard and Hertel, as depicted on 3-dimensional computed tomography (CT) after reconstruction with a hamstring graft. The graft tunnel center is 27% in the femoral deep-shallow direction and 35% in the femoral high-low direction (anatomic placement). (B) Tibial tunnel measurement according to Stäubli and Rauschning.^[@bibr42-2325967119832594]^ The graft tunnel center is 46% in the tibial anterior-posterior direction (anatomic placement). (C) Coronal angle measurement on CT (example of reconstruction with a bone--patellar tendon--bone graft), measured at 80° (nonanatomic placement). (D) Sagittal angle measurement, measured at 60° (nonanatomic placement).](10.1177_2325967119832594-fig1){#fig1-2325967119832594} The normal ranges for ACL coronal and sagittal angles were calculated from weighted means from the literature, as presented in [Table 1](#table1-2325967119832594){ref-type="table"}.^[@bibr3-2325967119832594],[@bibr4-2325967119832594],[@bibr36-2325967119832594]^ The normal coronal angle ranged from 66° to 74°, and the normal sagittal angle ranged from 47° to 59°. Graft placement in the coronal and sagittal planes was dichotomously recorded as "in" or "out" of the anatomic range. Within the revision group, the abovementioned analyses were also performed comparing the anteromedial portal and transtibial surgical approach. ###### Normal Ranges of Coronal and Sagittal Graft Angles  ------------------------------------------------------------------------------------------------------------- No. of Patients Coronal Angle, deg Sagittal Angle, deg ------------------------------------------------ ----------------- -------------------- --------------------- Ahn et al^[@bibr3-2325967119832594]^ (2007) 50 65.9 58.7 Ayerza et al^[@bibr5-2325967119832594]^ (2003) 30 -- 51.0 Reid et al^[@bibr36-2325967119832594]^ (2017) 188 74.3 46.9 Weighted mean (5th-95th percentile) \ \ 72.5 (66-74) 49.5 (47-59) ------------------------------------------------------------------------------------------------------------- Statistical Analysis {#section2-2325967119832594} -------------------- Dichotomous variables were assessed with the Pearson chi-square test. The combined assessments of grid in 2 directions or angles in 2 planes were classified in ordered categories (anatomic, partial anatomic, or nonanatomic), which were assessed with the weighted kappa. Continuous variables were assessed with the analysis of variance or Kruskal-Wallis test according to an assumed normality of data.^[@bibr48-2325967119832594]^ *P* \< .05 was considered significant, but the Bonferroni adjustment was used for multiple comparisons (*P* = .05, .017, or .08). All statistical analyses were performed with SPSS (v 25.0; IBM). Results {#section3-2325967119832594} ======= All of the primary ACL reconstructions were performed with the anteromedial portal approach. Within the study period, 137 patients were reviewed for revision surgery, of whom 6 were excluded from our study because of a ruptured graft and 14 were excluded because of multiple revisions. The final total of included revisions was 117 cases (REV group). Within the study period for primary ACL reconstruction, 100 cases of reconstruction with a hamstring graft (HAM group) and 91 cases of reconstruction with a BPTB graft (BPTB group) met the inclusion criteria. There were statistically significant differences in the distribution of sexes between the 3 groups, with more female patients in the REV group (62%) compared with the HAM (45%; *P* = .01) and BPTB (44%; *P* = .008) groups. There was no difference between laterality and age between groups ([Table 2](#table2-2325967119832594){ref-type="table"}). ###### Demographics of Study Groups*^a^*  Primary HAM (n = 100) Primary BPTB (n = 91) REV (n = 117)*^b^* *P* Value ------------------- ----------------------- ----------------------- -------------------- ----------------------- Age, y .037 (K-W)  Mean ± SD 29 ± 10 26 ± 10 29 ± 9  Median (range) 28 (14-54) 23 (14-53) 26 (15-55) Sex, n (%) **.009** *^c^* (χ^2^)  Female 45 (45) 40 (44) 73 (62)  Male 55 (55) 51 (56) 44 (38) Laterality, n (%) .588 (χ^2^)  Right 53 (53) 42 (46) 61 (52)  Left 47 (47) 49 (54) 56 (48) *^a^*Bolded *P* values indicate a statistically significant difference between groups (*P* \< .05). BPTB, bone--patellar tendon--bone; HAM, hamstring; K-W, Kruskal-Wallis; REV, revision anterior cruciate ligament reconstruction. *^b^*Months to revision surgery: mean ± SD, 50 ± 40; median (range), 36 (9-228). *^c^*Pairwise (overall ***P* \< .017**): HAM-BPTB: *P* = .88, HAM-REV: ***P*** **= .01**, BPTB-REV: ***P* = .008**. Grid Assessment {#section4-2325967119832594} --------------- In both the pairwise comparisons of grid measurements and in the rates of anatomic versus nonanatomic placement, there were significant differences in the femoral deep-shallow measurement between the HAM and BPTB groups and between the HAM and REV groups (*P* \< .001 for both), in which the mean graft placement in the REV group was shallower than in the HAM and BPTB groups (31% vs 24% and 28%, respectively). The nonanatomic rate in the HAM group was significantly worse than in the BPTB and REV groups (*P* ≤ .002 for both) (55% vs 20% and 34%, respectively). In the femoral high-low direction, for both the measurement and the rate of anatomic placement, there were significant differences between the HAM and REV groups and between the BPTB and REV groups (*P* \< .001 for all). A significant difference was seen in the tibial measurement between the HAM and REV groups but not in the comparison of the rate of tibial nonanatomic versus anatomic placement. The rate of anatomic placement according to the combined grid assessment (anatomic, partial anatomic, and nonanatomic) differed significantly between the BPTB and REV groups but not between the HAM and BPTB groups or between the HAM and REV groups ([Table 3](#table3-2325967119832594){ref-type="table"} and [Figure 2A](#fig2-2325967119832594){ref-type="fig"}). ###### Results of Grid Measurements*^a^*    Primary HAM (n = 100)*^b^* Primary BPTB (n = 91) REV (n = 117) *P* Value --------------------------------- ---------------------------- ----------------------- --------------- ------------------------- Femoral deep-shallow, % **\<.001** *^c^* (K-W)  Mean ± SD 24 ± 7 28 ± 5 31 ± 8  Median (range) 24 (7-49) 28 (18-47) 30 (11-56)  Graft placement, n (%) **\<.001** *^d^* (χ^2^)   Nonanatomic 55 (55) 18 (20) 40 (34)   Anatomic 45 (45) 73 (80) 77 (66) Femoral high-low, % **\<.001** *^e^* (K-W)  Mean ± SD 28 ± 9 30 ± 7 20 ± 14  Median (range) 29 (0-43) 30 (18-49) 20 (1-65)  Graft placement, n (%) **\<.001** *^f^* (χ^2^)   Nonanatomic 45 (45) 36 (40) 84 (72)   Anatomic 55 (55) 55 (60) 33 (28) Tibial, % **.010** *^g^* (ANOVA)  Mean ± SD 46 ± 6 47 ± 4 49 ± 8  Median (range) 46 (34-61) 48 (35-60) 48 (24-69)  Graft placement, n (%) .138 (χ^2^)   Nonanatomic 57 (58) 53 (58) 81 (69)   Anatomic 42 (42) 38 (42) 36 (31) Combined grid assessment, n (%) **\<.001** *^h^* (χ^2^)  Nonanatomic 10 (10) 4 (4) 20 (17)  Partial anatomic 79 (80) 67 (74) 89 (76)  Anatomic 10 (10) 20 (22) 8 (7) *^a^*Bolded *P* values indicate a statistically significant difference between groups (*P* \< .05). ANOVA, analysis of variance; BPTB, bone--patellar tendon--bone; HAM, hamstring; K-W, Kruskal-Wallis; REV, revision anterior cruciate ligament reconstruction. *^b^*n = 99 for tibial and combined grid assessment. *^c^*Pairwise (overall ***P* \< .017**): HAM-BPTB: ***P*** **\< .001**, HAM-REV: ***P*** **\< .001**, BPTB-REV: *P* = .11. *^d^*Pairwise (overall ***P* \< .008**): HAM-BPTB: ***P*** **\< .001**, HAM-REV: ***P*** **= .002**, BPTB-REV: *P* = .22. *^e^*Pairwise (overall ***P* \< .017**): HAM-BPTB: *P* = .413, HAM-REV: ***P*** **\< .001**, BPTB-REV: ***P*** **\< .001**. *^f^*Pairwise (overall ***P \<* .008**): HAM-BPTB: *P* = .447, HAM-REV: ***P*** **\< .001**, BPTB-REV: ***P*** **\< .001**. *^g^*Pairwise (overall ***P* \< .017**): HAM-BPTB: *P* = .620, HAM-REV: ***P*** **= .008**, BPTB-REV: *P* = .301. *^h^*Pairwise (overall ***P* \< .005**): HAM-BPTB: *P* \< .09, HAM-REV: *P* = .412, BPTB-REV: ***P*** **\< .001**. {#fig2-2325967119832594} Angle Assessment {#section5-2325967119832594} ---------------- The coronal angle measurement differed significantly between the HAM and BPTB groups (*P* \< .001) and between the BPTB and REV groups (*P* = .010), while the rate of anatomic versus nonanatomic placement differed between the HAM and BPTB groups (*P* = .001) and between the HAM and REV groups (*P* \< .001). The sagittal angle measurement did not differ between the 3 groups, but the rate of anatomic placement differed significantly between the BPTB and REV groups. The combined angle assessment (anatomic, partial anatomic, and nonanatomic) did not differ significantly between groups ([Table 4](#table4-2325967119832594){ref-type="table"} and [Figure 2, B and C](#fig2-2325967119832594){ref-type="fig"}). ###### Results of Angle Measurements*^a^*    Primary HAM (n = 100)*^b^* Primary BPTB (n = 91)*^c^* REV (n = 117) *P* Value ---------------------------------- ---------------------------- ---------------------------- --------------- ------------------------- Coronal angle, deg **\<.001** *^d^* (K-W)  Mean ± SD 72 ± 5 76 ± 5 74 ± 7  Median (range) 72 (59-86) 77 (53-86) 75 (51-87)  Graft placement, n (%) **\<.001** *^e^* (χ^2^)   Nonanatomic 44 (44) 61 (68) 81 (69)   Anatomic 56 (56) 29 (32) 36 (30) Sagittal angle, deg .019 (K-W)  Mean ± SD 65 ± 7 63 ± 5 62 ± 8  Median (range) 64 (51-89) 62 (53-74) 63 (27-82)  Graft placement, n (%) **.011** *^f^* (χ^2^)   Nonanatomic 83 (83) 67 (74) 76 (65)   Anatomic 17 (17) 24 (26) 41 (35) Combined angle assessment, n (%) .137 (χ^2^)  Nonanatomic 37 (37) 49 (54) 55 (47)  Partial anatomic 52 (52) 31 (34) 47 (40)  Anatomic 11 (11) 11 (12) 15 (13) *^a^*Bolded *P* values indicate a statistically significant difference between groups (*P* \< .05). BPTB, bone--patellar tendon--bone; HAM, hamstring; K-W, Kruskal-Wallis; REV, revision anterior cruciate ligament reconstruction. *^b^*n = 100 for combined angle assessment. *^c^*n = 91 for coronal angle and combined angle assessment. *^d^*Pairwise (overall ***P* \< .017**): HAM-BPTB: ***P*** **\< .001**, HAM-REV: *P* = .061, BPTB-REV: ***P*** **= .010**. *^e^*Pairwise (overall ***P* \< .008**): HAM-BPTB: ***P*** **= .001**, HAM-REV: ***P*** **\< .001**, BPTB-REV: *P* = .823. *^f^*Pairwise (overall ***P* \< .008**): HAM-BPTB: *P* = .115, HAM-REV: ***P*** **= .003**, BPTB-REV: *P* = .181. Agreement Between Measurement Methods {#section6-2325967119832594} ------------------------------------- The overall agreement, assessed with the weighted kappa, between the grid and angle measurements to identify anatomic, partial anatomic, or nonanatomic was low. The pairwise agreement between groups was also low ([Table 5](#table5-2325967119832594){ref-type="table"}). ###### Comparison of Grid Versus Angle Measurements*^a^*  Primary HAM (n = 100)*^b^* Primary BPTB (n = 91)*^c^* REV (n = 117) ----------------------------------- ---------------------------- ---------------------------- ------------------------- Grid vs angle assessment (95% CI)  Overall across groups 0.033 (--0.36 to 0.10)  Weighted kappa within group 0.009 (--0.11 to 0.127) 0.065 (--0.39 to 0.169) 0.041 (--0.74 to 0.156)  Pairwise kappa   HAM-BPTB 0.036 (--0.046 to 0.117)   HAM-REV 0.032 (--0.52 to 0.115)   BPTB-REV 0.046 (--0.035 to 0.128) *^a^*BPTB, bone--patellar tendon--bone; HAM, hamstring; REV, revision anterior cruciate ligament reconstruction. ^*b*^n = 99 for combined angle assessment. ^*c*^n = 91 for coronal angle and combined angle assessment. Comparison of Results Between Surgical Approaches {#section7-2325967119832594} ------------------------------------------------- Within the REV group, we further examined the differences between the transtibial and anteromedial portal approaches. In the femoral deep-shallow measurement and coronal angle measurement, there were no differences. In the femoral high-low measurement, tibial measurement, and sagittal angle measurement, there were significant differences (*P* \> .001, *P* = .001, and *P* = .002, respectively). The rate of anatomic graft placement according to the combined grid assessment also differed between the surgical approaches. However, no differences were observed in the combined angle assessment with regard to anatomic versus nonanatomic placement ([Table 3](#table3-2325967119832594){ref-type="table"} and [Figure 3](#fig3-2325967119832594){ref-type="fig"}). The agreement, assessed with the weighted kappa, between the grid and angle measurements to identify anatomic, partial anatomic, or nonanatomic placement was low in both the transtibial and anteromedial portal approach subgroups ([Table 6](#table6-2325967119832594){ref-type="table"}). {#fig3-2325967119832594} ###### Comparison of Transtibial Versus Anteromedial Portal Approach Within REV Group*^a^*  Anteromedial Portal (n = 66) Transtibial (n = 51) *P* Value ----------------------------------- ------------------------------ --------------------------- ------------------ Femoral deep-shallow, % .611 (K-W)  Mean ± SD 31 ± 9 31 ± 7  Median (range) 31 (11 to 56) 30 (20 to 51)  Graft placement, n (%) .864 (χ^2^)   Nonanatomic 23 (35) 17 (33)   Anatomic 43 (65) 34 (67) Femoral high-low, % **\>.001** (K-W)  Mean ± SD 24 ± 12 14 ± 15  Median (range) 25 (0 to 45) 12 (--1 to 65)  Graft placement, n (%) **.001** (χ^2^)   Nonanatomic 39 (59) 45 (88)   Anatomic 27 (41) 6 (12) Tibial, % **.001** (K-W)  Mean ± SD 46 ± 8 52 ± 7  Median (range) 47 (24 to 61) 52 (38 to 69)  Graft placement, n (%) **.021** (χ^2^)   Nonanatomic 40 (60) 41 (80)   Anatomic 26 (40) 10 (20) Combined grid assessment, n (%) **.004** (χ^2^)  Nonanatomic 10 (15) 10 (20)  Partial anatomic 48 (73) 41 (80)  Anatomic 8 (12) 0 (0) Coronal angle, deg .398 (K-W)  Mean ± SD 73 ± 7 74 ± 6  Median (range) 75 (53 to 87) 75 (52 to 86)  Graft placement, n (%) .082 (χ^2^)   Nonanatomic 50 (75) 31 (61)   Anatomic 16 (25) 20 (39) Sagittal angle, deg **.002** (K-W)  Mean ± SD 60 ± 8 65 ± 8  Median (range) 60 (27 to 73) 65 (49 to 82)  Graft placement, n (%) **.022** (χ^2^)   Nonanatomic 37 (56) 39 (76)   Anatomic 29 (44) 12 (34) Combined angle assessment, n (%) .639 (χ^2^)  Nonanatomic 29 (44) 26 (51)  Partial anatomic 29 (44) 18 (35)  Anatomic 8 (12) 7 (14) Grid vs angle assessment (95% CI)  Weighted kappa within approach 0.074 (--0.82 to 0.23) --0.006 (--0.86 to 0.163)  Overall across both approaches 0.041 (--0.74 to 0.156) *^a^*Bolded *P* values indicate a statistically significant difference between approaches (*P* \< .05). K-W, Kruskal-Wallis; REV, revision anterior cruciate ligament reconstruction. Discussion {#section8-2325967119832594} ========== The purpose of our study was to compare grid measurements and angle measurements with regard to the anatomic placement of grafts after ACL reconstruction. The major finding of our study was the lack of agreement between the 2 measurement methods in identifying anatomic graft placement. Technical errors, such as nonanatomic graft placement, are considered a common cause of graft failure, so one would expect the rate of nonanatomic placement to be higher in patients with failed ACL grafts.^[@bibr28-2325967119832594],[@bibr30-2325967119832594]^ The literature suggests that femoral tunnel placement is more important for a favorable outcome compared with tibial tunnel placement.^[@bibr6-2325967119832594],[@bibr28-2325967119832594]^ The results of the grid assessment confirmed this assumption, with no difference between the 3 groups in tibial tunnel placement but a difference in the overall femoral tunnel placement. Placement in the femoral deep-shallow direction did not differ significantly between the BPTB and REV groups. This may be explained by the fact that the tunnel aperture on CT in the BPTB group may not represent the true center of the graft, as the bony attachment is a few millimeters thick, thus slightly influencing the grid measurement ([Figure 4](#fig4-2325967119832594){ref-type="fig"}). Surprisingly, the rate of nonanatomic placement in the femoral deep-shallow direction was highest in the HAM group. In the femoral high-low direction, the nonanatomic placement rate was significantly higher in the REV group compared with the HAM and BPTB groups, as was to be expected. This finding might indicate that anatomic placement in the high-low direction in the femur is more important for outcomes than anatomic placement in the deep-shallow direction and that the surgical technique should aim to avoid nonanatomic placement in the high-low direction. {#fig4-2325967119832594} No difference in the groups was observed in tibial graft placement. Regarding overall nonanatomic graft placement, the highest rate was observed in the REV group (17% in REV vs 10% in HAM and 4% in BPTB). Pairwise comparisons showed no difference between the HAM and BPTB groups, as was expected. However, there was also no difference between the HAM and REV groups. This may be explained by the high rate of nonanatomic placement in the HAM group in the femoral deep-shallow direction, influencing the overall anatomic rate in the grid measurements. There was a significant difference in the combined grid assessment between the BPTB and REV groups. Angle measurements for assessing postoperative graft placement are often recommended in the literature.^[@bibr12-2325967119832594],[@bibr13-2325967119832594],[@bibr30-2325967119832594],[@bibr47-2325967119832594]^ In the coronal measurements, we found significant differences between the HAM and BPTB groups and between the BPTB and REV groups. In general, in the BPTB group, we observed a steeper (nonanatomic) coronal angle than in the other groups. The reason for this may again be the bony attachment of the BPTB graft, which when placed caudally in the tunnel will cause a more cranial exit for the tendon, causing it to run a steep slope in the coronal view. There were no differences in the sagittal angle measurements or in the combined angle assessment for anatomic placement between groups. The highest rate of overall nonanatomic placement was seen in the BPTB group (54% in BPTB vs 37% in HAM and 47% in REV). The 2 grid and the angle methods yielded significantly different rates of nonanatomic placement in the same patients. The explanation for this is a fundamental difference in measurement methods. The angle measurements were devised in the era of the transtibial surgical technique. Technical failure with the transtibial technique was related to "too high" placement of the femoral tunnel and was easily assessed on sagittal images, and a too steep graft angle was introduced as an imaging criterion.^[@bibr3-2325967119832594],[@bibr39-2325967119832594],[@bibr45-2325967119832594]^ However, it is known that flexion affects the ACL angle in the sagittal plane. A study showed that the sagittal angle of the ACL ranges from 45° to 20° with increasing knee joint flexion.^[@bibr16-2325967119832594]^ This factor affects the measurements in a clinical setting, as even the slightest flexion during CT or MRI will change the angle of the ACL graft. Furthermore, even if the angle is correct, the placement may still be faulty if the graft is placed too anteriorly or too posteriorly.^[@bibr32-2325967119832594]^ The grid measurements were devised so that the measurements in the femur and tibia are independent of the degree of knee flexion. Thus, methodological discrepancies relating to knee flexion may explain the poor agreement between the 2 measurement methods. CT is the more robust modality if one chooses to measure graft placement. Considering revision ACL reconstruction, previous studies have shown that the anteromedial portal technique yields higher rates of anatomic placement compared with the transtibial technique.^[@bibr39-2325967119832594],[@bibr45-2325967119832594]^ The transtibial technique is known to cause too high placement in the grid measurements, which was confirmed in our study.^[@bibr17-2325967119832594],[@bibr43-2325967119832594]^ The combined grid assessment differed significantly between the surgical techniques (*P* = .004), with no anatomic cases in the transtibial technique group. In addition, the sagittal angle differed between the surgical techniques (*P* = .002), while the coronal angle measurements and combined angle assessment did not differ within the REV group. Thus, the lack of agreement between the grid and angle measurement methods was also observed within the REV group. The clinical usefulness of (1) postoperative CT in primary reconstruction to improve a surgeon's learning curve and to serve as a baseline examination and (2) preoperative CT for planning revision surgery has been established.^[@bibr20-2325967119832594],[@bibr40-2325967119832594],[@bibr49-2325967119832594]^ MRI undoubtedly has a role in planning revision surgery for identifying recurrent ACL graft ruptures and missed concomitant lesions in other ligaments, menisci, or articular cartilage.^[@bibr2-2325967119832594],[@bibr22-2325967119832594],[@bibr37-2325967119832594]^ Ducouret et al^[@bibr10-2325967119832594]^ found that angle measurements did not differ between CT and MRI and suggested that MRI can be used to replace CT for identifying tunnel placement. Grasso et al^[@bibr14-2325967119832594]^ performed grid measurements on CT and MRI; however, the measurements were conducted on computer-generated models after adding digitized information acquired during revision surgery, not on actual CT or MRI scans. In our view, the clinical usefulness of angle and grid measurements on MRI has not been sufficiently established to date. No studies have examined the clinical benefit of angle measurements in reconstruction using the anteromedial portal technique. Furthermore, our results show that graft angles do not correlate with grid measurements, and they overestimate nonanatomic placement in ACL reconstruction with the anteromedial portal technique. Therefore, MRI currently cannot replace CT to identify anatomic graft placement in ACL reconstruction. This study highlights the problems that arise because of the lack of consensus on which measurement method to use when assessing ACL graft placement. Several studies have compared graft placement between the transtibial and anteromedial portal techniques, but the studies used several different measurement methods to assess graft placement.^[@bibr7-2325967119832594],[@bibr21-2325967119832594],[@bibr26-2325967119832594],[@bibr46-2325967119832594]^ This makes a comparison of surgical results difficult, as we now know that the reported rate of nonanatomic tunnel placement varies depending on the method used. This is the first study comparing grid and angle measurement methods after ACL reconstruction. We have laid bare the major discrepancy between these methods. As previous studies have shown low interrater and intrarater variability in both methods used in our study, we did not assess interrater variability and do not consider this a major limitation.^[@bibr18-2325967119832594],[@bibr23-2325967119832594]^ The normal ranges of grid and angle measurements are based on a relatively high number of anatomic and imaging cases (\>200-300).^[@bibr3-2325967119832594],[@bibr5-2325967119832594],[@bibr33-2325967119832594],[@bibr36-2325967119832594]^ This limits bias in identifying the appropriate cutoff in our study. As the purpose of our study was to compare 2 methods used for assessing anatomic tunnel placement on imaging, we did not correlate graft placement with clinical or functional assessments of graft laxity and cannot determine whether nonanatomic tunnel placement affects graft laxity. Conclusion {#section9-2325967119832594} ========== The agreement between angle and grid measurements to identify anatomic ACL graft placement was very low. Compared with grid measurements, angle measurements tended to overestimate nonanatomic tunnel placement. Grid measurements were better in identifying malpositioned ACL grafts. Orthopaedic surgeons and radiologists ought to be aware of the pitfalls of the angle measurement method when assessing ACL graft placement on imaging. The authors declared that there are no conflicts of interest in the authorship and publication of this contribution. AOSSM checks author disclosures against the Open Payments Database (OPD). AOSSM has not conducted an independent investigation on the OPD and disclaims any liability or responsibility relating thereto. Ethical approval for this study was obtained from Regional Committees for Medical and Health Research Ethics (No. 2017/2434/REK sør-øst).

Background ========== Several different bioinformatics technologies exist to quantify gene expression. Regardless of technological platform, laboratory assays of gene expression first extract mRNA from a test sample and a control sample. These samples may be labeled with a tag or dye and hybridized to amplified cloned sequences that represent a gene of interest. The amount of mRNA in each sample is usually measured by examining the amount of dye remaining after hybridization. Researchers use Q-RT-PCR to measure expression when there are only one or a few genes of interest. Several lab protocols from various companies exist to quantify gene expression such as RT-PCR assays using intercalating dyes like SYBR Green, the TaqMan Gene Expression Assays, LightCycler, and QuantiGene \[[@B1]-[@B3]\]. When genome-wide levels of expression are of interest, microarrays can measure expression for thousands of genes of interest. Microarray platforms employ either cDNA clones \[[@B4],[@B5]\] or *n*-mer oligonucleotide probes for many genes at once \[[@B6]\]. More recently, sequence-based technologies provide more efficient and accurate expression measurements on a genome-wide scale. Evolving from early techniques such as Serial Analysis of Gene Expression (SAGE) to modern techniques such as Massively Parallel Signature Sequencing (MPSS) and RNA Sequencing (RNA-Seq), these approaches now rival microarray-based gene expression analysis for efficiency, cost, and accuracy \[[@B7]\]. Sequence-based techniques are also more flexible, allowing for gene expression measurements on a genome-wide level from any organism with a published genome sequence \[[@B8]\]. Sequencing employs systems such as the 454 or Illumina platform with the latter demonstrating greater depth and coverage \[[@B9]\]. To illustrate the central motive of this paper, Figure [1](#F1){ref-type="fig"} demonstrates a two-color competitive hybridization assay of the kind used in TaqMan assays and cDNA microarrays. Other methods involve single-dye hybridization systems or intercalating dyes that bind to double-stranded DNA (dsDNA) product. The statistical models proposed below can be generalized to any scenario where gene expression is measured comparatively in a test sample and a reference sample. {#F1} Researchers commonly use the log~2~-ratio to measure relative mRNA expression between two samples. The estimate is as follows. Let *R~ij~*represent a summary expression value for gene *j*in the reference sample *i*where *i*= 1,\..., *n*and *j*= 1,\..., *K*. Let *G~ij~*represent a summary expression value for gene *j*in the test sample *i*. The value *n*is the number of paired samples or experiments and *K*is the number of genes studied. To summarize relative expression between two samples, the log~2~-ratio is or other similar variants on the theme. The log~2~-ratio is commonly interpreted as the average \"log-fold-change\" in gene expression between the reference sample and the test sample. Its estimate will be denoted by . If *r~j~*= 1, then the ratio between the two samples is 2^1^= 2, meaning that the expression of gene *j*in the test sample is two-fold that of the reference sample on average. If *r~j~*= 2, then the ratio between the two samples is 2^2^= 4, meaning that on average the expression in the test sample is four-fold that of the reference sample. Other values of *r~j~*are interpreted similarly. While the interpretation of the log~2~-ratio is appealing, the statistic has an important drawback. When expression in the reference sample is low, is numerically unstable because the denominators *R~ij~*are small. As *R~ij~*approaches zero, *r~j~*increases drastically, approaching infinity. When *R~ij~*= 0, then *r~j~*is undefined. Thus, when reference sample expression is low, we get extreme estimates or missing values for *r~j~*. This phenomenon is especially common when measuring gene expression in simple organisms. In bacteria, for example, transcription may be binary; either on or off. The log~2~-ratio is least reliable for these systems. This problem persists in human genomics research for certain experimental conditions and genes of interest. This article proposes a new estimate to compare mRNA expression in two samples. This estimate is the proportion of mRNA in the test sample *p~j~*for each gene. The proportion takes the amount of mRNA in the test sample and compares it to the total amount of expressed mRNA represented by the sum of the test and reference samples. One formula for estimating the proportion is The proportion is well-defined for all values of *R~ij~*and *G~ij~*. For example, when *R~ij~*= 0, then *G~ij~/*(*G~ij~*+ *R~ij~*) = 1. We can interpret the number as follows: mRNA expression is observed only in the test sample and not in the reference sample. Similarly, if *G~ij~*= 0, then *G~ij~/*(*G~ij~*+ *R~ij~*) = 0 and this means that mRNA expression is observed in the reference sample only. Figure [2](#F2){ref-type="fig"} demonstrates the relationship between the log~2~-ratio and the proportion estimates, which follows a logistic function. The relationship is roughly linear near the center point but non-linear at the extreme values. A detailed description for the estimate of the proportion, , and an alternative derived from a maximum likelihood estimate, is in the Results section. {#F2} More generally, *p~j~*can be interpreted as the proportion of mRNA from gene *j*expressed in the test sample. As *p~j~*deviates from 0.5, then there is differential expression between the test and reference samples. As *p~j~*approaches one, then gene *j*is up-regulated in the test sample. As *p~j~*approaches zero, then gene *j*is down-regulated in the test sample. The proportion statistic *p~j~*can also be transformed into a percentage: *p~j~*× 100% for reporting. For example, if *p~j~*= 0.75 then we can say that 75% of the mRNA expressed in the experiment comes from the test sample. The proportion estimate can easily be used to test for differential expression between groups. Under the null hypothesis of no gene expression, *p~j~*= 0.5. The alternative hypothesis is differential expression, *p~j~*≠ 0.5. The log~2~-ratio estimate requires a different hypothesis test. Under the null hypothesis, *r~j~*= 0 and under the alternative, *r~j~*≠ 0. Using a proportion *p~j~*to describe relative expression for gene *j*instead of the log~2~-ratio *r~j~*maintains the ability to interpret differential expression and test for differences. The added benefit of the proportion is the ability to preserve all data points, even for experiments with very low expression values. Typically when values of *R~ij~*are very small, researchers eliminate the *j^th^*probe of the *i^th^*experiment from their analysis. Eliminating missing data results in a loss of information and potential bias and loss of power. The proportion estimate does not require the removal of extreme, but legitimate, data points. The Results section provides details that describe the estimation of statistics for *p~j~*. The section also provides several test statistics for hypothesis tests of *p~j~*. Estimation and testing are developed in the frequentist context but the Bayesian context can also be used, as described in the Appendix. The Results section compares the testing scenarios in simulations and two datasets. The first dataset consists of expression data from a cDNA microarray platform and the second dataset uses RNA-Seq. Both the log~2~-ratio and proportion statistics achieve roughly equivalent results under usual conditions, but one of the proportion statistics performs better across a variety of distributional assumptions. Proportion statistics also detect differentially expressed genes that would typically be classified as missing data. Results ======= Parameter Estimates and Hypothesis Testing ------------------------------------------ We propose a new strategy for the comparison of expression values that is tied to the underpinnings of the hybridization process and its natural interpretation using a binomial distribution. Figure [1](#F1){ref-type="fig"} illustrates the hybridization process in a way that justifies the use of a binomial distribution. The description is specific to a two-color hybridization platform. The same concept extends to any system where both test samples and reference samples are assayed. For each gene sequence, suppose that researchers amplify sequences resulting in *M~ij~*clones, where *j*is the gene probe index and *i*is the sample number, in order to co-hybridize the extracted mRNA sequences from the reference and test samples. Usually *M~ij~*is in the millions, but the exact value will be unknown. For each probe, suppose it hybridizes to a test target with probability *p~j~*and to a reference target with probability 1 - *p~j~*. This reflects the proportion of available test sequences versus references sequences. We assume that each probe must hybridize to mRNA extracted from either the test or reference sample. Then, the number of hybridizing test target sequences *Y~ij~*follows a binomial distribution with size *M~ij~*and probability *p~j~*. We wish to estimate *p~j~*to calculate the proportion of hybridized test target sequences. The maximum likelihood estimate for *p~j~*is . In this scenario, *Y~ij~*= *G~ij~*and *M~ij~*= *G~ij~*+ *R~ij~*where *R~ij~*represents the expression value for gene *j*in the reference sample *i*and *G~ij~*represents an expression value for gene *j*in the test sample *i*when there are *i*= 1,\..., *n*paired experiments and *j*= 1,\..., *K*genes. Therefore, to summarize *n*experiments the estimated proportion for each gene *j*is To test for differential expression we set up a decision with the null hypothesis *H*~0~: *p*= 0.5 versus the alternative hypothesis *H*~1~: *p*≠ 0.5. The test derived for this binomial distribution has a test statistic The test statistic *z~j~*is compared to a quantile from the normal distribution *z*~1-α/2~. If the type I error is *α*= 0.05, then *z*~0.975~= 1.96. If *\|z~j~\| \>*1.96, then gene *j*is declared differentially expressed between test and reference samples. The *z*~1-α/2~quantile is replaced by a quantile when the variance estimate uses instead of *p~j~*. This test of binomial proportions, however, is not robust to deviations from the binomial distribution. Indeed, we do not believe that expression data will always follow a binomial distribution, but we include this derivation to motivate the choice of this statistic. Instead, we recommend an alternative test statistic that can be used whether distributional assumptions are met or not. The alternative test statistic simply uses a normal approximation to the binomial distribution and calculates a sample variance estimate. Then the test statistic for differential expression is where we estimate the proportion and the sample variance in the usual way, . If \|*t~j~*\| \>*t*~1-*α*/2,*n*-1~, then gene *j*is differentially expressed between test and reference samples. This test is valid for sufficiently large sample sizes (see Table [1](#T1){ref-type="table"}). ###### Simulation comparing test statistics for , , , limma/EBA, edgeR, and DESeq with a sample size of *n*= 20 under four distributional assumptions. Exponential Poisson Binomial Normal --------- ------------- --------- ---------- -------- ------- ------- ------- ------- 0.051 0.742 0.004 0.116 0.047 1.000 0.050 1.000 \+ 0.05 0.051 0.742 0.038 0.757 0.047 1.000 0.050 1.000 \+ 0.5 0.051 0.742 0.044 0.943 0.047 1.000 0.050 1.000 0.975 1.000 0.045 1.000 0.048 1.000 0.003 1.000 0.055 0.773 0.047 0.881 0.047 1.000 0.050 1.000 0.975 1.000 0.045 1.000 0.048 1.000 0.003 1.000 EBA 0.051 0.781 0.048 1.000 0.047 1.000 0.052 1.000 edger NA NA 0.033 1.000 0.014 1.000 NA NA DESeq NA NA 0.042 1.000 0.047 1.000 NA NA The exponential distribution has rate parameter 1/4000, the Poisson has rate parameter 3, the binomial has size 10000; and the normal has mean 10 and standard deviation 2. Each entry is proportion of times the null hypothesis was rejected at *α*= 0.05, out of 1000 simulations. The null hypothesis of no differential expression is equivalent to a fold change of one (*fc*= 1). When the fold change is three, we are calculating the power to detect differential expression (*fc*= 3). Tables for other distributional parameters may be found in Additional file [1](#S1){ref-type="supplementary-material"}. These tables also include a greater range of sample sizes and fold changes. Calculating corresponding confidence intervals for each of the test statistics above is straightforward. Previous research suggests adjusting confidence intervals for binomial proportions. The most popular adjustment of these intervals uses the Agresti-Coull procedure \[[@B10],[@B11]\]. We recommend this procedure to estimate confidence intervals for both of the proportion estimates above. The proposed statistics are evaluated within a frequentist framework. A Bayesian framework is provided in the Appendix. Simulation Results ------------------ We ran a series of simulations to compare the inference behavior of proportion based statistics, and , to log-ratio based statistics, and . The proportion statistics and are introduced in equations 2 and 3 above and their test statistics are given in equations 4 and 5. The ratio-based statistics that have been used in the literature previously are described in equation 1 () and equation 6 in the Methods section (). In preliminary simulation exercises, we found that the performance of some test statistics was heavily dependent on the distribution used to generate the expression data. Thus, we generated expression data under four different distributions. The simulation results in Table [1](#T1){ref-type="table"} present a subset of the sample sizes and fold changes examined. More extensive tables are in Additional file [1](#S1){ref-type="supplementary-material"}. The performance of the estimators under four different distributions is summarized in Table [2](#T2){ref-type="table"}. The table was created after examination of the empirical type I error and power for each statistic in each simulation (Additional file [1](#S1){ref-type="supplementary-material"}). The proportion performs at or above the others with the exception of for the Poisson, although still performs adequately there (cp. Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}). The statistics and the DESeq analysis exhibit unacceptable performance under one or more of the distributional assumptions. Statistics , and limma/empirical Bayes (EBA) and edgeR analysis are always good or acceptable, depending on the distributional assumptions. The edgeR and DESeq analyses have type I errors less than 0.05 in many instances. On the average, the and EBA tests have moderately better type I error and power than , (Additional file [1](#S1){ref-type="supplementary-material"}). Although might be expected to outperform the other methods under the binomial assumption, detection under this assumption is easy and all methods performed equally well. In conclusion, the statistic and limma/EBA have the best inference in our simulations overall. The empirical Bayes approach results in better power than on the average under the Poisson and normal distributions. ###### Comparison of estimators from the simulations. Good Acceptable Unacceptable ------------- --------- ---------------- -------------- Exponential , *EBA* Poisson , *EBA* , edgeR, DESeq Binomial , EBA edgeR DESeq Normal , EBA Four estimators (, and ) and three methods (EBA, edgeR, and DESeq) were used under four distributional assumptions (Exponential, Poisson, Binomial, and Normal). The performance rating (Good, Acceptable, Unacceptable) was judged on the basis of Type I error and power. See Table 1 for an example of the estimators and why the ratings were judged as shown. Additional data for judging the ratings is given in Additional file [1](#S1){ref-type="supplementary-material"}. Analysis of Gene Expression in Mice with apoAI Knockout ------------------------------------------------------- To examine the performance of our method on cDNA microarray data, we analyzed the expression values reported in Ge et al (2003) \[[@B12]\]. Since the apoAI experiment was a control-treatment experiment that used a third sample as a reference, this data exhibits how the methods of this paper can be extended to the case of a difference of two proportions. When testing for the difference between control and treatment, the p-values from and were very similar in magnitude. This was true for both raw p-values and p-values adjusted for multiple-testing. The order of the p-values was also similar, but not identical (see Figure [3](#F3){ref-type="fig"}). When using the limma/EBA method, the p-values from and were again similar in magnitude, although the order varied more after the 7th probe (Table [3](#T3){ref-type="table"}). The top 8 most differentially expressed probes from the original analysis differed from those selected by using t-statistics in the 8th probe, although the top 9 probes for both sets are the same (Table [3](#T3){ref-type="table"}). In the original analysis, the top 8 probes corresponded to four distinct genes, and were confirmed by real time quantitative PCR \[[@B13]\]. {#F3} ###### Table of raw p-values for Welch t-statistics and limma methods using and from the apoAI control treatment expression data. rank p-value () rank p-value () rank p-value(limma ()) rank p-value(limma ()) ---------- --------------- ---------- ------------------------- ---------- ------------------------- ---------- ------------------------- 1 7.3 × 10^-7^ 1 4.2 × 10^-6^ 1 3.8 × 10^-12^ 1 1.5 × 10^-9^ 2 2.4 × 10^-5^ 2 2.4 × 10^-5^ 4 5.2 × 10^-7^ 4 1.3 × 10^-6^ 3 3.4 × 10^-5^ 4 4.0 × 10^-5^ 2 2.6 × 10^-8^ 3 1.2 × 10^-7^ 4 5.0 × 10^-5^ 3 2.8 × 10^-5^ 3 5.1 × 10^-8^ 2 7.6 × 10^-8^ 5 1.0 × 10^-4^ 6 1.2 × 10^-4^ 5 1.4 × 10^-6^ 6 7.5 × 10 ^-6^ 6 1.0 × 10^-4^ 5 5.8 × 10^-5^ 7 9.6 × 10^-6^ 7 1.2 × 10^-5^ 7 2.9 × 10^-4^ 7 2.7 × 10^-4^ 12 **4**.**5**× **10**^-5^ 11 **3**.**7**× **10**^-5^ 8 5.9 × 10^-4^ 9 **6**.**4**× **10**^-4^ 8 **1**.**3**× **10**^-5^ 16 **4**.**8**× **10**^-5^ 9 7.4 × 10^-4^ 8 **5**.**8**× **10**^-4^ 10 **1**.**5**× **10**^-4^ 61 **3**.**3**× **10**^-4^ 10 1.3 × 10^-3^ 10 1.1 × 10^-3^ 6 **2**.**0**× **10**^-6^ 5 **1**.**4**× **10**^-6^ **rank** **t-stat ()** **rank** **t-stat ()** **rank** **t-stat(limma ())** **rank** **t-stat(limma ())** 1 -16.5 1 -12.8 1 -23.1 1 -14.4 2 -9.8 2 -9.8 4 -9.0 4 -8.3 3 -9.3 4 -9.1 2 -11.5 3 -10.1 4 -8.8 3 -9.6 3 -10.9 2 -10.5 5 -7.9 6 -7.7 5 -8.2 6 -7.0 6 -7.8 5 -8.5 7 -6.7 7 -6.7 7 6.6 7 6.7 12 5.9 11 6.0 8 -5.9 9 -5.8 8 -6.7 16 -5.9 9 -5.7 8 -5.9 10 -5.2 61 -4.8 10 5.2 10 5.3 6 8.0 5 8.2 The limma method was developed for log-ratios, not proportions, but we show the results using proportions for comparison. The first group of ten rows are the p-values and the second group is the t-statistics, for reference. In the original paper, the top 8 probes were selected using the maxT multiple testing procedure with using Welch t-statistics on in \[[@B12]\]. This selection is the first column, ranked 1 through 10. The p-values/t-statistics for the probes using and limma correspond to the first column, with their ranks shown. Using t-statistics with , the selection is similar to , but not identical, since probes ranked 8th and 9th switch places. Using limma with and , the selection begins to vary widely at the 7th probe. When using , there were 158 (2.5%) unanalyzable probes because one or more of the samples had both *G~ij~*= 0 and *R~ij~*= 0, which made undefined. The statistic was defined for all probes because *G~ij~*and *R~ij~*were never zero for all samples of a specific probe. For this data, none of the 158 unanalyzed probes were in the top eight when using , although if they were a potential discovery they would have been missed using . To avoid this problem, one may add an arbitrary constant to all probes before taking the log-ratio. If merely raw p-values were selected at *α*= 0.05, then would have selected 850, but there would have been 9 more significant p-values if an arbitrary 0.05 were added to the data to avoid zero denominators when using log-ratios. By comparison, would have selected 871 probes. Therefore, is able to give comparable results to for this cDNA microarray experiment, with the slight advantage that it provided information for 158 more probes in the study, without an arbitrary constant. Analysis of Differential Expression in Human Kidney and Liver Cells ------------------------------------------------------------------- To examine the performance of our methods on RNA-Seq data, we analyzed the expression values reported in Marioni et al (2008) \[[@B9]\]. This data compared the expression of human kidney and liver cells sampled from the same person. Concentrations of 3 pM of cDNA were sequenced using the Illumina platform in five lanes. The original paper analyzed the expression of 32,000 sequences and reported that 11,493 of the sequences were differentially expressed with q-values less than 0.001 (FDR \< 0.1%) \[[@B14]\]. Supplemental Table [3](#T3){ref-type="table"} from Marioni et al (2008) provides the results of 17,708 sequences analyzed with both RNA-Seq technology and Affymetrix microarrays. They reported that 8,113 of Affyymetrix probe sets were differentially expressed with q-values less than 0.001. In order to compare the methods in the original paper to those we are proposing, we used a type I error rate of *α*= 0.05/32000 for all tests. In this way, the threshold can be universally applied to all genes and methods while controlling the genomewide error rate. Table [4](#T4){ref-type="table"} shows the total number of significant genes detected for all methods as well as the overlap between each. The least powerful method to detect differential expression was the test of while the most powerful method was the test of . Of our proposed methods, the statistic gave conclusions that overlapped most with the original methods in the paper, the likelihood ratio test (LRT) based on the maximum likelihood estimate \[[@B9]\]. In fact, found all of the differentially expressed genes found by the LRT. Tests conducted using the edgeR package gave the second largest overlap. Comparing the proposed methods with the Affymetrix data reported in the original paper \[[@B9]\], the most overlap in calls was with the tests, followed by the LRT using . If we look at the most recent methods developed for RNA-Seq data, the results of the DESeq package overlap most with , followed by the edgeR package, the LRT and then . ###### Significant genes detected from the dataset in Marioni et al (2008). Significant Genes in Intersecting Sets ---------------------------------------- ---------------------- ----------- ------------------- ----------- ----------- ----- ------- ------ **Method** **EBA (Affymetrix)** **(LRT)** **EBA (RNA-Seq)** **edgeR** **DESeq** EBA (Affymetrix) 3641 3096 672 3127 3037 251 3183 960 (LRT) 8641 790 8461 5400 308 8641 1116 EBA (RNA-Seq) 790 790 789 275 790 700 edgeR 8697 5516 308 8697 1351 DESeq 7083 301 5644 1305 315 308 315 11915 3324 3331 For each row and column, the number gives the significant gene calls by both methods. The cut-off to determine significance was set at *α*= 0.05/32000. The acronym LRT denotes the likelihood ratio test based on the Poisson distribution, as described in the Methods. The acronym EBA denotes the empirical Bayes analysis performed on both the Affymetrix and RNA-Seq data. Log~2~-ratio estimates produced much missing data in the RNA-Seq data analysis, with 8,947 sequences eliminated from analysis. Of these missing sequences, the proposed methods detected 4,171 () or 2,406 () significant calls (see Table [5](#T5){ref-type="table"}). This means that using a test based on the log~2~-ratio may miss many possibly important differentially expressed genes. When using a LRT as suggested in Marioni et al (2008), 3,979 sequences were omitted from analysis because of missing data. The edgeR and DESeq packages did not generate any missing data. Of the 3,979 missing values generated by the LRT, our proposed methods detected 3,064 () or 2,208 () additional differentially expressed genes while the edgeR and DESeq packages detected an additional 235 and 197 genes. Since the true differential expression in this data is unknown, the differences between these methods are intriguing, but it is not clear whether one method is more accurate than another in this analysis. Overall, these findings suggest that test statistics based on a proportion statistic do not result in missing data, and more importantly, can detect possibly important differentially expressed genes that the log~2~-ratio based methods would miss. ###### A summary of the missing values for each of the tests and the number of significant genes detected by other methods within those missing values. Significant Genes from Sets of Missing Genes ---------------------------------------------- ---------------------- ----------- ------------------- ----------- ----------- ------ ------ ------ **Method** **EBA (Affymetrix)** **(LRT)** **EBA (RNA-Seq)** **edgeR** **DESeq** EBA (Affymetrix) 14292 561 17 589 450 16 2550 1316 (LRT) 75 3979 0 235 197 0 3064 2208 EBA (RNA-Seq) 81 0 4726 235 197 0 3064 2208 edgeR 0 0 0 0 0 0 0 0 DESeq 0 0 0 0 0 0 0 0 422 973 5 1166 997 8947 4171 2406 2 0 0 0 0 0 915 0 3 0 0 0 0 0 856 1832 The diagonal gives the number of genes with missing tests. The off-diagonals indicate those genes that are significant for one method amongst the missing calls for another method. The acronym LRT denotes the likelihood ratio test based on the Poisson distribution, as described in the Methods. The acronym EBA denotes the empirical Bayes analysis performed on both the Affymetrix and RNA-Seq data. Discussion ========== Although log~2~-ratios are widely used to compare two groups of expression data, there are limitations to using these statistics. The largest drawback to ratio statistics is that they are unstable as the denominator gets closer to zero. In addition, frequentist methods for constructing a corresponding variance and formally testing hypotheses of differential expression are unsatisfying and more complicated than typical scenarios \[[@B15],[@B16]\]. Due to these drawbacks, we proposed an alternative to testing for differential expression with all of the advantages of a log~2~-ratio statistic and none of its disadvantages. We examined the proposed alternative, a proportion statistic, in four sets of simulations and two different sets of expression data. In simulations, the statistic , plus a constant and limma/EBA were robust to changes in distributional assumptions and the others were not. For the case of the Poisson distribution with rate parameter λ = 3, the statistic was underpowered, but otherwise and performed similarly well in simulations. The simulations suggest the use of in differential expression analyses because it uniformly preserved type I error and had competitive power. Note, however, that is not uniformly most powerful, and statistics derived from specific distributions can beat it when the distributional assumptions hold. The performance of the empirical Bayes analysis was competitive with in simulations but not in the analysis of the RNA-Seq dataset. Future research of interest may extend the test within an empirical Bayes framework, akin to what already exists for the log~2~-ratio. This may even further extend the clearly demonstrated feasibility of to detect differentially expressed genes. Additionally, while the popular statistic performs sufficiently well under many simulation conditions, it suffers from problems with missing data in real data analysis problems when expression values are low. The addition of constants 0.05 and 0.5 appreciably improve simulation results and make the performance of nearly as good as (see Additional file [1](#S1){ref-type="supplementary-material"}). Nevertheless, this ad hoc procedure can be avoided using . In both the analysis of a cDNA microarray set and an RNA-Seq dataset, the log~2~-ratio based statistics led to missing values. Of these genes with missing ratio values, the proportion statistic was able to detect instances of statistically significant differential expression. We therefore recommend for general use over the other statistics discussed. Conclusions =========== The use of the log~2~-ratio statistic to compare two expression values is challenged by denominators with near zero values. Thus, a reasonable alternative is to suggest a statistic that is not constrained by problems with very low expression values that still provides a meaningful test of differential expression. Using a proportion estimate instead of a ratio estimate does exactly that. The methods of this paper may only be used when data is naturally paired in test and reference samples, i.e. when log-ratios have traditionally been used. Our research provides several alternatives based on estimates of a proportion in both a frequentist and a Bayesian inference framework. We showed the performance of these alternatives and compared them to log~2~-ratio based tests in simulations and two gene expression datasets. In the gene expression analysis, all of the proportion methods performed better than ratio based methods for genes with low expression. For normal expression levels, inferential conclusions are similar, with the average proportion method, , plus a constant and the augmented log~2~-ratio method in limma/EBA, performing the best overall. The statistic has the added advantage that it does not require adjusting for an arbitrary constant that introduces bias in the estimate. Thus, tests of differential expression should consider proportion statistics over log~2~-ratios in future scientific studies. Methods ======= This section describes the data generation process in our simulations and the data collection in the two datasets analyzed in this paper. Simulations =========== The proposed test statistics were evaluated under four different distributions. Though sophisticated simulations can be used to mimic expression data, the simulations below use simple scenarios so as to examine the performance of test statistics under basic distributions and to compare the eight different methods clearly and meaningfully. The first set of simulated intensity values were sampled from an exponential distribution that mimics the values from a 16-bit TIFF image of a cDNA microarray with respect to center and spread. The reference sample was taken from an Exp(1/4000) and the test sample was taken from a *c*× Exp(1/4000) where *c*was the fold-change value, *c*= 1,2,3,4,5. The four statistics, , and , were calculated for each value of *c*and sample sizes *n*= 3, 5, 10, 15, 20, 25, 30, 40, 50. Additionally we evaluated the ratio statistic after shifting values for an arbitrarily small constant set at either 0.05 and 0.5. For further comparison, a standard implementation of the limma/empirical Bayes method of Smyth (2004) was perfomed \[[@B17]\]. For simulations of count data values, we evaluated methods that account for overdispersion in the tests of differential expression using the edgeR and DESeq packages in Bioconductor \[[@B18],[@B19]\]. The implementation in both packages fixes a constant library size for each sample so that normalization is not executed. Sample sizes larger than *n*= 50 give simulation results similar to those for sample sizes of 50. In order to compare results, the p-value for an independent t-test was computed, with a null hypothesis of no difference between the two sample means. The null hypothesis was rejected if the p-value was below *α*= 0.05 and the proportion of rejections out of 1000 simulations was recorded (Table [1](#T1){ref-type="table"} and Additional file [1](#S1){ref-type="supplementary-material"}). The null hypothesis of no differential expression is equivalent to a fold change of one, *c*= 1. When the fold change is greater than one, we are calculating the power to detect differential expression. In this way, type I error and power were compared across the different methods. The results would be equivalent when using reciprocal fold changes instead. The simulations for an exponential distribution were repeated for an Exp(1/400) distribution, to study the effects of changing the scale. A second set of simulated sampled intensity values from a Binomial(*M*= 10000, *p*= 0.5) distribution were obtained. The choice of this distribution was motivated by the derivation behind the maximum likelihood estimate of the proportion . The size was chosen to mimic the values from a 16-bit TIFF image of a cDNA microarray with respect to center. Analogous simulations to the exponential above were conducted with respect to statistics, sample sizes, and fold changes. For the binomial distribution, fold-changes of 2, 3, 4, and 5 correspond to binomial probabilities of 2/3, 3/4, 4/5, and 5/6 respectively. The simulations were repeated for a Binomial(*M*= 100, *p*= 0.5) distribution, to study the effects of change in the size parameter. A third set of simulated sampled intensity values from a Poisson(λ = 3) distribution were obtained. This distribution is motivated by the derivation behind the likelihood ratio test used in Marioni et al (2008) \[[@B9]\]. The parameter λ = 3 was chosen to mimic the number of categories arising from a smooth histogram of values from the RNA-Seq data. The simulations were repeated for a Poisson(λ = 30) distribution, to study the effects of change in the rate parameter. Analogous simulations to the exponential above were conducted with respect to statistics, sample sizes, and fold changes. A fourth set of simulated sampled intensity values from a Normal(*μ*= 5, *σ*= 1) distribution were obtained. This distribution was included since many analyses assume expression data to be normally distributed. The center was chosen to mimic values from cDNA data with mean 5,000 and standard deviation 1,000, scaled to Normal(*μ*= 5, *σ*= 1). Analogous simulations to the exponential above were conducted with respect to statistics, sample sizes, and fold changes. For the normal distribution, fold-changes of 2, 3, 4, and 5 correspond to test samples of Normal(*c*× *μ*, *σ*= 1). The simulations were repeated for a Normal(*μ*= 10, *σ*= 2) distribution, to study the impact of changing the parameters. All simulations were conducted using R <http://www.r-project.org> and the code is available in Additional file [2](#S2){ref-type="supplementary-material"}. Gene Expression Data from Mice using cDNA Microarrays ----------------------------------------------------- We examined our proposed approach in a well-known and often cited set of cDNA microarrays. We chose this set because many research groups have evaluated their methods on this data and consequently the differential expression behavior in this data are better understood. The Apo AI experiment used cDNA microarrays to measure gene expression in the livers of 8 inbred control mice versus 8 mice with the Apo AI gene \"knocked out\" \[[@B20]\]. For each microarray, the reference sample was created from the pooled cDNA of the eight control mice. The goal of the experiment was to detect differential expression in the liver between control mice and the genetic knockout strain \[[@B12]\]. Since the Apo AI gene plays a role in HDL metabolism, differentially expressed genes are likely associated with lipid metabolism. The data can be obtained as an Rdata object from <http://www.bioconductor.org/help/course-materials/2005/BioC2005/labs/lab01/Data/apoai.zip> on the Bioconductor website. Welch two-sample t-statistics for each of the 6,384 probes were calculated, where *X~trt~*and *X~cont~*were either our proportion estimators, and or the usual log~2~-ratio estimators, and . Since the variability of the cDNA data resembles the exponential distribution, the assumptions for methods and do not hold and therefore they were not used. To account for multiple testing, the original analysis used the maxT step-down procedure based on the t-statistics and found eight significantly differentially expressed probe sequences \[[@B12]\]. In order to explore the performance of alternative methods with both of the test statistics, the limma/EBA method of Smyth (2004) was computed \[[@B17]\]. Although this method was developed for log~2~-ratio values, we used the same programs on the proportion values as well. Gene Expression Data from Human Kidney and Liver Cells using RNA-Seq -------------------------------------------------------------------- In order to examine the performance of our new approach on a sequence-based technology, we analyzed a set of RNA-Seq data discussed in Marioni et al (2008) \[[@B9]\]. This set of data compared the expression of 32,000 sequences in human kidney and liver cells extracted from the same person. The expression was also measured using Affymetrix U133 oligonucleotide arrays. Data was obtained from Supplemental Table [2](#T2){ref-type="table"} in the original manuscript. To compare our methods with those reported in the Supplemental Table [3](#T3){ref-type="table"} of their manuscript, we extracted the same five lanes of Illumina sequencing data corresponding to 3 pM concentrations of cDNA. We calculated both of the proportion tests outlined in the Results section, the ratio-based test provided in the Background section, and compared them to the methods from the original paper and more recent methods that account for overdispersion \[[@B18],[@B19]\]. The methods to test for differential expression from RNA-Seq data in the original paper used a likelihood ratio test (LRT) for inference \[[@B9]\]. Their test assumes that the expression data follows a Poisson distribution where the rate of expression λ is equivalent in kidney (K) and liver (L) cells under the null hypothesis. For gene *j*, the likelihood ratio test compares *H*~0~: λ*~Kj~*= λ*~Lj~*versus the alternative hypothesis that expression rates differ *H*~1~: λ*~Kj~*≠ λ*~Lj~*. The likelihood ratio test is for gene *j*. The maximum likelihood estimate for the alternative hypothesis from the above LRT is denoted by . The original paper also tested for differential expression on the Affymetrix platform for the same tissue samples. The methods employed were an empirical Bayes analysis with a false discovery rate of 0.1% \[[@B17]\]. More recent developments that account for overdispersion in the tests of differential expression were implemented using the edgeR and DESeq packages in Bioconductor \[[@B18],[@B19]\]. Appendix: Bayesian Estimation and Inference =========================================== In order to compare our proposed methods to previously suggested test statistics in the data analysis sections, we evaluated the proportion statistics within a frequentist testing framework. It is also possible to conceive the model in a Bayesian framework. Given the binomial assumption presented in the Results section, a Bayesian analysis can be conducted. Let the beta distribution be denoted by *β*(*a*, *b*), with density Where Γ (*a*) is the gamma function with parameter *a*. We denote the Bayesian estimator of *p~j~*by . Using a Beta prior for *p~j~*with parameters *a*and *b*, the posterior distribution of , is *β*(*y*+ *a*, *M*- *y*+ *b*), where and with density \[[@B21]\]. To compare the performance of the Bayesian with frequentist statistics , and , credible intervals and confidence intervals can be constructed and coverage can be examined in simulations. For data where the difference of two proportions is required, the posterior distribution derived in \[[@B22]\] can be used. Authors\' contributions ======================= TLB and JW contributed equally to the research. Both authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Additional materials**. Additional tables for each of the simulation scenarios are provided in the file exprPropSupp2011.pdf. This file was generated using LaTeX. ###### Click here for file ###### Additional file 2 **Additional materials**. The script written in R <http://www.r-project.org> to conduct simulations are provided in the file exprPropSimCode.R. ###### Click here for file Acknowledgements ================ The authors would like to thank Suzanne Grindle in the Department of Microbiology at the University of Minnesota for her feedback, as well as that of the anonymous reviewers, which has resulted in improvements to the manuscript. Support for this research was provided by the Institute for Pure and Applied Mathematics at UCLA.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ Type 1 diabetes (T1D) is a chronic inflammatory disease that is characterized by autoimmune destruction of insulin-producing pancreatic beta cells \[[@pone.0183908.ref001]--[@pone.0183908.ref002]\]. The triggering event (s) leading to beta cell damage are not known; however, multiple studies have linked a possible viral infection to islet autoimmunity and T1D. Our long term objective is to identify novel protein signatures that distinguish the Islets from patients with T1D from the pancreas from patients who are autoantibody positive without symptoms of diabetes (AAb+), and from patients with no evidence of disease (ND). Laser capture microdissection (LCM) allows for the isolation of enriched cell populations from heterogeneous tissue guided by microscopic visualization. LCM can be used to harvest the cells of interest directly resulting in histologically enriched cell populations \[[@pone.0183908.ref003]--[@pone.0183908.ref006]\]. In our current study, we have focused on proteomic characterization of laser capture microdissected islets that should be more reflective of the micro-environment of beta cells, as compared to total pancreatic sections. The ultimate objective is the identification of beta cell stress markers and any microorganisms that may have a role in the etiology of T1D. In this study, we have used LCM for isolation of pancreatic islets from the Network for Pancreatic Organ Donors with Diabetes (nPOD) pancreata. The LCM captured islets from normal and disease stratified were then investigated by high-resolution, accurate-mass (HR/AM) liquid chromatography mass spectrometry (LC/MS/MS) for qualitative and quantitative proteomic analysis to detect differential protein expression in disease stratified pancreas tissue obtained from the nPOD consortium. LCM was performed on pancreas tissue sections from organ donors without diabetes, with T1D-associated autoantibodies in the absence of diabetes, and with T1D. We identified proteins that are significantly differentially regulated in T1D LCM islets compared to the other two groups. These proteins have diverse biological functions, such as inflammation, metabolic regulation, and autoimmunity, all of which are pathways linked to the pathogenesis of T1D. The identified islet proteomes, protein-protein interaction networks, and enriched functional categories presented in the different disease states provide novel insights that should enhance our understanding of the etiology of T1D, and provide impetus for further investigations focusing on their expression profiles in beta cells to evaluate their role in T1D disease. Some of these proteins are likely involved in beta cell stress, as well as T1D development and pathogenesis. These molecules could also be potential biomarkers for T1D and targets for development of novel therapeutics. In addition, this optimized LCM-LC-MS method will be applicable to additional future targeted proteomic analysis of nPOD tissue samples in studies focusing on specific molecules and molecular pathways. Materials and methods {#sec002} ===================== Materials, chemicals and reagents {#sec003} --------------------------------- Laser Capture Microdissection slides and LCM caps were obtained from Thermo Fisher Scientific (Rockford, IL, USA). LCM Liquid Tissue MS Preparation Kit was purchased from Expression Pathology Inc (Rockville, MD, USA). Micro-BCA Protein Assay Kit, dithiothreitol (DTT) and high purity grade solvents including acetonitrile, water and formic acid were from Thermo Fisher Scientific (Rockford, IL, USA). Ammonium bicarbonate and Iodoacetamide were purchased from Sigma-Aldrich (St. Louis, MO, USA). Sequencing Grade Trypsin was purchased from Promega Corporation (Madison, WI, USA). Human tissue samples {#sec004} -------------------- Pancreatic tissue sections were obtained from optimal cutting temperature (OCT) embedded tissue from deceased organ donors provided by nPOD, University of Florida, Gainesville, Florida, USA, in accordance with ethical regulations \[[@pone.0183908.ref007]--[@pone.0183908.ref008]\]. These samples do not fall under human subjects research and the study was classified as exempt by the Eastern Virginia Medical School Institutional Review Board (IRB). Tissue samples were obtained from three groups: nondiabetic donors (ND), non-diabetic donors with type 1 diabetes-associated autoantibodies (AAB+) and donors with type 1 diabetes (T1D). Comprehensive qualitative and quantitative proteomic analyses were performed on five ND cases and four cases from AAB+ and T1D cases. The nPOD cases classification criteria are described under [S1 Table](#pone.0183908.s003){ref-type="supplementary-material"}. Laser capture microdissection and protein extraction {#sec005} ---------------------------------------------------- For LCM, serial 10 μm-thick OCT embedded frozen sections were prepared from pancreas blocks and attached to either polyethylene naphthalate (PEN) membrane slides or glass-membrane slides that allow for the utilization of both infrared (IR) laser capture and ultraviolet (UV) laser cutting. The tissue samples were processed as follows: Incubated in 70% ethanol for 1 min; rinsed in molecular biology grade water (5 dips); and then incubated sequentially in ethanol at the following concentrations 70%, 95%, and 100% for 1 min each followed by 100% ethanol for 5 min (all steps at 4^o^ C) to dehydrate the samples. The samples were then air-dried for 5--10 min at RT before capture. LCM of islets was performed using an Applied Biosystems ArcturusXT LCM System equipped with dual IR and UV lasers (Thermo Fisher Scientific) according to manufacturer's instructions. The utilization of the two lasers working together allows for the efficient and more specific isolation of cells and tissue without any detrimental effects on their morphology or integrity of their biological content. The IR laser helps to capture the cells of interest, and the UV laser microdissects the captured cells preventing any significant contamination of captured material with adjacent acinar tissue. For these studies, unstained tissue section islets were identified by their unique and specific intrinsic florescence behavior that is observed upon UV illumination of the specimen that can be detected using triple filter. Typically, the UV cut and IR capture settings were as follows: spot spacing, 110; spot diameter 30--50; spot power 60--70; spot duration 17--20; UV cut speed 900 microns/second. The islets were collected on CapSure LCM Caps (typically, about 1.5--2 mm^2^ of islet area for a single cap) which were then placed onto 0.5 ml PCR tubes and stored at -80°C until use. Protein extraction was performed using the Liquid Tissue MS Protein Prep Kit (Expression Pathology, Rockville, MD) according to the manufacturer\'s protocol. Briefly, the LCM cap films containing approximately 3 x10^4^ cell equivalents (estimates were based on the thickness and area of the captured islets tissue) were transferred into 20 μl of liquid tissue buffer in a 1.5 ml low protein binding reaction tube and centrifuged at 10,000 x g for 2 minutes to pellet the film. The islet proteins were extracted by heating the mixture at 95°C for 90 min with intermittent mixing at 20 min intervals. After 90 min, the samples were centrifuged at 10,000 x g for 1 minute before cooling in ice for 2 min. The equivalent of 1:50 trypsin was added to the extracted protein and incubated at 37°C for 18 hours to generate tryptic peptides. After the trypsin digestion step, an aliquot of the generated peptides was used in a Micro BCA assay to determine the peptide concentrations. The remaining peptides were reduced (10mM DDT) and alkylated (35mM iodoacetamide) and stored at -80°C prior to use for mass spectrometry analysis. Liquid chromatography and mass spectrometry {#sec006} ------------------------------------------- LC-MS analysis of digested samples was conducted as we have previously described \[[@pone.0183908.ref009]\]. Tryptic peptides were solubilized in normalized volumes of 0.1% formic acid (FA)/H2O before determining their concentrations using the Micro BCA assay. The concentrations of the samples were adjusted to 0.5 μg/μl in 0.1% formic acid. 2 μg of the tryptic peptides from each of the samples were injected and analyzed on a Q-Exactive Orbitrap mass spectrometer coupled to an Easy NanoLC-1000 system optimized under the following conditions: LC solvents: buffer A, 0.1% formic acid/water; buffer B, 0.1% formic/acetonitrile; column, Thermo Scientific™ EASY-Spray™, 75μm x 150mm C18 column; flow rate 400 nl/min; LC separation, total 120 min stepped linear gradient, 0--2 min 2% B; 2% to 40% B in 112 minutes, followed by 95% B in the next 3 minutes and holding at 95% B for 5 minutes. The MS data was acquired using data-dependent acquisition with following optimized settings: dynamic exclusion (DE = 1); top 12 higher energy collision induced dissociation (HCD); resolving power set at 70,000 for the full MS scan and 17,500 for the MS/MS scan at m/z 200 as we have previously described \[[@pone.0183908.ref009]\]. Each individual sample was run in randomized triplicate or duplicate LC/ESI-MS/MS analyses. Two blank runs injecting 0.1% formic acid were included before every sample injection to eliminate carry-over between the runs. The acquired high-resolution MS and MS/MS peptide spectra were imported to MaxQuant for further analysis. Database searching and label free quantification {#sec007} ------------------------------------------------ High resolution Q-Exactive.RAW MS files were imported to MaxQuant (version 1.5.2.8) \[[@pone.0183908.ref010]\] with Andromeda as search engine for protein identification/quantitation \[[@pone.0183908.ref011]\]. LCM samples from the three groups were analyzed as one set and aligned. Analysis of MS spectra was performed using the following parameters: acetylation of the protein N-terminus and oxidation of methionine as variable modifications; carbamidomethylation of cysteine as fixed modification. UniProt-SwissProt human canonical database (version 2016, canonical proteome; 20 198 identifiers) was selected as FASTA file. Seven amino acids were selected as minimum peptide length. Match between runs option was kept as default (match time window: 0.7 min; alignment time window: 4 min). Label free quantitation (LFQ) was enabled and LFQ minimum ratio count was set to 1. Remaining options were kept as default \[[@pone.0183908.ref010], [@pone.0183908.ref011]\]. Data and statistical analysis {#sec008} ----------------------------- Perseus, which comprises of a comprehensive portfolio of statistical tools for high-dimensional omics data analysis covering normalization, pattern recognition, time-series analysis, cross-omics comparisons and multiple-hypothesis testing of data generated in MaxQuant \[[@pone.0183908.ref012]\], was used to compare protein expression in the 3 sample groups. "ProteinGroups.txt" was imported from data generated in MaxQuant as described in the preceding section \[[@pone.0183908.ref012]\]. Peptides were filtered for posterior error probability (\< 0.05) and proteins for Q-value (\< 0.05). Possible contaminants and reversed sequences were excluded \[[@pone.0183908.ref012]\]. The detailed protocol for label free quantitation is described in detail by Luber et al. \[[@pone.0183908.ref013]\]. MaxQuant derived LFQ normalized protein/ peptide intensity were Log2 transformed before further analyses. This label free algorithm takes the maximum number of identified peptides for a specific protein identified between any two samples and compares the intensity of these peptides to determine peptide ratios \[[@pone.0183908.ref013]\]. Protein abundance was computed using median values of all peptide ratios of certain protein \[[@pone.0183908.ref013]\]. The peptides that were used for further analysis fulfilled the following criteria: (i) peptides were uniquely assigned to one protein group; (ii) peptides were not identified as in the decoy data base derived from reversed sequences of all peptides in the database. Only peptides with abundance data in at least 50% of the samples were included in the statistical analysis. Proteins without peptide quantitative values in more than half of the triplicate and duplicate measurements from the 13 biological replicates were excluded from the analyses. Data imputation to replace log2 transformed LFQ missing values was done by normal distribution. Using this approach, missing values were imputed using normal distribution to simulate peptide expressions below the detection limit. The approach makes the assumption that in proteomics experiments, low expression proteins give rise to missing values \[[@pone.0183908.ref012]\]. In this approach, a Gaussian distribution with a median shifted from the measured data distribution results in accurate imputation of such missing values \[[@pone.0183908.ref012]\]. Two sample t-test was performed on the abundances to confirm differentially expression patterns of these proteins (p\<0.05, permutation-based false discovery rates (FDR = 0.05) using Perseus \[[@pone.0183908.ref012], [@pone.0183908.ref014]\]. The combined score probability score was generated as the product of the two tests and the values expressed in a--log10 scale as calculated in Perseus \[[@pone.0183908.ref012]\]. Volcano plots were used to visualize the results of the t-test. Significantly differentially regulated proteins between AAB+, T1D and control ND islet samples were identified based on the permutation based FDR calculations. Protein pathway analysis {#sec009} ------------------------ Functional classification for the identified proteins was performed using the Panther database ([http://www.pantherdb.org](http://www.pantherdb.org/)) \[[@pone.0183908.ref015]\]. GO annotation of the identified proteins was done using the NCBI's DAVID analysis tool (V6.7: [http://david.abcc.ncifcrf.gov](http://david.abcc.ncifcrf.gov/)). GO_CC_FAT, GO_BP_FAT, and GO_MF_FAT algorithms were used \[[@pone.0183908.ref016]--[@pone.0183908.ref017]\]. The default parameters of the GO enrichment searches were used with a count threshold set at 2 and the EASE (p-value) threshold set at 0.1; fold enrichment as well as FDR values were displayed in search results. Differentially expressed proteins were analyzed using Ingenuity Pathway Analysis (IPA) (Ingenuity Systems: [http://www.ingenuity.com](http://www.ingenuity.com/)). Information stored in the Ingenuity Pathways Knowledge Base was used to identify and predict top canonical pathways, top upstream regulators, top diseases and bio functions and molecular networks. Immunofluorescence and confocal microscopy {#sec010} ------------------------------------------ Formalin fixed paraffin embedded tissue sections from nPOD were processed for immunofluorescence assay as follows. Sections were paraffinized in xylene followed by rehydration in descending concentrations of ethanol in water. Sections were then subject to heat induced epitope retrieval for twenty minutes at 95°C (HIER Buffer, Thermo Scientific). Slides were then blocked with 5% BSA for one hour, followed by primary antibody incubation overnight at 4°C in a low evaporation chamber. The primary antibodies that were used were: S100A9 (ab63818, 1:50), insulin (ab7842, 1:200), and glucagon (ab82270, dilution (1:50). All the antibodies were from Abcam (Cambridge, MA). After three 15 minute washes, sections were stained with the appropriate secondary antibodies conjugated with CY2 or Cy3 at 1:100 dilution for 90 minutes at room temperature in the dark. Slides were washed three times for 15 minutes and mounted with Cytoseal XYL (Thermo Scientific) before imaging using a Zeiss Axiophot epifluorescent microscope. Results and discussion {#sec011} ====================== In this study, we report a robust quantitative proteomic analysis approach for the identification of human islet proteins that are associated with type 1 diabetes. This pilot biomarker discovery pipeline is divided into 4 main stages. The first stage consists of laser capture microdissection to harvest highly enriched islet samples from disease stratified human pancreas tissues. The LCM enrichment minimizes interference from the background exocrine pancreas proteome and allows for a more precise quantitation of the islet proteome both in normal and diseased state. The second stage involves the generation of high resolution high mass accuracy raw nLC-MS/MS profiling data; the third stage includes both upstream sample processing steps that include multiple chromatographic peak and spectral alignment, peptide and protein identification, and quantitation and subsequent statistical analysis steps. The fourth and last stage involves GO enrichment and pathway analysis to identify molecular targets and pathways that are associated with T1D disease initiation and progression. A schematic chart summarizing the experimental strategy is shown in [Fig 1A](#pone.0183908.g001){ref-type="fig"}. {#pone.0183908.g001} LCM capture of islets and morphology and insulin content of islets {#sec012} ------------------------------------------------------------------ All the OCT embedded pancreatic sections from the nPOD cases are stained for pancreatic markers (including insulin and glucagon) as a standard histopathological evaluation. The insulin staining strongly correlated with the auto florescence on unstained sections that is specifically observed in islets using UV illumination. This intrinsic property was used to guide the capture of islets from pancreas tissue samples. It is important to note that the morphology of the pancreas tissues and the isolated islets that were used in this study were very well preserved. [Fig 1B](#pone.0183908.g001){ref-type="fig"} shows representative examples of images of pancreas tissue sections observed under bright field, UV illumination, H&E staining and insulin staining. Protein expression profiles of T1D, AAB+ and control ND islets {#sec013} -------------------------------------------------------------- The main objective was to characterize the proteome of islets derived from disease-stratified human pancreata to identify differentially expressed proteins that may play a role in the etiology and pathogenesis of T1D. It is important to note that known endocrine hormones including insulin, glucagon and somatostatin are detected in these analyses. We have identified a total of 2032 different protein groups and 1491 protein groups fulfilled the criteria for comparative quantitative analysis described under the data and statistical analysis in the materials and methods section. Perseus was used to analyze differential protein expression between the non-diabetic cases versus autoantibody positive and type 1 diabetes cases. The volcano plots summarizing the expression profiles of proteins in the three sample groups are shown in [Fig 2A and 2B](#pone.0183908.g002){ref-type="fig"}. For the AAb+ cases, 39 proteins were differentially regulated compared to the no-disease group, with 25 upregulated and 14 downregulated proteins ([S2 Table](#pone.0183908.s004){ref-type="supplementary-material"}). For the T1D samples, 63 proteins were differentially expressed, with 24 upregulated and 39 downregulated, compared to the no disease controls ([S3 Table](#pone.0183908.s005){ref-type="supplementary-material"}). Nine proteins are differentially regulated in both T1D and AAB+, with 5 upregulated and 4 downregulated proteins. A master table that shows the protein/peptide intensities for all the peptides that were quantified in the study is provided ([S4 Table](#pone.0183908.s006){ref-type="supplementary-material"}). [Fig 2C](#pone.0183908.g002){ref-type="fig"} shows the Venn diagram comparison of the differentially expressed proteins between AA+ and T1D islets. [Fig 2D](#pone.0183908.g002){ref-type="fig"} shows the comparison of upregulated proteins between AAB+ and T1D islets, and [Fig 2E](#pone.0183908.g002){ref-type="fig"} shows the comparison of down regulated proteins between AAB+ and T1D islets. {#pone.0183908.g002} The biological processes, molecular functions and cellular components of the 63 identified and quantifiable protein groups were determined using the Panther database \[[@pone.0183908.ref015]\] and the results are summarized in [Fig 3](#pone.0183908.g003){ref-type="fig"}. ![Gene ontology enrichment for differentially proteins between ND versus T1D cases.\ The GO analysis was performed using Panther \[[@pone.0183908.ref015]\]. GO-CC denotes Cellular Component, GO-MF denotes Molecular Function and GO-BP, Biological Processes.](pone.0183908.g003){#pone.0183908.g003} The differentially regulated proteins include those that are associated with different metabolic processes, including glycolysis/ gluconeogenesis, oxidative phosphorylation, and secretory granules. Functional annotation enrichment of these differentially expressed proteins was done using David \[[@pone.0183908.ref016]--[@pone.0183908.ref017]\]. The significance threshold for the enrichment values was set to 2.0. David functional annotation clusters with enrichment scores for differentially regulated proteins in T1D are shown in [S5 Table](#pone.0183908.s007){ref-type="supplementary-material"}. The significant annotations include: ***Cluster i*** includes GO: 0070062, extracellular exosome; GO: 1903561, extracellular vesicle; GO:0043230, extracellular organelle with an enrichment score of 7.46; ***Cluster ii*** includes GO:0016052, carbohydrate catabolic process with an enrichment score of 4.65; ***Cluster iii*** includes GO:0019752, carboxylic acid metabolic process with an enrichment score of 4.18; ***Cluster iv*** includes GO:0006090, pyruvate metabolic process with an enrichment score of 4.09; ***Cluster v*** includes GO:0046128, purine ribonucleoside metabolic process with an enrichment score of 3.31; ***Cluster vi*** includes GO:0046128, purine ribonucleoside metabolic process with an enrichment score of 3.10; ***Cluster vii*** includes GO:0006096, glycolytic process with an enrichment score of 2.94; ***Cluster viii*** includes GO:0002790, peptide secretion process with an enrichment score of 2.41; ***Cluster ix*** includes GO:0009167, purine ribonucleoside monophosphate metabolic process with an enrichment score of 2.34 and ***Cluster x*** includes GO:0006006, glucose metabolic process with an enrichment score of 2.02. The same analyses were performed using the proteins from AAB+ islets and the results are summarized in [S6 Table](#pone.0183908.s008){ref-type="supplementary-material"}. There are only two clusters that are beyond the set significance threshold ***Cluster i*** includes GO: 0070062, extracellular exosome; GO: 1903561, extracellular vesicle; GO: 0043230, extracellular organelle with an enrichment score of 6.77; ***Cluster ii*** includes GO: 0016052, carbohydrate catabolic process with an enrichment score of 2.34. We have performed a protein-protein interaction network analysis of the 63 differentially regulated proteins between T1D and non-diabetic cases. [Fig 4](#pone.0183908.g004){ref-type="fig"} shows the interaction networks for differentially expressed islets as determined by STRING analysis \[[@pone.0183908.ref018]\]. The main cluster (1) or node for the interactions is insulin which is downregulated in T1D islets versus ND. The other node or cluster (2) is associated with lactate dehydrogenase A and B which are both upregulated in the T1D cases compared to ND. These are enjoined to a cluster (3) that has ATP-citrate synthase (ACYL) which is the primary enzyme responsible for the synthesis of cytosolic acetyl-CoA in many tissues. ACYL has a central role in de novo lipid synthesis. These interact with a cluster (4) that comprises alcohol dehydrogenases (ALDHs) proteins that play a major role in the detoxification of alcohol-derived acetaldehyde and are also involved in the metabolism of corticosteroids, biogenic amines, neurotransmitters, and lipid peroxidation. The final cluster 5 comprises of SEC61B which is necessary for protein translocation in the endoplasmic reticulum. There are also a number of proteins which do not form clusters. These include BLVRB, C9, COL1A2, CPB1, CRP2, FAM3C, HNRNPH2, LAMA2, NPTX2, ORM1, PLCXD3, SELENBP1, and SCGN. We have also performed STRING analysis on the proteins that are differentially expressed in AAB+ versus non-diabetic cases. The interaction network and clusters are different from the ones that are observed in T1D other than the cluster that includes alcohol dehydrogenases ([Fig 5](#pone.0183908.g005){ref-type="fig"}). The central cluster is focused on SOD1, which destroys radicals which are produced within the cells and are toxic to cells. ) derived protein-protein interaction networks for 63 islet proteins that are differentially regulated in T1D compared to ND cases.\ The network nodes represent proteins. Splice isoforms or post-translational modifications are collapsed, i.e., each node represents all proteins produced by a single, protein coding gene. Edges represent protein-protein associations. The associations are meant to be specific and meaningful with associated proteins jointly contributing to a shared function. This does not necessarily mean that the proteins are physically binding each other \[[@pone.0183908.ref018]\] ([http://www.string-db.org](http://www.string-db.org/)).](pone.0183908.g004){#pone.0183908.g004} ) derived protein-protein interaction networks for 39 islet proteins that are differentially regulated in AAB+ compared to ND cases.\ The network nodes represent proteins. Splice isoforms or post-translational modifications are collapsed, i.e., each node represents all proteins produced by a single, protein coding gene. Edges represent protein-protein associations. The associations are meant to be specific and meaningful with associated proteins jointly contributing to a shared function. This does not necessarily mean that the proteins are physically binding each other \[[@pone.0183908.ref018]\] ([http://www.string-db.org](http://www.string-db.org/)).](pone.0183908.g005){#pone.0183908.g005} IPA network analysis was performed to identify potential markers that are differentially upregulated in only in AAB+ and T1D versus ND cases. The Top Canonical Pathways, Top Diseases and Bio Functions, Molecular and Cellular Functions, Physiological System Development and Function predicted for the proteins/genes that are differentially regulated in T1D and AAB+ and the probability values are summarized in Tables [1](#pone.0183908.t001){ref-type="table"} and [2](#pone.0183908.t002){ref-type="table"} respectively. The top 10 upregulated and downregulated genes that are included in the IPA analysis are shown in Tables [3](#pone.0183908.t003){ref-type="table"} and [4](#pone.0183908.t004){ref-type="table"}, and the entire list is given in [S2](#pone.0183908.s004){ref-type="supplementary-material"} and [S3](#pone.0183908.s005){ref-type="supplementary-material"} Tables. Some of these uniquely upregulated molecules and activated pathways may serve as novel targets for prevention and treatment of T1D. 10.1371/journal.pone.0183908.t001 ###### Summary of IPA Analysis AAB+ versus non-diabetic cases. {#pone.0183908.t001g} --------------------------------------------------- -------------------- ---------------------- **Top canonical pathways** Name p-value Overlap Retinol biosynthesis 2.86E-05 8.3% 3/36 Glycogen degradation III 1.75E-04 16.7% 2/12 Retinoate biosynthesis I 1.21E-03 6.5% 2/31 Triacylglycerol degradation 1.29E-03 6.2% 2/32 Antigen presentation pathway 1.72E-03 5.4% 2/37 **Top upstream regulators** Upstream regulator p-value of overlap Predicted activation NPC 1 8.22E-05 SMARCA4 4.91E-04 Activated HMGA1 7.30E-04 PRKCE 9.09E-04 TWIST1 1.18E-03 **Top diseases and bio functions** **Diseases and disorders** Name p-value \#Molecules Endocrine system disorders 4.23E-02--2.01E-06 15 Gastrointestinal disease 4.34E-02--2.01E-06 14 Inflammatory disease 4.34E-02--2.01E-06 12 Inflammatory response 4.34E-02--2.01E-06 12 Organismal injury and abnormalities 4.96E-02--2.01E-06 26 **Molecular and cellular functions** Name p-value \#Molecules Cellular growth and proliferation 4.23E-02--2.09E-05 9 Cell death and survival 4.74E-02--1.22E-04 9 Cellular movement 4.07E-02--4.15E-04 5 Free radical scavenging 4.54E-02--4.99E-04 2 Small molecule biochemistry 4.82E-02--1.38E-04 10 **Physiological system development and function** Name p-value \#Molecules Behavior 3.32E-03--2.40E-04 2 Cell-mediated immune response 2.63E-02--1.66E-03 1 Connective tissue development and function 3.27E-02--1.66E-03 6 Embryonic development 4.07E-02--1.66E-03 4 Hematological system development and function 3.77E-02--1.66E-03 4 --------------------------------------------------- -------------------- ---------------------- 10.1371/journal.pone.0183908.t002 ###### Summary of IPA analysis T1D versus non-diabetic cases. {#pone.0183908.t002g} --------------------------------------------------- -------------------- ---------------------- **Top canonical pathways** Name p-value Overlap LXR/RXR activation 1.52E-05 4.1% 5/121 FXR/RXR activation 3.09E-04 3.2% 4/125 Role of IL-17A in psoriasis 5.06E-04 15.4% 2/13 Acute phase response signaling 9.40E-04 2.4% 4/168 Acetyl-CoA biosynthesis III (from citrate) 2.60E-03 100.0% 1/1 **Top upstream regulators** Upstream regulator p-value of overlap Predicted activation PKM 1.16E-03 VAV 2.41E-03 ASB9 2.41E-03 ATP2B4 2.41E-03 MTOR 3.03E-03 **Top diseases and bio functions** **Diseases and disorders** Name p-value \#Molecules Hematological disease 1.35E-02--1.37E-06 8 Metabolic disease 4.57E-02--1.37E-06 16 Endocrine system disorders 4.57E-02--1.82E-05 9 Cancer 4.61E-02--3.95E-05 8 Organismal injury and abnormalities 4.93E-02--3.95E-05 30 **Molecular and cellular functions** Name p-value \#Molecules Carbohydrate metabolism 4.07E-02--2.49E-06 10 Molecular transport 2.43E-02--2.49E-06 13 Small molecule biochemistry 4.32E-02--2.49E-06 15 Lipid metabolism 4.32E-02--6.57E-05 8 Cell-to-cell signaling and interaction 4.82E-02--1.38E-04 11 **Physiological system development and function** Name p-value \#Molecules Humoral immune response 9.84E-05--9.84E-05 2 Hematological system development and function 4.82E-02--1.55E-04 7 Immune cell trafficking 4.82E-02--1.55E-04 6 Connective tissue development and function 2.82E-02--1.22E-03 5 Cell-mediated immune response 4.07E-02--2.60E-03 1 --------------------------------------------------- -------------------- ---------------------- 10.1371/journal.pone.0183908.t003 ###### Ten top differentially regulated molecules in IPA analysis T1D vs ND. {#pone.0183908.t003g} Molecules Expression Value Molecules Expression Value ----------- ------------------ ----------- ------------------ S100A9 **↑**2.9 PFKB2 **↓**-2.9 REG1B **↑**2.8 HADH **↓**-2.3 REG3A **↑**2.3 ATP2A3 **↓**-2.0 S100A8 **↑**2.3 STX1A **↓**-1.9 C9 **↑**1.8 INS **↓**-1.8 ABHD14B **↑**1.7 SEC61B **↓**-1.6 ALDH1B1 **↑**1.7 ASB9 **↓**-1.5 CKB **↑**1.6 PLCXD3 **↓**-1.5 HIST1H1B **↑**1.6 UCHL1 **↓**-1.5 ORM1 **↑**1.5 GSTM2 **↓**-1.5 10.1371/journal.pone.0183908.t004 ###### Ten top differentially regulated molecules in IPA analysis AAB+ vs ND. {#pone.0183908.t004g} Molecules Expression Value Molecules Expression Value ----------- ------------------ ----------- ------------------ REG1B **↑**3.3 SYCN **↓**-2.7 SPINK1 **↑**2.9 CTRL **↓**-2.6 PSMB6 **↑**2.1 BLVRB **↓**-2.1 SOD1 **↑**2.1 CAV1 **↓**-1.6 GSTP1 **↑**1.8 COA3 **↓**-1.4 CD99 **↑**1.8 HNRNPUL1 **↓**-1.3 GAA **↑**1.6 PNLIPRP1 **↓**-1.2 CMBL **↑**1.6 STAT3 **↓**-1.2 PGM1 **↑**1.6 CELA3B **↓**-1.2 CUZD1 **↑**1.6 ECHCD3 **↓**-1.0 The most significantly upregulated protein in T1D cases with a 2.9 fold increase when compared to controls was S100A9. This gene is a member of the S100 family of proteins containing 2 EF-hand calcium-binding motifs. S100A9 is involved in a wide variety of intracellular and extracellular functions \[[@pone.0183908.ref019]\]. The protein plays a prominent role in the regulation of inflammatory processes, immune response and antimicrobial response \[[@pone.0183908.ref020]--[@pone.0183908.ref023]\]. It can induce neutrophil chemotaxis, adhesion, and increase the bactericidal activity of neutrophils. S100A8/A9 protein complex also binds arachidonic acid was recently identified by in vitro studies as a novel partner of the phagocyte NADPH oxidase \[[@pone.0183908.ref024]\]. S100A9 has a secretory sequence and its extracellular functions involve proinfammatory, antimicrobial, oxidant-scavenging and apoptosis-inducing activities. Its proinflammatory activity includes recruitment of leukocytes, promotion of cytokine and chemokine production, and regulation of leukocyte adhesion and migration \[[@pone.0183908.ref025]--[@pone.0183908.ref035]\]. S100A9 is proposed to also direct selective inflammatory stimulus-dependent S-nitrosylation of multiple targets \[[@pone.0183908.ref036]\]. It is important to note that S100A8 which is also a member of the S100 family is also significantly upregulated with a 2.2 fold increase in T1D cases versus the non-disease controls. Our current data is supported by a recent report that demonstrates the elevation of both S100A9 and S100A8 in the salivary proteome and peptidome profile in selected subjects with type 1 diabetes \[[@pone.0183908.ref037]\]. It is important to note that both S100A8/S100A9 are potential biomarkers and targets for therapeutics against autoimmune diseases \[[@pone.0183908.ref019], [@pone.0183908.ref038]--[@pone.0183908.ref040]\]. S100A8/S100A9 have also been associated with host defence mediation in neuropathic foot ulcers in patients with type 2 diabetes mellitus \[[@pone.0183908.ref041]\]. We have performed immonohistochemical validation for the expression of S100A9 in pancreas tissue sections from non-diabetic and T1D cases. Whole tissue lysates of non-diabetic and T1D show no difference in their expression of S100A9, however, there is a significant upregulation of S100A9 in the islets of T1D donors upon isolation of proteins using LCM ([S1 Fig](#pone.0183908.s001){ref-type="supplementary-material"}). The other significantly upregulated proteins include Regenerating Islet-Derived 1 Beta (REG1B) and Regenerating Islet-Derived Protein III-Alpha (REG3A). The Reg gene family is a multigene family grouped into four subclasses, types I, II, III and IV based on the primary structures of the encoded proteins. Reg1B is associated pancreas regeneration is upregulated in human β cells under inflammatory conditions through the JAK/STAT pathway and upon upregulation may function as a growth factor for β cells to facilitate proliferation \[[@pone.0183908.ref042]\]. In addition REG3A is overexpressed during the acute phase of pancreatitis and in some patients with chronic pancreatitis \[[@pone.0183908.ref043]\]. We have previously reported the upregulation of REG3A in whole pancreata tissue lysates from nPOD samples from people with T1D \[[@pone.0183908.ref009]\]. The deterioration of β cells in the pancreas is a crucial factor in the progression of type 1 diabetes and therefore the recovery of β cells is critical for effective diabetic therapeutic strategies. The significant upregulation of REG1B and REG3A in the T1D cases would imply that these proteins may play a role in the persistence of beta cell mass in some people with T1D. Our current data is supports the subtractive hybridization studies by studies by Cho et al. that propose that REG3A is a potential target for genetic therapy in diabetes \[[@pone.0183908.ref044]\]. REG3A is also markedly increased in rats during pregnancy and is involved in islet regeneration and neogenesis \[[@pone.0183908.ref045]--[@pone.0183908.ref046]\]. The other significantly upregulated protein was complement 9 (C9), a sub-unit of the membrane attack complex (MAC) that is involved in innate and adaptive immune response mechanisms that destroy target cells by forming pores on their plasma membrane. C9 is 1.8 fold upregulated in T1D islets compared to the ND cases. Our observation is significant given that C9 and complement activation has been demonstrated to play a role in the lysis and destruction of transplanted pancreatic islets \[[@pone.0183908.ref047]\]. The other significantly upregulated protein includes ALDH1B1 which is a member of the alcohol dehydrogenase protein family that plays a role in the detoxification of alcohol-derived acetaldehydes. ALDH1B1 family of proteins are also involved in the metabolism of corticosteroids, biogenic amines, neurotransmitters, and lipid peroxidation. ALDH1B1 is 1.7 fold up regulated in T1D cases compared to the non-diabetic controls. The significance of this observation needs further studies although it has been demonstrated that there is a potential link between ALDH1B1and diabetes \[[@pone.0183908.ref048]\]. It is possible that its role in metabolism of lipid peroxidation is relevant given the role of 12-lipoxygenase in T1D \[[@pone.0183908.ref049]\]. The other significantly top upregulated proteins include abhydrolase domain containing 14B (ABHD14B)\], creatine kinase B-type (CKB), HIST1H1B (Histone H1.5) and ORM1 (also known as alpha-2-glycoprotein) are significantly upregulated in T1D cases compared to the controls. ORM1 is involved in modulating the activity of the immune system during the acute-phase reaction with the degranulation of neutrophils. In addition to insulin some of the other most significantly downregulated protein between T1D and the control cases is PFKB2 which is involved in glycolysis, hydroxyacyl-CoA dehydrogenase (HADH) which is involved in the beta-oxidation pathway and ATP2A3, magnesium-dependent enzyme catalyzes the hydrolysis of ATP coupled with the transport of calcium and STX1A, syntaxin 1A which is involved in peptide hormone metabolism and insulin secretion \[[@pone.0183908.ref050]\]. The others are SEC61B which is involved in the translocations of proteins to the ER, ASB9 which is involved in the ubiquitination pathway, PLCXD3 which has signal transducer and phosphoric diester hydrolase activity, UCHL1 which is a ubiquitin-protein hydrolase involved both in the processing of ubiquitin precursors and of ubiquitinated proteins and GSTM2 which is involved in the detoxification of electrophilic compounds and products of oxidative stress, by conjugation with glutathione. In the AAB+ cases the most significantly upregulated protein is REG1B which is also upregulated in T1D cases. In addition the other uniquely upregulated proteins include SPINK1, a serine protease inhibitor which exhibits anti-trypsin activity, PSMB6 that is involved in ATP-dependent proteolytic activity, SOD1 which is involved in the detoxification of free radicals and GSTP1 which performs similar functions. CD99 is involved in leukocyte migration and T-cell adhesion and the protein is a beta cell surface marker \[[@pone.0183908.ref051]\]. The other significantly upregulated protein is GAA (Lysosomal alpha-glucosidase) which is essential for the degradation of glycogen to glucose in lysosomes; CMBL which is a cysteine hydrolase, CUZD1 and PGM1 which participates in glucose synthesis and catabolism. Polymorphisms in PGM1 have been associated with gestational diabetes in expectant mothers and juvenile onset of diabetes \[[@pone.0183908.ref052]--[@pone.0183908.ref055]\]. The most significantly downregulated proteins in the AAB+ cases include SYCN, CTRL, BLVRB, CAV1, COA3, HNRNPUL1, PNLIPRP1, STAT3, CELA3B, and ECHCD3. SYCN localizes to zymogen granules and is associated with insulin secretion. Two previous studies have utilized LCM and mass spectrometry based proteomics to analyze differential protein expression in human islets obtained under different conditions and pathological states \[[@pone.0183908.ref056]--[@pone.0183908.ref057]\]. Unlike our current study, Nishida et al. \[[@pone.0183908.ref056]\], performed qualitative proteomic analysis to evaluate protein expression using the islets isolated from formalin-fixed paraffin-embedded pancreatic tissues from three autopsied cases of enterovirus-associated fulminant type 1 diabetes (FT1DM) with extensive insulitis compared to five autopsied non-diabetic men. On the other hand, Zhang et al. \[[@pone.0183908.ref057]\], recently performed comparative quantitative proteomic analyses of enzyme isolated cultured islets, and laser capture microdissected human islets and acinar tissue from fresh-frozen pancreas sections from three non-diabetic cadaveric donors. Unlike these two previous reports, our current proteomics data is obtained from 3 groups of disease stratified phenotype matched comprehensively annotated pancreas tissue samples from nPOD. Nishida et al, identified at total of 300 different proteins whereas Zhang et al identified at total of 1104 unique proteins. Therefore our current study which identifies a total of 2032 different unique protein groups, out of which 1491 proteins were quantified between ABB+ and T1D cases versus the cases with no evidence for disease, provides a very comprehensive proteome obtained from islets obtained after LCM. We have performed a comparative qualitative analysis between the data generated by Nishida et al, which analyzes disease stratified samples, and is most closely related to our current study. We have combined the list of proteins that Nishida, et al, identified in both islets affected by fulminant type 1 diabetes and in non-diabetic control pancreatic islets with those that they identified only in non-diabetic control pancreatic islets, and compared these proteins to the 2032 unique proteins that are identified in this study. [S2 Fig](#pone.0183908.s002){ref-type="supplementary-material"} summarizes the overlap between the two studies. There is a very highly significant overlap between the proteins that were identified by Nishida et al compared to our current study. 9 proteins KRT10, KRT2, RCL, YCN, SRSF8, ALB, PKM2, KRT9, HISTH2AB and KRT19 are exclusively identified in the previous study. It is important to note that a majority of these proteins including KRT10, KRT2, ALB, KRT9 and KRT19 are often considered as contaminants in proteomics experiments and are always excluded in comparative analyses. Taken together our data provides a very comprehensive proteomic analysis of disease stratified LCM islets to date. Conclusion {#sec014} ========== In this study, we report the successful analysis of protein expression in islets acquired from well annotated nPOD pancreas tissue samples by laser capture micro-dissection and in-depth proteomic analysis using mass spectrometry. The study provides comprehensive qualitative and quantitative information on the proteins expressed in islets and during T1D progression. This may provide pathomechanistic insights and provide the rationale and perspectives for novel therapeutic targets for intervention in T1D. The next steps would be to evaluate the presence and levels of the lead protein candidates in well characterized patient cohort samples from ongoing or completed studies evaluating the natural history and triggers of T1D. Supporting information {#sec015} ====================== ###### S100A9 localizes to islets in T1D donor pancreas tissues, but in the exocrine tissue of non-diabetic donors. Representative images from nPOD donors 6073 (ND), 6076 (T1D) and 6077 (T1D) are shown. Whole tissue lysates of non-diabetic and T1D show no difference in their expression of S100A9, however, there is a significant upregulation of S100A9 in the islets of T1D donors upon isolation of proteins using Laser Capture Microdissection. Glucagon (green) and S100A9 (red) immunofluorescent staining were carried out as described in the materials and methods section. (TIF) ###### Click here for additional data file. ###### Qualitative comparative analysis of 2032 unique proteins that are identified in this study (Nyalwidhe et al. 2017) and the proteins that are identified by Nishida et al. 2014. The proteins by identified by Nishida et al. include those identified in islets affected by fulminant type 1 diabetes and in non-diabetic control pancreatic islets those that are identified only in non-diabetic control pancreatic islets. A highly significant overlap exists between the proteins that are identified in the two studies. (TIF) ###### Click here for additional data file. ###### Donor phenotype and pancreas tissue samples used in the study. (DOCX) ###### Click here for additional data file. ###### Differentially expressed proteins in islets from non-diabetic and AAB+ cases. (XLSX) ###### Click here for additional data file. ###### Differentially expressed proteins in islets from non-diabetic and Type 1 diabetes cases. (XLSX) ###### Click here for additional data file. ###### Qualitative and quantitative comparisons between non-diabetes versus autoantibody cases and T1D cases. (XLSX) ###### Click here for additional data file. ###### Functional annotation and enrichment of differentially expressed proteins between T1D versus non-diabetes cases. (XLSX) ###### Click here for additional data file. ###### Functional annotation and enrichment of differentially expressed proteins between AAB+ versus non diabetes cases. (XLSX) ###### Click here for additional data file. We acknowledge Dr. Irina Kusmarteva from nPOD for assisting with the provision of the pancreas tissue sections for laser capture microdissection. We would also like to acknowledge the EVMS George L Wright Center for Biomedical Proteomics Facility for access to LC/MS/MS instrumentation. We would also like to acknowledge the EVMS Biorepository for access to laser capture microdissection instrumentation. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Background ========== The Dahl salt-sensitive (S) rat is a widely studied model of salt-induced hypertension \[[@B1],[@B2]\]. The Dahl S rat, but not the Dahl salt-resistant rat (R), develops hypertension, accompanied by hypovitaminosis D, when fed a high salt diet \[[@B3]-[@B5]\]. Plasma concentration of 25-hydroxyvitamin D (25-OHD), the liver metabolite of vitamin D, is a marker of vitamin D status. Plasma 25-OHD concentrations are similar in young S and R rats when the rats are fed a low salt diet (0.3% sodium chloride). High salt intake, however, causes significant decreases in plasma 25-OHD concentrations of S rats, but not R rats \[[@B3],[@B5]-[@B7]\]. Blood pressure was shown to be directly correlated and plasma 25-OHD concentration inversely correlated with the number of days that young S rats were fed an 8% salt diet \[[@B3],[@B6]\]. We demonstrated that Dahl S rats lose protein-bound vitamin D metabolites into urine and that this loss is markedly accelerated during high salt intake \[[@B8]\]. High salt intake, thus, creates a vitamin D deficiency state in Dahl S rats in the presence of standard amounts of dietary vitamin D \[[@B3],[@B5]-[@B7]\]. Since S and R rats were originally bred from the Sprague Dawley rat (SD) \[[@B1]\], we tested the hypothesis that R and SD rats would be similar in their vitamin D endocrine system response to high salt intake and not develop hypovitaminosis D, as does the S rat. Methods ======= Animals and diets ----------------- Male SD rats, Dahl S (SS/Jr) and R (SR/Jr) rats (130-150 g, 4-5 weeks old, twelve rats per type) were obtained from Harlan Sprague Dawley, Indianapolis, IN. All protocols involving animals were previously approved by the Morehouse School of Medicine Animal Care Committee. Guidelines followed were those of the Public Health Service and the revised animal welfare act as regulated by USDA. The rats were maintained as previously described \[[@B9]\]. They were housed in a room with 12 h light - dark cycles and, after one week of acclimation, six rats of each type were fed either a low (3 g/kg) or high (80 g/kg) salt diet (Harlan Teklad, Madison, WI) for three weeks. Diets, blood collection, and blood pressure measurements have been previously described \[[@B9]\]. Plasma 25-hydroxyvitamin D concentration ---------------------------------------- Plasma 25-OHD concentration was measured as previously described \[[@B3]\]. Plasma samples were purified using a dichloromethane/methanol liquid-liquid extraction, followed by solid-phase extraction. Fraction 1 from the solid phase extraction was used to assay for 25-OHD by radioimmunoassay kits available at the time of the study from Amersham Corp. (Arlington Heights, IL). Statistics ---------- A mean ± SEM was calculated for each group. Statistical significance (*P*\< 0.05) was evaluated by the Mann Whitney test (SigmaStat, SPSS, Inc., Chicago, IL.). Results ======= Blood pressure significantly increased from baseline to week 3 among all groups (Figure [1](#F1){ref-type="fig"}). S and SD rats, but not R rats, exhibited significantly higher mean blood pressures at week 3 of high salt intake than at week 3 of low salt intake. The blood pressures plotted (Figure [1](#F1){ref-type="fig"}) are means for six rats/dietary group (set 1) of 12 S and 12 R rats/dietary group in a previous report \[[@B9]\]. Set 1 of that study included 12 SD rats, 6/dietary group. Mean plasma 25-OHD concentrations of SD, R, and S rats were similar at week 3 of low salt intake (Figure [2](#F2){ref-type="fig"}). Mean plasma 25-OHD concentrations of SD and R rats were similar at week 3 of high and low salt intake, whereas mean plasma 25-OHD concentration of S rats at week 3 of high salt intake was only 20% of that at week 3 of low salt intake. {#F1} ![**Plasma 25-hydroxyvitamin D concentrations of Sprague Dawley and Dahl salt-sensitive and salt-resistant rats**. Sprague Dawley and Dahl salt-sensitive and salt-resistant rats were fed low (0.3%) or high (8%) salt diets for three weeks. Plasma samples were pre-purified by previously published methods \[[@B3]\] and 25-hydroxyvitamin D was assayed using a commercial kit (Amersham Corp., Arlington Heights, IL). Values are means ± SEM, n = 6. \*\**P =*0.002, high salt intake vs. low salt intake, salt-sensitive rats.](1756-0500-3-332-2){#F2} Discussion ========== The blood pressure of SD rats increased from baseline to week 3 during both high and low salt intake, and the mean blood pressure at week 3 of high salt intake was significantly higher than that at week 3 of low salt intake. Mean blood pressure of SD rats at week 3 of high salt intake was 114 ± 4 mm Hg, compared with means of 107 ± 2 mm Hg for R rats and 182 ± 6 mm Hg for S rats. Plasma 25-OHD concentrations of SD and R rats were unaffected by high salt intake, whereas mean plasma 25-OHD concentration of S rats at week 3 of high salt intake was decreased to 20% of that at week 3 of low salt intake. These data indicate that the vitamin D endocrine system response of SD rats to a high salt load (8%) for three weeks is more similar to that of R rats than to that of S rats. Dahl S rats, but not Dahl R rats, are insulin-resistant \[[@B10]-[@B12]\]. Ogihara et al. \[[@B13]\] have demonstrated that a high salt diet induces insulin resistance in both S and SD rats. It has thus been suggested that the SD rat is essentially the same as the Dahl S strain \[[@B14]\]. Channa et al. \[[@B12]\] have suggested that insulin resistance and hypertension may be inherited as separate traits. In this study, the blood pressure response of SD rats to high salt intake was between that of R and S rats. This and other studies suggest that SD rats are similar to S rats in the induction of insulin resistance by high salt intake \[[@B13],[@B14]\], but similar to R rats in the vitamin D endocrine response to high salt intake. The S rat, thus, appears to be a more appropriate model than the SD rat for assessing possible effects of salt-sensitivity on vitamin D status of humans. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= MT-P participated in the conception and design of the study and the vitamin D metabolite analysis and drafted the manuscript. TKT carried out the vitamin D metabolite analysis. NLE participated in the conception and design of the study. MAB participated in the conception, design, and coordination of the study and in the blood pressure measurements. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by the National Aeronautics and Space Administration grant NCC 9-53. Facilities and support services were partially funded by the National Institutes of Health/National Center for Research Resources grants RR03034 and 1 C06 RR18386. The authors thank Min Wang for blood pressure measurements and Stacy Cephas for statistical analyses.

Key points {#sec1} ========== •The ongoing coronavirus disease 2019 (COVID-19) pandemic has affected hundreds of thousands of people.•Children have so far accounted for 1.7% to 2% of diagnosed cases of COVID-19.•Children often have milder disease than adults, and child deaths have been rare.•Risk factors for severe disease from COVID-19 in children are reported to be young age and underlying comorbidities, although this is not confirmed in all studies.•It is unclear whether male gender and certain laboratory and imaging findings can also be considered as risk factors, because of insufficient data. Introduction {#sec2} ============ Until recently, 6 different coronaviruses (CoVs) had been identified in humans (human CoVs \[HCoVs\]): HCoV-OC43, HCoV-229E, HCoV-NL63, HCoV-HKU1, severe acute respiratory syndrome (SARS)-CoVs, and MERS-CoVs. Endemic HCoV-OC43 and HCoV-229E were described in the 1960s, and HCoV-NL63 and HCoV-HKU1 in 2004 and 2005, respectively.[@bib1] ^,^ [@bib2] The first serious CoV disease outbreak occurred in China in 2002, when the novel SARS-CoV emerged, which was thought to have been transmitted from civet cats or bats to humans.[@bib3] ^,^ [@bib4] The second novel CoV emerged in Saudi Arabia in 2012, the Middle East respiratory syndrome (MERS)--CoV,[@bib5] which is transmitted from dromedary camels to humans.[@bib6] Collectively, these 2 CoV diseases did not affect children widely, because of the short-term nature of the epidemic of SARS and the rigid transmission route of MERS. Since December 2019, SARS-CoV-2 has been recognized as the causal factor of severe pneumonia and potential damage to vital organs in humans. The first cases of SARS-CoV-2 originated in Wuhan in the Hubei province of China, and subsequently spread to other countries throughout the world.[@bib7] In February 2020, the World Health Organization (WHO) designated the disease CoV disease 2019 (COVID-19). A substantial number of studies have already been published on adults with COVID-19, but reports on children with COVID-19 are scarce. This article analyzes the current knowledge on the risk factors for the progression and severity of COVID-19 in infants and children. The possible mechanisms of aberrant clinical features of COVID-19 in children are also presented. To the best of our knowledge, this is the first review addressing the risk factors associated with the progression and severity of COVID-19 in children. Methods {#sec3} ======= Original research studies published in English between February 26, 2020 and June 10, 2020 were identified using PubMed and Scopus. The search used combinations of the key words "COVID-19," "SARS-CoV2," "mechanism," :risk factor," "severity," and "child." In addition, the reference lists of the retrieved articles were checked for other relevant articles. The initial search yielded 293 articles, of which, after screening of their titles, 72 studies were considered relevant to the aim of this review. Studies on adults and neonates were not included, and 7 studies were excluded because they were in Chinese. Pediatric case reports of COVID-19 were included only if they provided information about risk factors for severe disease. Thus, 23 studies were eventually selected, as shown in [Fig. 1](#fig1){ref-type="fig"} , and are discussed here. The factors that may introduce bias into the findings of this article are restriction to articles in English, together with database and citation bias.Fig. 1The literature search on risk factors for severe COVID-19 in childhood (February 26 to June 10, 2020). Most of the studies originated in China, the United States, Italy, Spain, and South Korea, despite the large number of patients diagnosed with COVID-19 throughout the world. Some published studies relating to COVID-19 in children do not provide detailed information on the mechanisms, triggering factors, or clinical features, which led to the deterioration of the status of the patients. In addition, the current studies do not provide a uniform definition of severe or critical disease. The information from all the studies related to the risk factors for severe COVID-19 in infants and children is summarized in [Table 1](#tbl1){ref-type="table"} .Table 1Studies on severity and risk factors of coronavirus disease 2019 in children (February 26 to June 10, 2020)First AuthorRegionStudy PeriodNumber of ChildrenMean Age (% of Young Children)Underlying Diseases Present (Diseases)SeverityRisk FactorsBialek et al[@bib9]United States (33% from New York City, 23% from the rest of New York State, 15% from New Jersey, 29% from other jurisdictions)February 12 to April 2, 2020257211 (\<1 y, 15%)23% (chronic lung disease, cardiovascular disease, immunosuppression)5.7%--20% hospitalized, 0.58%--2% admitted to ICU, aged \<1 y: 15%--62% hospitalized, 3 deathsChildren aged \<1 y, underlying conditionDong et al,[@bib10] 2020Chinese CDC, cases from Hubei province and Anhui, Henan, Hunan, Jiangxi, Shanxi, and ChongqingJanuary 16 to February 8, 20202135 suspected and confirmed cases7 (\<1 y, 17.6%)Not available90% had asymptomatic to moderate disease\ Severe or critical disease in 10.6% \<1 y, 7.3% 1--5 y, 4.1% 6--10 y, 3% \>16 y; 1 14-year-old boy diedYoung ageLu et al,[@bib30] 2020Wuhan Children's Hospital, ChinaJanuary 28 to February 26, 20201716.7 (\<1 y, 18%)3 patients (hydronephrosis, leukemia, intussusception)3 patients with invasive mechanical ventilation (all with underlying condition), 1 deathUnderlying conditionDeBiasi et al,[@bib22] 2020Children's National Hospital WashingtonMarch 15 to April 30, 20201779.639% (asthma, neurologic condition, DM, obesity, cardiac problem, hematological disease, oncological condition)9 critically ill patientsAdolescents and young adultsParri & Leng,[@bib32] 2020Italy, 17 pediatric emergency departments, the CONFIDENCE studyMarch 3 to March 27, 20201003.3 (40% \<1 y, 14% \<5 y)27%, cystic fibrosis; neurologic, hematological, cardiac, immunologic, oncological conditions; metabolic disease; prematurity syndrome1% had severe disease, 1% were in critical conditionUnderlying medical condition, young ageChao et al,[@bib44] 2020Single tertiary children's hospital, New York CityMarch 15 to April 13, 20206713.1Obesity and asthma33 admitted to ICUHigher levels of CRP, procalcitonin, and proBNP and platelet countWhittaker et al,[@bib24] 20208 hospitals in United KingdomMarch 23 to May 16, 20205893 had asthma, 1 neurodisability, 1 epilepsy, 1 sickle cell disease, 1 alopeciaAll had multisystem inflammatory syndrome, 50% developed shock, and 14% coronary artery aneurysmIncreased CRP and ferritin levels, older age, black or Asian raceShekerdemian et al,[@bib29] 202046 North American ICUsMarch 14 to April 3, 2020481383%All admitted to ICU, 23% had multiorgan failure, 2% needed extracorporeal membrane oxygenation, 4% diedUnderlying comorbiditiesTagarro et al,[@bib35] 202030 hospitals in Madrid, SpainMarch 2 to March 16, 202041127% had underlying disease60% hospitalized, 9.7% admitted to ICU, 9.7% needed respiratory support (1 had underlying condition)Perhaps young age, underlying conditionQiu et al,[@bib43] 20203 hospitals, Zhejiang, ChinaJanuary 17 to March 1, 2020368.3 (\<5 y, 28%)Not availableAll patients had mild or moderate typeRadiographic presentation, decreased lymphocyte counts, increased body temperature, high levels of procalcitonin, D-dimers, and creatine kinase-MBBelhadjer et al,[@bib49] 202014 ICUs in France and SwitzerlandMarch 22 to April 30, 2020351028% had comorbidities (asthma, overweight)Multisystem inflammatory syndrome--acute cardiac failureCytokine storm and macrophage activationBandi et al,[@bib23] 2020University COVID-19 clinic, Chicago, IL12 March to 20 April, 2020259.7 yNot available (1 sickle cell acute pain crisis)20% hospitalized, 12% admitted to ICU, 1 intubatedOlder age\ African American raceZheng et al,[@bib33] 202010 hospitals, Hubei, ChinaFebruary 1 to February 10, 2020253 (\<3 y 40%)8% (congenital heart disease, malnutrition, suspected hereditary metabolic diseases)Most patients had mild disease\ Two had critical disease (both with underlying disorder)Underlying disordersCheung et al,[@bib51] 2020Columbia University Irving Medical Center/New York-Presbyterian Morgan Stanley Children's Hospital in New York CityApril 18 to May 5, 20201783 mild asthmaMultisystem inflammatory syndromeInflammatory markers, troponin T, and NT-proBNP levelsVerdoni et al,[@bib48] 2020Bergamo province, ItalyFebruary 18 to April 20, 2020107.5NoneMultisystem inflammatory syndromeOlder age, features of macrophage activationRiphagen et al,[@bib47] 2020ICU, United KingdomMid-April, 202089NoneMultisystem inflammatory syndromeAfro-Caribbean descent\ Male genderSun et al,[@bib34] 2020ICU of Wuhan Children's Hospital, ChinaJanuary 24 to February 24, 202087 (3 children ≤1 y)1 acute lymphoblastic leukemiaAll admitted to ICUIncreased levels of CRP, LDH, procalcitonin, abnormal liver function, cytokine storm, abnormalities on chest CTLiu & Zhang,[@bib19] 20203 branches of Tongji Hospital, Wuhan, China)January 7 to January 15, 202063 (4 children ≤3 y)NoneAll 4 patients ≤3 y had pneumonia, 1 admitted to ICUΥoung ageCui et al,[@bib18] 2020Hubei Province, ChinaJanuary 28, 2020155 dNonePneumonia, myocardial injury, acute liver injuryYoung ageShi et al,[@bib42] 2020Hubei Province, ChinaFebruary 3, 202012 moNoneSevere pneumonia, need for noninvasive ventilationYoung age, coinfection with RSV[^2] Epidemiology of coronavirus disease 2019 {#sec4} ======================================== COVID-19 worldwide is less common in children than in adults. A review of 72,314 cases by the Chinese Center for Disease Control and Prevention showed that less than 1% of the cases were in children younger than 10 years and 1% of the cases were in children aged 10 to 19 years.[@bib8] In the United States, among 149,082 reported cases of COVID-19, 1.7% were in children aged less than 18 years.[@bib9] From the currently available data, it seems that children tend to have asymptomatic or mild disease more commonly than adults,[@bib8] ^,^ [@bib10] but severe cases and even deaths have been reported worldwide in patients younger than 18 years. In a cohort study of 32,583 confirmed cases of COVID-19 from Wuhan, China, 4.1% of severe and critical cases were in patients aged less than 20 years.[@bib11] According to a large retrospective study conducted in China, 4 HCoVs, HCoV-OC43, HCoV-229E, HCoV-NL63, and HCoV-HKU1, were more common in children, because their prevalence was 4.3%, and the highest prevalence was among infants aged 7 to 12 months.[@bib12] Infection by these 4 strains usually causes acute respiratory disease, with severe manifestations in some children.[@bib13] Regarding SARS-CoV, only 6 case series have been reported, including a total of 135 pediatric cases, from Canada, Hong Kong, Taiwan, and Singapore.[@bib14] A milder form of the disease was observed in children compared with adults, and no child death was recorded.[@bib15] In the MERS-CoV epidemic, pediatric cases were even fewer, because only 2 small case series of children were reported, both originating from Saudi Arabia, 1 of 31 children with a mean age of 10 years[@bib16] and 1 of 7 children with a mean age of 8 years.[@bib17] In both studies, 42% of the infected children were asymptomatic,[@bib16] ^,^ [@bib17] and in 1, 2 of the 7 had severe disease,[@bib17] whereas in the other, 2 of the 31 children died (6%).[@bib16] Risk factors for severity in coronavirus disease 2019 and other coronavirus infections {#sec5} ====================================================================================== The Impact of Age {#sec5.1} ----------------- ### Severe acute respiratory syndrome--coronavirus-2 {#sec5.1.1} In a series of 2135 children with suspected and confirmed COVID-19 from China, severe disease was defined as the occurrence of dyspnea, central cyanosis, and oxygen saturation of less than 92%. Critical disease was defined as progression to acute respiratory distress syndrome, shock, encephalopathy, myocardial injury, coagulation dysfunction, and acute kidney injury.[@bib10] Severe and critical cases were reported in 10.6% of the children aged less than 1 year, 7.3% of those aged 1 to 5 years, 4.1% of those aged 6 to 10 years, and 3% of the children aged greater than 16 years. One 14-year-old boy died, but no further information was provided about this patient, and the study gave no data on underlying comorbidity or other possible risk factors. It is of note that, of the 2135 children, only 728 had laboratory confirmation, and the severe symptoms in the suspected cases may have been be caused by pathogens other than SARS-CoV-2. Two case reports from the same country, China, referred to children with severe disease, a 55-day-old female infant and a 3-year-old girl with no apparent risk factor apart from the young age.[@bib18] ^,^ [@bib19] Cases have been reported of infants in China and in Vietnam that, despite their young age, had mild disease, including 10 diagnosed with COVID-19 who were otherwise healthy, with mild or no symptoms.[@bib20] ^,^ [@bib21] In a study of 177 children from the Children's National Hospital in Washington, DC, the adolescents and young adults were more commonly critically ill than the younger children.[@bib22] Another study from the United States reported that the mean age of COVID-19--positive children was significantly higher than those testing negative (9.72 vs 4.85 years). In that study, the ethnicity was examined, and African American children had a significantly higher rate of positive tests for COVID-19: 6.8% versus 1.7% of white children.[@bib23] In a study in the United Kingdom, among 58 children, race (black or Asian) was described as a risk factor for COVID-19.[@bib24] ### Other coronaviruses {#sec5.1.2} In the United States, in the case of other CoVs, specifically 229E, HKU1, NL63, and OC43, age less than 2 years has been reported as a risk factor for severe disease, defined as the need for respiratory support.[@bib25] In contrast, in a series of 44 children in China with SARS-CoV, an age of greater than 12 years was associated with severe illness, requiring methylprednisolone therapy and oxygen supplementation.[@bib15] In adults, older age has been reported to be an independent risk factor for severity and mortality, not only in SARS-CoV-2 but also in the previous epidemics of SARS and MERS.[@bib26] ^,^ [@bib27] The Impact of Male Gender {#sec5.2} ------------------------- Male gender is a risk factor for severe CoV disease in adults.[@bib28] A predominance of men was reported in all age subgroups among 2490 pediatric cases of COVID-19 in a series in the United States, but no details were given about the impact of gender on the severity of the disease.[@bib9] Among 2143 Chinese children with COVID-19 in the study of Dong and colleagues,[@bib10] no significant difference was reported in the number of cases between boys and girls, and no detailed information was given on the gender of the severe and critical cases. In a cross-sectional study of 48 children with COVID-19 admitted to US and Canadian intensive care units (ICUs), 52% were boys.[@bib29] Severe disease has been reported in girls and the current data suggest that, in children, male gender is not an independent risk factor of severe COVID-19. Underlying Medical Comorbidity {#sec5.3} ------------------------------ ### Severe acute respiratory syndrome--coronavirus-2 {#sec5.3.1} In a series of 171 children with COVID-19 from the city in China, Wuhan, where SARS-CoV-2 was first described, 3 patients required ICU support and invasive mechanical ventilation, all of whom had underlying comorbidities. One was a 10-month-old male infant with intussusception who developed multiorgan failure and died 4 weeks after admission.[@bib30] The second child had leukemia, in the maintenance chemotherapy phase, and the third, aged 13 months, had bilateral hydronephrosis and calculus of the left kidney.[@bib30] ^,^ [@bib31] It was not reported whether any of the 168 children who did not need ICU admission had an underlying condition. In the recently published The Coronavirus Infection in Pediatric Emergency Departments (CONFIDENCE) study from Italy, which included 100 children, 27% had an underlying medical condition. Of the 9 children needing respiratory support, 5 were aged less than 1 year and 6 had an underlying condition. The severe (1) and critical (1) cases were both in children with underlying medical conditions.[@bib32] Among 25 pediatric cases of COVID-19 from Hubei province in China, two 1-year-old boys needed invasive mechanical ventilation, both of whom had congenital heart disease. One of them also had malnutrition and a suspected hereditary metabolic disease, and the other had coinfection with *Enterobacter aerogenes*.[@bib33] The first report from the United States concerning children with COVID-19 is of 2572 pediatric cases. Among the children for whom hospitalization status was known, 20% were hospitalized. Because of lack of information on specific disease features, hospitalization was considered to be an indicator of serious illness, and it was most often reported in children younger than 1 year. An underlying medical condition was noted in 77% of hospitalized children, in contrast with 12% of those not hospitalized. The most common comorbidities were chronic lung disease (including asthma), cardiovascular disease, and immune suppression. Three deaths were reported, but their association with COVID-19 is still under investigation.[@bib9] In another US study, among 48 children admitted to an ICU, 83% had a significant preexisting comorbidity.[@bib29] Severe and critical cases have also been reported in children with no underlying comorbidity. Sun and colleagues[@bib34] reported 8 severe and critical cases of children in a hospital in Wuhan, 7 of whom were previously completely healthy. In this study, severe cases were defined as the coexistence of tachypnea, oxygen saturation less than 93%, and arterial partial pressure of oxygen less than or equal to 300 mm Hg, whereas critical cases were defined as the presence of septic shock or the need for mechanical ventilation or ICU admission. The age range of the patients in the 8 severe cases was from 2 months to 15 years, 6 were boys, and only 1 of them had an underlying medical condition (acute lymphocytic leukemia).[@bib34] Information from a registry of 310 hospitals in Madrid, Spain, showed that, of 41 children with COVID-19, 60% were hospitalized, 4 children were admitted to an ICU, and 4 needed respiratory support. Of these children, 1 had a previous condition (recurrent wheezing) and no patient died.[@bib35] In a recent report from Paris, France, of 27 children with severe COVID-19, 70% had an underlying medical condition. Of the 5 children who died, 3 had no underlying comorbidity, suggesting that comorbidities may be a risk factor for severe disease and fatality but that other mechanisms may also be implicated in the severity of the disease.[@bib36] It seems, therefore, that although underlying medical comorbidity may be a risk factor for severe disease in childhood, it is not the only risk factor for progression of the disease and development of complications. It would be of interest to gather further information on the children with underlying medical problems and assess the percentages with severe or mild disease, and their other risk factors. To date, there is lack of such data in the literature, although, in adults, specific comorbidities are well documented as risk factors not only for admission to the ICU but also for mortality.[@bib37] ### Other coronaviruses {#sec5.3.2} Severe pediatric disease from other CoVs reported in the United States, specifically 229E, HKU1, NL63, and OC43, defined as need for respiratory support or pediatric ICU admission, has been associated with underlying comorbidity, and, in particular, cardiovascular, chronic respiratory, and genetic/congenital conditions.[@bib25] Ogimi and colleagues[@bib38] in the United States showed that both an immunocompromised state and underlying pulmonary disorder were associated with lower respiratory tract disease or severe lower respiratory tract disease from HCoV. No significant difference was found regarding the severity of illness among hospitalized children with different HCoV types.[@bib25] The 2 deaths reported in children with MERS-CoV in Saudi Arabia were in a 2-year-old child with cystic fibrosis[@bib39] and a 9-month-old infant with infantile nephrotic syndrome,[@bib40] whereas a 14-year-old girl with Down syndrome needed hospital admission but eventually recovered.[@bib39] Coinfection with Another Pathogen {#sec5.4} --------------------------------- ### Severe acute respiratory syndrome--coronavirus-2 {#sec5.4.1} Coinfection with other pathogens may be a risk factor for severe disease. One child in Wuhan with a history of congenital heart disease and severe illness was found to have coinfection with *E aerogenes*.[@bib33] In a study of 20 pediatric cases from the same region, 40% had an underlying coinfection, but there was no report on their severity.[@bib41] A severe case of COVID-19 has been reported in a Chinese 2-month-old infant who had coinfection with respiratory syncytial virus (RSV).[@bib42] ### Other coronaviruses {#sec5.4.2} The presence of copathogens with more than 1 HCoV strain (229E, HKU1, NL63, and OC43) or other respiratory pathogens is a risk factor for febrile illness. Patients infected with a single strain of HCoV were more likely to present pulmonary rales than those infected by more than 1 HCoV strain or other respiratory pathogens.[@bib12] The presence of RSV has been associated with lower respiratory tract disease or severe lower respiratory tract disease from HCoV.[@bib38] Laboratory Findings {#sec5.5} ------------------- This article reports only the available laboratory information on the severe cases compared with mild cases, according to the current literature; several publications did not provide relevant data. ### Severe acute respiratory syndrome--coronavirus-2 {#sec5.5.1} Based on currently available data, it is not possible to document a pattern of laboratory values in pediatric COVID-19 according to the severity of the disease. In the study of Qiu and colleagues[@bib43] from China, no laboratory data were reported for severe cases, but only for 36 children with moderate and mild disease. Moderate cases (19 patients) compared with mild cases (17 patients) were associated with increased body temperature, a decrease in lymphocyte counts, higher levels of procalcitonin and creatine kinase-MB (myocardial band), and increased D-dimer levels.[@bib43] Laboratory data from 8 severe pediatric cases in the same country showed normal or increased leukocyte count, and high levels of C-reactive protein (CRP), procalcitonin, and lactate dehydrogenase, whereas half had abnormal liver function tests.[@bib34] In a study of 67 children in the United States, admission to an ICU was associated with higher levels of CRP, procalcitonin, and pro--B-type natriuretic peptide and an increased platelet count.[@bib44] Henry and colleagues[@bib45] reviewed 2020 case reports and case series providing laboratory data on pediatric cases of COVID-19. In that review, 69.6% of the children had a normal leukocyte count and the investigators commented that the absence of lymphopenia in children may in part be explained by the milder disease. Another assumption was that increased procalcitonin level could be caused by a bacterial coinfection as a complication of COVID-19.[@bib45] Procalcitonin level was increased in 80% of Chinese pediatric patients in the study of Xia and Shao,[@bib41] and, in that series, 40% of the children had a coinfection. ### Other coronaviruses {#sec5.5.2} Neutrophilia was a predictor of severe illness among 44 children with SARS.[@bib15] Lymphopenia was detected in 10 children with SARS, of whom 4 needed oxygen therapy and 2 needed assisted ventilation.[@bib46] Risk factors for pediatric multisystem inflammatory syndrome associated with severe acute respiratory syndrome--coronavirus-2 {#sec6} ============================================================================================================================= A syndrome of fever and multisystem inflammatory syndrome (MIS) has recently been described in children with COVID-19. Some of these children presented with shock and multiorgan failure and others had characteristics of Kawasaki disease or a combination of Kawasaki-like disease and shock, named the Kawasaki disease shock syndrome.[@bib47] ^,^ [@bib48] These children presented with acute cardiac decompensation,[@bib49] and some developed coronary artery aneurysms.[@bib24] Among 44 children hospitalized in the United States with MIS, 84.1% had gastrointestinal symptoms as the presenting clinical complaint.[@bib50] Most studies to date have reported that MIS presents in children at an older age, with a median age of 8 to 10 years.[@bib24] ^,^ [@bib49] ^,^ [@bib51] In a retrospective study of 35 children with MIS, admitted to ICUs in France and Switzerland, comorbidities were present in 28% of the children, including asthma and being overweight,[@bib49] but most of the children in other studies reported from Europe, specifically Italy and the United Kingdom, were previously healthy.[@bib24] ^,^ [@bib48] In a study of 8 children from the United Kingdom with MIS, 6 were Afro-Caribbean and 5 were male.[@bib47] It has been suggested that black and Asian races may be predisposed to this clinical complication.[@bib24] These limited data indicate a possible gender and race predilection for MIS. The laboratory findings in children with MIS were characterized by a marked increase in levels of inflammatory markers such as CRP and ferritin,[@bib24] and a cytokine storm, with specific increase in the level of interleukin (IL)-6 and macrophage activation.[@bib49] ^,^ [@bib51] The patients often had a significant increase in B-type natriuretic peptide and troponin T.[@bib48] MIS is considered to be a result of a continuous immune response rather than injury from an acute SARS-CoV-2 infection. The disease presented 2 to 3 weeks after the peak of the infection and most children had negative COVID-19 polymerase chain reaction but positive viral serology.[@bib52] What mechanisms play a role in the atypical picture of coronavirus disease 2019 in children? {#sec7} ============================================================================================ The SARS-CoV-2 is a β CoV of group 2B, with more than 70% similarities in genetic sequence to SARS-nCoV.[@bib53] The established scientific evidence on SARS-novel coronavirus has enabled elucidation of the host defense mechanisms against SARS-CoV-2 and helped to explain the lower susceptibility of children to the virus and the variability between children. The reasons for the different pattern of COVID-19 in children are still unclear, but several hypotheses have been put forward. Environment-Epigenetics {#sec7.1} ----------------------- The effect of the environment must be considered a factor with significant impact on infection with COVID-19. Children have healthier airway tracts, because of having less exposure to cigarette smoke, air pollution, chemicals, and industrial pollutants than adults. In adults, these environmental factors, and especially smoking, have a negative epigenetic impact on epithelial and immune cells, leading to increased vulnerability to all respiratory viruses, including SARS-CoV-2.[@bib54] ^,^ [@bib55] CoVs are known to alter the epigenetic cellular mechanisms of the host associated with viral entry, replication, and innate immune control.[@bib56] Most children hospitalized with COVID-19, especially those in the ICU, were less than 3 years of age.[@bib33] ^,^ [@bib35] This finding may be explained by the immaturity of the immune system in this age period, the low likelihood of wearing face masks in this age group, and the subsequent high viral load.[@bib57] Another reason for the different clinical picture of COVID-19 in children is that they have fewer underlying disorders that may predispose to severe COVID-19 than adults.[@bib58] The severity of COVID-19 is higher in children with preexisting conditions, such as asthma, malignancies, cardiovascular disorders, and immunosuppression.[@bib33] ^,^ [@bib35] In certain chronic diseases, including systemic lupus erythematosus (SLE), epigenetic dysregulation might enable viral entry, replication, and a disproportionate immune response to SARS-CoV-2.[@bib59] Entry of the Virus into the Cells {#sec7.2} --------------------------------- Angiotensin-converting enzyme 2 (ACE2) is a zinc-containing metalloenzyme located on the surface of endothelial and other cells that counters the activity of the related angiotensin-converting enzyme (ACE) by reducing the amount of angiotensin-II.[@bib60] ACE2 serves as the entry point into cells for NL63 and SARS-CoV, and recent studies indicate that ACE2 is also likely to be the receptor for SARS-CoV-2 and the key region responsible for the interaction.[@bib61] ^,^ [@bib62] Differences in the distribution, maturation, and functioning of ACE2 in the developing phase of childhood is a possible reason for milder SARS-CoV-2 infection. Newborn infants and children have higher ACE activities, with serum levels showing an increase until puberty and progressive reduction after maturity.[@bib63] In contrast, ACE2 expression in rat lung has been found to decrease dramatically with age.[@bib64] Studies have provided evidence that ACE2 also protects against the severe acute lung injury that can be activated by sepsis, SARS, and avian influenza A H5N1 virus infection.[@bib65] It may be that children are protected against SARS-CoV-2 because ACE2 is less mature at younger ages. Epigenetic alteration of ACE2, which is further exacerbated by virus infections, is another potential mechanism in the severity of COVID-19 in patients with chronic diseases such as SLE.[@bib59] Another aspect in the variability of severity is the genetic variation of ACE among different populations. The polymorphism D/I in ACE1, an enzyme with amino acid identity and function similar to ACE2, could explain the varying rate of COVID-19 infection between European countries, and, specifically, the prevalence of COVID-19 infections has been shown to be correlated with the ACE D allele frequency.[@bib66] Immune Antiviral Response {#sec7.3} ------------------------- Frequent exposure of children to viral infections boosts the immune system and possibly enhances the response to SARS-CoV-2, and the presence of other concurrent viruses in the airway mucosa may limit the replication and the viral load of SARS-CoV2.[@bib67] It has been shown that the number of viral copies is correlated with the severity of COVID-19.[@bib68] The immune system undergoes significant changes from birth to adulthood, especially in lymphocyte biology,[@bib69] and the interaction of lymphocytes with SARS-CoV-2 may be different in children from that in adults. It is of note that, when documented, lymphocytopenia is frequent in adults with COVID-19 (83%)[@bib70] but not in children (3%).[@bib30] ^,^ [@bib45] However, in the 2003 SARS epidemic, lymphocytopenia was reported in 77% of infected children.[@bib15] The changing level of T lymphocytes with age may also be a reason for the mild disease phenotype in childhood.[@bib71] Interferon-mediated response to HCoVs is essential for the disease course. Virus-induced suppression of interferon-induced pathways leads to viral replication and disease progression, along with the production of other proinflammatory cytokines, such as IL-2, IL-6, and tumor necrosis factor, in the lower respiratory tract and other tissues.[@bib72] In some cases, the increase of cytokine levels is uncontrolled, leading to the detrimental cytokine syndrome, with a poor outcome.[@bib73] The percentage of children with COVID-19 with increased levels of inflammatory markers is reported to be low, and this could be a cofactor for nonsevere disease.[@bib45] In contrast, an unusual immune response accompanied by cytokine storm and macrophage activation is thought to result in MIS, which has been linked to COVID-19 in children.[@bib24] Another immunologic aspect that could be related to the mild disease in children is trained innate immunity, because of the routine use of various vaccines, including bacillus Calmette-Guérin(BCG). BCG vaccination induces epigenetic changes in monocytes, and increased cytokine production in response to several different pathogens.[@bib74] In mice, BCG also enhances nonspecific defense against influenza virus infection.[@bib75] Several studies have identified links between inadequate vitamin D concentrations and the development of upper and lower respiratory tract infections in infants and young children. Although the mechanism of the vitamin D effect on immunity is complex, currently available data support the hypothesis that cathelicidins and defensins can reduce viral replication rates and the levels of proinflammatory cytokines.[@bib76] Studies in small children with influenza have shown that high doses of vitamin D resulted in fast relief from symptoms, a rapid decrease in viral load, and early disease recovery. In addition, high daily doses of vitamin D have been shown to be effective in the prevention of seasonal influenza.[@bib77] Summary {#sec8} ======= Although children are less susceptible to COVID-19, and the clinical picture in childhood is often distinct from that in adults, in both age groups chronic underlying medical problems can predispose to severe disease. In contrast with adults, in whom older age is an independent risk factor for severity and mortality, very young age is considered a risk factor for severity in children, although this has recently been questioned, and MIS occurs in older children. Although a distinct pattern of laboratory findings has not emerged as being associated with severity of the disease in pediatric cases of COVID-19, lymphopenia seems to be a risk factor for severe disease in children. Increased levels of the inflammatory markers procalcitonin and CRP could be caused by a bacterial coinfection as a complication of COVID-19. The recently described pediatric MIS seems to be the result of continuous immune response rather than an injury from an acute SARS-CoV-2 infection, but further studies are needed to reach definitive conclusions. Several other aspects could be implicated in the severity of COVID-19 in children, such as coinfection with RSV, responsiveness of the immune system, vaccination history, levels of vitamin D, and genetic polymorphisms, but the present paucity of data limits the ability to draw such conclusions. It is important to further study the potential risk factors for severe disease in children and to clarify the underlying mechanisms in order to improve the management of children with COVID-19 and to help in the development of new forms of treatment. Contributors {#sec9} ============ S. Tsabouri and A. Makis designed the study, and S. Tsabouri, A. Makis, and C. Kosmeri did the literature search. A. Makis, C. Kosmeri, and E. Siomou were responsible for the data collection. S. Tsabouri and C. Kosmeri collected and analyzed the data. S. Tsabouri, A. Makis, C. Kosmeri, and E. Siomou analyzed data and wrote the article. Funding: None. Conflicts of interest: The authors have no conflicts to disclose. [^1]: Equal contributors. [^2]: *Abbreviations:* CDC, Center for Disease Control and Prevention; CRP, C-reactive protein; CT, computed tomography; DM, diabetes mellitus; ICU, intensive care unit; LDH, lactate dehydrogenase; MB, myocardial band; NT-proBNP, N-terminal pro--b-type natriuretic peptide; proBNP, pro--b-type natriuretic peptide; RSV, respiratory syncytial virus.

Following ocular injury, innate immune responses lead to deleterious inflammatory tissue damage, primarily through the recruitment of activated neutrophils. Despite their role as an essential arm of innate immunity, activated neutrophils contribute greatly to the collateral tissue damage associated with acute injury and inflammation.^[@i1552-5783-59-3-1191-b01]^ Effector molecules released by activated neutrophils, including the enzymes myeloperoxidase (MPO) and N-elastase (ELANE), have been implicated in inflammatory damage of the cornea.^[@i1552-5783-59-3-1191-b02][@i1552-5783-59-3-1191-b03]--[@i1552-5783-59-3-1191-b04]^ Current nonspecific anti-inflammatory treatments (such as corticosteroids) are associated with serious side effects, including infection, cataract, and glaucoma.^[@i1552-5783-59-3-1191-b05],[@i1552-5783-59-3-1191-b06]^ Thus, there is a pressing need for the development of novel therapeutic strategies that can curb tissue damage by selectively inhibiting pathogenic neutrophil pathways. Mesenchymal stromal cells are tissue-resident cells that have been shown to regulate the function of various innate immune cells (including neutrophils, natural killer cells, and mast cells) in a wide array of inflammatory disorders.^[@i1552-5783-59-3-1191-b07],[@i1552-5783-59-3-1191-b08]^ Stromal cells have been reported to suppress neutrophil apoptosis, which leads to the increased survival of neutrophils at the site of injury.^[@i1552-5783-59-3-1191-b09]^ Moreover, studies have shown that stromal cells inhibit neutrophil infiltration of inflamed tissues, in part through the secretion of TNF-α-stimulated gene 6 protein (TSG-6).^[@i1552-5783-59-3-1191-b10]^ Previous work from our group has shown that in vitro--expanded stromal cells suppress ocular inflammation following corneal injury.^[@i1552-5783-59-3-1191-b11],[@i1552-5783-59-3-1191-b12]^ Although the propensity of stromal cells to modulate neutrophil survival and migration has been reported, the regulation of neutrophil effector functions by stromal cells at the ocular surface has not been fully investigated. In the present study, we investigated the effect of stromal cells in regulating the secretion of neutrophil effector molecules, including proteases and cytokines, which cause tissue damage. Using both an in vivo murine model of ocular inflammaton and in vitro coculture assays, we demonstrate that mesenchymal stromal cells limit ocular tissue damage by inhibiting the release of MPO and ELANE by neutrophils in a cell--cell contact-dependent manner. Methods {#s2} ======= Animals {#s2a} ------- Six- to 8-week-old C57BL/6NCrl wild-type mice (Charles River Laboratories, Wilmington, MA, USA) were used in these experiments. Given that previous studies have shown similar corneal inflammation in male and female mice,^[@i1552-5783-59-3-1191-b13][@i1552-5783-59-3-1191-b14]--[@i1552-5783-59-3-1191-b15]^ we used male mice to maintain homogeneity in this study. The protocol was approved by the Schepens Eye Research Institute Animal Care and Use Committee, and all animals were treated according to the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. Ocular Injury Model {#s2b} ------------------- Mice were deeply anesthetized, and corneal injury was created as previously described.^[@i1552-5783-59-3-1191-b12],[@i1552-5783-59-3-1191-b16]^ Briefly, the central cornea was marked by a 2-mm trephine, and using the tip of a handheld motor brush (AlgerBrush II; Alger Company, Inc., Lago Vista, TX, USA), the corneal epithelium and anterior stroma were removed mechanically (approximately one third of total corneal thickness). Upon completion of the procedure, triple antibiotic ointment (neomycin and polymyxin B sulfates and bacitracin zinc ophthalmic ointment USP; Bausch + Lomb, Wilmington, MA, USA) was applied to the injured eyes, and a subcutaneous injection of buprenorphine was given to mice to minimize injury-induced pain. To study the in vivo effects of stromal cells, mice were randomly divided into saline-treated control and stromal cell--treated group (*n* = 5--6 mice/group). In vitro expanded and characterized stromal cells (0.5 × 10^6^ cells/100 μL sterile saline) were injected into the tail veins of mice at 1 hour post injury. Mice were euthanized at two separate time points following injury; corneas were harvested at 24 hours post injury to examine neutrophil function, and eyeballs were harvested at 48 hours post injury to evaluate corneal thickness. Corneal Tissue Digestion {#s2c} ------------------------ Single-cell suspensions were prepared from corneas as previously described.^[@i1552-5783-59-3-1191-b11]^ In brief, corneas were digested in RPMI media (Lonza, Walkersville, MD, USA) containing 2 mg/mL collagenase type IV (Sigma-Aldrich Corp., St. Louis, MO, USA) and 2 mg/mL DNase I (Roche, Basel, Switzerland) for 45 minutes at 37°C and then filtered through a 70-μm cell strainer. Cell Culture Assays {#s2d} ------------------- Due to the cornea harboring very low numbers of stromal cells and neutrophils, these cells were isolated from bone marrow for our in vitro experiments. Neutrophils were isolated from bone marrow of C57BL/6 mice using a neutrophil isolation kit (purity ≥ 95%) (MACS; Miltenyi Biotec, Inc., San Diego, CA, USA).^[@i1552-5783-59-3-1191-b17],[@i1552-5783-59-3-1191-b18]^ Purified neutrophils were cultured alone or stimulated with fMLP (formyl-methionyl-leucyl-phenylalanine, 1 μM; Sigma-Aldrich Corp.) for 1 hour.^[@i1552-5783-59-3-1191-b19],[@i1552-5783-59-3-1191-b20]^ Bone marrow--derived mesenchymal stromal cells (stromal cells) were generated by culturing bone marrow cells using the plastic adherence method and characterized as described previously.^[@i1552-5783-59-3-1191-b11],[@i1552-5783-59-3-1191-b12]^ Stromal cells were passaged every 3 to 5 days and were used for experiments at passage three. Stromal cells were stimulated with IL-1β (100 ng/mL; Biolegend, San Diego, CA, USA) for 24 hours.^[@i1552-5783-59-3-1191-b12]^ For coculture assays, neutrophils were cultured alone or on stromal cell monolayer at the ratio of 1:1 for 1 hour. For TSG-6 neutralization experiments, cocultures were pretreated with a standard maximal concentration (10 μg/mL) of anti-TSG-6 antibody (AF2326; R&D Systems, Minneapolis, MN, USA) for 1 hour and were then stimulated with fMLP for an additional 1 hour. Two mice were used in each experiment, and each experiment was repeated three times. Transwell Experiments {#s2e} --------------------- To perform the Transwell coculture assays, Transwell inserts with polycarbonate membrane (0.4-μm pore size; Corning, NY, USA) were used to prevent neutrophil--stromal cell contact in 24-well plates. Neutrophils stimulated with fMLP were placed in the lower chambers, and stromal cells were cultured in the upper chambers with a 1:1 stromal cell-to-neutrophil ratio. After 1 hour, supernatants were collected for the analysis of MPO and ELANE secretion using ELISA described below (*n* = 3 well/group, and repeated three times in three independent experiments). Enzyme-Linked Immunosorbent Assay {#s2f} --------------------------------- Levels of MPO and ELANE in culture supernatants from neutrophil and stromal cell coculture assays were analyzed using commercially available murine ELISA kits (R&D Systems; Abcam, Cambridge, MA, USA) per the manufacturer\'s instructions. Flow Cytometry {#s2g} -------------- Single-cell suspensions were prepared and stained with fluorochrome-conjugated monoclonal antibodies against CD11b, Ly6G for their cell surface expression, and MPO for intracellular expression of neutrophils. Appropriate isotype controls were used. Antibodies against CD45, CD34, and CD29 were used for the phenotypic characterization of stromal cells. For cell survival assays, neutrophils were stained with propidium iodide (PI). Stained cells were analyzed using a flow cytometer (LSR II; BD Biosciences, San Jose, CA, USA) and FlowJo software (FlowJo LLC, Ashland, OR, USA). All antibodies and isotypes controls were purchased from Biolegend. Real-Time PCR {#s2h} ------------- Total RNA was isolated using a kit (RNeasy Micro Kit; Qiagen, Valencia, CA, USA) and reverse transcribed into cDNA using reverse transcriptase (Superscript III; Invitrogen, Carlsbad, CA, USA). Quantitative real-time PCR was then performed using preformulated Taqman-based probes for murine *Mpo* (Mm01298424-m1), *Elane* (Mm00469310_m1), *Il-1b* (Mm00434228_m1), and glyceraldehype-3-phosphate dehydrogenase (*Gapdh*, Mm99999915_gl) (Taqman Universal PCR Mastermix; Thermo Fisher Scientific, Walthan, MA, USA) in a Mastercycler Realplex2 (Eppendorf, Hamburg, Germany). The results were analyzed by the comparative threshold cycle method and normalized to *Gapdh* as an internal control.^[@i1552-5783-59-3-1191-b11],[@i1552-5783-59-3-1191-b12]^ Histology {#s2i} --------- Whole eyeballs were harvested after 48 hours of injury. Paraformaldehyde-fixed cross sections were stained with hematoxylin and eosin (H&E) and examined for corneal tissue structures using a bright-field microscope (Nikon Eclipse E800; Nikon Instruments, Melville, NY, USA) at 20× magnification.^[@i1552-5783-59-3-1191-b11]^ Statistical Analysis {#s2j} -------------------- A Mann-Whitney *U* test was performed to determine significance, which was set at *P* ≤ 0.05. Data are presented as the mean ± standard deviation (SD). Results shown are representative of three independent experiments. Sample sizes were estimated on the basis of previous experimental studies on corneal injury and inflammation^[@i1552-5783-59-3-1191-b12],[@i1552-5783-59-3-1191-b21],[@i1552-5783-59-3-1191-b22]^ and using Harvard-MGH Biostatistics Center software (available free of charge at <http://hedwig.mgh.harvard.edu/sample_size/js/js_parallel_quant.html>). Results {#s3} ======= Corneal Injury Promotes the Infiltration of Neutrophils at the Ocular Surface {#s3a} ----------------------------------------------------------------------------- For this study, we utilized a well-established murine model of corneal inflammation.^[@i1552-5783-59-3-1191-b03],[@i1552-5783-59-3-1191-b12]^ Low vascularity and paucity of resident immune cells in the cornea make this model an excellent in vivo system in which to study the recruitment and function of immune cells in corneal inflammation. We first investigated the infiltration of inflammatory cells to the cornea after injury. Corneal injury was created by mechanical removal of corneal epithelium and anterior stroma ([Fig. 1](#i1552-5783-59-3-1191-f01){ref-type="fig"}A). Noninjured corneas served as control. Corneas were harvested 24 hours post injury for further analysis. Flow-cytometry analysis of corneal cells revealed increased frequencies of CD45^+^ inflammatory cells at the ocular surface compared to control noninjured corneas ([Figs. 1](#i1552-5783-59-3-1191-f01){ref-type="fig"}B, [1](#i1552-5783-59-3-1191-f01){ref-type="fig"}C). Our results further showed that the majority of CD45^+^ cells were CD11b^+^Ly6G^+^MPO^+^ neutrophils ([Figs. 1](#i1552-5783-59-3-1191-f01){ref-type="fig"}B, [1](#i1552-5783-59-3-1191-f01){ref-type="fig"}C). There was also a moderate increase in the frequencies of CD11b^+^Ly6G^−^ cells (macrophages) in injured corneas compared to noninjured controls ([Fig. 1](#i1552-5783-59-3-1191-f01){ref-type="fig"}C). {#i1552-5783-59-3-1191-f01} Stromal Cells Suppress the Neutrophil Effector Functions Without Inducing Cell Death {#s3b} ------------------------------------------------------------------------------------ Next, we investigated the effect of stromal cells on neutrophil effector functions. To this aim, neutrophils were cocultured with stromal cells, and the secretion of MPO and ELANE by neutrophils was assessed using ELISA. Given that the cornea harbors very low numbers of stromal cells and neutrophils, we isolated these cells from bone marrow for our in vitro experiments. Characterization of in vitro--expanded stromal cells using flow cytometry revealed these cells to be positive for the stromal cell marker CD29 and negative for the hematopoietic cell markers CD45 and CD34 ([Fig. 2](#i1552-5783-59-3-1191-f02){ref-type="fig"}A). Additionally, neutrophils were isolated from the bone marrow using magnetic activated cell sorting (purity ≥ 95%) ([Fig. 2](#i1552-5783-59-3-1191-f02){ref-type="fig"}B). Neutrophils were stimulated with fMLP, a neutrophil stimulant that is produced by necrotic cells during sterile inflammation, and cultured with or without stromal cells for 1 hour. ELISA analysis of culture supernatants revealed that fMLP treatment significantly enhanced the secretion of MPO (1609 ± 159 pg/mL) and ELANE (722 ± 66 pg/mL) by neutrophils. It is interesting that the secretion of these effector molecules was dramatically suppressed in fMLP-stimulated neutrophils cultured with stromal cells (MPO: 1090 ± 67; ELANE: 353 ± 21 pg/mL) compared to fMLP-stimulated neutrophils cultured alone ([Figs. 2](#i1552-5783-59-3-1191-f02){ref-type="fig"}C, [2](#i1552-5783-59-3-1191-f02){ref-type="fig"}D). Next, we investigated whether stromal cells suppress neutrophil effector functions by promoting cell death, using PI viability staining. No significant difference in the frequencies of PI-positive neutrophils (dead cells) between stromal cell--treated (3.81 ± 0.79) and untreated (4.59 ± 1.2) cultures ([Fig. 2](#i1552-5783-59-3-1191-f02){ref-type="fig"}E) was observed, further indicating that stromal cell--mediated suppression of neutrophil function is not due to induction of cell death. {#i1552-5783-59-3-1191-f02} Stromal Cells Inhibit Neutrophil Function in a Contact-Dependent and TSG-6-Independent Manner {#s3c} --------------------------------------------------------------------------------------------- Next, we set to delineate the mechanism by which stromal cells suppress neutrophil functions. Mesenchymal stromal cells regulate the immune response through the secretion of paracrine inhibitory factors as well as by cell--cell interactions via membrane-bound receptor molecules.^[@i1552-5783-59-3-1191-b23]^ Indeed, the soluble factor TSG-6, which is primarily produced by human^[@i1552-5783-59-3-1191-b21]^ and mouse^[@i1552-5783-59-3-1191-b24]^ mesenchymal stromal cells, has been shown to suppress neutrophil infiltration of the injured cornea.^[@i1552-5783-59-3-1191-b21]^ Our observation that stromal cells express *Tsg-6* mRNA in the steady state as well as in inflammatory conditions ([Fig. 3](#i1552-5783-59-3-1191-f03){ref-type="fig"}A) led us to investigate whether stromal cell--mediated suppression of neutrophil function is TSG-6 dependent. To determine this, fMLP-stimulated neutrophils and stromal cells were cocultured in the presence of TSG-6-neutralizing antibody for 1 hour. ELISA analysis of culture supernatants demonstrated that TSG-6 neutralization did not alter the secretion of MPO (1330 ± 163 pg/mL) or ELANE (320 ± 48 pg/mL) by fMLP-stimulated neutrophils cultured with stromal cells, compared to neutrophil--stromal cell cocultures without TSG-6 neutralization (MPO: 1299 ± 79; ELANE: 331 ± 28 pg/mL) ([Figs. 3](#i1552-5783-59-3-1191-f03){ref-type="fig"}B, [3](#i1552-5783-59-3-1191-f03){ref-type="fig"}C). These results suggest that stromal cell--mediated suppression of neutrophil effector functions is independent of TSG-6 secretion. To determine whether direct cell--cell interactions contribute to the inhibitory effect of stromal cells on release of these effector molecules, fMLP-stimulated neutrophils were cultured with stromal cells with or without Transwell inserts. ELISA analysis of culture supernatants demonstrated that unlike direct cocultures, stromal cells cultured in Transwells failed to suppress the secretion of MPO (2274 ± 209 pg/mL) and ELANE (638 ± 12.6 pg/mL) by neutrophils ([Figs. 3](#i1552-5783-59-3-1191-f03){ref-type="fig"}D, [3](#i1552-5783-59-3-1191-f03){ref-type="fig"}E). {#i1552-5783-59-3-1191-f03} Stromal Cells Suppress Neutrophil Effector Functions and Tissue Damage During Ocular Inflammation {#s3d} ------------------------------------------------------------------------------------------------- Finally, using our in vivo model of injury-induced corneal inflammation, we determined whether stromal cells could regulate the neutrophil effector functions in the inflamed cornea. We have shown previously that bone marrow--derived stromal cells home specifically to the injured cornea.^[@i1552-5783-59-3-1191-b22],[@i1552-5783-59-3-1191-b25]^ Here, we intravenously injected in vitro--expanded stromal cells to injured mice at 1 hour following injury and harvested corneas after 24 hours. Saline-treated injured mice and mice without injury served as controls. Flow-cytometry analysis of corneal cells revealed that stromal cell--treated mice showed reduced levels of MPO (2-fold decrease) in the infiltrated CD11b^+^ Ly6G^+^ neutrophils compared to the control group ([Figs. 4](#i1552-5783-59-3-1191-f04){ref-type="fig"}A, [4](#i1552-5783-59-3-1191-f04){ref-type="fig"}B, [4](#i1552-5783-59-3-1191-f04){ref-type="fig"}C). We confirmed the stromal cell--mediated suppression of MPO and ELANE expression by neutrophils at the mRNA level, with an approximate 2.5-fold decrease in *Mpo* and *Elane* mRNA observed in mice treated with stromal cells relative to the control group ([Figs. 4](#i1552-5783-59-3-1191-f04){ref-type="fig"}D, [4](#i1552-5783-59-3-1191-f04){ref-type="fig"}E). We also evaluated the ocular surface expression of the inflammatory cytokine IL-1β, which is expressed at higher levels in activated neutrophils.^[@i1552-5783-59-3-1191-b26]^ Real-time PCR analysis showed significantly reduced expression of IL-1β in the stromal cell--treated group compared to untreated injured mice ([Fig. 4](#i1552-5783-59-3-1191-f04){ref-type="fig"}F). Consistent with previous reports,^[@i1552-5783-59-3-1191-b03],[@i1552-5783-59-3-1191-b16]^ we found reduced frequencies of neutrophils in stromal cell--treated mice compared to control groups ([Fig. 4](#i1552-5783-59-3-1191-f04){ref-type="fig"}G). Given the central role of neutrophil-derived MPO and elastase in tissue damage during inflammation, we evaluated corneal tissue architecture after injury using H&E staining ([Fig. 4](#i1552-5783-59-3-1191-f04){ref-type="fig"}H). Histopathologic analysis of injured corneas harvested at 48 hours post injury demonstrated restoration of normal corneal tissue structures, including stromal thickness and reduced inflammatory cell infiltration in stromal cell--treated mice compared to untreated injured mice. Collectively, these findings indicate that stromal cells suppress the neutrophil effector functions and subsequent tissue damage after corneal injury. {#i1552-5783-59-3-1191-f04} Discussion {#s4} ========== Dysregulated neutrophil activation leads to persistent inflammation and subsequent tissue damage. In this study, we investigated the effect of stromal cells in regulating neutrophil effector functions during eye inflammation. Using a murine model of ocular injury, we report that stromal cells inhibit the secretion of the tissue-degrading enzymes MPO and ELANE by neutrophils and limit ocular inflammation. Moreover, we demonstrate that the observed stromal cell--mediated suppression of neutrophil function is primarily dependent on direct cell--cell interactions and is independent of stromal cell--secreted TSG-6. The role of certain immune cells in curbing the inflammatory response has been established in a wide range of immune disorders.^[@i1552-5783-59-3-1191-b27],[@i1552-5783-59-3-1191-b28]^ For example, regulatory T cells are crucial for modulating the antigen-specific immune response,^[@i1552-5783-59-3-1191-b28]^ and myeloid-derived suppresser cells and M2 macrophages are involved in regulating non--antigen-specific innate inflammation of nonocular tissues such as the liver, kidneys, and lungs.^[@i1552-5783-59-3-1191-b29],[@i1552-5783-59-3-1191-b30]^ Our study reveals that stromal cells, a type of nonimmune cell, are also critical for regulating nonspecific inflammation through their suppression of neutrophil effector functions. Mesenchymal stromal cells inhibit neutrophil apoptosis and promote their survival through secretion of IL-6.^[@i1552-5783-59-3-1191-b09]^ However, our study provides novel evidence that stromal cells also regulate neutrophil secretion of the tissue-damaging molecules MPO and ELANE without promoting neutrophil cell death. We thus decided to further delineate the mechanisms of stromal cell suppression of neutrophil function. Previous studies have shown that stromal cell--derived TSG-6 interacts with CXCL8 and suppresses the infiltration of neutrophils in inflammatory conditions such as acute pancreatitis.^[@i1552-5783-59-3-1191-b10],[@i1552-5783-59-3-1191-b31]^ At the eye, TSG-6 has been shown to attenuate the recruitment of neutrophils to the cornea after chemical and mechanical injuries.^[@i1552-5783-59-3-1191-b03]^ It is interesting that neutralization of TSG-6 in our stromal cell--neutrophil coculture assays did not abrogate the suppressive effects of stromal cells on MPO and ELANE secretion by neutrophils. Stromal cells have been reported to regulate the function of myeloid cells by direct cell-to-cell contact and by secreting soluble factors such as prostaglandin E2, indoleamine 2,3-dioxygenase transforming growth factor-β1, and IL-10.^[@i1552-5783-59-3-1191-b32][@i1552-5783-59-3-1191-b33]--[@i1552-5783-59-3-1191-b34]^ Using the Transwell system, we further demonstrated that suppression of neutrophil effector functions is mediated by direct cell-to-cell interactions between stromal cells and neutrophils. Our observations are in agreement with a number of previous studies,^[@i1552-5783-59-3-1191-b35][@i1552-5783-59-3-1191-b36]--[@i1552-5783-59-3-1191-b37]^ including one in a model of corneal inflammation^[@i1552-5783-59-3-1191-b35]^ showing a direct interaction between keratocytes (a type of stromal cells) and neutrophils through ICAM1-CD18 binding. Consistent with our in vitro observations, systemic treatment of mice with stromal cells results in significantly decreased expression of tissue-damaging factors, including MPO and the inflammatory cytokine IL-1β, by infiltrating neutrophils at the ocular surface. In addition, stromal cell--treated mice exhibited reduced ocular infiltration of neutrophils, which is consistent with previous reports showing that stromal cell--derived TSG-6 inhibits neutrophil migration to inflamed tissues.^[@i1552-5783-59-3-1191-b10]^ Moreover, this stromal cell--mediated suppression of neutrophil activation is accompanied by reduced ocular inflammation and a faster normalization of corneal tissue structure. The current study elucidates the novel function of mesenchymal stromal cells in regulating neutrophil effector functions and limiting tissue damage in ocular inflammation. Our study provides new insights that may be utilized in the development of stromal cell--based therapeutic strategies for the prevention and treatment of ocular and nonocular tissue damage caused by excessive neutrophil activation. Supported in part by the National Institutes of Health Grant EY024602 (SKC) and Core Grant P30EY003790. Disclosure: **S.K. Mittal**, None; **A. Mashaghi**, None; **A. Amouzegar**, None; **M. Li**, None; **W. Foulsham**, None; **S.K. Sahu**, None; **S.K. Chauhan**, None [^1]: License: CC YB

Introduction {#s1} ============ The concept of the Neotropical orchid genus *Vargasiella* C.Schweinf. was proposed by Schweinfurth in 1952, along with the description of *Vargasiella peruviana* C. Schweinf., the only species known at that time. Six years later Schweinfurth [@pone.0098472-Schweinfurth1] described another species, *Vargasiella venezuelana* C. Schweinf. Currently, the genus comprises only two species with a disjunctive distribution. They are terrestrial plants growing in wet, dense, submontane or montane forest, rich in epiphytes and lianas ([Fig. 1](#pone-0098472-g001){ref-type="fig"}). Populations of *Vargasiella peruviana* occasionally form quite large clumps in seasonally inundated meadows next to the forest edge. This plant\'s habit is unusual for orchids ([Fig. 2a](#pone-0098472-g002){ref-type="fig"}). Its stem is long, reaching about 1 m in length, creeping and climbing to nearby woody plants or robust grass, rooted occasionally. It appears to be monopodial which could be misleading; however, it becomes very clear that the stem consists of 15--25 cm-long segments if studied carefully. Each segment succeeds one arising from the apical part of the preceding one. The segments do not form any pseudobulbs, each of them is enclothed with 3--5 leaves. The leaf blades are oblong- or elliptic-lanceolate, convolute, thin-textured with 3--5 prominent, longitudinal nerves on the underside. The leaf blade is set on a short petiole, sheathed basally. The inflorescence is produced in the upper part of the segments ([Fig. 2b](#pone-0098472-g002){ref-type="fig"}). The peduncle is much longer than the laxly few-flowered raceme, the flowers are medium-sized and, in *V. peruviana*, rather attractive. The basic color is a mixture of deep purple and pink. The sepals and petals of *V. peruviana* are subsimilar, whereas in *V. venezuelana* the petals are wider than the sepals. The lip in both species is slightly different; in *V. peruviana* it is ovate-lanceolate to ligulate, undulate along upper margins with a thickened disc ([Fig. 2c](#pone-0098472-g002){ref-type="fig"}). The lip of *V. venezuelana* is elliptic-ovate to almost elliptic-orbicular, undulate above the cordate base, without any thickening. {#pone-0098472-g001} {#pone-0098472-g002} The only other orchid genus that might possibly be confused with *Vargasiella* is *Palmorchis* Barb. Rodr.; however, this is possible only when exclusively vegetative characteristics are considered. Both genera are easily separable at the anthesis. Since the description of the genus, the taxonomic position of *Vargasiella* has remained vague. Dressler [@pone.0098472-Dressler1] classified *Vargasiella* in his broadly defined tribe Maxillarieae Pfitzer, in the subtribe Zygopetalinae Schltr., into which he included a further 25 genera in four closely related alliances. He stated, however, that the relationship of monopodial *Vargasiella* to other genera of the subtribe is uncertain. Later, he suggested raising *Vargasiella* to subtribal rank [@pone.0098472-Dressler2], but upheld its position in the Maxillarieae. Simultaneously, he expressed doubts regarding the monopodial type of growth reported in that genus. Such a subtribe was validated in the same year by Romero and Carnevali [@pone.0098472-Romero1]. In turn, Senghas [@pone.0098472-Senghas1] placed *Vargasiella* in the subtribe Liparidinae Lindl. *ex* Miq., based on the presence of naked pollinia which had already been reported by Schweinfurth [@pone.0098472-Schweinfurth2], but without taking into consideration, however, the numerous differences in the gynostemium organization noted in these two taxa. In 1996, Senghas modified his opinion on that genus in response to Romero and Carnevali [@pone.0098472-Romero1] who had noted the presence of tegula and viscidium in *Vargasiella peruviana.* Senghas [@pone.0098472-Senghas2] included the genus in the subtribal rank in the tribe Oncidieae Pfitzer, along with, inter alia, Dichaeinae Schltr., Pachyphyllinae Pfitzer, Ornithocephalinae Schltr. and Oncidiinae Benth. The combination of the gynostemium structure and the monopodial type of growth misled Szlachetko [@pone.0098472-Szlachetko1] into coming to the conclusion that Vargasiellinae, Dichaeinae and Pachyphyllinae might be cognate and he accepted the tribe Dichaeeae Pfitzer. Pridgeon et al. [@pone.0098472-Pridgeon1] included Vargasiellinae in the broadly defined Cymbidieae Pfitzer. While the taxonomy of numerous orchid genera are still discussed and their systematic position is debated by scientists, especially when morphological data are in conflict with molecular analyses, the inconsistency in the classification of *Vargasiella* has mostly been related to the lack of material for genetic studies and the insufficient number of specimens for morphological examination. Both *Vargasiella* species have found growing in remote areas of Bolivia, Peru and Venezuela. Unfortunately, all efforts to keep them alive via cultivation have failed. During the scientific expeditions (2007, 2008) conducted to the Peruvian Departments of Loreto and Amazonas, we had the opportunity to study specimens of *Vargasiella peruviana in situ*. Based on macro- and micromorphological studies, analyses of selected molecular markers and ecological niche distribution modeling methods, we made an attempt to clarify the taxonomic placement of *Vargasiella*. Materials and Methods {#s2} ===================== Field study {#s2a} ----------- The population of *Vargasiella peruviana* has been found in the Peruvian District Molinopampa (Amazonas, Prov. Chachapoyas), about 25 km E of Chachapoyas, at the altitude of about 2400 m, in the mountainous area covered by patches of low forest separated by pasture and wet meadows. The plants have been found growing both inside and along edges of a forest. The standard measurements of vegetative parts and photographic documentation was taken in the field. The samples of the specimens collected in the Departments of Loreto and Amazonas for morphological and molecular analyses were collected thanks to the courtesy of Mr. Manolo Arias and his Peruana Orchid Nursery. Molecular study {#s2b} --------------- ### DNA Isolation {#s2b1} Fresh leaf samples were preserved in silica gel. Total genomic DNA was extracted from 20 mg of dried leaves [@pone.0098472-Chase1] using the DNA Mini Plant (A&A Biotechnology, Poland) following the manufacturer\'s protocol. Plant tissues were homogenized in a sterilized mortar and kept at −65°C for 12 hours. ### Amplification {#s2b2} Nuclear ITS region (ITS1-5.8S-ITS2) was amplified by Polymerase Chain Reaction (PCR) method using the primers 17SE and 26SE [@pone.0098472-Sun1] in a T1 thermocycler (Biometra, Germany). PCR was carried out in a volume of 50 µl. The PCR mixture contained: dd H~2~O, 5 µl 10x Polymerase buffer with 15 mM MgCl~2~, 2 µl of 5 mM mix of each dNTP (200 µM), 0.3 µl of 20 mM primers, 2 µl DMSO, 2.0 units of Taq DNA Polymerase (EURx, Gdańsk, Poland) and genomic DNA. The thermal cycling protocol of the PCR consisted of 5 min initial denaturation at 94°C, 28 cycles, each with 45s denaturation at 94°C, 45 s annealing at 53°C and an extension of 60 s at 72°C, ending with an extension of 5 min at 72°C. The *mat*K gene (about 1300 bp) was amplified with the following two primers: - 19F [@pone.0098472-Molvray1] and 1326R [@pone.0098472-Cunoud1]. The PCR mixture contained: dd H~2~O, 5 µl 10x Polymerase buffer with 15 mM MgCl~2~, 1 µl of 10 mM mix of each dNTP (200 µM), 0.3 µl of 20 mM primers, 2 µl MgCl~2~ \[50 mM\], 2.0 units of Taq DNA Polymerase and genomic DNA. The thermal cycling protocol comprised 28 cycles, each with 45 s denaturation at 94°C, 45 s annealing at 52°C, an extension of 2 min 30 s at 72°C, concluding with an extension of 5 min at 72°C. The *trn*L-F region containing the *trn*L intron and the *trn*L-*trn*F intergenic spacer was amplified using primers *trn*L-c and *trn*L-F as described by Taberlet [@pone.0098472-Taberlet1]. Polymerase chain reaction (PCR) amplifications were carried out in a total volume of 25 µl containing 5 µl 5x buffer, 1 µl 50 mM MgCl~2~, 1 µl 5 mM dNTPs, 0.5 µl of 10 µM of each primers, and 1.0 unit of Blue Perpetual DNA polymerase (EURx, Gdańsk, Poland). The thermal cycling protocol of the PCR consisted of 5 min initial denaturation at 80°C, 30 cycles, each with 60 s denaturation at 94°C, 60 s annealing at 51°C and an extension of 2 min at 65°C, ending with an extension of 7 min at 65°C. Amplified products were cleaned with High Pure PCR Product Purification Kit (Roche Diagnostic GmbH, Mannheim, Germany) following the manufacturer protocol. ### Sequencing {#s2b3} Cycle sequencing was carried out directly on the purified product using a Big Dye Terminator v 3.1 Cycle Sequencing Kit (Applied Biosystems, Warrington, Cheshire, UK): 2 µl of sequencing buffer, 2 µl of Big Dye terminator with 2 µl of 0,08 mM primer (1.6 pmol), 2--4 µl of amplified product (30--90 ng/ µl), and dd H~2~O in a total reaction volume of 10 µl. The PCR primers were used to sequence both strands of ITS and *trn*L-F. Two PCR primers (-19F, 1326R) and one internal primer (390F) were used [@pone.0098472-Cunoud1] to sequence the *mat*K gene. Cycle sequencing condition for two strands (all regions) were as follows: 20 s initial denaturation followed by 25 cycles each with 15 s denaturation (94°C), 5 s annealing (50°C) and 4 min elongation (60°C) using T1 thermocycler (Biometra, Germany). Sequencing reactions were purified with an ExTerminator Kit (A&A Biotechnology, Gdynia, Poland) following the manufacturer\'s protocol. Pelleted samples were sequenced on an Applied Biosystems 377 automated sequencer. Both strands were sequenced to enssure accuracy in base calling. A Sequence Navigator (Applied Biosystems) was used to edit the sequences and each individual base position was examined for agreement between the two strands using AutoAssembler (Applied Biosystems) software. The *mat*K, *trn*L-F (containing *trn*L intron and *trn*L-*trn*F intergenic spacer), and ITS1-5.8S-ITS2 (nuclear ribosomal DNA; referred to as ITS) sequences for *Vargasiella peruviana* were deposited in the International Nucleotide Sequence Databases (INSD) under accession numbers KF938587/KF938588/KF938586 respectively. ### Phylogenetic analysis {#s2b4} To enlarge the dataset for all representatives of the subtribe Zygopetalinae, additional 84 sequences were taken from INSD. All sequences for *mat*K, *trn*L-F, and ITS are derived from the publication of Whitten [@pone.0098472-Whitten1] (INSD PopSet no. 58200482/580042546/580042630 respectively) except for *Warrea warreana* (AF239417 -- *mat*K, AF239513 -- *trn*L, AF239321 -- ITS). The list of analysed sequences downloaded from INSD is provided in [Annex S1](#pone.0098472.s003){ref-type="supplementary-material"}. The DNA sequences were aligned by ClustalX [@pone.0098472-Thompson1] and adjusted by eye using Seaview [@pone.0098472-Gautier1]. Maximum parsimony analyses were undertaken using heuristic searches in PAUP\* version beta 10 [@pone.0098472-Swofford1] with tree-bisection-reconnection (TBR) branch swapping and the MULTREES (holding multiple trees) option in effect with 1000 replicates of random sequence addition. Only 10 trees were saved for each replicate to reduce the time spent in swapping large numbers of suboptimal trees. All characteristics were treated as unordered and as equally weighted [@pone.0098472-Fitch1]. Tree length, consistency index (CI) and retention index (RI) were estimated. The internal support of clades was evaluated using non-parametric bootstrapping [@pone.0098472-Felsenstein1] with 1000 replicates and the same settings as above, except for the simple sequence addition. Incongruence between data sets was tested using the incongruence-length difference (ILD) test [@pone.0098472-Farris1] using PAUP\*. Time calibration {#s2c} ---------------- The *mat*K matrix was used to estimate the calibration point for MRCA (most recent common ancestor) of the *Warrea-Vargasiella* clade. The Bayesian uncorrelated relaxed molecular clock approach implemented in BEAST [@pone.0098472-Drummond1] was used. The age for the root of the tree was set to a normal prior distribution with a mean of 19 Ma and a standard deviation of 5.0 (giving a 95% CI ranging from *c*. 11--27 Ma) corresponding to the resulting age estimate for the *Oncidium-Lycaste* clade from a calibration based on the analysis of *mat*K+*rbc*L for Orchidaceae [@pone.0098472-Gustafsson1]. The Yule process was chosen as the speciation process for the *mat*K data matrix. The Akaike Information Criterion in a ModelTest v.3.7 [@pone.0098472-Posada1] was used to choose the best-fitting evolutionary model for *mat*K (GTR+Γ+I). Runs were performed in BEAST with 20 million generations each. Log files were analysed with Tracer v1.5 [@pone.0098472-Rambaut1]. All resulting trees were then combined with LogCombiner v1.7.3 [@pone.0098472-Drummond1], with a burn-in of 20%. A maximum credibility tree was then produced using TreeAnnotator v1.5.3 [@pone.0098472-Drummond1]. Macromorphological features {#s2d} --------------------------- The examined plant material was obtained from the AMES, MO, UGDA and USM herbaria. Moreover, the observations and remarks made during two scientific expeditions (2007, 2008) to the Peruvian departments of Loreto and Amazonas were included into the analysis. The herbarium material was prepared according to the standard classical taxonomy procedure and examined under a stereomicroscope. The morphological descriptions were based on the observation of the type and non-type specimens deposited at AMES and preserved in liquid (UGDA, USM). The following vegetative characters of individual plants were analysed: stem (height, type of growth), leaves (number, size, shape), sheaths (number, shape, size), inflorescence (size, density), floral bracts (shape), flowers taken from the middle part of the inflorescence (size and surface of pedicel and ovary, presence of mentum, size and shape of perianth parts), as well as gynostemium (size and shape of the column, presence of column foot). The most discriminative characters have been selected and used in both phenetic as well as cladistic analyses ([Annex S2](#pone.0098472.s004){ref-type="supplementary-material"}, [Annex S3](#pone.0098472.s005){ref-type="supplementary-material"}). Particular parts of the flower were softened in boiling water prior to dissection followed by stereomicroscope examination and drawing. The database of the line-drawings and photographs of all studied specimens is available in the first author\'s archives. A key for determination of the *Vargasiella* species is provided. The line drawings were prepared using CorelDRAW Graphics Suite 12. Macromorphological analyses {#s2e} --------------------------- To create hierarchic phenograms and cladograms, the PAST program [@pone.0098472-Hammer1] was used. A distance matrix was created using the most common Euclidean measure and middle links rule unweighted pair-group average (UPGMA) as an amalgamation rule. Neighbor joining is a bottom-up clustering method for the creation of phylogenetic trees proposed by Saitou and Nei [@pone.0098472-Saitou1]. The principle of this method is to find pairs of taxa (neighbors) that minimize the total branch length at each stage of the clustering of taxa. Finally, a character-based cladistical method - maximum parsimony with Fitch maximum parsimony algorithm -- was used, which rapidly and precisely evaluates the minimum number of changes. This method infers a phylogenetic tree by minimizing the total number of evolutionary steps required to explain a given set of data or, in other words, by minimizing the total tree length (Occam\'s razor). *Vargasiella* was treated as a output group and for searching for the optimal tree structure we used heuristic algorithm tree bisection and reconnection (TBR), that detaches a subtree from the main tree at an interior node and then attempts all possible connections between branches of the two trees thus created. For phylogenic estimation, we used a bootstrap calculation with 1000 sampling. Micromorphological study {#s2f} ------------------------ Fresh flowers were gathered and preserved in Kew Mixture in the following proportions: 70% of 95% ethanol, 5% of glycerin, 5% of formaldehyde and 20% of distillate water. The plant material was examined using a Nikon SMZ 1500 light microscope with a Nikon Digital Sight DS-Fi1 camera. For observations in the scanning electron microscope (SEM) the samples kept in the Kew mixture were dehydrated through ethanol series. To preserve the surface structure of the specimens, the procedure of critical-point drying in an Emitech K850 Critical Point Dryer apparatus was applied prior to mounting the samples on SEM stubs with an SPI Carbon Conductive Double Sided Adhesive Disc. Then, the plant material was gold-coated (Sputter Coater Spi-Module). The samples were examined and imaged in a Phillips XL-30 Scanning Electron Microscope operating at an accelerating voltage of 15kV. Ecological niche modeling {#s2g} ------------------------- The ecological niche modeling (ENM) was applied to locate the glacial refuge areas and to estimate the potential range of *Vargasiella* and *Warrea* based on the suitable habitat distribution. Because the biogeographic data about both studied taxa are poor, we decided to include *Warreopsis* into the ENM analysis as an additional datum. The list of localities used in the analysis was compiled based on the data taken from the herbarium specimens\' labels complemented by information from the electronic database of Missouri Botanical Garden (available at [www.tropicos.org](http://www.tropicos.org)). Only those localities which could be precisely placed on the map were used in the ecological niche modeling. Six locations of *Vargasiella*, 19 *Warrea* and 10 *Warreopsis* populations were used ([Annex S4](#pone.0098472.s006){ref-type="supplementary-material"}), which is more than the minimum number required to obtain reliable predictions in the MaxEnt application [@pone.0098472-Pearson1]. As input data of 19 climatic variables in 2.5 arc minutes (±21.62 km^2^ at the equator) developed by Hijmans and others [@pone.0098472-Hijmans1] as well as the elevation data were used ([Table S1](#pone.0098472.s001){ref-type="supplementary-material"}). The bioclimatic data for the last glacial maximum (LGM, between 26500 and 19000--20000 years ago) were developed and mapped by Paleoclimate Modeling Intercomparison Project Phase II [@pone.0098472-Braconnot1]. To the accuracy of the modeling, the maximum iterations was set to 10,000 and convergence threshold to 0.00001. For each run 20% of the data were used to be set aside as test points [@pone.0098472-UrbinaCardona1]. The \"random seed\" option which provided random test partition and background subset for each run was applied. Each run was performed as a bootstrap with 1000 sampling and the output was set to logistics. Models were evaluated using the threshold-independent area under the Receiver Operator Characteristic curve (AUC), which is a measure of discrimination. All operations on the GIS data were carried out on ArcGis 9.3 [@pone.0098472-ESRI1]. Range and niche overlap {#s2h} ----------------------- The overlapping of the ranges and niches calculated for LGM and present time were determined using ENMTools application. Since the ecological amplitude of the studied taxa seems to be narrow, the threshold used in the calculation of the range overlap was 0.8. The niche overlap was measured using Schoener\'s D (D) [@pone.0098472-Schoener1] and I statistic (I) [@pone.0098472-Warren1]. Schoener\'s D was initially developed to compare diet and microhabitats and here it is used with the assumption that direct measures of local species density are comparable with each other. The I statistic is based on the Hellinger distance and measures the ability of the model to estimate the true suitability of a habitat. Both metrics range from 0 (niches completely different) to 1 (total overlap). Results {#s3} ======= Molecular analyses {#s3a} ------------------ The result of the individual and combined analyses are presented in one of the most parsimonious trees. Bootstrap support (BS) above 50% is given for supported clades below branches. Branches that collapse in the strict consensus tree are indicated by arrows. Statistics for one of the most parsimonious trees from each analysis is shown in the [Table 1](#pone-0098472-t001){ref-type="table"}. Statistics for *mat*K, *trn*L-F, ITS and combined data matrices are separated by "/". The number of analysed taxa were 85/82/85/85 respectively. The aligned matrix comprised 1327/1389/842/3558 characters of which 233/286/374/1392 were variable and 134/132/263/529 were potentially parsimony informative. The number of the most parsimonious trees were 48/416/3258/10.000. Tree-length was 374/995/818/1640, consistency index (CI) = 0.72/0.77/0.77/0.68, and retention index (RI) = 0.83/0.86/0.86/84. Results from the *partition homogeneity test* indicate incongruence between all (plastid and nuclear partitions): *mat*K/*trn*L-F P = 0.01; *mat*K/ITS P = 0.01; *trn*L-F/ITS P = 0.01; *mat*K/*trn*L-F/ITS P = 0.01. A visual comparison of the topology and bootstrap support between plastid and ITS data sets shows incongruence between low supported clades. In this case combined analyses were performed. On a combined ITS-*mat*K-*trn*L-F tree ([Fig. 3](#pone-0098472-g003){ref-type="fig"}), the analysed genera are arranged in groups congruent with Szlachetko\'s subtribes [@pone.0098472-Szlachetko1], i.e. Huntleyinae Schltr. (pseudobulbs lacking or obscure), Zygopetalinae *s.str.* (heteroblastic pseudobulbs) and Warreinae Szlach. (homoblastic pseudobulbs). Dichaeinae (monopodial plants) and Cryptarrheninae Dressl. (pseudobulbs obscure or absent) are nested in this clade as well. In all analyses, *Vargasiella* composes a highly supported subclade together with *Warrea* and *Warreopsis* (ITS 97 BS; *mat*K 95 BS; *trn*L-F 80 BS (only *Warrea-Vargasiella*); combined 100 BS). The subclade is situated at the base of the Zygopetalinae *sensu lato* clade and it is sister to the rest of the taxa (except for the *mat*K tree where it forms a basal polytomy with the other subclades). The relationships between *Warrea, Warreopsis* and *Vargasiella* are not resolved. In all strict consensus trees (not shown) the *Warrea-Vargasiella* clade present in the *mat*K and combined trees is collapsed. {#pone-0098472-g003} 10.1371/journal.pone.0098472.t001 ###### Statistics for one of the most parsimonious trees from each analysis (CI - consistency index; RI - retention index). {#pone-0098472-t001-1} matrix *mat*K trnL-F ITS *mat*K/trnL-F/ITS ------------------------------ -------- -------- ------ ------------------- No. of taxa 85 82 85 85 Included positions in matrix 1327 1389 842 3558 Variable site 233 286 374 1392 Parsimony-informative sites 134 132 263 529 Trees (MPT) 48 416 3258 10000 Fitch tree length 374 995 818 1640 CI 0.72 0.77 0.77 0.68 RI 0.83 0.86 0.86 0.84 Tree calibration {#s3b} ---------------- The topology of the tree within Zygopetalinae estimated from the *mat*K gene are generally congruent with those from the parsimony analysis ([Fig. 4](#pone-0098472-g004){ref-type="fig"}). The most recent common ancestor of *Vargasiella, Warrea* and *Warreopsis* originated in the Miocene around 7.5 Ma ago. The divergence time for *Vargasiella* and *Warrea* was estimated for approximately 5.4 Ma ago (Miocene/Pliocene). {#pone-0098472-g004} Macromorphological characters {#s3c} ----------------------------- Regardless of the method of clustering *Vargasiella* is well separated from all other taxa used in used in phenetic and cladistic analyses ([Fig. 5](#pone-0098472-g005){ref-type="fig"}). The genus either occupies the basal position in Zygopetalinae s.l. comprising a branch (UPGMA based phenogram) or the basal position of the whole tree (Neighbor joining and Maximum parsimony), which clearly indicates that there is a morphological gap between *Vargasiella* and other genera. In [Table 2](#pone-0098472-t002){ref-type="table"} we present the most important morphological characters of *Vargasiella, Warrea* and *Warreopsis*. Admittedly, both taxa could be regarded as related as far as molecular analyses are concerned;, however, based on morphological characters, their relation can be questioned. {#pone-0098472-g005} 10.1371/journal.pone.0098472.t002 ###### Comparison of main morphological characters between *Vargasiella*, *Warrea*, *Warreopsis, Maxillaria* and *Oncidium*. {#pone-0098472-t002-2} Character *Vargasiella* *Warrea* *Warreopsis* *Maxillaria* *Oncidium* ------------------- ----------------- ---------------- ----------------- ------------------ ------------------ Pseudobulbs absent homoblastic homoblastic heteroblastic heteroblastic Leaves convolute convolute convolute conduplicate conduplicate Inflorescence subapical, subbasal, subbasal, basal, single- basal, many- many- many- many- flowered flowered flowered flowered flowered Lip simple, simple, simple, more or less sessile, oblong to flabellate flabellate, prominently variously elliptic, sessile shortly 3-lobed, lobed sessile to clawed sessile clawed Callus obscure radiating prominent, simple, often a from the base flabellate oblong complex along nerves structure Gynostemium erect, rather elongate, elongate, elongate or slightly arched short slightly slightly short, slender or erect, rather arched, arched, or stout, stout robust robust usually gently arched, occasionally suberect Column foot short short, but obscure usually half absent robust, gently as long as narrowing column part, towards the occasionally apex longer than column part Stigma large, deltoid, rather large, rather large, large, elliptic, large, ovate to deeply elliptic, elliptic, deeply elliptic, deeply concave deeply deeply concave concave concave, concave, partially partially hidden by hidden by rostellum rostellum Anther ventral, ventral, ventral, subapical, subapical, transversely incumbent, incumbent, incumbent, incumbent, ellipsoid, ellipsoid- ellipsoid- ellipsoid- ovoid dorsiventrally ovoid, ovoid, ovoid or flattened dorsiventrally dorsiventrally obovoid compressed compressed Pollinia 4 in two 4, in two 4, in two 4, in two 2, subglobose, pairs, pairs, almost pairs, almost pairs, slightly subequal, superposed, superposed, unequal in dorsiventrally obovoid- unequal in unequal in size, flattened, ellipsoid size, size, dorsiventrally hard, dorsiventrally dorsiventrally compressed, unequally and compressed, compressed, almost flat or deeply cleft, ellipsoid- ellipsoid- concave on empty inside ovoid, rather ovoid, rather ventral hard hard surface and slightly convex on the outer one, obovoid to ellipsoid, rather soft Tegula single, single, single, single, single, oblong, oblong, oblong, oblong, elliptic-ovate thin, lamellate lamellate widest near attenuate to crepiform, the base, towards acute thin, narrowed apex, lamellate gradually lamellar towards the apex, thin, lamellar Viscidium elliptic- single, single, single, very single, oblong cordate, elliptic-ovate elliptic-ovate, narrow, elliptic, thick, distinctly 2- or elliptic- lamellar transversely fleshy lobed at the rhombic, elliptic, apex attenuate triangular or towards both crepiform, apex and thin lamellate base, lamellar thin Rostellum short and rather large, relatively dome-like, short, conical- broad dome-like, 3- short, broad and digitate in the lobed, the dome-like, 3- short, middle, obtuse middle lobe lobed, the shallowly ligulate- middle lobe sinuous in triangular, ligulate, side front acute, side lobes obscure lobes obscure, separated by shallow sinus from the middle lobe Rostellum remnant with short 3-lobed, the 3-lobed deeply bilobulate at apiculus in middle lobe notched at the middle, the middle subulate, the apex with oblique acute, rigid, shallowly both side concave plate lobes much between acute reduced, lobules obtuse Micromorphological features {#s3d} --------------------------- The characteristic feature of the lip of *Vargasiella peruviana* is the presence of the central fleshy disc which is divided into two calli ([Fig. 6A](#pone-0098472-g006){ref-type="fig"}) starting from the base ([Fig. 6A-B, D](#pone-0098472-g006){ref-type="fig"}). The epidermis of the lip base is built by irregularly sized rounded cells to obpyriform and conical papillae ([Fig. 6C](#pone-0098472-g006){ref-type="fig"}). Between the thickenings ([Fig. 6D-E](#pone-0098472-g006){ref-type="fig"}), obpyriform to slightly conical papillae are noticed ([Fig. 6E-F](#pone-0098472-g006){ref-type="fig"}). The bigger conical papillae are distinctly visible on the calli ([Fig. 6G](#pone-0098472-g006){ref-type="fig"}). The cuticle swellings are noticed on the whole inner lip surface ([Fig. 6C, F, G, I](#pone-0098472-g006){ref-type="fig"}). The involute an undulate margins ([Fig. 6H](#pone-0098472-g006){ref-type="fig"}) consist of groups of elongated conical papillae, covered by undulated cuticle ([Fig. 6I](#pone-0098472-g006){ref-type="fig"}). The stout, short gynostemium with a large, deltoid, deeply concave stigma ([Fig. 7A-C](#pone-0098472-g007){ref-type="fig"}) is covered by flat cells with a regularly ridged cuticle ([Fig. 7E-G](#pone-0098472-g007){ref-type="fig"}) or conical papillae ([Fig. 7D-F](#pone-0098472-g007){ref-type="fig"}). The rounded to conical papillae occurring on the edge of the clinandrium ([Fig. 7C-D](#pone-0098472-g007){ref-type="fig"}) are covered by a cuticle with thick ridges. Under the stigma, close to the column margins, a few longer papillae are noticed ([Fig. 6E](#pone-0098472-g006){ref-type="fig"}) with visible drops of secretory remnants on their surface ([Fig. 7F](#pone-0098472-g007){ref-type="fig"}). The paracytic type of stomata is present ([Fig. 7G](#pone-0098472-g007){ref-type="fig"}). The four obovoid-ellipsoid pollinia are grouped in two pairs ([Fig. 7H](#pone-0098472-g007){ref-type="fig"}). {#pone-0098472-g006} {#pone-0098472-g007} Several differences were observed in the morphology of the lip and gynostemium between *V. peruviana* ([Fig. 7A-H](#pone-0098472-g007){ref-type="fig"}), *Warrea costaricensis* Schltr. ([Fig. 8A-D](#pone-0098472-g008){ref-type="fig"}) and *W. warreana* (Lodd. *ex* Lindl.) C. Schweinf. ([Fig. 8E-F](#pone-0098472-g008){ref-type="fig"}). In both *Warrea* species the central, single callus starts from the smooth lip base ([Fig. 8A, E](#pone-0098472-g008){ref-type="fig"}), but the surface of the lip apex is deeply undulate ([Fig. 8A, F](#pone-0098472-g008){ref-type="fig"}). The gynostemium is erect, rather short, clavate ([Fig. 8B, G](#pone-0098472-g008){ref-type="fig"}, [Fig. 9](#pone-0098472-g009){ref-type="fig"}), with a short column foot. The tegula is thin, narrowing gradually towards the apex ([Fig. 8B-C, G-H](#pone-0098472-g008){ref-type="fig"}, [Fig. 9C, D, E](#pone-0098472-g009){ref-type="fig"}). The rather large, elliptic and deeply concave stigma is partially hidden by the rostellum ([Fig. 8B-C, G-H](#pone-0098472-g008){ref-type="fig"}). The four ellipsoid-ovoid pollinia are arranged in two unequal pairs ([Fig. 8D, I](#pone-0098472-g008){ref-type="fig"}, [Fig. 9G](#pone-0098472-g009){ref-type="fig"}). {#pone-0098472-g008} {#pone-0098472-g009} Ecological niche modeling evaluation and the distribution factors {#s3e} ----------------------------------------------------------------- All repeated ecological niche models received AUC values above 0.9 ([Table S2](#pone.0098472.s002){ref-type="supplementary-material"}), which confirms the very high sensitivity of the analysis - the models accurately predict the value of an observational response [@pone.0098472-Hosmer1]. The high AUC scores are partially related to the small sample size, especially in *Warreopsis* models; however; the low standard deviation for all studied taxa observed between the replicate runs for all the studied taxa indicates that all the created models are clustered closely to the mean. From the studied bioclimatic variables the isothermality (bio3) and altitude (alt) were the most influential bioclimatical factors for *Vargasiella* distribution. Some contribution to the models derived also from the temperature seasonality. The variables limiting the distribution of *Warrea* and *Warreopsis* species are similar; however, a larger impact is observed in the case of precipitation (upon the impact of the elevation recognized in *Vargasiella*). The potential range of *Warrea warreana* and *Warreopsis pardina* is related to the amount of rainfall in the warmest quarter of the year (bio18), while the total annual precipitation (bio12) defines the geographical limits of *Warrea costaricensis* and *Warreopsis parviflora*. A comparison of the percentage contribution of the three most important variables to the models created for the studied taxa is given in [Table 3](#pone-0098472-t003){ref-type="table"}. 10.1371/journal.pone.0098472.t003 ###### Contribution of most important bioclimatic variables to the ecological niche models of the studied taxa. {#pone-0098472-t003-3} Taxon Model Var_1 (% contribution) Var_3 (% contribution) Var_3 (% contribution) ------------------------- -------------------- ------------------------ ------------------------ ------------------------ *Vargasiella* sp. LGM model Bio3 (56.5) alt (22.8) Bio4 (8.4) *Vargasiella* sp. Present-time model Bio3 (53.4) alt (22.5) Bio4 (8.5) *Warrea costaricensis* LGM model Bio12 (26.1) Bio4 (22.9) Bio3 (14.7) *Warrea costaricensis* Present-time model Bio12 (31.9) Bio4 (18.9) Bio3 (12.8) *Warrea warreana* LGM model Bio3 (44.2) Bio18 (13.9) Bio4 (13.1) *Warrea warreana* Present-time model Bio3 (45) Bio18 (16.2) Bio4 (12) *Warreopsis pardina* LGM model Bio3 (49.6) Bio4 (17.1) Bio18 (12.9) *Warreopsis pardina* Present-time model Bio3 (54.1) Bio4 (15.4) Bio18 (9.4) *Warreopsis parviflora* LGM model Bio3 (24.6) Bio12 (18.8) Bio4 (18) *Warreopsis parviflora* Present-time model Bio3 (23.6) Bio4 (21.9) Bio12 (17.8) Postulated LGM refuge areas {#s3f} --------------------------- The most suitable habitats for *Vargasiella* during the last glacial maximum (LGM) were probably distributed in the montane regions from the southern part of the Colombian Central Cordillera to the Bolivian Cordillera Real ([Fig. 10A](#pone-0098472-g010){ref-type="fig"}). Some potentially available but less suitable niches were probably also located around Pico da Neblina on the border between Venezuela and Brazil as well as near the Brazilian Mato Grosso do Sul. All the identified areas correspond, during LGM, to rock deserts and semi-deserts characterized by low vegetation cover [@pone.0098472-Olson1]. {#pone-0098472-g010} On the other hand the most suitable habitats for both *Warrea* species were located in the northern part of the South American Pacific coast (*W. warreana*, [Fig. 9C](#pone-0098472-g009){ref-type="fig"}) and along the Panamanian Isthmus (*W. costaricensis,* [Fig. 10B](#pone-0098472-g010){ref-type="fig"}). The designated refuge areas were covered in LGM with the tropical rainforest that corresponds to Olsons\' seasonal tropical forest and broad-leaved humid forest (including swamp forest). The estimated glacial refugia of *Warreopsis* species differ significantly between the studied species. The suitable ecological niches of *W. pardina* were limited to the lower parts of the Andean slopes from Peru to Colombia ([Fig. 10D](#pone-0098472-g010){ref-type="fig"}), which correspond to Olsons\' savannas and woodlands [@pone.0098472-Olson1] characterized by the presence of leaf cover above 80 cm from the ground. The suitable habitats for *W. parviflora* were more widespread and they were located both in the lower montane regions (Andes, Santa Marta, Talamanca) and highlands (Guiana Highlands) as well as in lowland areas (Darién Gap, [Fig. 10E](#pone-0098472-g010){ref-type="fig"}). The estimated refugia were covered by Olsons\' [@pone.0098472-Olson1] woodlands, seasonal tropical forest and swamp forest. Based on the niche overlap statistics the habitats occupied by *Vargasiella* species varied significantly from all other studied taxa. The most similar niches were occupied by *Warrea warreana* (D = 0.4830, I = 0.7632) while the greatest differences are observed in comparison with *Warreopsis parviflora* (D = 0.2148, I = 0.4565). The glacial habitats of both *Warrea* species were similar (D = 0.4753, I = 0.7436) whereas the suitable niches of *Warreopsis* varied slightly between the studied species (D = 0.3404, I = 0.6065). All calculations of the similarities between the geographical distribution of the niches occupied in LGM by all studied taxa are given in [Tables 4](#pone-0098472-t004){ref-type="table"}--[5](#pone-0098472-t005){ref-type="table"}. 10.1371/journal.pone.0098472.t004 ###### The glacial niche overlap statistics -- Schoener\'s D statistic. {#pone-0098472-t004-4} *W. warreana* *W. costaricensis* *Vargasiella* sp. *Warreopsis pardina* *Warreopsis parviflora* ------------------------- --------------- -------------------- ------------------- ---------------------- ------------------------- *W. warreana* x 0.4753 0.4830 0.2713 0.2770 *W. costaricensis* 0.4753 x 0.2901 0.2779 0.4737 *Vargasiella* sp. 0.4830 0.2901 x 0.3125 0.2148 *Warreopsis pardina* 0.2713 0.2779 0.3125 x 0.3404 *Warreopsis parviflora* 0.2770 0.4737 0.2148 0.3404 x 10.1371/journal.pone.0098472.t005 ###### The glacial niche overlap statistic - I statistic. {#pone-0098472-t005-5} *W. warreana* *W. costaricensis* *Vargasiella* sp. *Warreopsis pardina* *Warreopsis parviflora* ------------------------- --------------- -------------------- ------------------- ---------------------- ------------------------- *W. warreana* x 0.7436 0.7632 0.5138 0.5610 *W. costaricensis* 0.7436 x 0.5429 0.5377 0.7677 *Vargasiella* sp. 0.7632 0.5429 x 0.5800 0.4565 *Warreopsis pardina* 0.5138 0.5377 0.5800 x 0.6065 *Warreopsis parviflora* 0.5610 0.7677 0.4565 0.6065 x Current potential range {#s3g} ----------------------- Compared to the LGM range, the current distribution of the habitats suitable for the occurrence of the *Vargasiella* species is significantly limited both longitudinally as well as latitudinally. The available niches are concentrated from the Ecuadorian and Peruvian Andes to the Altiplano in the south. There is no additional suitable location outside the Andean region ([Fig. 11A](#pone-0098472-g011){ref-type="fig"}). {#pone-0098472-g011} The current potential ranges of both studied *Warrea* species are discontinuous. The suitable niches of *W. costaricensis* are distributed along lower parts of Cordillera de Talamanca to Cordillera de Salamanca ([Fig. 11B](#pone-0098472-g011){ref-type="fig"}). The other available habitats are located in the Darién region and the northern foothills of the Western Andean Cordillera as well as in eastern slopes of the Eastern Cordillera. The most suitable habitats for *W. warreana* are located on the Pacific coast of northern South America; however, the essential part of the Amazon basin is characterized by somewhat less favorable, but possibly tolerable bioclimatic conditions ([Fig. 11C](#pone-0098472-g011){ref-type="fig"}). The suitable habitats for *Warreopsis pardina* are restricted to the lower parts of the northern Andean range from Colombia to the Chamaya river in the south, but the species has not been reported from Peru so far ([Fig. 11D](#pone-0098472-g011){ref-type="fig"}). The niches of *Warreopsis parviflora* are distributed irregularly along the northern Andes and Cordillera de Talamanca. Additional potentially available habitats are located in the Santa Marta mountains (Colombia) and Guiana Highlands ([Fig. 11E](#pone-0098472-g011){ref-type="fig"}). It is noteworthy that so far this species has only been collected in Central America. The similarities between the geographical distribution of the suitable niches of all studied taxa are given in [Tables 6](#pone-0098472-t006){ref-type="table"}--[7](#pone-0098472-t007){ref-type="table"}. 10.1371/journal.pone.0098472.t006 ###### The niche overlap statistics -- Schoener\'s D statistic. {#pone-0098472-t006-6} *Vargasiella* sp. *Warrea warreana* *Warrea costaricensis* *Warreopsis pardina* *Warreopsis parviflora* ------------------------- ------------------- ------------------- ------------------------ ---------------------- ------------------------- *Vargasiella* sp. x 0.3952 0.3812 0.2590 0.2576 *Warrea warreana* 0.3952 x 0.5650 0.2085 0.3272 *Warrea costaricensis* 0.3812 0.5650 x 0.2300 0.5434 *Warreopsis pardina* 0.2590 0.2085 0.2300 x 0.3859 *Warreopsis parviflora* 0.2576 0.3272 0.5434 0.3859 x 10.1371/journal.pone.0098472.t007 ###### The niche overlap statistics - I statistic. {#pone-0098472-t007-7} *Vargasiella* sp. *Warrea warreana* *Warrea costaricensis* *Warreopsis pardina* *Warreopsis parviflora* ------------------------- ------------------- ------------------- ------------------------ ---------------------- ------------------------- *Vargasiella* sp. x 0.6754 0.6459 0.5219 0.4840 *Warrea warreana* 0.6754 x 0.8001 0.4337 0.5863 *Warrea costaricensis* 0.6459 0.8001 x 0.4779 0.8145 *Warreopsis pardina* 0.5219 0.4337 0.4779 x 0.6610 *Warreopsis parviflora* 0.4840 0.5863 0.8145 0.6610 x Range and niche shift {#s3h} --------------------- While the current and LGM estimated distribution of the suitable habitats seem to be similar, the calculated values indicate the significant differences in both, ranges and niches occupied by *Vargasiella* in the studied time periods. The whole potential range overlap value is just 0.361 and the niche overlap statistics are: D = 0.279, I = 0.572. The same calculations made for models of the studied *Warrea* and *Warreopsis* species revealed a rather insignificant shift in the occupied areas and niches between LGM and the present time ([Table 8](#pone-0098472-t008){ref-type="table"}). 10.1371/journal.pone.0098472.t008 ###### Overlapping of the ranges and occupied niches between LGM and present time. {#pone-0098472-t008-8} Taxon Range overlap test D I -------------------- -------------------- ------- ------- *Vargasiella* sp. 0.361 0.279 0.572 *W. costaricensis* 0.761 0.716 0.924 *W. warreana* 0.934 0.930 0.994 *W. pardina* 0.748 0.708 0.922 *W. parviflora* 0.620 0.592 0.831 Discussion {#s4} ========== The taxonomic position of *Warrea* and *Warreopsis* is rather obvious. Almost all orchid taxonomists classify them usually within the variously defined Zygopetalinae ([@pone.0098472-Dressler1] [@pone.0098472-Dressler2] [@pone.0098472-Senghas2] [@pone.0098472-Pridgeon1]). However, Szlachetko [@pone.0098472-Szlachetko1] separated *Warrea*, *Otostylis* Schltr., *Warreella* Schltr. and *Warreopsis* Garay from Zygopetalinae and united them in the subtribe Warreinae. This subtribe has been proposed as a taxon clustering those genera different from all other Zygopetaleae *sensu* Szlachetko [@pone.0098472-Szlachetko1] by having slender, homoblastic pseudobulbs and convolute, plicate leaves. All the genera share a similar gynostemium structure, i.e., usually a short column foot, 4 superposed, unequal pollinia, a long rostellum, lamellar tegula and viscidium forming together a kind of sheath around the rostellum core. *Warrea/Warreopsis* and *Vargasiella* differ one from another in their stem architecture, and leaf and lip structure. Even though they are characterized by the presence of four superposed pollinia arranged in two pairs, the rostellum structure is essentially different. In *Warrea, Warreopsis* and other genera of Zygopetaleae *sensu* Szlachetko [@pone.0098472-Szlachetko1] the rostellum is 3-lobed, with the middle lobe being ligulate-triangular, acute, springy and obscure lateral lobes, separated from the middle one by a shallow sinus. The tegula and viscidium are thin, lamellar, more or less elongate and together they form a sheath around the rostellum core. The stigma is very narrow, hidden in the most part by the rostellum. In contrast, the rostellum of *Vargasiella* is short and broad and its major part becomes transformed into the viscidium and tegula, and when removed with the pollinia drops out as accessories. The tegula is small and the viscidium even smaller, and distinctly bilobed. The receptive surface is broad, easily accessible to pollen mass. Additionally, the genera in question are distinguishable by the form of the rostellum remnant. In *Warrea*, the rostellum after removal of the pollinarium is 3-lobed, the middle lobe is subulate, acute, rigid, both lateral lobes are much reduced, obtuse. In the closely related *Warreopsis* the rostellum remnant is similar, with the middle lobe being shorter. In contrast in *Vargasiella*, the rostellum remnant consists of a short, fleshy apiculus in the middle. The question is why we consider the rostellum structure in detail? The rostellum evolved from an altered middle lobe of the stigma and it plays a crucial role in the pollination of orchid flowers. First, it produces important accessories for pollinia, i.e., the viscidium and tegula, the structures that enabled their transfer by a pollinator. Secondly, in many species the rostellum forms a kind of barrier between the anther and the receptive surface suppressing autogamy. The rostellum of *Warrea* and *Warreopsis* are such cases. The short and erect rostellum of *Vargasiella* is not a sufficient barrier to prevent autogamy. A similar type of gynostemium is found in *Maxillaria*. The rostellum is short and broad, dome-like and the stigma is relatively large, not hidden by the rostellum. Both rostellar products are similar to those of *Vargasiella*, i.e. the tegula is rather small and the viscidium distinctly bilobed, hippocrepiform and does not form a sheath. The pollinators\' observations would be useful to understand the function of these floral structures. Unfortunately, nothing is known about pollination biology of *Vargasiella*, or that of *Warrea* or *Warreopsis*. Pupulin [@pone.0098472-Pupulin1] considered that *Warrea* is probably pollinated by male euglossine bees. Based on the flower colour and its morphology ([Fig. 12](#pone-0098472-g012){ref-type="fig"}) and the knowledge that most Zygopetalinae are pollinated by euglossinae bees [@pone.0098472-Davies1], we can predict that *Vargasiella* are exploits the same pollinators. The cuticle swellings visible on the lip surface may be the fragrance which is collected by male euglossinae bees, but this hypothesis needs more detailed studies. {#pone-0098472-g012} The number of known populations of *Vargasiella* is very small, and the information about its ecological interactions and habitat requirements is equally limited. The quick organic matter decomposition in the tropical regions together with the anatomical structure of the studied species (i.a. the thin coat of the microscopic seeds, and thin cuticule) have made them undetectable in fossil material and this precludes conducting traditional paleobotanic research including palynological procedures. The presented results of the ENM analysis based on presence-only data clearly indicate the Central Andes as the most probable glacial refugium of *Vargasiella*. The statistical calculations of the ranges and niches overlap between models for the studied time periods indicate not only the current regression of *Vargasiella* range, but also the niche shifting in response to warming climate after the LGM. This kind of range reduction is primarily the result of distributional shifts and fragmentation of primary habitats as well as the expansion of unsuitable climatic conditions exceeding species\' ecological tolerance [@pone.0098472-Hewitt1] [@pone.0098472-Hewitt2]. The exact opposite situation is observed with respect to the other two studied genera. The estimated suitable habitats for the occurrence of *Warrea* and *Warreopsis* species during LGM were probably located in warmer lowland to lower montane regions of Mesoamerica and Northern South America and no significant changes in their geographic ranges or occupied niches are observed in the present time. The finding is important especially in terms of plant speciation. Wright [@pone.0098472-Wright1] suggested that ecological shifts are the most likely basis for the origin of genera and these shifts seem to be punctuational rather than gradual [@pone.0098472-Eldredge1] [@pone.0098472-Bateman1]. We decided to include the results of niche modeling in this study, despite the fact that we are aware that the results of ENM analysis may overestimate the distribution of the suitable niches of the studied taxa, mainly due to the small sample size. The models created for the present time are rather consistent with the occurrence data of the species, even though not all of them were included in the analysis because it was not possible to georeference their locations. We believe that in the case of poorly known taxa, such as *Vargasiella*, which are untraceable in fossil material, the use of the algorithms for computing the distribution of their suitable niches in the past is the only method to recognize their areas of historical stability and hereby to analyse the changes in their ecological niches. *Vargasiella* species can be defined as psychrophytes usually growing in wet and fairly cool conditions, in submontane and montane forest and forest edges, about 2100--3400 m a.s.l. The occurrence of *V. venezuelana* has been reported from *Bonnetia* forest (*Steyermark 74914*, F), while *V. peruviana* has also been reported from stunted cloud forest (*Foster & Smith 9094*, MO; *Gentry & Smith 35984*, MO). *Warrea* species occupy dense, rain and moist forest, growing in decaying leaf litter in the shade, usually below 1500 m a.s.l, and often near rivers and streams (*Binacional 1104*, MO; *Dressler 3250,* MO). *Warrea warreana* has been reported growing epiphytically on tree ferns (*Zuloaga & al. 6898*, SI). Studied *Warreopsis* species are found in lower montane and montane areas growing terrestrially (rarely epiphytically) in wet and very wet forests (*Haber & al. 4510, 4484, MO*; *Davidse & Pohl 1692A*, MO). The ENM results indicate that the occurrence of *Vargasiella* and *Warrea* species is limited by similar bioclimatic variables, mainly by isothermality. However, the distribution of *Vargasiella* is seriously influenced by the elevation above sea level and this is probably the crucial environmental factor resulting in the separation of suitable habitats and thereby the ranges of the studied taxa. The habitat requirements of *Vargasiella clearly* differ from those of *Warrea* and *Warreopsis*, as indicated by the scores of the niche overlap test. The clocktree of Zygopetalinae indicated that the most recent common ancestor of *Vargasiella, Warrea* and *Warreopsis* could probably originate from around 7.5 Ma ago, while the divergence time for *Vargasiella* and *Warrea* was estimated for approximately 5.4 Ma ago. These events may be easily linked to the major geological changes in South America. The rapid rise of the Andean plateau took place in the late Miocene (10.4--5 Ma ago) and this was related with the establishment of dramatic precipitation gradients perpendicular to the orogen, and changes in tectonic processes in the Andean orogenic wedge [@pone.0098472-Garzione1]. These changes also influenced the Amazon region and during the Pliocene the lowland fluvial systems of southwestern Amazonia become isolated from the Andes by the newly formed fluvial systems (Ucayali and Madre de Dios). In the early Pliocene the Amazon fluvial system integrated regionally and acquired its present appearance [@pone.0098472-Latrubesse1]. Conclusions {#s5} =========== In the genetic analyses *Vargasiella* is placed within a highly supported subclade together with *Warrea* and *Warreopsis* that is situated at the base of the Zygopetalinae *sensu lato* clade. However, the relationships between *Warrea, Warreopsis* and *Vargasiella* are not resolved. Significant differences in the presence of storage organs, leaf structure, inflorescence type and lip form were observed in the macromorphological studies of *Vargasiella* and related taxa. Moreover, differences in the micromorphology of the lip and gynostemium between *V. peruviana, Warrea costaricensis* and *W. warreana* were found. The ecological niche modeling suggests regression of *Vargasiella* range since LGM. Moreover, unlike in *Warrea* and *Warreopsis* a niche shift in response to postglacial climate changes was observed in *Vargasiella*. The genus *Vargasiella* appears to be an outshoot of the main branch of evolution of Zygopetaleae. Interestingly, the *Vargasiella*-*Warrea* dichotomy could have taken place ca 5.4 Ma ago, later than the divergence of *Warreopsis* from the mutual lineage (ca 7.5 Ma ago, [Fig. 4](#pone-0098472-g004){ref-type="fig"}). Considering the morphological data and the results of molecular analyses we therefore formulate a hypothesis that *Vargasiella* and *Warrea* may have evolved from a common ancestor. Accumulation of morphological differences and acceleration of the evolution of *Vargasiella* were faster than in other Warreinae, a feature which could probably be synchronized with adaptation to cooler and wetter conditions. This can be supported by the observations made on *Warrea* and *Warreopsis*, both inhabiting similar ecological niches. Both are more reminiscent of one another considering morphological characters than could be speculated based on molecular analysis outcomes. Taxonomic Treatment {#s6} =================== ***Vargasiella*** C.Schweinf., Bot. Mus. Leafl., Harvard Univ. 15: 150. 1952. Type: *Vargasiella peruviana* C.Schweinf. The genus includes two species, they can be distinguished as follows: 1\. Sepals and petals subsimilar, lip sessile, base truncate-subcordate, ovate-lanceolate, with calli ................***V. peruviana*** 1\. Sepals and petals dissimilar, lip clawed, base cordate, elliptic-suborbicular, ecallose .......... ***V. venezuelana*** ***Vargasiella peruviana*** C.Schweinf., Bot. Mus. Leafl., Harvard Univ. 15: 150, tab. 47. 1952; Type: Peru, Convención, hills of Pintobamba, in humus forest, perianth white with pinkish lip. *Vargas 3288* (holotype AMES! - type illustration). Cuzco: Paucartambo, Pillahuata, floral segments white lined with pink. *Vargas 3010* (paratype AMES!). Plant epiphytic or terrestrial, slender. Stem elongate, decumbent, producing scattered fibrous, short, densely tomentose and stout roots in the lower part, leafy in the upper part, entirely concealed by tubular sheaths. Leaves 5.6--13.5 cm long, 2--2.5 cm wide, several, distichous, convolute, elliptic to oblong-elliptic or elliptic-lanceolate, acuminate, cuneate below, sessile or indistinctly petiolated, articulated to close tubular sheaths, membranaceous, 3- to 5-nerved. Inflorescence up to 33.5 cm long, arising from the axil of an upper leaf, erect, racemose; peduncle 21.5 cm long with several remote, tubular below and lanceolate above sheaths; raceme loosely up to 15-flowered. Flowers medium-sized, subfleshy, perianth white with pinkish lip. Floral bracts up to 14 mm long, oblong, acute, spreading. Ovary prominently 6-sulcate. Dorsal sepal 13.2 mm long, 6 mm wide, ovate-oblong, acute to mucronate, concave, 5-nerved, with margins very minutely cellular-erose. Lateral sepals 14.5 mm long, 7 mm wide, similar, ovate-oblong, acute, cymbiform, dorsally carinate with the keel produced into a conspicuous mucronate apex, 5- or 6-nerved, lightly oblique. Petals 12 mm long, 6 mm wide, elliptic-ovate, acute to apiculate, 5-nerved. Lip 10 mm long, 6 mm wide, simple, arcuate-recurved and parallel to the column with the sides erect in natural position, articulated to the column foot, with the anterior margins strongly undulate, disc when expanded ovate-oblong, cordate at base, rounded and acute or apiculate at apex, furnished with two fleshy thickenings in the lower half. Gynostemium 7 mm long, stout, with a narrow fleshy wing on each side throughout, subtruncate above. Anther relatively small, 1-celled, galeate. Pollinia 4, in two unequal pairs, without appendages, strongly complanate-subglobose, waxy. Flowering time: Throughout the year. Distribution: Peru, Bolivia. Elev. 2400--3400 m a.s.l. ***Vargasiella venezuelana*** C.Schweinf., Bot. Mus. Leafl., Harvard Univ. 18: 219, tab. 44. 1958; Type: Venezuela, Bolívar. Chimantá Massif, northwestern part of summit of Abácapa-tepuí, in Bonnetia forest. *Steyermark 74914* (holotype AMES! - type illustration, isotype F!). Plant terrestrial, robust, up to 187.5 cm long with decumbent stem, sparsely rooting. Roots solitary, very few, fibrous, rather stout, sparsely pubescent. Stem entirely concealed by appressed, imbricating, evanescent, leaf-bearing, tubular sheaths; the lower ones scarious and disintegrating into fibers, the middle and the upper ones green and leaf-bearing. Leaves 14-17 cm long, up to 3.3 cm wide, elliptic to elliptic-oblong, acuminate, long-narrowed below to an articulated case, submembranaceous, plicate, 5- to 7-nerved. Inflorescence up to 52 cm long, arising from the axil of one of the middle leaves; peduncle 43 cm long, dull lavender, glabrous, remotely concealed with six short, tubular, acute sheaths; raceme up to 9 cm long, very loosely few-flowered. Flowers purple, subfleshy, with the sepals projecting backwards and the petals erect. Floral bracts small, narrowly lanceolate, concave, equaling about half of the pedicellate ovary. Ovary up to 23 mm long. Dorsal sepal 15 mm long, 4.4 mm wide, lanceolate-oblong, obtuse to acute. Lateral sepals 16 mm long, 6 mm wide, similar, obliquely lanceolate-oblong, obtuse to acute. Petals 12 mm long, 7 mm wide, distinctly shorter and broader than the sepals. Lip 11 mm long, 10 mm wide, simple, clawed, with involute undulate margins; claw short but distinct, abruptly dilated from an oblong base, 3 mm long, furnished with a central fleshy callus dividing into two branches; lamina gently recurved, triangular-ovate, rounded at the apex, conspicuously cordate at the base, fleshy thickened in the middle. Gynostemium 5 mm long, stout. Flowering time: April. Distribution: Venezuela. Elev. 2125--2300 m a.s.l. Supporting Information {#s7} ====================== ###### **Variables used in the modelling.** (DOC) ###### Click here for additional data file. ###### **The average training AUC for the replicate runs (AUC - area under the curve, SD -- standard deviation).** (DOC) ###### Click here for additional data file. ###### **List of analysed taxa downloaded from International Nucleotide Sequence Databases.** INSD accession numbers for DNA sequences are listed in following order: ITS matK trnL-F. (DOC) ###### Click here for additional data file. ###### **Set of the characters used in the phenetic analysis.** (DOC) ###### Click here for additional data file. ###### **Morphological character matrix.** (DOC) ###### Click here for additional data file. ###### **Localities of the specimens collection of** ***Vargasiella*** **C.Schweinf. and** ***Warrea*** **Lindl. used in the ENM analysis.** (DOC) ###### Click here for additional data file. The curators and staff of the cited herbaria are thanked for their kind hospitality and assistance during visits and for making specimens for making specimens available on loan. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: DS MG MK AK. Performed the experiments: DS MG MK JM-E AK PR TK. Analyzed the data: DS MG MK JM-E AK PR TK. Contributed reagents/materials/analysis tools: DS MG MK JM-E AK PR TK. Wrote the paper: DS MG MK JM-E AK.

Introduction {#sec1} ============ Highly concentrated solutions occur in a wide range of industrial and natural processes including brines, carbon capture, atmospheric aerosols, and food drying. These solutions are typically labeled as exhibiting nonideal behavior because the vapor pressures of the solute and solvent and their corresponding activities in solution are not proportional to their mole fraction. Raoult first proposed this linear mole fraction--vapor pressure relationship in 1887^[@ref5]^ and Raoult's law has become a paradigm in chemistry and chemical engineering. For dilute solutions, this relation holds but deviations grow as the solution becomes more concentrated. In 1908, Callendar^[@ref6]^ explained some of the deviations from Raoult's model by positing that in aqueous solutions, solute molecules are hydrated such that some of the water is bound to the solute, so it does not contribute to the vapor pressure. That is, only the so-called "free water" contributes to the vapor pressure. This paradigm extended the validity of Raoult's law to higher concentrations but it still failed at yet higher concentrations. In 1973, Stokes and Robinson^[@ref7]^ extended this paradigm further by assuming that for electrolytes there is an equilibrium between the free and bound water and a power law relationship between the equilibrium constants. As the concentration of the solution increases, the water activity decreases and some of the bound water falls off the solute and becomes free water. Starting in 2011, we published a series of papers that used statistical mechanical techniques to describe solutions over the full range from the infinitely dilute limit to the pure solute limit.^[@ref8]−[@ref10]^ These models were successful at describing solutions of organics and electrolytes in water by assuming that one water was bound to the solute or ion, the next water bound to that water, and so on. Although a huge step forward, this model has a shortcoming of not necessarily capturing hydration the way we understand it. From the prior work of Callendar,^[@ref6]^ Stokes and Robinson,^[@ref7]^ and many others, we know that multiple waters can bind to solute molecules. They may also bind to each other in a secondary layer, but the single stacked water model of Dutcher and colleagues does not capture what is known from the physics and chemistry of these solutions. For instance, raffinose is thought to have a hydration number (the average number of water molecules bound to the solute molecule in dilute solutions) in excess of 10 but we do not expect 10 waters to be stacked on top of each other to hydrate raffinose. Results and Discussion {#sec2} ====================== Here, we apply two assumptions to derive equations that describe the "nonideality" of solutions to show that they are actually ideal and fit Raoult's law: Assumption 1: Raoult's law applies rigorously. That is, the ratio of partial pressure to vapor pressure, the activity of all constituents, *a*~*i*~, be they pure solute, hydrated solute, or free water, is simply their mole fraction given bywhere *n*~*i*~ is the number of moles of constituent *i* in solution and *n*~t~ is the total number of moles of all of the constituents in solution. These constituents consist of free water, bare solute, and hydrated solute. Assumption 2: Each solute may be hydrated and this hydration is governed by the following equilibria\...where *a*~w~ is the activity of water, *a*~*ij*w~ is the activity of solute *i* hydrated with *j* water molecules, and *K*~*ij*w~ is the equilibrium constant between the solute hydrated with *j* water molecules and (*j* -- 1) water molecules. All activities are given by [eq [1](#eq1){ref-type="disp-formula"}](#eq1){ref-type="disp-formula"}. We will now use [eqs [1](#eq1){ref-type="disp-formula"}](#eq1){ref-type="disp-formula"} and [2.1](#eq2){ref-type="disp-formula"}--[2.4](#eq5){ref-type="disp-formula"} to calculate the water associated with each solute. First, we multiply all of these equilibria in [eqs [2.1](#eq2){ref-type="disp-formula"}](#eq2){ref-type="disp-formula"}--[2.4](#eq5){ref-type="disp-formula"} by each other, so that the activities of the hydrated solutes cancel, except for the most hydrated solute, givingMultiply both sides by *n*~t~ and applying [eq [1](#eq1){ref-type="disp-formula"}](#eq1){ref-type="disp-formula"}, we obtain the number moles of each hydrated solute in terms of the number of moles of free solute, water activity, and the equilibrium constantswhere the number of moles of each constituent, *n*~*ij*w~, contains one mole of solute *i* and *j* moles of water. By adding all of these together plus the number of moles of bare solute, we obtain the total number of moles of solute *i* in solutionwhere *S*~*i*~ is the total number of moles of solute *i* per mole of unhydrated solute *i* and is only a function of water activity. Similarly, the total amount of water hydrated to solute *i* is given bywhere *W*~*i*~ is the total number of moles of water hydrated to solute *i* per mole of bare solute *i* and like *S*~*i*~ is only a function of water activity. Dividing [eq [6](#eq9){ref-type="disp-formula"}](#eq9){ref-type="disp-formula"} by [eq [5](#eq8){ref-type="disp-formula"}](#eq8){ref-type="disp-formula"} gives the hydration number as a function of water activityThat is, *H*~*i*~(*a*~w~) is the total number of water molecules hydrated to all of the hydrated forms of solute *i* per total number of moles of solute *i*. The hydration number defined by Callendar (1908) is the value of *H*~*i*~ at the limit of infinite dilution, *H*~*i*~(*a*~w~ = 1). Marcus^[@ref11]^ used isothermal compressibility to deduce how hydration number depends on concentration fitting a linear curve through his data. [Equation [7](#eq10){ref-type="disp-formula"}](#eq10){ref-type="disp-formula"} is the number of bound water molecules per solute molecule. We also need to know the number of free waters in solution. Starting with the definition of water activity from [eq [1](#eq1){ref-type="disp-formula"}](#eq1){ref-type="disp-formula"}where ∑~*i*~*n*~*iT*~ is the total number of all solutes in solution regardless of their hydration state. Remember that *n*~t~ is the total number of moles of free water, free solute, and hydrated solute in solution. Solving [eq [8](#eq11){ref-type="disp-formula"}](#eq11){ref-type="disp-formula"} for the number of moles of free water givesThe term *a*~w~/(1 -- *a*~w~) appears in other derivations of the thermodynamics of solutions, such as that by Dutcher and co-workers.^[@ref8]−[@ref10]^ Here, we see that it represents the number of moles of free water per mole of solute. Now, we have the tools that we need to derive the relationship between molality and water activity. The simplest case is an aqueous solution containing a single solute *A* that does not dissociate or associate. Many alcohols and sugars fit this description. The total amount of water, *n*~wTot~, in a solution containing solute *A*, its hydrated forms, and free water isCombining this with [eqs [7](#eq10){ref-type="disp-formula"}](#eq10){ref-type="disp-formula"} and [9](#eq12){ref-type="disp-formula"} givesNoting that the molality of solute in solution is given by *M*~w~*m = n*~*AT*~/*n*~wTot~, where *M*~w~ is the molar mass of water, yields an equation for the molality as a function of water activityThe denominator is the amount of free water per total solute plus the amount of water bound in hydrates per total solute. The solute activity is obtained by taking the total number of constituents in solutionDividing both sides by *n*~t~ and combining with [eq [5](#eq8){ref-type="disp-formula"}](#eq8){ref-type="disp-formula"} to obtainwhich can also be obtained using the Gibbs--Duhem equation, [eq [12](#eq15){ref-type="disp-formula"}](#eq15){ref-type="disp-formula"}, and noting that *W*~*A*~ = *a*~w~d*S*~*A*~/d*a*~w~. The values of the equilibrium constants, *K*~*iq*w~, govern the performance of [eqs [12](#eq15){ref-type="disp-formula"}](#eq15){ref-type="disp-formula"} and [14](#eq17){ref-type="disp-formula"} via the functions *H*~*A*~(*a*~w~) and *S*~*A*~(*a*~w~). For each water hydrated to each solute, there is a *K*~*iq*w~ so that say for raffinose that has a hydration number of about 12, at least 12 *K*~*iq*w~ values are needed. For organic solutes, we expect that roughly one water molecule will hydrogen bond to each of the OH moieties and that the equilibrium constant will be roughly the same for each hydrogen bond. As a result, the *K*~*iq*w~ values should be roughly the same until *q* (the number of hydrated waters) reaches a value, where all of the bonding locations are taken, then it should precipitously drop in value to zero. The logistic function neatly captures this behaviorwhere *K*~*i*w~^*x*^ is the equilibrium constant for each water hydrogen bonded to the solute, *q*~0~ is the number of bound waters where the value of *K*~*iq*w~ is half *K*~*i*w~^*x*^, and Δ*q* governs how rapidly *K*~*iq*w~ drops to zero as *q* increases. Since we expect the value of *K*~*iq*w~ to drop precipitously to zero when *q* becomes larger than the number of hydrogen bonding sites, *q*~0~, we set Δ*q* = 0.1, an arbitrary but small value. We now have only two fit parameters, *K*~*i*w~^*x*^ and *q*~0~, instead of the 12 or more equilibrium constants needed for say raffinose. [Figure [1](#fig1){ref-type="fig"}](#fig1){ref-type="fig"} shows the fit for raffinose. For this and subsequent solutes, osmotic coefficient defined as φ = −ln(*a*~w~)/(*M*~w~*m*) is displayed on the *y*-axis because it is a very sensitive indicator of deviations between activity data and fits to these data. The square root of the solute mole fraction is displayed on the *x-*axis to better show the behavior at low concentration, which is not relevant for raffinose but is for other more-soluble solutes. ![Osmotic coefficient as a function of solute mole fraction for raffinose in water. Freezing point depression data from ref ([@ref2]). *K*~*i*w~^*x*^ was fixed at 2, the fit value for *q*~0~ is 13.8, and the resulting hydration number is 12.3.](ao-2019-01707e_0001){#fig1} Due to scatter in the raffinose data, many combinations of *K*~*i*w~^*x*^ and *q*~0~ fit the data well, so here we picked *K*~*i*w~^*x*^ = 2 and just fit *q*~0~. For glycerol and NaCl, the values were simple fits to the data. The fits minimized the sum of the squared error in osmotic coefficient between the model and the data. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}a shows the fit for glycerol that is miscible in water, so data exist for the full range of solute mole fractions. For both raffinose and glycerol, the hydration number deduced from these fits agrees well with the values reported in the literature. [Figure [2](#fig2){ref-type="fig"}](#fig2){ref-type="fig"}b shows the change in hydration as a function of water activity for glycerol. The corresponding figure for raffinose and NaCl is not shown because the change over the water activity range is extremely small. ![(a) Osmotic coefficient as a function of solute mole fraction for glycerol in water. Data from triangle,^[@ref3]^ diamond.^[@ref4]^ Parameter values are *K*~*i*w~^*x*^ = 0.553 and *q*~0~ = 5.1; the resulting hydration number is 1.0. (b) Predicted hydration as a function of water activity for glycerol using the same parameters as in (a).](ao-2019-01707e_0002){#fig2} For two or more solutes in solutions *A* and *B*, what kind of mixing rule applies? The total amount of water in solution is nowDividing both sides by *n*~wTot~ and rearranging givesComparing this to [eq [12](#eq15){ref-type="disp-formula"}](#eq15){ref-type="disp-formula"}, we see that the square-bracketed quantities in the denominator are simply the single-solute molalities; that is, the numerator in each term is the molality in the solution mixture, whereas the denominator is the molality in the single-solute solution. This is the ZSR mixing rule posited independently by Zdanovski^[@ref12]^ and Stokes and Robinson,^[@ref13]^ which has been shown to be valid for a wide range of solutes that do not associate in solution. It also shows the role that free water *a*~w~/(1 -- *a*~w~) and hydrated water *H*~*A*~ play. In the work of Dutcher and colleagues,^[@ref9]^ free water was assumed to be associated with each solute. Here, we obtain the same result without assumption. Electrolytes can be thought of as multicomponent solutions, wherein the salt dissociates into its ionic components and each of these ions comprises a solute in solution. In contrast to mixtures of nonionic solutes, the ions project their electrostatic fields beyond the vicinity of the ion altering the thermodynamics of the solution. This effect was first recognized by Debye and Hückel in 1923.^[@ref14],[@ref15]^ Pitzer^[@ref16]^ evaluated a number of forms for representing these long range forces and found that it is well represented bywhere *I* is the ionic strength, *A~x~* = 2.917 at 298 K and ρ is a fit parameter related to the ion size and charge. The factor *K*~w~^DH^ modifies the water activity in the equations derived here. Employing the stoichiometry of the electrolyte, the equations derived above and this Debye--Huckel term yield an expression for the molality as a function of water activity for electrolyteswhere *v*~*A*~ and *v*~*B*~ are the number of cations and anions in the electrolyte, respectively, assuming that the electrolyte is only comprised of one type of cation and one type of anion. More complex electrolytes are an easy extension. [Figure [3](#fig3){ref-type="fig"}](#fig3){ref-type="fig"} shows the fit for NaCl to saturation. ![Osmotic coefficient as a function of solute mole fraction for sodium chloride in water. Data from ref ([@ref1]). Parameter values are given in the table.](ao-2019-01707e_0003){#fig3} Conclusions {#sec3} =========== In summary, the theory presented here shows that Raoult's original idea, that the activity of solutes in solution is proportional to the mole fraction, needs no correction if we consider the hydrated forms of the solute in solution as Callendar suggested and employ simple equilibria between the hydrated forms. Most models of solutes in solution use one of the forms derived by Pitzer^[@ref17]^ because they accurately model the molality--activity relationship over a wide range of conditions. For instance, the Clegg and Pitzer formulation is employed by E-AIM, our popular model of aerosol thermodynamics that performs about 100 000 calculations annually for scientists worldwide working in the fields of air pollution and climate change. The shortcoming of the Pitzer family of models is that they do not extrapolate well to concentrations higher than the available data because they are semiempirical. In fact, beyond the physical data, they can give unphysical results. The formulation by Dutcher and co-workers^[@ref8]−[@ref10]^ extrapolates well at concentrations beyond the available data but does not capture conceptually how we understand that solutes hydrate in solution. The model presented here, based on the work published more than 100 years ago, extrapolates well to concentrations higher than the available data, represents hydration of the solute realistically, covers both organic and electrolyte solutes in water, and due to its simplicity shows the role of free and bound water in solution. The author declares no competing financial interest. I would like to thank Ahmad Ikram, a student at Franklin High School, for his help with the calculations and figures and Simon Clegg, University of East Anglia, for his valuable guidance.

The decrease in partial pressure of oxygen on ascent to high altitude has been reported to adversely affect the cognitive performance in both animals and human subjects. Occurrence of hippocampal atrophy and hypoxia mediated neurodegeneration has been shown following ascent to high altitude. There have been reports on increase in latency of P3 component reflecting sensory discrimination and delay in evaluation process and impairment in short term memory following ascent to high altitude[@ref1]. Recent studies on a small group of human volunteers residing at high altitude revealed impairment in verbal working memory which could be associated with chronic hypoxia exposure[@ref2]. However, most of these studies have either been performed in mountaineers or on subjects who have stayed at high altitude for shorter durations. No demographic study has ever been conducted to assess the prevalence of mild cognitive impairment (MCI) in the considerable size of human population that stay at high altitude for prolonged durations owing to professional or livelihood reasons. MCI, as the name suggests, is the early stage of cognitive impairment where only a single or a few of the cognitive domains are affected usually to a lesser extent. MCI is characterized by impairment in cognitive domain tests after correction for age and education, but with preserved general cognitive functioning, intact activities of daily living and absence of dementia[@ref3]. It is however, an excellent indicator for preponderance and progression of an individual towards dementia and cognitive impairment[@ref4]. The commonly used tests for screening MCI are Mini Mental State Examination (MMSE)[@ref5], Montreal cognitive Assessment (MoCA)[@ref6], Mini-Cog[@ref7], Computer Administered Neuropsychological Score (CANS-MCI)[@ref8] and the recently developed Patient Reported Outcomes in Cognitive Impairment (PROCOG)[@ref9]. With the recent findings on involvement of several brain regions in cognitive functions and low predictive accuracy of several of the tests for MCI that are currently in use, there is a necessity to increase the scope of cognitive assessment to cognitive domains like procedural memory, mind-body co-ordination, attention and learning of complex tasks through improvised and customized psychometric tests. MMSE which is most widely used for demographic screening of MCI, has been shown to have low sensitivity and is highly influenced by age and education[@ref10][@ref11]. MoCA on the other hand, has better sensitivity than MMSE, takes a short duration of 10 min to be administered and has a wide application in routine clinical practice. However, MoCA covers only 10 cognitive domains and does not include procedural memory, mind-body co-ordination and learning of complex tasks. PROCOG comprises 55 questions, making it too lengthy to be administered routinely and is thus less preferred for demographic studies. It can rather be used as a confirmatory test following preliminary screening. Further, clinical practice of computer assisted tests like Computer Assessment of MCI (CAMCI)[@ref10] or CANS-MCI is difficult in remote locations and specifically in the high altitude regions owing to the technological requirements and administration skills. The present study was, therefore, designed to investigate the percentage prevalence and extent of MCI in a young and healthy ALL population staying at altitudes above 4,300 m in the trans-Himalayan regions for durations longer than 12 months and to validate a newly designed multi-domain cognitive screening test (MDCST) for demographic studies in remote locations at high altitude. The MDCST scores were compared with MMSE and MoCA scores and the specificity and reliability were determined using electrophysiological recordings. Standard Medical Questionnaire for medical history, Everyday ability Scale for India (EDAS-I) and Clinical Dementia Rating (CDR) formed a part of the neuropsychologic battery and functional assessments performed during the investigation. All these parameters were compared with age, sex and education matched control subjects who had never been to high altitude. A sensitive, rapid, easy to administer and customized neuropsychological battery was used for the study. Material & Methods {#sec1-1} ================== *Subjects*: The study was conducted in High altitude Physiology division of Defence Institute of High Altitude Research, Leh, India, during August 2009 - January 2011 after obtaining ethical clearance from the institutional ethics committee of Defence Institute of High Altitude Research. A total of 2000 volunteers comprising of male subjects with education of 10-12 yr and between age group of 25-45 yr enrolled voluntarily at different army unit locations after being explained about the study purpose, protocol, and expected outcomes. Amongst the total population, 1000 ALL volunteers staying at altitudes above 4,300 m for duration of more than 12 months were enrolled, of whom 843 qualified the inclusion criteria and participated in the study ([Table I](#T1){ref-type="table"}). Similarly, 1000 subjects comprising male volunteers of the same age group who resided at a MSL of \<230 m in and around the metropolitan area of Delhi and had never been to high altitude were enrolled at Delhi and 862 subjects who met the inclusion criteria participated for the study. Thus, a total of 1705 healthy young male volunteers between the age group of 25-45 yr with no previous history of drug abuse, stroke, epileptic seizures or brain injury, CDR \<0.5 and with 10-17 yr of education participated in the demographic study. Following recruitment of subjects, the studies were carried out in make shift set ups in the locations of availability of the subjects that also included remote residential areas in the high altitude region. ###### Inclusion criteria of subjects in the study (n=843)  The draft MDCST was first tested on a small group of 28 individuals residing at MSL \<230 m of whom 10 individuals between the age group of 35-45 yr had MMSE scores \<25, with 8 in the age group of 60-65 yr had MMSE scores \<20 and were previously diagnosed with mild Alzheimer\'s disease using Clinical Dementia Rating (CDR 1.0), and 10 individuals in the age group of 35-45 yr who were considered normal with MMSE scores \>29. Following the evaluation of MDCST score on the test group, the score was further validated on the total population of 1705 volunteers and those having scores \<35 were considered with MCI while those \< 25 were considered moderate cognitively impaired. The correlation of outcome of MDCST with MMSE scores was determined. *Psychometric assessments*: A medical questionnaire comprising questions related to occurrence of chronic diseases, physical and physiological ailments, heart problems, stroke, epilepsy, head injury, drug abuse, psychological disorders and general health status was administered to all the volunteers. Core behavioural measures (CBM) like core alcohol consumption (section A), core tobacco use (section C), core diet (section D) and core physical activity (section P) were also applied to all subjects in accordance with WHO guidelines[@ref12]. Following short listing of the subjects who qualified the inclusion criterion, the EDAS-I[@ref13] and Beck Depression Inventory (BDI)[@ref14] were applied to assess activities of daily living and to investigate the presence of hitherto undetected depression. The information was verified from a close acquaintance of the subject. Lake Louis Score for acute mountain sickness (AMS) was administered only to the participants enrolled at Leh to negate possible occurrence of AMS symptoms. The neuropsychological assessment for MCI was performed through independent administration of MMSE, MoCA and MDCST batteries. The tests were administered by field investigators under the supervision of a clinical psychiatrist. General, physical and neurologic examination was performed by a clinician and the information was documented in a structured performa. Subjects with underlying heart disease, chest pain, stroke/infarction/cerebral haemorrhage, renal failure, diabetes, viral hepatitis, chronic disease and gastroesophageal reflux disease (GERD) were excluded. Subjects with previous neurologic/psychiatric symptoms, major surgery and familial disorders were also excluded. This ensured inclusion of only healthy subjects in this study. The MDCST was designed to evaluate the cognitive domains involved in Orientation, Memory registration, Immediate recall, Visuo-spatial executive, Object recognition, Remote memory, Mind-body coordination, Learning, Attention, Problem solving, Language fluency, Delayed recall and Procedural memory ([Table II](#T2){ref-type="table"}). The score assessed larger number of cognitive domains as compared to MoCA and MMSE ([Table III](#T3){ref-type="table"}). Accepted methods for creation of a new patient-reported outcomes instrument were followed while designing the MDCST[@ref15]. Each of the cognitive parameter was assigned equal weightage with respect to the total score and was to be assessed independently. The complete test was to be administered in a single session lasting for 5-10 min depending on individual variations. The procedure for administration of MDCST and the scoring awarded for each parameter was as follows: ###### Psychometric behavioral paradigms  ###### Domain specific paradigms in cognitive assessment scores  *Orientation*: To assess orientation on the MDCST score, the subject had to answer his/her date of birth, present address, day, last festival, and name of head/director/employer. One point was awarded for each correct answer. *Memory registration*: The subject was made to remember 5 logically unrelated words. The points awarded were equal to the number of words the subject remembered after repeating the words twice. *Visuospatial executive*: The subject was made to match the words to numerical in the order A to 1, B to 2 and so on. The subject was made to replicate the cube and draw a clock with hour and minute hand showing 0800 h. On correct matching of the alphabets with the numbers, cube drawing and correct positioning of either the hour hand or minute hand, fetched the subject one point each. Drawing a clock with numbers written fetched the subject one point. *Object recognition and Remote memory*: The subject was made to identify four unrelated objects. The subject was awarded with 1 point for each correct identification. The subject was also made to recall the name of two school teachers and was awarded one point for correct recall. *Mind-body co-ordination and learning*: The subject was made to learn a complex task like tying a tie knot in 5 steps. The number of times the subject failed to learn the task was subtracted from 5 to obtain the final score. If the number of attempts were more than 5, the final score for learning the task was 0. If the subject learnt the task in the first attempt but took more than 3 min to complete it in the second and third attempt he was awarded 3 points. *Attention and problem solving*: The subject was made to tap a finger depending on number of times 'A' appeared on different sites of the computer screen or number of times the administrator uttered the alphabet 'A' while reading a jumbled alphabet sequence. The subject was awarded one point on tapping the finger for correct number of times 'A' appeared in the jumbled alphabet sequence. The subject was made to read a sequence of numbers forward and then backward. On correct reading of the sequence in both the directions; the subject was awarded one point. For assessment of the problem solving ability, the subject was made to subtract 7 sequentially for four times, starting with 100 and one point was awarded for each correct result. *Language*: The subjects were made to read 5 sentences in English or in the local dialect as preferred by the subject. Correct reading of each sentence resulted in awarding of one point. *Delayed recall*: Delayed recall was assessed by making the subject recall the words that were presented during memory registration. Each correct recall led to award of one point. If the subject failed to recall 2 words, he was assessed for cued recall by asking to identify the correct words from a list of ten. On correct identification of all the words from the list, the subject was awarded with one point. *Procedural memory*: The subject was given a single attempt to perform the complex task that he was made to learn in five steps for assessment of procedural memory. For each correct step the subject was awarded one point. All the individual parameter scores were then summed to determine the total MDCST score. Subjects scoring \>35 were considered to be normal, those scoring \<35 were considered to have MCI. The MDCST was administered to all the volunteers independently by a field investigator and a clinical psychiatrist and the scores were compared to determine the subjective reliability of the scoring. Percentage of volunteers having MCI on the MMSE, MoCA and MDCST test batteries were independently determined. Percentage overlaps of MCI individuals between the scores and those exclusive to each test were also computed. For test-retest reliability, 50 subjects found to have MCI on the MDCST and 50 normal subjects were randomly chosen from amongst the volunteers and the questionnaire was administered three months and six months after the initial test. The subjects were re-examined with the questionnaire for medical health, BDI and CBM prior to determination of test-retest reliability for MDCST. *Electroencephalography*: Standard EEG electrode placement (10-20 system) was followed and recordings were acquired from eight electrodes using NeuroTravel-Mini EEG data recorder (ATES MEDICA Device, Italy) at a sampling rate of 500 Hz. Prior to the recordings it was ensured that the subject had a good sleep the previous night and did not take any medication for the last 24 hours. The electrodes were placed at Fp1, Fp2, T3, T4, C3, C4, O1 and O2 positions and the impedance of the electrode-skin interface was maintained below 5Ω. To monitor eye movements, horizontal and vertical electro-oculogram (EOG) was simultaneously recorded. Additional electrodes were placed on the lower jaw to gather electromyogram (EMG) signals resulting due to jaw movements. Data were acquired in eyes closed sitting position for 5 min after an initial latency of 30 sec for stabilization of the EEG waves. The data acquired from each channel were filtered using a 0.5 to 50 Hz filter and corrected for EOG and EMG artifacts, and 30 sec segments of artifact free data were used for power spectrum analysis of alpha, beta, theta and delta frequency bandwidths using AcqKnowledge 4.0.0 software (Biopac Systems, USA). The maximum amplitude of each bandwidth with respect to the source of origin was determined and spectral maps were computed. *Data analysis*: Correlation coefficient for draft MDCST score with MMSE and CDR was determined. Scoring of the field investigators and that of the clinical psychiatrist was compared by paired 't' test. The subjects were divided into four groups *viz*., Group 1: staying at MSL \<230 m and normal, Group 2: staying at MSL \<230 m and MCI, Group 3: staying at MSL \>4,300 m and normal, Group 4: staying at MSL \>4,300 m and MCI. Percentage prevalence of MCI was determined in all the populations based on the MDCST scores. One way analysis of variance (ANOVA) was performed to determine the level of significance between the groups. Newman-Keul statistics was calculated for post-hoc analysis. 'Z' test of proportion was performed to test the difference in proportion of population under the median scoring for MDCST, MMSE and MoCA. Correlation of MDCST scores with that of MMSE and MoCA was also determined. Internal consistency reliability was assessed for the cognitive parameters by Cronbach\'s formula for coefficient α on a scale of 0 to 1. Higher scores were indicative of more reliability and consistency. Test-retest reliability to measure the reliability of the scale and discriminant validity between groups was also determined[@ref9]. *P*\<0.001 was considered to be significant for all tests. Digital Fast Fourier transform (FFT) based power spectrum analysis was evaluated to calculate mean amplitude in the alpha, beta, theta and delta range. Peak values between the groups were compared independently for different regions of interest using one way ANOVA. Results {#sec1-2} ======= The study population consisted of two cohorts viz. ALL and LL comprising 1000 volunteers each with similar age, education and sex. Among the 1000 volunteers (ALL) staying at MSL \>4,300 m for \>12 months who were initially enrolled for the study, 96 were chronic alcoholics, 13 had severe brain injury in the past, 31 had history of chronic illness in recent past and 17 were unavailable for the complete test procedures. Thus, 843 healthy ALL of Indian ethnicity participated in the study, Similarly, among the 1000 volunteers (LL) staying at MSL \<230 m and who had never been to high altitude, 23 had severe head injury, 29 had history of chronic illness in recent past, 12 did not complete the test procedure, 72 were chronic alcoholics and 2 subjects had an history of epileptic seizures. The population thus comprised of 862 low lander (LL) volunteers of Indian ethnicity. The draft MDCST score in the test cohort of 28 individuals showed high correlation coefficient (0.96) with MMSE and CDR scores. The MDCST score was then applied to both ALL and LL groups. Comparison of the MDCST scores with MMSE and MoCA showed correlation coefficients of 0.62 and 0.84, respectively. The mean score for EDAS-I, BDI, MMSE, MoCA and MDCST are summarized in [Table II](#T2){ref-type="table"}. The average EDAS-I score for the ALL group was 0 ± 0.31 while that for the LL group was 0 ± 0.14. The average Lake Louis Score which is an indicator of AMS was 0 ± 0.43 for the ALL population. The average BDI score for the ALL population was 7.13 ± 0.64 while that for the LL was 5.19 ± .94. Based on the BDI scores, percentage prevalence of depression in the ALL population was calculated to be 8.69 per cent as compared to 1.04 per cent in the LL population. Similarly, the mean MMSE score for the ALL population was 27.86 ± 1.48 and percentage prevalence of MCI was 4.18 per cent while that of the LL population was 28.67 ± 1.63 and percentage prevalence of MCI was 0.42 per cent. The mean MoCA scores for the ALL and LL population was 26.49 ± 1.76 and 28.71 ± 1.04, respectively. The difference between both the populations was more evident in the MDCST in which ALL had a population mean of 36 ± 3.17 as compared to 39.43 ± 4.38 in the LL population. The percentage prevalence of MCI based on calculations from the MDCST scores was 12.4 per cent in the ALL population as compared to 1.19 per cent in the LL population. [Table III](#T3){ref-type="table"} shows a comparative evaluation of the cognitive domains assessed by MDCST and other tests. The average time of a session for administration of MDCST was determined to be 8.34 ± 0.24 min. Subjective reliability of administration of MDCST by field investigator showed 97.8 per cent accuracy with *P*\<0.001 when compared to that by the clinical psychiatrist. As shown in [Table IV](#T4){ref-type="table"}, six cognitive domains *viz*. Visuo-spatial executive, Attention, Mind-body co-ordination, Immediate recall, Delayed Recall and Procedural memory contributed more towards the MCI in the ALL population while MCI in the LL population was primarily due to impairment in Visuo-spatial executive, Attention and Delayed recall. While 28 per cent of the total MCI subjects of the ALL group showed amnestic MCI, 63 per cent showed multi-domain MCI and 9 per cent had single non-memory domain MCI. On the contrary, 42 per cent of the total MCI subjects of the LL group had amnestic MCI and 31 per cent showed multi-domain MCI, and the remaining 27 per cent had single non memory domain MCI. No statistically significant difference was observed between the scores of MDCST during the test and retest. Discriminant validity test for MDCST showed significant difference between the scores of normal and MCI individuals. All the 79 subjects found positive for MCI on the basis of MMSE scores also had positive scores on MoCA and MDCST. However, 40 subjects were found to be normal on both MMSE and MoCA but had MCI positive scores on MDCST. The average population mean for the MDCST score of normal subjects was 39.43 while that of the MCI individuals was 29.31 ([Fig. 1](#F1){ref-type="fig"}). ###### Domain specific scoring on administration of MDCST  {#F1} Power spectrum analysis of the EEG data ([Fig. 2](#F2){ref-type="fig"}) acquired from resting subjects in eyes closed position showed decrease in alpha wave amplitude at the T3 and T4 sources in MCI subjects belonging to the LL group while there was increase in amplitude values for alpha wave in these regions in the ALL groups ([Table V](#T5){ref-type="table"}). Interestingly, the subjects diagnosed with MCI in the ALL group also showed decreased amplitude values in the beta range in the T3 (3.2 ± 0.98, *P*\<0.001) and the T4 (4.32 ± 1.22, *P*\<0.001) sources when compared to the normal LL subjects as well as LL diagnosed with MCI. In contrast to the LL population, the MCI subjects in the ALL population also had increased alpha amplitudes in the C3 (27.92 ± 8.20, *P*\<0.001) and C4 (21.88 ± 5.69, *P*\<0.001) regions. {#F2} ###### Amplitudes of EEG waves  Discussion {#sec1-3} ========== The present study was an attempt to investigate the effect of prolonged stay at high altitude on the cognitive performance of ALL. The study also differs from previous investigations in terms of domain specific evaluation of cognitive performance at high altitude in a considerably large population. Besides that, the psychological assessment was coupled with electrophysiological recordings to obtain quantitative measures for high altitude induced cognitive impairment. Only healthy subjects with no clinical antecedents of depression, alcoholism or drug abuse were recruited to negate the influence of these factors on the cognitive performance. The subjects were recruited randomly in both the high altitude and low altitude locations according to their qualification in the inclusion criterion. Keeping in mind the limitations of MMSE and MoCA, MDCST was designed to establish an easy to administer and more sensitive test for detection of early stages of MCI. The test was validated in a small group by adhering to standard procedures as mentioned in previous literatures. The key requirements of psychological tests including good test retest reliability[@ref16], validity, practice effects and equivalent forms were evaluated prior to being implemented during this demographic study. MDCST exhibited a correlation coefficient of 0.96 at 3 months and 0.93 at 6 months indicating high test retest reliability. The test also had a Cronbach\'s alpha of 0.87 indicating excellent internal consistency[@ref17]. Psychological assessment of the ALL and LL populations with MMSE, MoCA and MDCST showed increased percentage prevalence of MCI in the ALL population when compared to the LL population. Domain specific assessment by MDCST showed decline in visuo-spatial executive, attention and delayed recall in the subjects showing MCI in both the LL and ALL populations. The MCI subjects in the ALL population also showed decline in immediate recall, procedural memory and mind body co-ordination which was negligible in the LL population. This is perhaps the first study on high altitude related cognitive impairment that reports the specific domains affected due to global hypobaric hypoxia. Though our previous studies[@ref1][@ref18] have shown a decrease in vigilance and an increase in reaction time after stay at high altitude for a short duration of 3 months, the other cognitive domains remained to be studied. Diagnosis of MCI was also established by analysis of EEG data acquired from resting subjects in eyes closed position. Consistent to the findings of Babiloni *et al*[@ref19] for progressive MCI, we observed a decrease in alpha wave amplitude in the T3 and T4 regions of subjects showing MCI in the LL group. MCI subjects on the other hand showed increased alpha wave amplitude at the T3, T4, C3 and C4 sources along with decrease in beta wave amplitude in the T3 and T4 regions. The study however, had a few limitations. Longitudinal study on the progression of mild MCI to moderate MCI and cognitive impairment needs to be performed in a clinical set up to extend the utility of the test for clinical practice in detection of Alzheimer\'s disease and vascular dementia. Since all the subjects in the present study had education of 10-12 years, the efficacy of the test in illiterate population also needs to be validated. Another limitation in the study was that the EEG recordings were performed with 8 channels which was due to logistic problems in the remote locations in the high altitude regions. Also, since a major part of the study was conducted in actual field conditions greater than 4300 m, visual and auditory evoked potential could not be studied due to the logistic problems like availability of adequate power source, portable equipment, *etc*. In summary, MDCST exhibited excellent psychometric properties in terms of sensitivity, and test-retest reliability qualifying it to be used as a more effective cognitive measure for assessment of MCI in demographic studies in comparison to traditional measures. The domain specific evaluation of cognitive performance using MDCST could also assist in detection of early stages of MCI and in clinical assessment of its progression as well as the effectiveness of the therapy being implemented to reverse or decrease its progression. The study also showed the increased prevalence of MCI in ALL staying for longer durations at high altitude which is neurophysiologically distinct from MCI leading to Alzheimer\'s disease. We hypothesize that the effect of global hypoxia at high altitude could be different from vascular dementia and more adversely affect the cognitive performance due to its manifestation on all the brain regions. Further investigations on the brain function and electrophysiology in both these conditions could provide information that would be helpful in devising clinical management strategies for hypobaric hypoxia induced cognitive impairment. The study was supported by Defence Research and Development Organization (DRDO), Ministry of Defence, Government of India. The authors acknowledge the contribution of Ladakh Institute of Prevention, Leh, India.

Part I -- Diagnosis and risk stratification =========================================== Introduction ------------ These guidelines aim to assist physicians, particularly cardiologists, to identify adults at high risk of coronary disease as early as possible, and to highlight its most common symptoms, especially coronary arery disease (CAD) symptoms. According to Brazilian's Unified Health System database (DATASUS), cardiovascular causes represent nearly 30% of all causes of death in Brazil^[@r01]^. - Class I: conditions for which there is conclusive evidence or general agreement that the procedure is useful/effective; - Class II: conditions for which there is conflicting evidence and/or divergence of opinion about the usefulness/efficacy of the procedure; - Class IIa: weight of evidence/opinion in favor of usefulness/efficacy. Approved by the majority of the professionals; - Class IIb: safety and usefulness/efficacy is less well established, with no predominance of opinion in favor of the procedure; - Class III: conditions for which there is evidence and/or general agreement that the procedure is not useful or effective and in some cases may be harmful; ```{=html} <!-- --> ``` - Level A: data derived from multiple consistent, large randomized clinical trials and/or robust systematic meta‑analysis of randomized clinical trials. - Level of evidence B: data derived from a less robust meta-analysis, a single randomized trial or nonrandomized (observational) studies. - Level of evidence C: data derived from consensus opinion of experts. Diagnosis --------- ### Diagnosis of subclinical coronary artery disease The risk of atherosclerotic disease may be measured by the sum of individual risks and by the synergism between the known risk factors for cardiovascular disease. Due to these complex interactions, an intuitive approach of risk attribution frequently lead to underestimation or overestimation of cases with higher or low risk, respectively. #### Diagnosis of symptomatic patients The approach proposed by Diamond and Forrester^[@r02],[@r03]^ ([Table 1](#t01){ref-type="table"}): Level of recommendation I, evidence level B was considered for diagnosis. ###### Pre-test probability of coronary artery disease in symptomatic patients by age and sex (Diamond/Forrester e CASS Data)  Age (years) Nonanginal chest pain Atypical angina Typical angina -------------- ----------------------- ----------------- ---------------- ------- ------- ------- 35 3-35 1-19 8-59 2-39 30-88 10-78 45 9-47 2-22 21-70 5-43 51-92 20-79 55 23-59 4-25 25-79 10-47 80-95 38-82 65 49-69 9-29 71-86 20-51 93-97 56-84 For the assessment of cardiovascular risk, the Brazilian Guidelines for Atherosclerosis Prevention and the V Brazilian Guidelines on Dyslipidemia and Atherosclerosis Prevention were used^[@r04],[@r05]^. (Level of recommendation IIa, evidence level B). ### Diagnosis of manifest coronary artery disease #### History, physical examination, differential diagnosis ##### Definition of angina Angina is a clinical syndrome characterized by pain or discomfort in any of the following regions: chest, epigastrium, mandible, shoulder, dorsum, or upper limbs. It is triggered or aggravated by physical activity or emotional stress and attenuated by nitroglycerin and its derivatives. ##### Clinical assessment of patients with chest pain **a) Clinical history:** Detailed clinical history. Some characteristics should be carefully investigated to determine the probability of the presence of angina: quality: constriction, tightness, heaviness, distress, suffocation, discomfort, burning, and stabbing; location: precordium, retrosternal area, shoulder, epigastrium, neck, hemithorax and dorsum; irradiation: upper limbs (right, left, or both), shoulder, mandible, neck, dorsum, and epigastrium; duration: seconds, minutes, hours, or days; triggering factors: exertion, sexual activity, position, eating habits, breathing, emotional component , and spontaneous; relieving factors: rest, sublingual nitrates, analgesic, food, antacids, position, and apnea; associated symptoms: sweating, nausea, vomiting, pallor, dyspnea, hemoptysis, cough, presyncope, and syncope. An episode of angina lasts for a few minutes. It is generally triggered by exertion of emotional stress, and relieved by rest. The use of nitroglycerin, such as sublingual nitrate, relieves angina within approximately 1 min. Pain in the chondrosternal joints is rarely of cardiac origin. The Canadian Cardiovascular Society (CCS) grading of angina pectoris^[@r06]^ is the most widely used classification of angina ([Chart 1](#t02){ref-type="table"}). ###### Canadian Cardiovascular Society grading of angina pectoris ----------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Class I Habitual physical activity, such as walking and climbing sairs, does not cause angina. Angina occurs during prolonged or strenuous physical activity. Class II Slight limitation for habitual activities. Angina during walking or climbing stairs rapidly, walking uphill, walking or climbing stairs after meals or in the cold, in the wind or under emotional stress, or within a few hours after waking up. Angina occurs after walking two blocks or climbing more than 1 flight of stairs in normal conditions. Class III Limitation of habitual activities. Angina occurs after walking one block or climbing 1 flight of stairs. Class IV Unable to carry on any habitual physical without discomfort. Angina symptoms may be present at rest. ----------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **b) Physical examination:** Physical examination is usually normal in patients with stable angina. However, during an episode of angina, it may provide important evidence about the presence of absence of CAD. When physical examination is performed during an episode of pain, third heart sound (S3), fourth heart sound (S4) or gallop, mitral regurgitation, paradoxical splitting of the second heart sound (S2), and bibasilar crackles are suggestive and predictive indicators of DAC^[@r07]^. The occurrence of atherosclerosis in other regions, including decreased pulse in lower limbs, arterial hardening, and abdominal aneurysm, increase the likelihood of CAD. ##### Differential diagnosis of chest pain: associated conditions, and provoking and relieving factors of angina In all patients, especially in those with typical angina, associated (simultaneous) diseases that can precipitate \"functional\" angina, i.e. myocardial ischemia in the absence of significant anatomic coronary obstruction, should be considered. These diseases generally cause myocardial ischemia either by increasing myocardial oxygen consumption or by decreasing the oxygen supply. An increase in oxygen consumption may be caused by hyperthermia, hyperthyroidism, and cocaine use. Obstructive sleep apnea should be seriously considered in patients with significant nocturnal symptoms. #### Noninvasive tests Additional tests in stable angina are based on the probability of CAD. After estimating the probability, it is categorized as low, intermediate, or high according to established values: 10%--90% in intermediate probability, \< 10% in low probability, and \> 90% in high probability cases. Since overall mortality of patients with stable angina varies from 1.2% to 2.4% per year^[@r08]^, a diagnostic method that leads to a higher incidence of complications and death would be inappropriate. ##### Electrocardiogram The test is indicated when a cardiac cause of chest pain is suspected (level of recommendation I, evidence level B). ##### Chest radiography Chest radiography is indicated for patients with CAD and signs or symptoms of congestive heart failure (level of recommendation I, evidence level B), and patients with signs and symptoms of pulmonary disease (level of recommendation IIa, evidence level B). ##### Exercise treadmill test The most predictive variables in the diagnosis of coronary obstruction are ST-segment depression ≥ 1 mm (measured at 0.80 seconds from the J-point), with a horizontal or descending pattern, and presence of anginal pain. ###### [Exercise treadmill test for the diagnosis of coronary obstruction]{.ul} 1. **Level of recommendation I, evidence level B** 2. 1\. Intermediate probability 3. **Level of recommendation IIa, evidence level B** 4. 1\. Suspected vasospastic angina. 5. 2\. Coronary angiography for assessment of intermediate lesions. 6. 3\. Asymptomatic individuals with more than two risk factors. 7. **Level of recommendation IIb, evidence level B** 8. 1\. A high or low pretest probability of coronary obstruction, based on age, sex and symptoms. 9. 2\. Risk assessment for noncardiac surgery (in low cardiovascular risk). **Level of recommendation III:** abnormalities: pre-excitation syndrome or Wolff-Parkinson-White syndrome, pacemaker rhythm, ST-segment depression \>1 mm at rest, and complete left bundle-branch block. ####### Echocardiography Echocardiography may help in the diagnosis^[@r09]^, by showing reversible and irreversible abnormalities in segmental motion in patients with clinical features of CAD. **a) Stress echocardiography in chronic coronary atherosclerotic disease:** the test is used in diagnosis and prognosis, to assess the impact of revascularization therapies and myocardial viability, and to support therapeutic decisions. The test has good accuracy for induced myocardial ischemia in patients with intermediate or high pretest probability, with higher diagnostic sensitivity and specificity as compared with the exercise treadmill test^[@r10]^. **b) Preoperative evaluation:** according to recommendations of the American College of Cardiology/American Heart Association (ACC/AHA) and the European Association of Cardiovascular Imaging (EACVI), dobutamine stress echocardiography has been valuable in preoperative risk stratification in patients with CAD^[@r11]^. ####### Radioisotopes Aspects of myocardial perfusion, cellular integrity, myocardial metabolism, myocardial contractility, and global or segmental ventricular function are evaluated^[@r12]^. The radioisotope thallium-201 is less frequently used because of its association with higher radiation, and is indicated for the detection of ischemia concomitant with viable myocardium. ####### Coronary angiography Coronary lesions are significant when one or more epicardial arteries are obstructed, with at least 70% stenosis and/or stenosis greater than 50% of the left main coronary artery. Assessment and measurement of obstructions are performed using coronary angiography ([Chart 2](#t03){ref-type="table"}). ###### Recommendations for coronary angiography in patients with coronary artery disease ------------------------------------------------------------------------------------------------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------- Class I Stable angina (CCS III or IV) despite clinical treatment (B) High risk in noninvasive tests, regardless of angina (B) Angina and cardiac arrest or severe ventricular arrhythmia survivors (B) Angina and symptoms/signs of congestive heart failure (C) Class IIa Patients with uncertain diagnosis after noninvasive tests, when the benefits of an accurate diagnosis outweigh the risks and costs of coronary angiography (C) Unable to undergo noninvasive tests due to physical disability, illness, or obesity (C) High-risk jobs that require an accurate diagnosis (C) Patients with uncertain prognostic information after noninvasive tests (C) Class IIb Multiple hospitalizations for chest pain, when a definitive diagnosis is considered necessary (C) Class III Significant comorbidities, when the risks of angiography outweigh the benefits of the procedure (C) Stable angina (CCS I or II) that responds to drug treatment and no evidence of ischemia in noninvasive tests (C) Preference to avoid revascularization (C) ------------------------------------------------------------------------------------------------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------- CCS: Canadian Cardiovascular Society. ####### Cardiac computed tomography There are two main modes of examinations using cardiac computed tomography that use different techniques and provide different information: the calcium score and coronary computed tomography angiography. **a) Calcium score** Quantification of coronary artery calcification using calcium score correlates with the atheroscleroctic load^[@r13]^. **Level of recommendation I, evidence level A** Asymptomatic individuals at intermediate risk using the overall risk score. **Level of recommendation IIa, evidence level B** Asymptomatic individuals at low risk using the overall risk score and family history of early CAD. 1. Asymptomatic patients at high risk of CAD or with known CAD. 2. Follow-up of coronary calcification progression. 3. Symptomatic patients. **b) Coronary computed tomography angiography** Coronary computed tomography angiography enables the noninvasive evaluation of the lumen of coronary arteries^[@r14]^. The test is clinically indicated for symptomatic patients with conflicting results between ischemia and clinical tests. **Level of recommendation IIa, evidence level A** 1. Suspected chronic CAD using: 2. a\) Previous conflicting or inconclusive ischemia tests; 3. b\) Continuous symptoms and ischemia tests with normal or inconclusive results. **Level of recommendation IIa, evidence level B** 1. To assess the patency of grafts for myocardial revascularization in symptomatic individuals with pretest probability. **Level of recommendation IIb, evidence level B** 1. Symptomatic individuals with intermediate probability of CAD and positive ischemia tests. 2. Symptomatic individuals with low probability of CAD and negative ischemia tests. 3. Assessment of in-stent restenosis in symptomatic individuals with intermediate pretest probability. **Level of recommendation III, evidence level B** 1. Symptomatic individuals with high probability of CAD. 2. Initial evaluation of CAD in asymptomatic individuals, able to exercise and with interpretable electrocardiogram. 3. Follow-up of coronary atheroscleroctic lesions in asymptomatic individuals. ####### Cardiovascular magnetic resonance imaging Magnetic resonance imaging is an excellent diagnostic method; it allows the assessment of cardiac and vascular anatomy, ventricular function, myocardial perfusion, and tissue characterization in an accurate, reproducible manner, in a single test^[@r15]^. **a) Myocardial ischemia** The protocols for the investigation of ischemia by magnetic resonance with myocardial perfusion are similar to those used in scintigraphy. **b) Delayed enhancement** The diagnosis and characterization of areas of myocardial infarction/necrosis/fibrosis using CMR is based on the delayed enhancement technique^[@r16]-[@r18]^. c\) Coronary magnetic resonance angiography The clinical use of the test has been focused on the assessment of congenital anomalies and the origin and course of the coronary arteries^[@r19]^. ###### [Recommendations for magnetic resonance imaging]{.ul} ####### Level of recommendation I, evidence level A Evaluation of the global (left and right) ventricular function, volume, and mass Detection of ischemia: - Assessment of myocardial perfusion under stress using vasodilators. - Assessment of ventricular contractility using dobutamine stress magnetic resonance. - Detection and quantification of myocardial fibrosis and infarction. - Assessment of myocardial viability. ####### Level of recommendation I, evidence level B Differentiation between ischemic and nonischemic cardiopahty - Coronary magnetic resonance angiography: - Assessment of congenital anomalies. ### Cardiovascular risk stratification in CAD The strategies and methods used in the diagnosis of CAD also provide information on disease severity, with implications for complementary invasive methods, including coronary angiography, and therapeutic decision-making. #### Exercise treadmill test for the prognosis of coronary atherosclerosis **Level of recommendation I, evidence level B** Patients with intermediate or high probability of CAD after initial evaluation; patients showing changes in symptoms. **Level of recommendation IIb, evidence level B** Patients with pre-excitation, ST-segment depression \> 1 mm in echocardiogram at rest, pacemaker rhythm, and complete left bundle-branch block. **Level of recommendation IIa, evidence level C** Revascularized patients with symptoms suggestive of ischemia. **Level of recommendation III, evidence level C** Patients with severe comorbidities. In patients with CAD who are able to reach stage 3 of the Bruce protocol, the annual mortality rate is approximately 1%, whereas in those unable to exceed 5 METs, the annual mortality rate is approximately 5%^[@r20]^. Other high-risk variables include ST-segment depression in multiple leads, persistent ST-segment depression in recovery phase \> 5 min, inadequate chronotropic response, fall in systolic blood pressure during physical exertion or a flat curve, and severe ventricular arrhythmia at low level of exercise in the presence of ST-segment depression or anginal pain. ##### [Stress echocardiography]{.ul} Echocardiography for CAD prognosis takes into account mainly the left ventricle function, and the presence or absence of myocardial ischemia induced by physical or pharmacological stress on echocardiography. In asymptomatic patients who have successfully undergone coronary artery bypass graft surgery (CABG), routine evaluation using stress echocardiography is not indicated. Other important variables for risk stratification include pulmonary uptake of thallium in myocardial perfusion scintigraphy, and the transient increase in the left ventricle. ### Strategies for the diagnosis and stratification of coronary artery disease The prognosis of CAD may also be based on the direct anatomical visualization of the coronary lesion by coronary angiography. Normal functional testing, performed with appropriate stress protocol yields the same prognosis as compared with the standard coronary angiography test. Part II -- Drug Treatment ========================= The main objectives of the treatment of CAD are to prevent myocardial infarction and decrease mortality, and to reduce symptoms and the incidence of myocardial ischemia, providing a better quality of life. Drug treatments to reduce the risk of myocardial infarction and mortality ------------------------------------------------------------------------- ### Antiplatelet drugs **a) Acetylsalicylic acid (ASA):** Level of recommendation I, evidence level A. b\) Thienopyridine derivatives: **Clopidogrel:** Level of recommendation I, evidence level B. Indicated when aspirin is absolutely contraindicated, and associated with aspirin after stent implant for at least 30 days. **Ticlopidine:** Level of recommendation IIa, evidence level B. Indicated when aspirin is absolutely contraindicated, and associated with aspirin after stent implant for at least 30 days. **c) Dipyridamole:** Level of recommendation III, evidence level B. **d) Anticoagulants:** should be used in combination with aspirin in case of high risk of thrombosis, especially after myocardial infarction. Level of recommendation I, evidence level A. As an alternative to aspirin intolerance: Level of recommendation IIa, evidence level A. For specific situations and after implantation of antiproliferative drugs-coated stent, follow the Brazilian Guidelines of Antiplatelet Agents and Anticoagulants in Cardiology. ### Secondary prevention: Hypolipidemic agent Lifestyle change (LC) is recommended for all patients with CAD ([Chart 3](#t04){ref-type="table"}). ###### Recommendations for drug therapy in dyslipidemias Indications Class-level of evidence ------------------------------------------------------------------------------------------------------------------ ------------------------- Statins are first choice treatment in primary and secondary prevention I-A Fibrate monotherapy or in combination with statins to prevent microvascular diseases in type 2 diabetes patients I-A Associations of ezetimibe or resins with statins when LDL-C target levels are not achieved IIa-C Association of niacin with statins III-A Omega-3 fatty acids for cardiovascular prevention IIII-A Source: Brazilian guidelines for cardiovascular disease prevention^[@r10]^. ### Blockade of the renin--angiotensin system **a) ACE inhibitors:** the benefits of ACE inhibitors in the treatment of CAD have been shown in clinical trials involving asymptomatic patients with reduced ejection fraction^[@r21]^ and patients with ventricular dysfunction after acute myocardial infarction^[@r21],[@r22]^. They should be used routinely for ventricular dysfunction, and/or heart failure, and/or diabetes mellitus management^[@r23],[@r24]^. Level of recommendation I, evidence level A. It should be used routinely in all patients with CAD: Level of recommendation IIa, evidence level A. **b) Angiotensin receptor blockers:** alternative therapy for patients intolerant to ACE inhibitors, since no study has been conducted on the use of this group of drugs in stable coronary disease. In other situations, angiotensin receptor blockers have provided no additional benefits over those of ACE inhibitors, which can decrease the incidence of infarction. Treatment to reduce symptoms and myocardial ischemia ---------------------------------------------------- **a) Beta-blockers:** beta-blockers are drugs of choice, to be administered alone or in combination with other antianginal drugs. Indicated as first-line agents in patients with stable angina without previous myocardial infarction and/or left ventricle dysfunction^[@r25]^. Level of recommendation I, evidence level B. 1. \- First-line agents in patients with stable angina within 2 years of myocardial infarction and/or left ventricle. Level of recommendation III, evidence level C. 2. \- For symptomatic relief in patients with vasospastic angina: Level of recommendation III, evidence level C. **b) Calcium-channel blockers:** heterogeneous group of drugs with pharmacological effects that include smooth muscle relaxation, afterload reduction, and negative inotropic effects (some formulations). On the other hand, they are contraindicated in case of ventricular dysfunction (verapamil and diltiazem)^[@r26]^. 1. -- First-line agents for symptomatic relief in patients with vasospastic angina. Level of recommendation IIa, evidence level B. 2. -- In symptomatic patients with stable angina on beta-blockers (dihydropyridines). Level of recommendation I, evidence level B. 3. -- In symptomatic patients with stable angina on beta-blockers (verapamil or diltiazem). Level of recommendation III, evidence level B. 4. -- In patients with stable angina and contraindications to beta-blockers (preferably verapamil or diltiazem). Level of recommendation I, evidence level B. 5. -- In symptomatic patients with stable angina (fast-acting ihydropyridines). Level of recommendation III, evidence level B. **c) Nitrates:** 1. -- **Fast-acting nitrates:** for symptomatic relief of acute angina. Level of recommendation I, evidence level B. 2. -- **Long-acting nitrates:** continuous use of long-acting nitrates leads to drug tolerance. 3. -- First-line agents in patients with stable angina. Level of recommendation III, evidence level C. 4. -- Third-line agents in stable angina patients who still have symptoms even after using other antianginal agents associated. Level of recommendation IIa, evidence level B. 5. -- For symptomatic relief in patients with vasospastic angina after using calcium-channel blockers. Level of recommendation IIa, evidence level B. **d) Trimetazidine:** drug with metabolic and anti-ischemic effects and no effect on cardiovascular hemodynamics^[@r27]^. 1. -- In symptomatic patients with stable angina on beta-blockers alone or in combination with other antianginal agents. Level of recommendation IIa, evidence level B. 2. -- In patients with stable angina and left ventricle dysfunction associated with optimized medical therapy. Level of recommendation IIa, evidence level B. 3. -- In patients with stable angina during myocardial revascularization procedures (percutaneous or surgical). Level of recommendation IIa, evidence level B. **e) Ivabradine:** a specific sinus node I~*f*~ current i inhibitor, which specifically decreases the heart rate^[@r28]^. 1. -- In symptomatic patients with stable angina on beta-blockers alone or with other antianginal agents, and heart rate \> 70 bpm. Level of recommendation IIa, evidence level B. 2. -- In symptomatic patients with stable angina who are intolerant to beta-blockers alone or with other antianginal agents. Level of recommendation IIb, evidence level B. 3. -- In patients with stable angina, left ventricle dysfunction (LVEF \< 40%) and heart rate ≥ 70 bpm under optimized medical therapy. Level of recommendation IIa, evidence level B. **f) Ranolazine:** piperazine derivative. Similar to trimetazidine, it protects patients from ischemia by increasing glucose metabolism and decreasing fatty acids oxidation. However, its major effect appears to be the inhibition of late sodium current^[@r29]^. [Figures 1](#f01){ref-type="fig"} and [2](#f02){ref-type="fig"} depict algorithms that facilitate understanding of drug therapy options in stable CAD. {#f01} {#f02} Part III -- Treatment with invasive measures ============================================ Treatment with invasive measures -------------------------------- ### Direct surgical revascularization The Guidelines on Myocardial Revascularization^[@r30]^ cover the procedure techniques, alternatives, and current practices. They also briefly review classic studies, comparing surgical treatment strategies with clinical treatment and percutaneous coronary intervention. #### Main indications for direct revascularization **Level of recommendation I** Left main coronary artery stenosis ≥ 50% or equivalent conditions (left descending anterior and circumflex arteries in the ostium, or before the exit of important branches). Evidence level A. Proximal stenosis (\> 70%) in the three main arteries with or without involvement of proximal left anterior descending artery, especially in patients with ejection fraction \< 50% or functional evidence of moderate to severe ischemia. Evidence level B. Stenosis in two main vessels, with proximal left anterior descending artery lesion in patients with ejection fraction \< 50% or functional evidence of moderate to severe ischemia. Evidence level B. **Level of recommendation IIa** Left internal mammary artery graft in patients with significant stenosis (\> 70%) in proximal left anterior descending artery and evidence of extensive ischemia, aiming to improve survival. Evidence level B. Coronary artery by-pass surgery instead of percutaneous coronary intervention in patients with multivessel CAD and diabetes mellitus, particularly in those who underwent internal mammary artery grafting with revascularization to the left anterior descending artery. Evidence level B. **Level of recommendation III** Asymptomatic patients with normal ventricular function, without extensive areas of ischemia or involvement of the left anterior descending artery. Evidence level C. Asymptomatic patients without significant anatomical lesions (\< 70%, or \< 50% of the left main coronary artery) or functional lesions (e.g., fractional flow reserve \> 0.8 or mild ischemia in noninvasive tests). Evidence level C. Involvement of one or two arteries, except for the proximal left anterior descending artery, with no evidence of relevant ischemia in functional tests, and presence of perfusion in small areas of viable myocardium. Evidence level B. Moderate lesions (between 50% and 60%) except in left main coronary artery, without moderate ischemia in functional tests. Insignificant lesions (\< 50%). #### The \"Heart Team\" concept for myocardial revascularization ##### [Class I]{.ul} A team made up of clinical cardiologists, cardiac surgeons and interventional cardiologists is recommended to individualize the indication for the treatment of left main coronary artery lesions or complex CAD. Evidence level C^[@r31]^. ### Catheter-based revascularization: clinical indications #### Revascularization vs. drug treatment ([Figure 3](#f03){ref-type="fig"}) {#f03} ##### Percutaneous coronary intervention vs. clinical treatment To date, no study has demonstrated that percutaneous coronary intervention in patients with CAD improves survival rates^[@r32]^. ##### Appropriate use of revascularization ###### Patients with three-vessel disease The coronary artery bypass surgery is the preferred strategy for three-vessel disease patients with increased age, low ejection fraction, renal dysfunction, peripheral vascular disease, diabetes mellitus, or Syntax score \> 22. Special situations ------------------ ### Patients with diabetes mellitus Diabetes mellitus is an increasingly prevalent condition associated with increased risk of cardiovascular complications, especially late mortality. #### Indications for myocardial revascularization ##### Comparison of revascularization strategies in diabetic patients with multi-vessel CAD Sensitivity analysis showed that the superiority of coronary artery bypass surgery was more evident in individuals with high Syntax score (\> 33), with no significant difference between the low score and intermediate score groups^[@r33]^. ##### Aspects of percutaneous coronary intervention in diabetes mellitus patients Drug-eluting stents are recommended to reduce restenosis and the need of a new target vessel revascularization^[@r34],[@r35]^. The dual antiplatelet therapy with aspirin and a P2Y12 receptor blocker is an essential component of drug regiments for perioperative and postoperative periods. Patients who receive drug-eluting stents should use the therapy for 12 months, and those who receive non-drug-eluting stents should use it for 1 month. #### Patients with previous revascularization The main indications for revascularization are persistence of symptoms, despite optimized medical therapy and/or prognosis. **Author contributions** Writing of the manuscript and Critical revision of the manuscript for intellectual content: César LAM, Mansur AP, Ferreira JFM. **Potential Conflict of Interest** Drs. Luiz Antonio Machado César and João Fernando Monteiro Ferreira participated in clinical studies and / or experimental trials supported by Servier e Astra-Zeneca. Dr. Luiz Antonio Machado César has spoken at events or activities sponsored by Servier e Astra-Zeneca; It was (is) advisory board member or director of a Servier e Astra-Zeneca; Committees participated in completion of research sponsored by Servier e Astra-Zeneca; Personal or institutional aid received from Servier e Astra-Zeneca; Produced scientific papers in journals sponsored by Servier e Astra-Zeneca. **Sources of Funding** There were no external funding sources for this study. **Study Association** This study is not associated with any thesis or dissertation work.

Coronary artery bypass graft (CABG) surgery to restore normal blood flow to the heart is one of the most common surgical procedures worldwide. However, many patients suffer from a frequent complication: post-CABG pain (PCP) syndrome. PCP is characterized by chronic, atypical pain that restricts physical activity in daily living and causes depression \[[@B1]\]. Accurate, early diagnosis and active treatment of PCP are therefore very important. The onset of PCP is related to various factors, including brachial plexus traction, scar pain, costochondral junction pain, and upper limb complex regional pain syndrome. This pain usually occurs in the left anterior chest wall, although PCP can also be located at the mid-line scar and the right chest wall \[[@B2]\]. However, we report a case of sympathetically mediated pain occurring in the left upper back area accompanied by autonomic dysfunction in the left arm after the CABG. A 55 year-old male patient was diagnosed with coronary artery disease (CAD) when he visited the hospital with a myocardial infarction in March 2010, as he showed 60% stenosis in the left anterior descending artery, 90% stenosis in the left circumflex artery, and 90% stenosis in the right coronary artery from coronary angiography. However, the diameter of the lesion was ≤ 1.5 mm, making it difficult to perform percutaneous coronary intervention (PCI); thus, medical therapy was administered. In March 2012, while still on therapy, he was admitted to the hospital via the emergency room with the main complaints of sudden chest pain and radiating facial pain. One week later, he underwent CABG surgery using the left internal mammary artery and right saphenous vein. Ten days after the surgery, chest pain and facial radiating facial pain disappeared, but the patient was referred to our pain clinic still complaining of left upper back pain on the thoracic 1-4th dermatome as well as numbness with a cold sensation in the left arm. Intercostal nerve block of the pain sites on the left thoracic 2-3th and 3-4th level was performed with 0.375% ropivacaine. Despite the block, the patient continued to complain of extreme pain at the same site as well as numbness in the left arm and tingling sensation. At the pain clinic, the pain intensity in the left upper back area was assessed as having a visual analogue scale (VAS) of 9-10, while the left arm and hand showed slight weakness and pale skin on the ulnar side. The patient and his guardian were given information about PCP and its likelihood after CABG surgery. Left stellate ganglion block (SGB) was carried out seven times every other day using 0.375% ropivacaine 6 ml. For pain management, NSAIDs, tri-cyclic antidepressants (TCAs), and anticonvulsants were prescribed, and a single epidural block was conducted two times on the left T2 and T3 using a mixture of 0.09% ropivacaine 6 ml and triamcinolone 10 mg. After the SGB and epidural block, the upper back pain improved to a VAS of 2-3, but the patient still complained of left hand numbness and weakness. After the patient\'s discharge, the pain clinic performed an epidural block twice on the left C7 and T1, as well as left SGB seven times over two weeks. While receiving treatments as an out-patient at the pain clinic, the left upper back pain resolved, and the symptoms of autonomic dysfunction such as numbness and skin color change in the left arm noticeably improved without sequelae. The patient remained symptom-free and returned to work after a month. Many studies have investigated the prevalence of PCP in connection to its onset mechanism. Eng and Wells \[[@B1]\] reported a PCP prevalence of 23%, while another report put the figure at 56% \[[@B3]\]. The mechanism of PCP is believed to be chest wall trauma-induced musculoskeletal nociceptive pain and neuropathic pain from secondary formation. During harvesting of the internal mammary artery, the anterior branch of the intercostal nerve is most likely to be injured, triggering stepwise central neuropathic pain \[[@B2]\]. In many cases, the characteristic profile of PCP accompanies sensation abnormalities with hypoesthesia, allodynia and hyperalgesia, which are mainly expressed from the thoracic 2-4th dermatome. In addition, autonomic dysfunction can accompany symptoms such as swelling, skin color change, and temperature change \[[@B4]\]. In our case, tingling and cold sensations were expressed in the left arm, the same side, in which sympathetic activity increased excessively in the same segmental site during the course of central sensitization of the peripheral nerve damage, causing neuropathic pain. In addition, there was simultaneous, sympathetically mediated radiating pain in the upper back area. It is important to differentiate PCP from anginal pain because, in many cases, patients themselves are already aware of them, and detailed pain profiling and history taking are important in making an accurate diagnosis. Other causes of chronic chest pain after CABG surgery are sternal wound infection, brachial plexus injury, and shoulder girdle pain. Despite the small number of studies on the treatment of PCP, there are some case reports of successful SGB \[[@B4]\]. Our case, involved a successful treatment experience with several SGBs and thoracic epidural block. It is thought that SGB and thoracic epidural block reduce the secretion of catecholamine and various pain substances \[[@B5]\]. In conclusion, peripheral nerve damage during harvesting of a blood vessel for CABG may cause autonomic dysfunction at the same segmental site, and sympathetically mediated radiating pain may occur in the left upper back area. Because PCP has high prevalence and can last from several months to several years, it is important to preemptively reduce the development of PCP at the initial stage.

{#sp1 .491} {#sp2 .492} {#sp3 .493} {#sp4 .494} {#sp5 .495} {#sp6 .496} {#sp7 .497} {#sp8 .498} {#sp9 .499} {#sp10 .500} {#f1 .492} {#f2 .492} {#f3 .496} {#f4 .496} {#f5 .497} {#f6 .497} {#f7 .497} {#f8 .498} {#f9 .498} {#f10 .498} {#f11 .498} {#f12 .499} {#f13 .499} {#f14 .499}

Background ========== The induction of broadly neutralizing antibodies (bNAbs) against the envelope protein is thought to be an essential part of an effective HIV vaccine. A part of gp41 which is conserved through all HIV-1 clades, the membrane-proximal external region (MPER), bears epitopes for two of the most potent known neutralizing monoclonal antibodies. As HIV-1 virus-like particles (VLPs) provide a natural membrane setting for gp41-derived proteins, we take advantage of VLPs as carriers for membrane-anchored, truncated gp41 as immunogen. Methods ======= Several versions of gp41 including the C-terminal heptad repeat (CHR), MPER, transmembrane domain and a shortened cytoplasmic tail have been designed and characterised. To stabilize the trimeric state of gp41, heterologous zipper domains have been introduced at appropriate positions. Additionally, constructs with a further truncated CHR have been designed in order to focus the immune response to the highly conserved MPER. The most truncated version only consists of the transmembrane domain and the MPER. Results ======= All designed gp41 derivatives were expressed in a human cell line and incorporated into VLPs. Most of them show enhanced binding to bNAbs 2F5 and 4E10 compared to full-length gp160 and similar binding compared to full-length gp41, as measured in FACS analysis and VLP capture assay. Also the most truncated version of gp41 still seems to be properly transported to the cellular membrane and recognized by 2F5 and 4E10. All described immunogens are currently tested in the rabbit model for immunogenicity and elicitation of neutralizing antibodies. The animals are primed with DNA and boosted two times with VLPs presenting gp41 fragments. Conclusion ========== Thus, we created several fragments of gp41 with preserved binding affinity for broadly neutralizing antibodies, to serve as immungens on the surface of VLPs for the elicitation of antibodies directed against the membrane proximal region of gp41.

1. Introduction {#sec1} =============== Recently, optimization becomes one of the most interesting issues in different life aspects, such as engineering designs, browsing the Internet, and business management. Time reduction, high quality, and financial profit can be challenging for the most real-world applications. Therefore, most optimization methods try to find a perfect method in order to deal with limited resources problem within various restrictions. Many effective search algorithms, which are using mathematical formulae and computational simulations, have been implemented to solve optimization problems. Metaheuristic algorithms try to balance between randomization and local search. So, most of these algorithms are used for global optimization \[[@B1], [@B2]\]. Metaheuristic algorithms have two basic elements, which are exploitation and exploration; in exploration, different solutions are found to explore the search space to find the global optimal, but in exploitation, local search is used by exploiting information about the best solutions that have been recently found. This combination with choosing the best solutions will guarantee that solutions reach the optimality, also exploration bypasses the local optima problem through randomization and raises the diversity of the solutions \[[@B1], [@B3]\]. Swarm-based nature metaheuristic algorithms are used to solve optimization problems by imitating the biological behavior of certain animals. Mirjalili and Lewis proposed the whale optimization algorithm to simulate the hunting behavior of humpback whales, and this is done by two main attacking mechanisms; first by chasing the prey with random or the best search agent and second by simulating the bubble net hunting strategy. Humpback whales like to hunt a group of small fish close to the surface. So, they swim around the target inside and alongside a thin circle to make a winding-shaped way, creating distinct blebs along a circle or '9\' shaped ways altogether. Humpback whales have a very remarkable hunting method; this hunting behavior is called the bubble net feeding method. It has been observed that foraging is done by creating unique bubbles along a circle or '9\' shaped path as shown in [Figure 1](#fig1){ref-type="fig"} \[[@B5]\]. The aim of this research consists of several aspects: first of all, highlighting all studies and researches conducted on WOA, where metaheuristic hybridization models have been used to combine WOA with other techniques to enhance the performance of the resulting algorithm. Second, this work has focused on all modification methods, which have been applied on WOA to improve its ability to search for the best solution. Third, we have collected most of the research works related to various applications applied on WOA. Finally, a new hybridizing of WOA and BAT algorithms is presented. The proposed algorithm is used to overcome the problems of staying in the local optimum and increase the speed of convergence to the best solution. Consequently, this research work in return will pave the way for researchers to make other modifications on the WAO algorithm to suit their different purposes. The rest of the paper outline starts with describing WOA, its characteristics, and limitations followed by providing various WOA modifications and hybridizations, which have been applied to different problems. Next, various applications of WOA are presented. After that, results from different benchmark functions and experiments are analyzed and compared to WOA modifications and other metaheuristic optimization algorithms. Then, the BAT algorithm is presented, and the WOA-BAT is proposed. The results of the WOA-BAT are evaluated against the original WOA. WOA-BAT is happened to be very competitive and better than WOA in 16 out of 23 benchmark test functions, 13 out of 25 CEC2005 test functions, and 7 out of 10 CEC2019 test functions. Finally, the conclusion is presented with future works on WOA and WOA-BAT. 1.1. Whale Optimization Algorithm {#sec1.1} --------------------------------- This algorithm consists of two main phases; in the first phase, encircling prey and spiral updating position are implemented (exploitation phase). However, searching for a prey is done randomly in the second phase (exploration phase) \[[@B5]\]. The mathematical model of each phase is illustrated in the following subsections. ### 1.1.1. Bubble Net Attacking Method {#sec1.1.1} Two approaches are designed in order to mathematically model the bubble net behavior of humpback whales that is called the exploitation phase. The two approaches are described as follows: *(1) Encircling Prey*. After the humpback whales discover the position of the prey, they encircle around them. Firstly, the location of the optimal design in the search space is unidentified; thus, the WOA algorithm assumes that the present leading candidate solution is the target prey or near to the optimum. Then the other search agents will attempt to change their locations to the best search agents. This behavior is represented by the following equations:$$\begin{matrix} {\overset{\rightarrow}{X}\left( {t + 1} \right) = \overset{\rightarrow}{X^{\ast}}\left( t \right) - \overset{\rightarrow}{A} \cdot \overset{\rightarrow}{D},} \\ \end{matrix}$$$$\begin{matrix} {\overset{\rightarrow}{D} = \left| {\overset{\rightarrow}{C} \cdot \overset{\rightarrow}{X^{\ast}}\left( t \right) - \overset{\rightarrow}{X}\left( t \right)} \right|,} \\ \end{matrix}$$where $\overset{\rightarrow}{X^{\ast}}\left( t \right)$ indicates the whale\'s earlier best location at iteration *t*. $\overset{\rightarrow}{X}\left( {t + 1} \right)$ is the whale\'s current position, $\overset{\rightarrow}{D}$ is the distance vector between whale and prey, and \| \| denotes absolute value. The *C* and *A* are coefficient vectors calculated as follows:$$\begin{matrix} {\overset{\rightarrow}{A} = 2 \cdot \overset{\rightarrow}{a} \cdot \overset{\rightarrow}{r} + \overset{\rightarrow}{a},} \\ \end{matrix}$$$$\begin{matrix} {\overset{\rightarrow}{C} = 2 \cdot \overset{\rightarrow}{r}.} \\ \end{matrix}$$ To apply shrinking, the value of $\overset{\rightarrow}{a}$ is reduced in Equation ([3](#EEq3){ref-type="disp-formula"}); thus, the oscillating range of $\overset{\rightarrow}{A}$ is also reduced by $\overset{\rightarrow}{a}$. The value of $\overset{\rightarrow}{A}$ could be in the interval (−*a*, *a*), where the value of *a* is decreased from 2 to 0 through iterations. By selecting random values for $\overset{\rightarrow}{A}$ in (−1, 1), the new position of any search agent can be determined anywhere in the range between the original position of the agent and the position of the current best agent. *(2) Spiral Updating Position*. After calculating the distance between the whale located at (*X*, *Y*) and prey located at (*X*^*∗*^, *Y*^*∗*^). At that point, a spiral equation is generated between the location of the whale and prey to imitate the helix-shaped movement of humpback whales as follows:$$\begin{matrix} {\overset{\rightarrow}{X}\left( {t + 1} \right) = e^{bk}\, \cdot \,\cos\left( {2\pi k} \right) \cdot \overset{\rightarrow}{D^{\ast}} + \overset{\rightarrow}{X^{\ast}}\left( t \right),} \\ \end{matrix}$$$$\begin{matrix} {\overset{\rightarrow}{D^{\ast}} = \left| {\overset{\rightarrow}{X^{\ast}}\left( t \right) - \overset{\rightarrow}{X}\left( t \right)} \right|,} \\ \end{matrix}$$where *b* is a constant value for identifying the shape of logarithmic spiral and *k* is a random number in the range \[−1, 1\]. This behavior is represented in WOA to change the position of whales during optimization. There is a 50% chance for selecting between the shrinking encircling mechanism and the spiral model, and their components are designed as follows:$$\begin{matrix} {\overset{\rightarrow}{X}\left( {t + 1} \right) = \begin{cases} {\overset{\rightarrow}{X^{\ast}} - \overset{\rightarrow}{A} \cdot \overset{\rightarrow}{D},} & {\text{if }p < 0.5,} \\ {e^{bk}\, \cdot \,\cos\left( {2\pi k} \right) \cdot \overset{\rightarrow}{D^{\ast}} + \overset{\rightarrow}{X^{\ast}}\left( t \right),} & {\text{if }p \geq 0.5,} \\ \end{cases}} \\ \end{matrix}$$where *p* is a random number in (0, 1). ### 1.1.2. Search for Prey {#sec1.1.2} In the search phase for the prey, which is called the exploration phase, a similar method depending on the variance of the $\overset{\rightarrow}{A}$ vector can be used. The whales actually use random search to discover their prey depending on the position of each other. Therefore, to oblige the search agents to move far away from the local whale, WOA uses the $\overset{\rightarrow}{A}$ vector with random values greater or less than 1. Throughout the exploration phase, the location of a search agent is reorganized according to randomly selected search agent rather than the best search agent (exploitation phase). This procedure aids the WOA algorithm to perform the global search and overcome the local optimal problem. The mathematical model is expressed as follows:$$\begin{matrix} {\overset{\rightarrow}{X}\left( {t + 1} \right) = \overset{\rightarrow}{X_{\text{rand}}}\, - \,\overset{\rightarrow}{A} \cdot \overset{\rightarrow}{D},} \\ \end{matrix}$$$$\begin{matrix} {\overset{\rightarrow}{D} = \left| {\overset{\rightarrow}{C} \cdot \overset{\rightarrow}{X_{\text{rand}}}\, - \,\overset{\rightarrow}{X}} \right|,} \\ \end{matrix}$$where $\overset{\rightarrow}{X_{\text{rand}}}$ is the random position vector (a random whale) chosen from the current population. 1.2. Operation of Whale Optimization Algorithm {#sec1.2} ---------------------------------------------- The WOA algorithm starts by assigning whales population with random solutions and assuming the best optimal value of the objective function is a minimum or maximum value (depending on the problem), then the objective function for each search agent is calculated. At each iteration, each search agent updates their location depending on either the best solution found so far when $\left| \overset{\rightarrow}{A} \right| < 1$ or on a randomly chosen search agent when $\left| \overset{\rightarrow}{A} \right| > 1$. In order to achieve exploration and exploitation phases, respectively, the value of *a* parameter is decreased from 2 to 0. Also, WOA has the feature to select either a spiral or circular movement through the value of another parameter, which is *p* (a random number in \[0, 1\]) with a probability of 50% to select one of these two mechanisms, so if its value is greater than 0.5, then the search agents change their positions using Equation ([5](#EEq5){ref-type="disp-formula"}), otherwise they use Equation ([1](#EEq1){ref-type="disp-formula"}). Finally, the WOA algorithm ends by implementing of the termination condition \[[@B5]\] ([Algorithm 1](#alg1){ref-type="fig"}). 1.3. Whale Optimization Algorithm Pseudocode {#sec1.3} -------------------------------------------- 1.4. Characteristics of WOA {#sec1.4} --------------------------- The process of obtaining a suitable equivalence between exploitation and exploration in the improvement of any metaheuristic algorithm is a topmost challenge due to the arbitrary nature of the optimization algorithm. WOA has the highest significance compared to the different optimization approaches through the following:Exploitation abilityExploration abilityAbility to get rid of the local minima The WOA has an important capability of exploration due to the position updating mechanism of whales by using Equation ([7](#EEq7){ref-type="disp-formula"}). Throughout the initial step of the algorithm, this equation forces the whales to move randomly around each other. In the next steps, Equation ([8](#EEq8){ref-type="disp-formula"}) makes the whales update their positions rapidly and move along a spiral-shaped route in the direction of the best path that has been found so far. Since these two stages are done independently and in half iteration each, the WOA avoids local optima and achieves convergence speed at the same time through the iterations. But most of the other optimization algorithms (like PSO and GSA) do not have operators to consecrate a particular iteration to the exploration or the exploitation because they use only one format to update the position of search agents, so the probability of falling into local optima is more likely increased \[[@B2]\]. [Figure 2](#fig2){ref-type="fig"} shows the number of publication that has been published about WOA since 2016. 1.5. Limitation of WOA {#sec1.5} ---------------------- Metaheuristic algorithms have both efficiency and limitation for convergence speed and obtaining optimal solution. Thus, the limitation of WOA should also be found out depending on \[[@B6]\]. Randomization has a crucial role in exploration and exploitation, so using the current randomization technique in WOA would increase computational time especially for the highly complex problem \[[@B4]\]. Besides, convergence and speed depend on one control parameter, which is *a*. This parameter has an excessive impact on the performance of WOA \[[@B7]\]. For that reason, WOA has poor convergence speed in both exploration and exploitation phases \[[@B8], [@B9]\]. Thus, a balancing formulation between exploration and exploitation requires proper enhancement \[[@B10]\]. In addition, WOA uses the encircling mechanism in the search space, and this mechanism has less capability to jump out from local optima. Accordingly, it results in poor performance \[[@B11]\]. It also has a drawback when improving the best solution after each iteration \[[@B12]\]. It is worth mentioning that WOA cannot work in the fields of classification and dimensionality reduction as it is not suitable for the binary space \[[@B13]\]. Likewise, the original WOA cannot deal with complex environmental constraint as for the vehicle fuel consumption problem \[[@B14]\]. It cannot solve single and multidimensional 0--1 knapsack problems with different scales as it requires additional functions \[[@B15]\]. 2. WOA Modifications and Hybridizations {#sec2} ======================================= In this review, we focus on reporting the developments of WOA, which have been published recently; this section is separated into three parts:Modifications of WOA: including AWOA, IWOA, chaotic WOA, ILWOA, and MWOA.Hybridizations of WOA: with metaheuristic algorithms, such as SA, PSO, local search, EWGC, and BS-WOA.Problems solved by WOA 2.1. Modifications of WOA {#sec2.1} ------------------------- There are different types of WOA, which have been modified. The following subsections are the summary of the modifications of WOA. ### 2.1.1. AWOA and SAWOA {#sec2.1.1} Randomizations have an essential influence in exploration and exploitation in optimization algorithms. Therefore, there are different techniques, which have been used in randomization, for example, Markov chain, Lévy flight, and normal distribution or Gaussian. Despite using these techniques, the adaptive technique has been used in WOA, which is called adaptive WOA (AWOA). This technique has also been used in the cuckoo search algorithm. This technique is crucial due to decreasing computational times for highly complicated problems \[[@B4], [@B16]\]. Having fewer parameters dependency is the best feature of this technique, which is useful, and does not need to initialize parameters and step sizes. Therefore, these parameters change regarding its fitness values during iterations. As a result, AWOA reached an optimum solution in less computational time and local optimum was avoided with the fast convergence \[[@B16], [@B17]\]. Trivedi et al. \[[@B4]\] proved that AWOA was better than WOA in terms of computational times and convergence speed. Another AWOA was proposed in \[[@B18]\] for cluster head selection based on the Internet of things (IoT). IoT is another vital area, which can be researched on due to improvement in its performance \[[@B19], [@B20]\]. Regardless of using parameters, such as distance energy and delay of sensor nodes in a wireless sensor network, self-adaptiveness WOA (SAWOA) considers temperature and load parameters of IoT devices. Results proved that SAWOA performance was better than other algorithms like GSA, GA, ABC, PSO, and WOA \[[@B21]--[@B23]\]. ### 2.1.2. IWOA {#sec2.1.2} Distance control parameter *a* has value, which affects the ability of exploration and exploitation. However, this parameter is started from 2 and then decreased to 0 during iterations \[[@B7]\]. This parameter resulted in fast convergence and obtained accurate results for most problems. Despite these effects, it is linear and cannot adapt to the search process of WOA, which is nonlinear and complex \[[@B24]\]. Therefore, in an improved WOA (IWOA), some strategies defined for distance control parameter in order to adapt the nonlinear search process to achieve better results. There are five kinds of IWOA regarding its distance control variable, and those are SinWOA, CosWOA, TanWOA, LogWOA, and SquareWOA. Because of having a poor balance between exploration and exploitation, researchers in \[[@B8]\] proposed a novel constitutional appraising approach based on WOA. Thus, results from clinical data analysis showed that IWOA had better efficiency in terms of convergence performance compared to the original WOA. IWOA proposed in \[[@B25]\] used a new control parameter, which was inertia weight. This parameter was used to adjust the impact on the current best solution. To evaluate the performance of IWOA, 31 benchmark functions were used to test it in \[[@B5]\]. Then, IWOA was compared with WOA, FOA, and ABC algorithms. The size of the population was 1000 and with 30 iterations. IWOA outperformed compared to ABC, FOA, and WOA in terms of mean and standard deviations. According to \[[@B26]\], the mean values of ABC and FOA were greater than IWOA and WOA for functions *f*1, *f*2, *f*3, *f*7, *f*10, *f*11, *f*12, *f*13, *f*16, and *f*27. The mean value of IWOA was the least compared to FOA and ABC for functions *f*4*′*, *f*8*′*, *f*14, *f*15, *f*17, *f*18, *f*19, *f*20, *f*21, and *f*26. However, the mean value for functions *f*5 and *f*9 was equal for all algorithms. The mean value of FOA was more than other algorithms for function *f*6. IWOA and FOA had greater mean values for functions *f*23, *f*24, and *f*25 compared to other algorithms. The ABC mean value was better than IWOA, WOA, and FOA for functions *f*^″^4 and *f*4. Function *f*22 was unfit with all algorithms because the mean values were far away from the optimal value. The WOA mean value for *f*^″^8 had the least value. Similarly, the ABC mean value for *f*8 was the least. In addition, IWOA convergence was faster and obtained a lower value compared to WOA, FOA, and ABC. It can be said that IWOA was better than ABC and FOA. IWOA also enriched the original of WOA. ### 2.1.3. Chaotic WOA {#sec2.1.3} Metaheuristic algorithms have problems due to convergence speed and obtaining better performance. The theory of nonlinear chaos has widely been used in different applications \[[@B9]\]. Dynamical chaotic systems are able to control status and unsteady periodic motions. Chaos can be used in stochastic and deterministic algorithms. Due to developing the performance and convergence speed, chaos was used with WOA \[[@B27]\]. This theory has been used by variety of algorithms such as genetic algorithm \[[@B24]\], harmony search \[[@B28]\], PSO \[[@B29]\], ABC \[[@B30], [@B31]\], FA \[[@B32]\], KH \[[@B33]\], BOA \[[@B34]\], and GWO \[[@B35]\]. Different types of chaotic maps used with WOA in order to control the main parameter of WOA to provide stability between exploration and exploitation. Chaos means features of a complicated system, which have unpredictable behavior and map means relating parameters by using functions with chaos behavior. Ten unidimensional chaotic maps were used with WOA \[[@B36]\]. These ten maps were used to produce a chaotic set. The initial point was crucial because it had an impact on chaotic maps. As a result, 0.7 was chosen as an initial point, which ranges between 0 and 1 \[[@B37]\]. Therefore, 20 benchmark functions were tested by CWOA. As a result, chaotic maps enhanced the efficiency of WOA. ### 2.1.4. ILWOA {#sec2.1.4} Cloud computing is a computing system, which provides services via the Internet to clients \[[@B26]\]. Cloud computing is divided into two parts, which are the front end and back end. The front end includes all the software, which clients need, and the back end is related to the server and data storage \[[@B38]\]. Effectiveness and intelligent usages of cloud datacentre resources are the strategies of the consolidation of the virtual machine (VM). The most significant problem due to VM consolidation is VM replacement. Researchers aim was to minimize the number of physical machines, which were running in cloud datacentre. Abdel-Basset et al. \[[@B39]\] proposed improved Lévy WOA (ILWOA) to solve the minimization problem regarding the available bandwidth. Cloudsim toolkit was used to test the ILWOA on 25 different datasets and then compared with WOA, first fit, best fit, particle swarm optimization, genetic algorithm, and intelligent tuned harmony search. As a result, ILWOA showed better performance compared to other algorithms. ### 2.1.5. MWOA {#sec2.1.5} Because of developing technologies, protecting information is crucial in order to transmit it to the Internet. A modified version of WOA (MWOA), which was used for cryptanalysis of the Merkle--Hellman knapsack cryptosystem (MHKC), was developed in \[[@B40]\]. The continuous value was converted to discrete by a sigmoid function. Then, the evaluation function was dealt with an infeasible solution by adding a penalty function. Mutation operation was added to improve the solution. MHKC was the first public key cryptosystem (PKC) invented in 1987. Two keys were used by MHKC. The keys were private and public. Encrypting the plaintext was done by a public key, and decrypting was done by a private key \[[@B11]\]. MWOA was used to breakdown the MHKC by knowing ciphertext. Consequently, an attacker could reach the plaintext by using MWOA with the ciphertext \[[@B40]\]. ### 2.1.6. Memetic WOA {#sec2.1.6} WOA is very competing with other common metaheuristic algorithms. However, WOA performance is restricted because of having search dynamics. Thus, the encircling mechanism mostly focuses on the exploration in the search space. As a result, WOA has poor performance to jump out from local optima. To solve this problem, memetic WOA (MWOA) was proposed in \[[@B12]\] by using chaotic local search inside WOA in order to extend the exploration capacity. MWOA was used to create stability between exploration and exploitation phases in the search space. To achieve a balance, MWOA was tested on 48 benchmark functions; then, results showed that MWOA performed better compared to its competitors with regard to accuracy and convergence speed. 2.2. Hybridization of WOA {#sec2.2} ------------------------- WOA was used with common metaheuristic algorithms to achieve better solutions and get rid of the weakness of WOA and other algorithms. The summaries of hybridizations of WOA are explained in the following subsections. ### 2.2.1. With SA {#sec2.2.1} Majdi and Mirjalili \[[@B41]\] presented WOA with simulated annealing (SA). SA was embedded inside WOA to improve the best solution, which was found at the end of each iteration. WOA was able to search efficiently for finding the best solution. The blind operator was used by WOA in the exploitation phase, so this technique was replaced by using SA. Despite the effectiveness of WOA, SA was used to enhance the exploitation phase and overcome the stagnation in local optima. ### 2.2.2. With PSO {#sec2.2.2} Trivedi et al. \[[@B42]\] used PSO and WOA to obtain a superior solution for global numerical functions. They used PSO for the exploitation phase, and WOA was used in exploration phase in an environment, which was not certain. WOA used the algorithmic spatial path to explore a possible solution in less computational time to avoid local optima \[[@B5]\]. The result showed the efficiency of PSO-WOA compared to PSO and WOA individually. ### 2.2.3. With Local Search {#sec2.2.3} According to \[[@B40]\], the authors proposed WOA, a strategy that is called Local Search, in order to reduce permutation flow shop scheduling problem (FSSP). FSSP is an NP-hard problem, which is hard to find a result in polynomial time. Despite its essentiality, several algorithms have been developed to achieve two goals: reducing the time complexity and decreasing the duration of the best schedule. Other algorithms, which solved FSSP, had some drawbacks due to high computational cost and local optima \[[@B40]\]. Therefore, an algorithm was required for the largest rank value (LRV) to deal with the search space of the problem, which is discrete. As a result, a hybrid whale algorithm (HWA) was presented and able to achieve optimal solution quickly by using various techniques, for example, swap mutation, insert-reversed block operation, local search strategy, and integrated with a heuristic algorithm that is known as Nawaz--Enscore--Ham (NEH). Swap mutation operation was used to improve the diversity of the candidate schedule. Local optima were also avoided by using the insert-reversed block operation. As a result, HWA was combined with NEH to develop basic WOA performance. The proposed algorithm showed better results compared to the basic WOA \[[@B40]\]. ### 2.2.4. With EWGC {#sec2.2.4} Data are increasing nowadays; hence, controlling data becomes a difficult task. Therefore, data might be too complex. As a result, decision-making is affected by the way of organizing data. Thus, data clustering is essential to extract knowledge and make an efficient decision about knowledge. Exponential grey wolf optimization (EGWO) with whale optimization for data clustering (EWGC) was proposed to identify optimal centroid through the clustering process. WGC used the hybridization of WOA and WEGWO \[[@B43]\]. WGC used the WOA algorithm hunting mechanism to find centroid and position updates by using the EGWO algorithm in the exploration phase. Three datasets were used to test the proposed algorithm, and the results were compared with the particle swarm clustering (PSC), modified PS (mPS), grey wolf optimization (GWO), exponential GWO, kernel-based EGWO, and WOA. WGC showed better results compared to those algorithms. ### 2.2.5. With BS {#sec2.2.5} Cloud computing has a major role in the digital era because it serves a large number of users at the same time. Besides of having many advantages, security of data, which are stored in the cloud platform, is a big challenge. Brainstorm WOA (BS-WOA) is a hybridized algorithm which is based on brainstorm optimization and WOA. Thus, BS-WOA was used to identify the secret key of the database because the privacy of users should be preserved. Therefore, BS-WOA generated a key for the data, which came from the data owner, in order to protect the data from being used by the third-party user. As a result, BS-WOA improved the privacy and utility of the data in the cloud while the secret key was identified during the optimization process \[[@B44]\]. 3. Applications of WOA {#sec3} ====================== WOA has been used in several areas in various academic and industrial fields so far and the most important application classes are shown in the following subsections. 3.1. Electrical and Electronics Engineering {#sec3.1} ------------------------------------------- In the last years, the distribution systems of electric power are requiring extensive voltage ratio to supply inductive loads, which cause more power losses in the distribution networks and weakness in the power factor. To control these problems, appropriate distribution of capacitors is provided and eliminating the network line losses could enhance the constancy and the accuracy of the system. In order to find the optimum sizing and status of the capacitors for standard radial distribution systems, WOA was proposed as a solution, and several aspects were taken into consideration, such as decreasing the cost of operating and minimizing the losses in the power with disparity limitation on the voltage range. The suggested algorithm was confirmed by applying it to standard radial systems: IEEE-34 and IEEE-85 bus radial distribution test systems. The obtained results were efficient compared with the existing algorithms \[[@B45]\]. The main function of the economic operation of power plants is scheduling the generating units to obtain minimum generation cost for the power utilities that means low-cost electricity. WOA is one of the most important new strategies to solve the economic dispatch problem. The execution of the utilized algorithm was verified using standard test system of IEEE 30-Bus; the obtained results from the proposed algorithm was compared with other metaheuristic approaches, such as PSO, ant colony optimization, and genetic algorithm and the comparison indicated that the obtained results were somewhat similar \[[@B46], [@B47]\]. 3.2. Economic Scheduling {#sec3.2} ------------------------ With a massive amount of real-world applications, the flow shop scheduling problem (FSSP) has increased intensely. FSSP is regarded as an NP-hard problem since finding a solution in polynomial time is a difficult issue. In order to decrease the makespan of the best schedule and reduce the required time, WOA was merged with the local search technique for handling the flow shop scheduling problem. Swap mutation operation was utilized to enhance the diversity of item schedules, and the local optima problem was overcome by using insert-reverse block operation. The hybrid whale algorithm (HWA) obtained competitive results compared with the previous algorithms \[[@B10]\]. 3.3. Civil Engineering {#sec3.3} ---------------------- The enhanced whale optimization algorithm (EWOA) was suggested to deal with sizing and optimization problems of truss and frame structures. EWOA was used to solve four structural optimization problems: two truss optimization problems (spatial 72-bar truss and spatial 582-bar tower) and two frame optimization problems (3-bay 15-story frame and 3-bay 24-story frame). The obtained numerical results showed that the suggested EWOA had better efficiency than the standard whale optimization algorithm \[[@B48]\]. 3.4. Fuel and Energy {#sec3.4} -------------------- WOA is widely used in the fields of saving, processing, and improving energy and fuel sources, and the following are some of these applications:The need for cleaning source of energy caused a rise in the using of solar energy; thus, researchers have given great importance to the design of photovoltaic cells. They faced two important problems; the first one was finding a beneficial model to describe the solar cells, and the second one was the lack of information about photovoltaic cells, which badly influences the efficiency of the photovoltaic modules (panels). The chaotic whale optimization algorithm (CWOA) for the parameter estimation of solar cells was made and used for calculating and automatically adapting the internal parameters of the optimization algorithm. The improved technique was able to optimize difficult and multimodal objective purpose. The experimental results of the proposed approach showed higher performance regarding accuracy and robustness \[[@B49]\].More recently, researchers have been searching for alternative energy sources, such as solar, wind, and biomass because of the lack of conventional energy sources, such as petrol and coal, and these sources are among the main causes of environmental pollution. At different circumstance and under variance conditions, it is very important to exploit the maximum solar power from the photovoltaic panels; thus, a modified artificial killer whale optimization algorithm (MAKWO) to trace and find the highest power region of the photovoltaic module in the partially cloudy atmosphere was suggested. The obtained findings from MAKWO were compared with different metaheuristic algorithms (modified wolf pack algorithm (MAWP), artificial bee colony (ABC), and particle swarm optimization (PSO)) with a significant performance for the proposed algorithm (MAWP) over all the other algorithms \[[@B50]\]. 3.5. Medical Engineering {#sec3.5} ------------------------ Lately, analysis of medical images has become the focus of many researchers because they highly depend on these images for diagnosis and surgery. The liver is one of the organs most used in the computer-aided diagnosis system in order to detect the correct position of the organ inside the abdominal and also to avoid the intensity values overlapping with other organs. The whale optimization algorithm was proposed for liver segmentation in MIR images. To do the segmentation process, many clusters in the abdominal were determined. WOA had split the image into a number of clusters. After converting it to a binary image, it was multiplied by the previously clustered image with WOA in order to delete several parts of other organs; then, the required clusters were represented, which led to the liver area. A set of 70 images were tested using the suggested method illustrated and agreed by radiology specialists. Some measures like structural similarity index measure (SSIM), similarity index (SI), and other five measures were used to verify the correctness of the image. The final resolution of the processed image showed 96.75% accuracy using SSIM and 97.5 using SI% \[[@B51]\]. 3.6. Problems Solved by WOA {#sec3.6} --------------------------- WOA is a metaheuristic optimization algorithm that can be used to solve different problems, such as engineering problems, binary problems, multiobjective problems, and scheduling problems. [Table 1](#tab1){ref-type="table"} summarizes several problems, which have been solved by the WOA. 4. Benchmark Functions Experiment {#sec4} ================================= According to \[[@B21]\], WOA was compared with different algorithms, such as GSA, PSO, FEP, and DE. These algorithms were tested on 29 benchmark functions, it can be said that the benchmark functions are separated into four types: unimodal, multimodal, fixed-dimension multimodal, and composite functions, as shown in [Table 2](#tab2){ref-type="table"}. These benchmark functions are used as a validation procedure to test WOA, and then the results are compared with other common algorithms to ensure whether WOA is better or not. Each algorithm runs 30 times in order to obtain the optimum solution. The following subsections include comparison and discussion, solving classical engineering problem by WOA, comparison of WOA with IWOA, comparing WOA with other algorithms for feature selection, and finally, the evaluation performance of WOA is compared against AWOA and ILWOA. 4.1. Comparison and Discussion {#sec4.1} ------------------------------ WOA characteristics were assessed based on 29 benchmark functions. These benchmark functions are standard functions that are used as a validation procedure to assess WOA and its modifications. The following sections have Tables [3](#tab3){ref-type="table"}[](#tab4){ref-type="table"}--[5](#tab5){ref-type="table"}, which show the average and standard deviation. The following points explain the exploitation, exploration, escaping from local minima, and convergence behavior. ### 4.1.1. Capability Exploitation Assessment {#sec4.1.1} F1 to F7 are unimodal functions that have only one local optimum. Therefore, by using them, we can evaluate the performance of exploitation of each algorithm. [Table 3](#tab3){ref-type="table"} shows that WOA is as good as other optimization algorithms for unimodal functions in exploration capability. Specifically, for F1 and F2, WOA is the most efficient optimizer and it has the second rank in almost all functions. Therefore, WOA is good at exploitation behavior \[[@B5]\]. ### 4.1.2. Capability Exploration Assessment {#sec4.1.2} Multimodal functions have various local optima, while unimodal has one local best. Therefore, the local optimal number increases when the number of design variables increases. As a result, these types of function are vital to evaluate the exploration capability over other optimal algorithms. [Table 4](#tab4){ref-type="table"} shows that WOA has a good capability for exploration. Because of the integration mechanism of exploration, WOA has the second rank compared with other optimization algorithms. ### 4.1.3. Escaping from Local Minima {#sec4.1.3} Balancing between exploration and exploitation is the only way to avoid local optima because of challenging of mathematical computation of a composite function. [Table 5](#tab5){ref-type="table"} shows that the WOA algorithm ranked as the first optimizer in three tests and is as good as other optimization algorithms. It also demonstrates that WOA works well to make a balance between exploration and exploitation phases. ### 4.1.4. Convergence Behavior Analysis {#sec4.1.4} When comparing different metaheuristic algorithms (WOA, PSO, and GSA) for some problems, it can be seen that the convergence rate of WOA is well competitive with other algorithms when it is tested on 29 benchmarks functions \[[@B5]\]. WOA has many main characteristics that make it faster than other algorithms. In the initial steps of iterations, the search agents try to relocate their positions randomly around each other through Equation ([8](#EEq8){ref-type="disp-formula"}), which gives WOA high exploration capability, while using Equation ([7](#EEq7){ref-type="disp-formula"}), the search agents reposition their locations in a spiral-shaped path toward the best solution found so far. Each phase is done in almost half of iterations and simultaneously; thus, the WOA has the highest local optima avoidance capability and fast convergence rate than other similar metaheuristic algorithms. However, PSO and GSA have a greater probability of falling into starvation in local optima simply because they do not have parameters to determine specific iterations to the exploration or exploitation phases. In other words, they use only one equation to update the search agents\' positions, and also WOA requires less iteration to obtain global optimum compare to other algorithms. 4.2. WOA for Classical Engineering Problem {#sec4.2} ------------------------------------------ Mirjalili and Lewis \[[@B5]\] used WOA to solve the following engineering problems, which are shown in [Table 6](#tab6){ref-type="table"}. 4.3. WOA Feature Selection Experiment {#sec4.3} ------------------------------------- 16 datasets were chosen in this paper \[[@B28]\]. Training, validation, and testing were steps in which datasets were used. Each dataset randomly separated into three parts. The classification was done by the training part, and the validation part was used to assess the classification capability. Finally, the test part was required for evaluating the selected features. WOA, PSO, and GA were used on this test in order to achieve the comparison results. The results were computed in the Matlab environment 20 times. Overall, WOA outperformed feature selections, which approved the ability of wrapper-based approach and premature convergence while searching for optimal feature subset in the search space. WOA was better than PSO and GA in terms of ability to search for optimal features. Occurring local optima that may happen because of premature convergence can be avoided by WOA. Moreover, the results proved that WOA can find an optimal solution, which had maximum classification accuracy. It was also capable to make stability between exploration and exploitation. 4.4. Performance Evaluation on Benchmark Functions between Several Variants of WOA {#sec4.4} ---------------------------------------------------------------------------------- WOA has different types of modification. Therefore, it was compared with AWOA and ILWOA in the following subsections. ### 4.4.1. WOA and AWOA Comparison {#sec4.4.1} AWOA was tested on different unconstraint benchmark functions. It can be said that AWOA had a better result compared to WOA. AWOA improved solution by fast convergence, randomness, and stochastic behavior. It was also used as a random search in workspace while no optimal solutions exist. Thus, AWOA was an effective technique to solve the problem within unknown search space \[[@B4]\]. ### 4.4.2. WOA and ILWOA {#sec4.4.2} Statistical results from \[[@B12]\] show the difference between ILWOA and WOA performance, Friedman test is used with the experimental result to analysis. Friedman test can be executed on more than two dependent samples because it is a non-parametric and rank-based version of one-way ANOVA with respected measures. Figures [3](#fig3){ref-type="fig"} and [4](#fig4){ref-type="fig"} show datasets with three and five types of hosts with Friedman ranked mean for each algorithm. The result showed that ILWOA had the best performance in minimizing utilization of host machines \[[@B39]\]. Friedman test and datacentre utilization host were performed to analyze the obtaining result. It is clear that ILWOA had been tested on three and five bin datasets. As a result, the efficiency of ILWOA increased as the number of bin datasets increased. 5. Standard Bat Algorithm {#sec5} ========================= The bat algorithm is a metaheuristic algorithm developed by XinShe Yang in 2010 \[[@B57]\]. It was based on the echolocation capabilities of microbats. Before illustrating the details of this algorithm, we summarize echolocation briefly. 5.1. Echolocation of Microbats {#sec5.1} ------------------------------ Bats are mammals with echolocation capabilities. They use echolocation sonar to detect prey or to avoid obstacles. These bats send a very loud sound pulse and receive the echo that rebounds back from the surrounding objects. In zones with identical atmospheric air pressure, these sound pluses emit at a constant velocity while they get changed if the atmospheric pressure is changed \[[@B57]\]. Bats can estimate the positions of any surrounding objects using the time delay of the returning pulse, and also they determine the shape and the direction of the objects using comparative amplitudes of the sound pulses collected at each ear. Finally, the data obtained so far are analyzed and interpreted in the brain to construct image about their surroundings \[[@B58]\]. 5.2. Bat Algorithm {#sec5.2} ------------------ Using the concept of bat echolocation abilities, Yang developed various bat-inspired algorithms or bat algorithms \[[@B57]\]. He simulated this behavior to solve different optimization problems. Bats can determine the position of their preys, objects, or food exactly through very loud sound wave emission and receiving echo that comes back from these objects. Bats use the advantages of time delay concept to find their preys, whereas the time delay is calculated as space between bats\' ears and the echo wave variations. On a trip finding the prey, bats fly randomly in the search space with a speed *v*~*i*~ and change their positions *x*~*i*~ at a constant frequency *f*~min~, different wavelengths *β*, and loudness *A*~0~. In the algorithm, to update the value of these parameters, Yang used the following three equations \[[@B57]\]:$$\begin{matrix} {f_{i} = f_{\text{min}} + \left( {f_{\text{max}} - f_{\text{min}}} \right)\beta,} \\ \end{matrix}$$$$\begin{matrix} {v_{i}^{t + 1} = v_{i}^{t} + \left( {x_{i}^{t} - x_{\ast}} \right)f_{i},} \\ \end{matrix}$$$$\begin{matrix} {x_{i}^{t + 1} = x_{i}^{t} + v_{i}^{t},} \\ \end{matrix}$$where *x*~*i*~ is the position of the bats, *v*~*i*~ is the velocity of bats, *f*~*i*~ is the frequency of waves, and *β* is a random vector in the range \[0, 1\] taken from regular distribution. However, *x*~*∗*~ refers to the global best solution found so far among all bats in the search space. Based on the domain size of the optimization problem, the upper and the lower limits of the frequency are determined. Usually, the upper boundary is assigned 100 and the lower boundary is assigned 0. Initially, each bat takes a random value of the frequency within the range (*f*~min~, *f*~max~). The velocity of the search agent is compatible with the frequency, and the position of the new solution is located depending on its new velocity. When a bat finds its prey or food, the rate of the loudness reduces while the ratio of pulse emission rises. A pseudocode listed by Yang \[[@B57]\] is shown in [Algorithm 2](#alg2){ref-type="fig"}. 6. Hybrid WOA-BAT Algorithm {#sec6} =========================== WOA is an optimization algorithm, which showed high performance in solving many optimization problems. Despite all the results, WOA showed slow convergence speed due to finding the global optimum \[[@B53]\]. Therefore, the BAT algorithm is used to improve the exploration capability of WOA. In this approach, two basic techniques are used: (1) the BAT algorithm is partially embedded inside WOA search phase and (2) the condition technique is used after changing the position of each search agent; for example, if the new position is better than the old position, then the old position is replaced. As a result, the WOA-BAT algorithm can obtain better results in fewer iterations compared to WOA. The detail of this modification can be seen in the WOA-BAT algorithm pseudocode ([Algorithm 3](#alg3){ref-type="fig"}) and [Figure 5](#fig5){ref-type="fig"}. 7. Implementation and Results {#sec7} ============================= The proposed algorithm WOA-BAT is implemented and evaluated by using different benchmark functions. Three different benchmark functions are used to test the proposed algorithm; these are 23 mathematical optimization problems ([Table 2](#tab2){ref-type="table"} and \[[@B5]\]), CEC2005 ([Table 7](#tab7){ref-type="table"}), and CEC2019 ([Table 8](#tab8){ref-type="table"}). The code of WOA-BAT is available at the following link: <https://github.com/Hardi-Mohammed/WOA-BAT-modification>. In order to improve the WOA code which have been implemented by Mirjalili and Lewis, WOA and BAT algorithms are hybridized using the Matlab code. The following subsections include a description of benchmark functions, experimental setup, evaluation criteria, comparison of WOA-BAT with WOA, and comparison of WOA-BAT with common algorithms. 7.1. Benchmark Functions {#sec7.1} ------------------------ First, the implementation of 23 mathematical benchmark functions is conducted \[[@B5]\]. The test functions can be classified into two groups: *f*1*--f*7 (unimodal benchmark functions) and *f*8*--f*23 (multimodal benchmark functions). Second, CEC2005 includes four types of benchmark functions; these are unimodal functions, multimodal functions, expanded multimodal functions, and hybrid composition functions \[[@B59]--[@B61]\]. CEC2005 function details are in [Table 7](#tab7){ref-type="table"} \[[@B62]\]. Each part includes the numbers, respectively, 5, 7, 2, and 11. Third, 10 benchmark functions are used from CEC2019. All the benchmark functions in CEC2019 are multimodal functions and can be seen in [Table 8](#tab8){ref-type="table"} \[[@B63]\]. 7.2. Experimental Setup {#sec7.2} ----------------------- To obtain an accurate result, population size is randomly generated in order to make the best comparison with other common algorithms. The population size is 30, which were randomly generated, maximum iteration for the population size is 500, and the dimension is 30. The population size and iterations are executed 30 times then the average result is taken. 7.3. Evaluation Criteria {#sec7.3} ------------------------ Three ways are used to evaluate algorithms to obtain better comparison, and the following points are the criteria of evaluation:Obtaining average and standard deviationComparing WOA-BAT by building a box and whisker plot with WOACompare WOA-BAT with other metaheuristic algorithms 7.4. Comparison with Original WOA Algorithm {#sec7.4} ------------------------------------------- ### 7.4.1. Evaluation of F1--F7 Exploitation {#sec7.4.1} These are unimodal functions as they have a single optimum global value. By using these functions, we can easily investigate the exploitation capability of the developed algorithm. Therefore, [Table 9](#tab9){ref-type="table"} and [Figure 6](#fig6){ref-type="fig"} show that WOA-BAT as a better optimizer for *f*3, *f*4, *f*5, *f*6, and *f*7, while WOA is better for *f*1 and *f*2. As a result, WOA-BAT has an effective ability in exploitation. ### 7.4.2. Evaluation of F8--F23 Exploration {#sec7.4.2} Functions *f*8*--f*23 are multimodal functions, which can have many local optima. These numbers of local optima are increased depending on the design variables. Therefore, these functions can be used to test the exploration capability of the WOA-BAT algorithm. [Table 9](#tab9){ref-type="table"} and [Figure 6](#fig6){ref-type="fig"} illustrate that the averages of 10 benchmark functions (*f*8, *f*11, *f*12, *f*13, *f*14, *f*15, *f*17, *f*19, *f*20, *f*22, and *f*23) are very efficient in WOA-BAT, while WOA has the optimum value in 4 functions (*f*9, *f*10, *f*18, and *f*21). 25 benchmark functions of CEC2005 are used to test on WOA-BAT and WOA. [Table 10](#tab10){ref-type="table"} and [Figure 7](#fig7){ref-type="fig"} show the comparison results of WOA and WOA-BAT in box and whisker plot. WOA-BAT outperforms well in 13 functions. [Table 10](#tab10){ref-type="table"} shows that WOA-BAT has better performance compared to WOA original in *f*1, *f*2, *f*3, *f*4, *f*6, *f*9, *f*10, *f*12, *f*13, *f*18, *f*19, *f*22, and *f*25. However, WOA outperforms in other functions while WOA-BAT and WOA have the same performance in *f*7 and *f*8, which can be seen in [Figure 7](#fig7){ref-type="fig"}. Overall, it can be said that the proposed algorithm improved the WOA original to obtain a better result in approximately 13 functions. Like CEC2005, CEC2019 is used to test the WOA-BAT algorithm and WOA. [Table 11](#tab11){ref-type="table"} and [Figure 8](#fig8){ref-type="fig"} show that WOA-BAT has lower average result compared to WOA in seven functions *f*1, *f*2, *f*3, *f*5, *f*7, *f*8, and *f*10. However, WOA-BAT is not very competitive with WOA in *f*4, *f*6, and *f*9. Overall, WOA-BAT could improve the WOA in 7 benchmark functions from CEC2019. 7.5. Comparison with Metaheuristic Algorithms {#sec7.5} --------------------------------------------- The results from different papers included and presented in this paper in order to compare WOA-BAT with other well-known evolutionary algorithms, for example, GA, DE, ABC, and BSO. The results of these algorithms are obtained from CEC2005, which includes 25 benchmark functions \[[@B59]--[@B61]\]. The results for CEC2005 as shown in [Table 12](#tab12){ref-type="table"} indicate that WOA-BAT has the first rank because it outperforms well in 13 functions. The function, which WOA-BAT has a better result, are *f*3, *f*11, *f*12, *f*15, *f*16, *f*17, *f*18, *f*19, *f*20, *f*21, *f*22, *f*23, and *f*25. BSO and DE have the second and third ranks, respectively. WOA-BAT outperforms well in 13 functions, while BSO is well in 8 functions. Performance of DE is sufficient in 3 functions, which are *f*4, *f*5, and *f*6. However, in terms of standard deviation, the ABC result is the best in 8 functions. GA has the worst results in all functions and does not perform well compared to other algorithms. DE is the second worse algorithm. [Table 13](#tab13){ref-type="table"} is created in order to obtain the ranking result of the optimization algorithms from [Table 12](#tab12){ref-type="table"}. As a result, [Table 13](#tab13){ref-type="table"} illustrates that WOA-BAT has the best ranking among the five optimization algorithms. Overall ranking WOA-BAT is 1.6. However, BSO has 2.6. Accordingly, the difference between WOA-BAT and BSO is 1, so it can be said that the difference is significant. WOA-BAT and BSO have approximately the same ranking result for *f*1*--f*12. However, there is a significant difference between *f*13*-f*14 and *f*15*--f*25.WOA-BAT is better than BSO in *f*15*--f*25 by 1.9. Overall, it is believed that WOA-BAT has better ranking compare to GA, DE, ABC, and BSO. 8. Conclusion {#sec8} ============= In this study, WOA was explained in detail. WOA characteristics and its functionality were presented. In addition, the use of WOA was described in different areas, such as electrical and electronics engineering, automatic control system, civil engineering, fuel and energy, and medical engineering. Furthermore, researchers have modified and hybridized WOA in order to overcome optimization problems in the above areas. WOA was tested on 23 benchmark functions in order to determine the capability of exploitation, exploration, escaping from local minima, and convergence behavior. WOA had better performance of exploitation when it was tested on unimodal functions. It was also performed well in exploration on multimodal functions. In addition, testing WOA on composite functions can be viewed as the best way to stabilize between exploration and exploitation. Therefore, it can be said that WOA would increase convergence speed during iterations, while the majority of optimization algorithms (like PSO and GSA) do not have operators to consecrate a particular iteration to the exploration or the exploitation because they use only one format to update the position of search agents; thus, the probability of falling into local optima is more likely increased. It is safe to say that WOA achieves convergence speed and avoids local optima at the same time through iterations because of having two independent stages (exploration and exploitation). Both exploration and exploitation are done in each iteration. It is obvious that WOA cannot solve every optimization problems. However, it is very competitive with other common optimization algorithms. Another limitation of WOA is that WOA has poor convergence speed while searching around the global optimum. It is established that there are many types of WOA modifications and hybridizations. It is impossible to compare each new proposed WOA with all other types, and there are different benchmark functions, which can be used to test any new modifications. Therefore, it is believed that creating a platform for the researchers is essential in order to upload their program. After that, it will be easy to conduct and compare all the modifications and hybridizations and decide which one is the best. WOA demonstrated high performance in solving many optimization problems. Regardless of all the results, WOA exhibited slow convergence speed due to finding the global optimum. As a result, the BAT algorithm is used to recover the exploration capability of WOA. Thus, the WOA-BAT algorithm presented to obtain better results in fewer iterations compared to WOA. In this paper, WOA-BAT and WOA were tested on 25 functions from CEC2005. The results indicate that WOA-BAT performance is much better than WOA in 13 functions and have the same result in two functions. Also, WOA-BAT is tested on CEC2019 and compared with WOA. WOA-BAT has a lower average than WOA in 7 out of 10 functions. WOA-BAT was evaluated against other competitive algorithms using CEC2005. The results showed that WOA-BAT has the first rank among GA, DE, ABC, and BSO. There are several areas in WOA that can be further researched in the future. Therefore, the following areas might be interesting for researchers:Hybridization of WOA with other population metaheuristic algorithm, such as the ant-lion algorithmInvestigation on the adaptive value, which is responsible for the exploration and exploitation ability of WOA-BATSolving real-world problems in health care filed by hybridizing WOA-BAT with another optimization algorithm would be interestedHybridization of other optimization algorithms with WOA-BAT for cluster head selection for IoTIt is recommended to use WOA-BAT to train other advanced types of machine learning techniques such as Capsule Net, LSTM, and CNNApplying WOA-BAT for constrained optimization problemsApplying WOA-BAT for discrete optimization problemsSolving different business applications by using the WOA-BAT algorithmUsing WOA-BAT for feature selection in data miningUsing WOA-BAT in the text mining field. The authors wish to express their deep thanks to the University of Kurdistan Hewler (UKH) for providing funds for conducting this research study. Conflicts of Interest ===================== The authors declare that they have no conflict of interest. ![Spiral shape bubble net \[[@B4]\].](CIN2019-8718571.001){#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#fig8} {#alg1} {#alg2} {#alg3} ###### Problems solved by WOA. Method Year, references Problem Purpose Conclusion ---------------------------------------------------------------------------------------------------- ------------------ ------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- WOA for constrained economic load dispatch problems 2018, \[[@B13]\] Economic load dispatch problem constraining Giving reliable and constant electricity, whereas obtaining the best production with least cost and system operations Solving the ELD problem resulted in the fast convergence and appropriate execution time Binary WOA (bWOA) 2018, \[[@B14]\] Dimensionality reduction and classifications problem Selecting the optimal feature subset, which can be the optimal solution based on the sigmoid transfer function (S shape) bWOA could find optimal features, which have vital performance in terms of accuracy and execution time Multiobjective method for vehicle traveling based on WOA (MOWOA) 2017, \[[@B52]\] Vehicle fuel consumption problem Optimizing vehicle fuel consumption in terms of vehicle direction and traffic status MOWOA satisfied the performance within the vehicle traveling optimization, and the performance increased slightly compared to Dijkstra\'s and A^*∗*^ algorithm Using WOA 2018, \[[@B53]\] Nonuniformity in speed communication and illumination Optimizing the position of the light emitting diodes (LEDs) The result showed that this approach has given the higher uniformity compared to another result achieved by PSO MOWOA 2018, \[[@B54]\] Multilevel threshold as a multiobjective function problem Determining the multilevel threshold value for image segmentation The result showed that WOA had better performance for solving this problem within faster convergence and lower execution time Multiobjective task scheduling algorithm using WOA 2017, \[[@B15]\] The multiobjective task scheduling problem Availability of low cost for each service and minimizing the execution time The result showed great improvement in the proposed algorithm compared to original WOA Improved whale optimization algorithm (IWOA) for solving both single and multidimensional problems 2017, \[[@B55]\] 0--1 knapsack problem Handling infeasible solutions are the aim of this modification by adding penalty function to the evaluation function and sigmoid function to take the input parameter, which is the real-valued, and then produce the output IWOA is able to give a balance between exploration and exploitation by using local search strategy (LSS) and the Lévy flight walks. The result indicated that IWOA is robust, effective, and efficient for solving this problem compared to other metaheuristic algorithms, which were used to solve this problem The time-optimal memetic whale optimization algorithm 2017, \[[@B56]\] Hypersonic vehicle re-entry trajectory optimization problem with no-fly zones Improving the robustness of IWOA to extend its strong ability on global search and improve the nonsensitivity of the initial values. Improve IWOA poor searching convergence speed by using Gauss pseudospectral methods (GPM) Compared to the initial guess solution results of this hybridized technique, it concluded that it is very competitive and has better search accuracy, convergence speed, and robustness ###### Description of unimodal, multimodal, fixed-dimension multimodal, and composite benchmark functions used in this work. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------   Function V_no Range *f* ~min~ -------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------ ----------------- --------------- *Unimodal benchmark functions* 1 *f* ~1~(*x*)=∑~*i*=1~^*n*^*x*~*i*~^2^ 30 \[−100, 100\] 0 2 *f* ~2~(*x*)=∑~*i*=1~^*n*^\|*x*~*i*~\|+∏~*i*=1~^*n*^\|*x*~*i*~\| 30 \[−10, 10\] 0 3 *f* ~3~(*x*)=∑~*i*=1~^*n*^(∑~*j*−1~^*i*^*x*~*j*~)^2^ 30 \[−100, 100\] 0 4 *f* ~4~(*x*)=max~*i*~{\|*x*~*i*~\|,  1 ≤ *i* ≤ *n*} 30 \[−100, 100\] 0 5 *f* ~5~(*x*)=∑~*i*−1~^*n*−1^\[100(*x*~*i*+1~ − *x*~*i*~^2^)^2^+(*x*~*i*~ − 1)^2^\] 30 \[−30, 30\] 0 6 *f* ~6~(*x*)=∑~*i*=1~^*n*^(\[*x*~*i*~+0.5\])^2^ 30 \[−100, 100\] 0 7 *f* ~7~(*x*)=∑~*i*=1~^*n*^*ix*~*i*~^4^+random\[0, 1\] 30 \[−1.28, 1.28\] 0 *Multimodal benchmark functions* 8 $f_{8}\left( x \right) = {\sum_{i = 1}^{n}{- x_{i}\sin\left( \sqrt{\left| x_{i} \right|} \right)}}$ 30 \[−500, 500\] −418.9829 × 5 9 *f* ~9~(*x*)=∑~*i*=1~^*n*^\[*x*~*i*~^2^ − 10  cos(2*πx*~*i*~)+10\] 30 \[−5.12, 5.12\] 0 10 $f_{10}\left( x \right) = - 20\text{ exp}\left( {- 0.2\sqrt{\left( {1/n} \right){\sum_{i = 1}^{n}x_{i}^{2}}}} \right) - \exp\left( {\left( {1/n} \right){\sum_{i = 1}^{n}{\cos\left( {2\pi x_{i}} \right)}}} \right) + 20 + e$ 30 \[−32, 32\] 0 11 $f_{11}\left( x \right) = \left( {1/4000} \right){\sum_{i = 1}^{n}x_{i}^{2}} - {\prod_{i = 1}^{n}{\cos\left( {x_{i}/\sqrt{i}} \right) + 1}}$ 30 \[−600, 600\] 0 12 *f* ~12~(*x*)=(*π*/*n*){10sin(*πy*~1~)+∑~*i*=1~^*n*−1^(*y*~*i*~ − 1)^2^\[1+10  sin^2^(*πy*~*i*+1~)\]+(*y*~*n*~ − 1)^2^}+∑~*i*=1~^*n*^*u*(*x*~*i*~, 10, 100, 4)\ 30 \[−50, 50\] 0 \ $y_{i} = 1 + \left( {x_{i} + 1/4} \right)u\left( {x_{i},a,k,m} \right) = \left\{ \begin{matrix} {k\left( {x_{i} - a} \right)^{m}x_{i} > a} \\ {0 - a < x_{i} < a} \\ {k\left( {- x_{i} - a} \right)^{m}x_{i} < - a} \\ \end{matrix} \right.$ 13 *f* ~13~(*x*)=0.1{sin^2^(3*πx*~1~)+∑~*i*=1~^*n*^(*x*~*i*~ − 1)^2^\[1 + sin^2^(3*πx*~*i*~+1)\]+(*x*~*n*~ − 1)^2^\[1 + sin^2^(2*πx*~*n*~)\]}  +∑~*i*=1~^*n*^*u*(*x*~*i*~, 10, 100, 4) 30 \[−50, 50\] 0 *Fixed-dimension multimodal benchmark functions* 14 *f* ~14~(*x*)=((1/500)+∑~*j*=1~^25^(1/(*j*+∑~*i*=1~^2^(*x*~*i*~ − *a*~*ij*~)^6^)))^−1^ 2 \[−65, 65\] 1 15 *f* ~15~(*x*)= ∑~*i*=1~^11^\[*a*~*i*~ − (*x*~1~(*b*~*i*~^2^+*b*~*i*~*x*~2~)/*b*~*i*~^2^+*b*~*i*~*x*~3~+*x*~4~)\]^2^ 4 \[−5, 5\] 0.00030 16 *f* ~16~(*x*)=4*x*~1~^2^ − 2.1*x*~1~^4^+ (1/3)*x*~1~^6^+*x*~1 ~*x*~2~ − 4*x*~2~^2^+4*x*~2~^4^ 2 \[−5, 5\] −1.398 17 *f* ~17~(*x*)=(*x*~2~ − (5.1/4*π*^2^)*x*~1~^2^+(5/*π*)*x*~1~ − 6)^2^+10(1 − (1/8*π*))cos  *x*~1~+10 2 \[−5, 5\] 0.398 18 *f* ~18~(*x*)=\[1+ (*x*~1~+*x*~2~+1)^2^(19 − 14*x*~1~+3*x*~1~^2^ − 14*x*~2~+6*x*~1~*x*~2~+3*x*~2~^2^)\] ×  \[30+(2*x*~1~ − 3*x*~2~)^2^ ×  (18 − 32*x*~1~+12*x*~1~^2^+48*x*~2~ − 36*x*~1~*x*~2~+27*x*~2~^2^)\] 2 \[−2, 2\] 3 19 *f* ~19~(*x*)=−∑~*i*=1~^4^*c*~*i*~exp(−∑~*j*=1~^3^*a*~*ij*~(*x*~*j*~ − *p*~*ij*~)^2^) 3 \[1, 3\] −3.86 20 *f* ~20~(*x*)=−∑~*i*=1~^4^*c*~*i*~exp(−∑~*j*=1~^6^(*x*~*j*~ − *p*~*ij*~)^2^) 6 \[0, 1\] −3.32 21 *f* ~21~(*x*)=−∑~*i*=1~^5^\[(*x* − *a*~*i*~)(*x* − *a*~*i*~)^*T*^+*c*~*i*~\]^−1^ 4 \[0, 10\] −10.1532 22 *f* ~22~(*x*)=−∑~*i*=1~^7^\[(*x* − *a*~*i*~)(*x* − *a*~*i*~)^*T*^+*c*~*i*~\]^−1^ 4 \[0, 10\] −10.4028 23 *f* ~23~(*x*)=−∑~*i*=1~^10^\[(*x* − *a*~*i*~)(*x* − *a*~*i*~)^*T*^+*c*~*i*~\]^−1^ 4 \[0, 10\] −10.5363 ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ###### Result comparison among optimization algorithms \[[@B2]\]. F DE GSA PSO FEP WOA ---- ------------- ------------- -------------- -------------- ---------- -------------- --------- --------- -------------- -------------- F1 8.2*E* − 14 5.9*E* − 14 2.53*E* − 16 9.67*E* − 17 0.000136 0.000202 0.00057 0.00013 1.41*E* − 30 4.91*E* − 30 F2 8.2*E* − 14 5.9*E* − 14 2.53*E* − 16 9.67*E* − 17 0.000136 0.000202 0.00057 0.00013 1.41*E* − 30 4.91*E* − 30 F3 1.5*E* − 09 9.9*E* − 10 0.055655 0.194074 0.042144 0.045421 0.0081 0.00077 1.06*E* − 21 2.39*E* − 21 F4 6.8*E* − 11 7.4*E* − 11 896.5347 318.9559 70.12562 22.11924 0.016 0.014 5.39*E* − 07 2.93*E* − 06 F5 0 0 7.35487 1.741452 1.086481 0.317039 0.3 0.5 0.072581 0.39747 F6 0 0 67.54309 62.22534 96.71832 60.11559 5.06 5.87 27.86558 0.763626 F7 0 0 2.5*E* − 16 1.74*E* − 16 0.000102 8.28*E* − 05 0 0 3.116266 0.532429 ###### Result comparison among optimization algorithms \[[@B5]\]. F DE GSA PSO FEP WOA ----- ---------- ------------- ---------- -------------- ---------- -------------- ------------- ------------- ---------- -------------- F8 −11080.1 574.7 −2821.07 493.0375 −4841.29 1152.814 −12554.5 52.6 −5080.76 695.7968 F9 69.2 38.8 25.96841 7.470068 46.70423 11.62938 0.046 0.012 0 0 F10 7.4043 4.2*E* − 08 0.06207 0.23628 0.27605 0.50901 0.018 0.0021 7.4043 9.897572 F11 0.000289 0 27.70154 5.040343 0.009215 0.007724 0.016 0.022 0.000289 0.001586 F12 0.339676 8*E* − 15 1.799617 0.95114 0.006917 0.026301 9.2*E* − 06 3.6*E* − 06 0.339676 0.214864 F13 1.889015 4.8*E* − 14 8.899084 7.126241 0.006675 0.008907 0.00016 0.000073 1.889015 0.266088 F14 2.111973 3.3*E* − 16 5.859838 3.831299 3.627168 2.560828 1.22 0.56 2.111973 2.498594 F15 0.000572 0.00033 0.003673 0.001647 0.000577 0.000222 0.0005 0.00032 0.000572 0.000324 F16 −1.03163 3.1*E* − 13 −1.03163 4.88*E* − 16 −1.03163 6.25*E* − 16 −1.03 4.9*E* − 07 −1.03163 4.2*E* − 07 F17 0.397887 9.9*E* − 09 0.397887 0 0.397887 0 0.398 1.5*E* − 07 0.397914 2.7*E* − 05 F18 3 2*E* − 15 3 4.17*E* − 15 3 1.33*E* − 15 3.02 0.11 3 4.22*E* − 15 F19 N/A N/A −3.86278 2.29*E* − 15 −3.86278 2.58*E* − 15 −3.86 0.000014 −3.85616 0.002706 F20 N/A N/A −3.31778 0.023081 −3.26634 0.060516 −3.27 0.059 −2.98105 0.376653 F21 −10.1532 0.0000025 −5.95512 3.737079 −6.8651 3.019644 −5.52 1.59 −7.04918 3.629551 F22 −10.4029 3.9*E* − 07 −9.68447 2.014088 −8.45653 3.087094 −5.53 2.12 −8.18178 3.829202 F23 −10.5364 1.9*E* − 07 −10.5364 2.6*E* − 15 −9.95291 1.782786 −6.57 3.14 −9.34238 2.414737 ###### Composite benchmark functions comparison result \[[@B5]\]. F DE GSA PSO WOA ----- ------------- ------------- ------------- -------------- -------- -------- ---------- ---------- F24 6.75*E* − 2 6.75*E* − 2 6.75*E* − 2 2.78*E* − 17 100 81.65 0.568846 0.505946 F25 28.759 8.6277 200.6202 67.72087 155.91 13.176 75.30874 43.07855 F26 144.41 19.401 180 91.89366 172.03 32.769 55.65147 21.87944 F27 324.86 14.784 170 82.32726 314.3 20.066 53.83778 21.621 F28 10.789 2.604 200 47.14045 83.45 101.11 77.8064 52.02346 F29 490.94 39.461 142.0906 88.87141 861.42 125.81 57.88445 34.44601 ###### Different engineering problem comparison result. Problems Aim Result -------------------------------------------- ----------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------ Tension/ compression spring design problem Minimizing the weight of tension/ compression spring is the goal of this design problem WOA had better performance over PSO and GSA on average, and both PSO and GSA required more function evaluation than WOA \[[@B5]\] Welded beam design problem Minimizing the fabrication cost of the welded beam is the objective WOA outperformed over PSO and GSA on average and required the least number of function evaluations to find the best optimal solution Pressure vessel design The objective is to minimize the total cost of a cylindrical vessel WOA performed better compared to PSO and GSA on average and the required of a number of evaluation function \[[@B5]\] 15-bar truss design problem Minimizing the weight of the 15-bar truss is the goal of this problem WOA had similar performance, which would find a similar structure with other algorithms. WOA had the second rank for the number of the evaluation function ###### Summary of the 25 CEC2005 test functions. No. Functions *F* ~1~(*x*^*∗*^)=*f*\_*bias*~*i*~ *D* Search range ------------------------------------------------------------------------------ -------------------------------------------------------------------------------- ------------------------------------ ------------ --------------- *Unimodal benchmark functions* ([5](#EEq5){ref-type="disp-formula"}) 1 Shifted sphere function −450 10, 30, 50 \[−100, 100\] 2 Shifted Schwefel\'s problem 1.2 −450 10, 30, 50 \[−100, 100\] 3 Shifted rotated high conditioned elliptic function −450 10, 30, 50 \[−100, 100\] 4 Shifted Schwefel\'s problem 1.2 with noise in fitness −450 10, 30, 50 \[−100, 100\] 5 Schwefel\'s problem 2.6 with global optimum on bounds −310 10, 30, 50 \[−100, 100\] *Multimodal functions Basic functions* ([7](#EEq7){ref-type="disp-formula"}) 6 Shifted Rosenbrock\'s function 390 10, 30, 50 \[−100, 100\] 7 Shifted rotated Griewank function without bounds −180 10, 30, 50 \[0, 600\] 8 Shifted rotated Ackley\'s function with global optimum on bounds −140 10, 30, 50 \[−32, 32\] 9 Shifted Rastrigin\'s function −330 10, 30, 50 \[−5, 5\] 10 Shifted rotated Rastrigin\'s function −330 10, 30, 50 \[−5, 5\] 11 Shifted rotated weierstrass function 90 10, 30, 50 \[−0.5, 0.5\] 12 Schwefel\'s problem 2.13 −460 10, 30, 50 \[−*π*, *π*\] *Expanded functions* ([2](#EEq2){ref-type="disp-formula"}) 13 Expanded extended Griewank plus Rosenbrock\'s function (F8F2) −130 10, 30, 50 \[−3, 1\] 14 Shifted rotated expanded Scaffer\'s F6 −300 10, 30, 50 \[−100, 100\] *Hybrid composition functions* ([11](#EEq11){ref-type="disp-formula"}) 15 Hybrid composition function 120 10, 30, 50 \[−5, 5\] 16 Rotated hybrid composition function 120 10, 30, 50 \[−5, 5\] 17 Rotated hybrid composition function with noise in fitness 120 10, 30, 50 \[−5, 5\] 18 Rotated hybrid composition function 10 10, 30, 50 \[−5, 5\] 19 Rotated hybrid composition function with a narrow basin for the global optimum 10 10, 30, 50 \[−5, 5\] 20 Rotated hybrid composition function with the global optimum on the bounds 10 10, 30, 50 \[−5, 5\] 21 Rotated hybrid composition function 360 10, 30, 50 \[−5, 5\] 22 Rotated hybrid composition function with high condition number matrix 360 10, 30, 50 \[−5, 5\] 23 Noncontinuous rotated hybrid composition function 360 10, 30, 50 \[−5, 5\] 24 Rotated hybrid composition function 260 10, 30, 50 \[−5, 5\] 25 Rotated hybrid composition function without bounds 260 10, 30, 50 \[2, 5\] ###### Summary of the basic CEC2019 functions. No. Functions *F* ~*i*~ ^*∗*^=*F*~*i*~(*x*^*∗*^) *D* Search range ----- ----------------------------------------------- ------------------------------------ ----- ------------------- 1 Storn\'s Chebyshev polynomial fitting problem 1 9 \[−8192, 8192\] 2 Inverse Hilbert matrix problem 1 16 \[−16384, 16384\] 3 Lennard-Jones minimum energy cluster 1 18 \[−4, 4\] 4 Rastrigin\'s function 1 10 \[−100, 100\] 5 Griewangk function 1 10 \[−100, 100\] 6 Weierstrass function 1 10 \[−100, 100\] 7 Modified Schwefel\'s function 1 10 \[−100, 100\] 8 Expand Schaffer\'s F6 function 1 10 \[−100, 100\] 9 Happy Cat function 1 10 \[−100, 100\] 10 Ackley function 1 10 \[−100, 100\] ###### Comparison result of WOA-BAT and WOA. Function WOA WOA-BAT ---------- ---------------------- -------------- ---------------------- -------------- 1 **1.2*E*** − **74** 5.89*E* − 74 1.34*E* − 06 4.07*E* − 07 2 **2.37*E*** − **51** 8.82*E* − 51 0.0074 0.0013 3 50945 13527.44 **0.000216** 0.001126 4 52.426 24.97794 **0.001** 7.1*E* − 05 5 28.02927 0.452718 **11.1554** 13.89905 6 0.4356 0.212037 **1.58*E*** − **06** 7.15*E* − 07 7 0.0026 0.002513 **0.0037** 0.0076 8 −10424 1668.107 −**12214** 1084.168 9 **1.89*E*** − **15** 1.04*E* − 14 2.9852 9.107663 10 **4.8*E*** − **15** 2.35*E* − 15 0.1201 0.652436 11 0.011 0.042629 **7.2*E*** − **08** 3.4*E* − 08 12 0.02 0.010083 **1.27*E*** − **08** 5.89*E* − 09 13 0.5672 0.296041 **2.1*E*** − **07** 1.07*E* − 07 14 3.258 3.214906 **0.998** 4.52*E* − 16 15 0.000566 0.000369 **0.000384** 0.000354 16 −**1.0316** 6.78*E* − 16 −**1.0316** 6.78*E* − 16 17 0.3979 9.73*E* − 06 **0.39789** 1.69*E* − 16 18 **3** 6.15*E* − 05 7.5 10.23432 19 −3.856 0.009536 −**3.8623** 0.002004 20 −3.225 0.101675 −**3.2822** 0.057156 21 −**8.746** 2.324189 −8.4675 2.42464 22 −7.6138 2.858764 −**9.697** 1.830619 23 −6.7571 3.587922 −**9.9988** 1.640539 ###### Comparison result of WOA-BAT and WOA on CEC2005. Function WOA WOA-BAT ---------- -------------------------- -------------- -------------------------- -------------- 1 8.83*E* + 00 1.55*E* + 01 **3.94*E*** **+** **00** 5.47*E* + 00 2 1.09*E* + 04 3.96*E* + 03 **6.92*E*** **+** **03** 2.83*E* + 03 3 3.02*E* + 06 3.18*E* + 06 **1.33*E*** **+** **06** 1.72*E* + 06 4 1.83*E* + 04 6.56*E* + 03 **1.64*E*** **+** **04** 3.89*E* + 03 5 **2.87*E*** **+** **03** 2.89*E* + 03 6.60*E* + 03 3.68*E* + 03 6 1.39*E* + 05 2.42*E* + 05 **6.17*E*** **+** **04** 2.01*E* + 05 7 **1.27*E*** **+** **03** 2.98*E* − 01 **1.27*E*** **+** **03** 7.36*E* − 02 8 **2.03*E*** **+** **01** 9.86*E* − 02 **2.03*E*** **+** **01** 1.84*E* − 01 9 4.22*E* + 01 1.43*E* + 01 **3.46*E*** **+** **01** 1.10*E* + 01 10 6.23*E* + 01 2.22*E* + 01 **5.76*E*** **+** **01** 1.89*E* + 01 11 **8.87*E*** **+** **00** 1.22*E* + 00 1.00*E* + 01 1.75*E* + 00 12 1.60*E* + 04 1.38*E* + 04 **1.29*E*** **+** **04** 1.60*E* + 04 13 4.42*E* + 00 2.33*E* + 00 **4.17*E*** **+** **00** 2.44*E* + 00 14 **3.92*E*** **+** **00** 3.18*E* − 01 4.01*E* + 00 3.37*E* − 01 15 **2.19*E*** **+** **01** 3.88*E* + 01 4.71*E* + 01 4.38E + 01 16 **3.35*E*** **+** **01** 7.19*E* + 01 5.51*E* + 01 5.17*E* + 01 17 **2.29*E*** **+** **01** 3.78*E* + 01 3.88*E* + 01 3.62*E* + 01 18 2.94*E* + 02 1.41*E* + 02 **2.50*E*** **+** **02** 1.07*E* + 02 19 2.77*E* + 02 1.32*E* + 02 **2.67*E*** **+** **02** 9.22*E* + 01 20 **206.6743** 170.0525 250 135.8244 21 **223.3337** 212.8355 267.9505 122.4694 22 330.573 165.6103 **2.59*E*** **+** **02** 105.2167 23 **253.7131** 242.8109 294.9476 126.9669 24 **199.7812** 27.13892 215.8896 71.95231 25 137.1858 20.53592 **1.34*E*** **+** **02** 28.38144 ###### Comparison results of WOA-BAT and WOA CEC2019. Function WOA WOA-BAT ---------- -------------------------- -------------- -------------------------- -------------- 1 2.10*E* + 10 3.57*E* + 10 **7.60*E*** **+** **07** 4.16*E* + 08 2 1.84*E* + 01 1.61*E* − 02 **1.75*E*** **+** **01** 1.21*E* − 01 3 1.37*E* + 01 7.23*E* − 15 **1.27*E*** **+** **01** 9.53*E* − 04 4 **3.48*E*** **+** **02** 1.72*E* + 02 2.12*E* + 03 1.01*E* + 03 5 3.03*E* + 00 4.86*E* − 01 **2.44*E*** **+** **00** 6.67*E* − 01 6 **1.03*E*** **+** **01** 1.39*E* + 00 1.11*E* + 01 1.55*E* + 00 7 6.14*E* + 02 2.98*E* + 02 **6.06*E*** **+** **02** 3.90*E* + 02 8 6.03*E* + 00 5.66*E* − 01 **5.72*E*** **+** **00** 7.18*E* − 01 9 **5.93*E*** **+** **00** 6.85*E* − 01 2.28*E* + 01 4.92*E* + 01 10 2.13*E* + 01 1.35*E* − 01 **2.12*E*** **+** **01** 2.26*E* − 01 ###### Comparison of WOA-BAT with GA, DE, ABC, and BSO. Function GA DE ABC BSO WOA-BAT ---------- -------------- -------------- -------------------------- -------------- -------------------------- -------------- --------------------------- ------------------ -------------------------- -------------- 1 2.45*E* + 03 7.30*E* + 02 1.79*E* − 04 1.31*E* − 04 2.20*E* − 02 4.08*E* − 02 **−4.50*E*** **+** **02** 3.50*E* **−** 14 3.94*E* + 00 5.47*E* + 00 2 3.26*E* + 04 1.08*E* + 04 2.12*E* + 02 9.29*E* + 01 2.73*E* + 04 4.05*E* + 03 **−4.48*E*** **+** **02** 9.36*E* − 01 6.92*E* + 03 2.83*E* + 03 3 1.56*E* + 08 6.85*E* + 07 6.28*E* + 06 2.09*E* + 06 1.22*E* + 08 2.90*E* + 07 2.04*E* + 06 7.23*E* + 05 **1.33*E*** **+** **06** 1.72*E* + 06 4 1.30*E* + 05 6.17*E* + 04 **1.15*E*** **+** **03** 7.23*E* + 02 3.38*E* + 04 4.49*E* + 03 2.78*E* + 04 8.05*E* + 03 1.64*E* + 04 3.89*E* + 03 5 1.47*E* + 04 2.76*E* + 03 **5.63*E*** **+** **02** 2.84*E* + 02 8.30*E* + 03 8.00*E* + 02 4.70*E* + 03 1.22*E* + 03 6.60*E* + 03 3.68*E* + 03 6 6.71*E* + 07 3.87*E* + 07 **3.94*E*** **+** **01** 2.98*E* + 01 3.65*E* + 05 2.58*E* + 05 1.26*E* + 03 9.48*E* + 02 6.17*E* + 04 2.01*E* + 05 7 5.34*E* + 03 8.55*E* + 01 4.70*E* + 03 9.01*E* − 11 4.89*E* + 03 2.88*E* + 01 **6.25*E*** **+** **02** 3.25*E* + 02 1.27*E* + 03 7.36*E* − 02 8 2.10*E* + 01 6.64*E* − 02 2.10*E* + 01 7.75*E* − 02 2.10*E* + 01 6.86*E* − 02 −**1.20*E*** **+** **02** 9.90*E* − 02 2.03*E* + 01 1.84*E* − 01 9 7.86*E* + 01 1.36*E* + 01 1.46*E* + 02 2.87*E* + 01 2.10*E* + 02 1.35*E* + 01 −**2.86*E*** **+** **02** 1.27*E* + 01 3.46*E* + 01 1.10*E* + 01 10 3.39*E* + 02 4.09*E* + 01 2.15*E* + 02 1.13*E* + 01 2.46*E* + 02 9.04*E* + 00 −**2.93*E*** **+** **02** 8.79*E* + 00 5.76*E* + 01 1.89*E* + 01 11 3.56*E* + 01 2.71*E* + 00 4.04*E* + 01 1.35*E* + 00 4.05*E* + 01 1.37*E* + 00 1.10*E* + 02 2.51*E* + 00 **1.00*E*** **+** **01** 1.75*E* + 00 12 1.95*E* + 05 5.95*E* + 04 1.82*E* + 04 1.19*E* + 04 4.02*E* + 05 5.17*E* + 04 2.84*E* + 04 1.99*E* + 04 **1.29*E*** **+** **04** 1.60*E* + 04 13 1.49*E* + 01 2.53*E* + 00 1.79*E* + 01 1.49*E* + 00 2.31*E* + 01 1.45*E* + 00 −**1.26*E*** **+** **02** 1.05*E* + 00 4.17*E* + 00 2.44*E* + 00 14 1.34*E* + 01 2.28*E* − 01 1.37*E* + 01 1.32*E* − 01 1.36*E* + 01 1.34*E* − 01 −**2.87*E*** **+** **02** 3.78*E* − 01 4.01*E* + 00 3.37*E* − 01 15 5.47*E* + 02 6.41*E* + 01 2.70*E* + 02 9.66*E* + 01 3.06*E* + 02 5.76*E* + 00 5.43*E* + 02 7.94*E* + 01 **4.71*E*** **+** **01** 4.38*E* + 01 16 4.33*E* + 02 8.20*E* + 01 2.54*E* + 02 4.05*E* + 01 2.63*E* + 02 9.94*E* + 00 2.87*E* + 02 1.34*E* + 02 **5.51*E*** **+** **01** 5.17*E* + 01 17 8.34*E* + 02 2.25*E* + 02 2.81*E* + 02 4.62*E* + 01 2.86*E* + 02 1.72*E* + 01 3.10*E* + 02 1.57*E* + 02 **3.88*E*** **+** **01** 3.62*E* + 01 18 9.60*E* + 02 1.42*E* + 01 9.06*E* + 02 7.56*E* − 01 9.60*E* + 02 5.84*E* + 00 9.17*E* + 02 1.36*E* + 00 **2.50*E*** **+** **02** 1.07*E* + 02 19 9.57*E* + 02 1.62*E* + 01 9.06*E* + 02 8.12*E* − 01 9.63*E* + 02 7.72*E* + 00 9.16*E* + 02 1.07*E* + 00 **2.67*E*** **+** **02** 9.22*E* + 01 20 9.58*E* + 02 1.17*E* + 01 9.06*E* + 02 4.04*E* − 01 9.60*E* + 02 6.53*E* + 00 9.16*E* + 02 1.36*E* + 00 **250** 135.8244 21 1.01*E* + 03 1.72*E* + 02 5.59*E* + 02 1.79*E* + 02 5.10*E* + 02 3.45*E* + 00 9.27*E* + 02 1.37*E* + 02 **267.9505** 122.4694 22 1.20*E* + 03 8.52*E* + 01 8.77*E* + 02 1.04*E* + 01 1.08*E* + 03 2.19*E* + 01 1.21*E* + 03 1.99*E* + 01 **2.59*E*** **+** **02** 105.2167 23 1.01*E* + 03 1.71*E* + 02 5.91*E* + 02 1.72*E* + 02 5.49*E* + 02 2.56*E* + 01 9.48*E* + 02 1.38*E* + 02 **294.9476** 126.9669 24 9.17*E* + 02 1.56*E* + 02 9.20*E* + 02 1.70*E* + 02 **2.00*E*** **+** **02** 3.48*E* − 02 4.67*E* + 02 6.23*E* + 00 215.8896 71.95231 25 1.79*E* + 03 3.92*E* + 01 1.64*E* + 03 3.33*E* + 00 1.51*E* + 03 8.75*E* + 00 1.88*E* + 03 4.44*E* + 00 **1.34*E*** **+** **02** 28.38144 ###### Ranking of WOA-BAT optimization compared to GA, DE, ABC, and BSO. Functions 1^st^ 2^nd^ 3^rd^ 4^th^ 5^th^ Rank Subtotal BSO ------------------ ------------- ------------------ ------------- ------- ------- ------ ---------- ----- 1 BSO WOA-BAT GA DE ABC 2     2 BSO DE WOA-BAT ABC GA 3     3 WOA-BAT BSO DE ABC GA 1     4 DE WOA-BAT BSO ABC GA 2     5 DE BSO WOA-BAT ABC GA 3 11 9 6 DE BSO WOA-BAT ABC GA 3     7 BSO WOA-BAT DE ABC GA 2     8 BSO WOA-BAT GA DE ABC 2     9 BSO WOA-BAT GA DE ABC 2     10 BSO WOA-BAT DE ABC GA 2     11 WOA-BAT GA DE ABC BSO 1     12 WOA-BAT DE BSO GA ABC 1 13 14 13 BSO WOA-BAT GA DE ABC 2     14 BSO WOA-BAT GA ABC DE 2 4 2 15 WOA-BAT DE ABC BSO GA 1     16 WOA-BAT DE ABC BSO GA 1     17 WOA-BAT DE ABC BSO GA 1     18 WOA-BAT DE BSO GA ABC 1     19 WOA-BAT DE BSO GA ABC 1     20 WOA-BAT DE BSO GA ABC 1     21 WOA-BAT ABC DE BSO GA 1     22 WOA-BAT DE ABC GA BSO 1     23 WOA-BAT ABC DE BSO GA 1     24 ABC WOA-BAT BSO GA DE 2     25 WOA-BAT ABC DE GA BSO 1 12 42   **Total** 40 **Total** 67 **Overall rank** 40/25 = 1.6 **Overall rank** 67/25 = 2.6 **F1--F5** 11/5 = 2.2 **F1--F5** 9/5 = 1.8 **F6--F12** 13/7 = 1.8 **F6--F12** 14/7 = 2 **F13--F14** 4/2 = 2 **F13--F14** 2/2 = 1 **F15--F25** 12/11 = 1.9 **F15--F25** 42/11 = 3.8 [^1]: Academic Editor: Maciej Lawrynczuk

1. Introduction {#s0005} =============== The cytoplasmic proteins NOD (nucleotide-binding oligomerisation domain containing) 1 and NOD2 are members of the NLR (nucleotide-binding, leucine-rich repeat containing receptor) family of pattern recognition receptors. They act as immune sentinels and play an important role in combating bacterial infection and maintaining cellular homeostasis [@b0005]. NOD1 and NOD2 recognise fragments of bacterial peptidoglycan via their C-terminal leucine rich repeats [@b0010; @b0015; @b0020; @b0025]. Activation causes conformational change and relocalisation to the plasma membrane [@b0030; @b0035; @b0040; @b0045]. NF-κB (nuclear factor kappa B)-mediated pro-inflammatory signalling pathways are activated following interaction between the CARDs (caspase activation and recruitment domain) of NOD1/2 with their adaptor protein RIP2 (receptor interacting protein 2). Understanding how NOD1/2 signalling is regulated is important for the future development of therapeutic treatments targeting inflammatory disorders such as Crohn's Disease. CARD9 is an important adaptor molecule in the innate immune response. CARD9 is predominantly associated with NF-κB signalling pathways following stimulation of C-type lectin receptors like DECTIN-1 [@b0050; @b0055]. Receptor ligation upregulates Spleen tyrosine kinase (SYK) and activates Protein Kinase C delta which phosphorylates Thr231 in CARD9 [@b0060]. This causes formation of a CARD9/B Cell lymphoma 10 (Bcl-10)/Mucosa-associated lymphoid tissue lymphoma translocation protein 1 (Malt1) 'signalosome' which activates NF-κB [@b0050; @b0065]. CARD9 is also involved in NF-κB signalling downstream of the Retinoic acid-inducible gene I receptor (RIG-1) family [@b0070]. The biological function of CARD9 is conserved between mice and humans. Human CARD9 is able to restore signalling in murine *Card9* knock-out cells [@b0075] and both inactive human *CARD9* mutant cells and murine *Card9* knockout cells display a defective response to β-glucan stimulation [@b0075; @b0080] Recently, Hsu and colleagues demonstrated in mice that *CARD9* is required for the synergistic activation of p38 and JNK (c-Jun N-terminal kinase) following stimulation of NOD2 by either muramyl dipeptide or *Listeria monocytogenes* [@b0085]*.* The association between NOD2 and CARD9 was enhanced by the presence of RIP2 in both over-expression and endogenous systems [@b0085]. The relationship between CARD9 and NOD2 is particularly intriguing as the genes for both these proteins contain polymorphisms influencing susceptibility to Crohn's Disease in humans [@b0090; @b0095]. Multiprotein complexes play a key role in innate immune signalling. Complexes such as the inflammasome and Myddosome are formed through interactions between members of the death domain superfamily [@b0100], which includes CARDs. NOD2 and CARD9 have two and one N-terminal CARDs respectively. We have used cell-based immunoprecipitation and co-purification of overexpressed recombinant protein to study the molecular details of the interaction between NOD2 and CARD9. Unexpectedly, we did not find any evidence for an interaction between the CARDs of NOD2 and CARD9 as previously suggested. Instead, we show that the region in NOD2 responsible for the interaction with CARD9 involves the NACHT domain and the preceding linker to the CARDs. 2. Materials and methods {#s0010} ======================== 2.1. Plasmids {#s0015} ------------- Full length murine CARD9 (GENBANK: [NP_001032836.1](ncbi-p:NP_001032836.1){#ir030}) with a C-terminal V5-His epitope tag in the pEF6 expression vector (pEF6-mCARD9-V5) was a kind gift from David Underhill [@b0055]. Full-length human NOD1 (GENBANK: [AAD28350.1](ncbi-p:AAD28350.1){#ir035}) and NOD2 (GENBANK: [AAG33677.1](ncbi-p:AAG33677.1){#ir040}) with an N-terminal FLAG tag in a pCMV backbone (pCMV-FLAG-NOD1 and pCMV-FLAG-NOD2) were kindly provided by Thomas Kufer [@b0105]. Single nucleotide polymorphisms (SNP) across the NACHT domain were identified in the NCBI SNP database and cloned into full-length pCMV-FLAG-NOD2 using site-directed mutagenesis. N-terminal and C-terminal NOD2 constructs were also generated by site-directed mutagenesis. GB1-RIP2--CARD (human) and the tandem human NOD2 CARD construct used for NMR chemical shifts have been described previously [@b0110]. The CARD of Card9 (residues 6-100), the tandem CARDs of NOD2 (residues 28-215) and the RIP2 CARD (residues 432-534) were amplified by PCR and inserted, using Gateway® cloning, into the expression plasmid pDEST-HisMBP [@b0115]. In addition to the N-terminal His6-MBP (maltose binding protein) fusion each construct included a C-terminal FLAG tag to further aid expression and stability. The CARD of murine CARD9 was also inserted into pDEST-17 to generate an N-terminally 6His tagged construct. 2.2. Immunoprecipitation {#s0020} ------------------------ HEK 293T cells were maintained in DMEM (Sigma) supplemented with 10% FCS, 100 μg/ml Penicillin/Streptomycin and 2 mM [l]{.smallcaps}-glutamine at 37 °C and 5% CO~2~. Cells were seeded in 6 well plates and transfected using jetPEI (Polyplus Transfection) with 1 μg/well of full-length, mutated, or truncated pCMV-FLAG-NOD2, or full-length pCMV-FLAG-NOD1, and 1 μg pEF6-mCARD9-V5. After 24 h cells were washed twice in 1 × PBS and lysed in 300 μl RIPA buffer (50 mM Tris--HCl pH 7.6, 150 mM NaCl, 0.25% Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate) supplemented with 1 × Protease Inhibitor Cocktail set V (Calbiochem) and 7.5 units of Benzonase nuclease (Sigma) per well. Lysates were incubated on ice for 10 min with shaking and clarified by centrifugation (16 000×*g*; 2 min; 20 °C). Cells were lysed and incubated with Protein G coated magnetic Dynabeads (Life Technologies) coupled to mouse anti-FLAG antibody (Sigma). Immunoprecipitated proteins and inputs were detected by western blot analysis using: rabbit anti-flag (Sigma), mouse anti-V5 (Abcam) and mouse anti-GAPDH (Abcam) primary antibodies and goat anti-rabbit (Abcam) or goat anti-mouse (Sigma) secondary antibodies. 2.3. Recombinant protein expression and co-sonication pull-down assays {#s0025} ---------------------------------------------------------------------- Recombinant proteins were expressed in 10 ml cultures of Rosetta2 *Escherichia coli* for 16 hr at 22 °C except for GB1-RIP2--CARD which used 4 h at 37 °C. Cultures were pelleted, frozen overnight and resuspended in standard lysis buffer (100 mM NaCl, 25 mM sodium phosphate, 20 mM imidazole, 5 mM β-mercaptoethanol, 0.1 % Triton-X100). Appropriate samples were combined and sonicated on ice. Insoluble debris was removed by centrifugation (16 000×*g*, 4 °C, 10 min) and proteins purified using amylose affinity chromatography. Eluted samples were visualised by Coomassie Brilliant Blue staining. 2.4. NMR chemical shift assays {#s0030} ------------------------------ Samples of unlabelled murine CARD9 CARD and ^15^N-labelled human NOD2 (28--218) were buffer exchanged overnight into 20 mM sodium phosphate (pH 7.1), 100 mM NaCl and 5 mM DTT. 1D (^1^H) and 2D (^1^H/^15^N) HSQC NMR spectra were recorded on a Bruker Avance spectrometer at 600 MHz proton frequencies and processed using an associated software package. All spectra were recorded at 180 μM sample concentration (with 10% added D~2~O) and at 298 K. 2.5. Homology modelling and bioinformatics {#s0035} ------------------------------------------ Residues 217--820, corresponding to residues encoded by the large, central exon of human *NOD2*, were submitted to the PHYRE2 server for automated modelling [@b0120]. Only one NLR NACHT structure, the 4KXF structure of NLRC4 [@b0125], is available in the PDB at present and this was used as a template. Inspection of the resulting alignment and model showed that while the NBD, HD1 and WH domains (217--632) were clearly homologous between NOD2 and NLRC4, the rest of the exon, which forms HD2 and part of the LRR domain in NLRC4, showed a more ambiguous alignment and so was not considered further. The final model represents NOD2 residues A217--C632. Homology models of the CARD of murine CARD9 and human CARD9 from residue 6--100 were built using the CARD from CARD11 (PDB 1D: [4I16](pdb:4I16){#ir045}) [@b0130] as a template. Models were generated using the model-default script from the MODELLER package v9.12 ([http://salilab.org/modeller/](http://www.salilab.org/modeller/){#ir050}). The amino acid sequences of full length murine CARD9 (GENBANK: [NP_001032836.1](ncbi-p:NP_001032836.1){#ir055}) and human CARD9 (GENBANK: NP_43470.2) were aligned and the percentage identity and similarity calculated using Clustal Omega [@b0135]. 3. Results {#s0040} ========== 3.1. There is no interaction between the CARDs of NOD2 and CARD9 {#s0045} ---------------------------------------------------------------- CARD-mediated protein--protein interactions play an important role in numerous immune signalling pathways such as RIG-1 mediated viral sensing, inflammasome formation and facilitating NOD1/2-mediated NF-κB signalling via RIP2 [@b0140; @b0145; @b0150]. NOD2 and CARD9, which both possess CARDs, work synergistically to drive p38 and JNK signalling following activation of NOD2 and have been shown to interact [@b0085]. Currently nothing is known about how this interaction is mediated, although theoretical models have suggested the involvement of CARD:CARD interactions [@b0155]. To test whether the NOD2 CARD9 interaction is CARD-mediated we expressed the CARDs of human NOD2, murine CARD9 and human RIP2 as recombinant proteins fused to solubility-enhancement and epitope tags. Interactions were assessed using a modified version of the protocol established to study interactions between the CARDs of NOD2 and RIP2 [@b0110]. Specifically, bacterial cultures containing overexpressed recombinant protein were resuspended and combined prior to sonication. Amylose resin based affinity chromatography was used to purify proteins fused to MBP and co-purifying interacting proteins were detected by Coomassie Brilliant Blue staining following SDS--PAGE. Consistent with previous work the NOD2 CARDs successfully pulled down the RIP2 CARD, which was also capable of mediating homomeric RIP2 interactions ([Fig. 1](#f0005){ref-type="fig"}A). However, we detected no interaction between the CARD of CARD9 and either the NOD2 tandem CARDs or the RIP2 CARD; although the CARD9 CARD could facilitate weak self-association as has been previously reported [@b0160] ([Fig. 1](#f0005){ref-type="fig"}A and B). The inability of the tandem CARDs of NOD2 and the CARD9 CARD to interact was further confirmed using NMR chemical shift studies. 1D (^1^H) NMR spectrum confirmed that the CARD of murine CARD9 is tertiary structured as shown by the ring-current shifted methyl signals (\<0.5 ppm) resulting from formation of a hydrophobic core ([Fig. 1](#f0005){ref-type="fig"}C). ^15^N-labelled NOD2 (28--218) is also tertiary structured, displaying a generally good dispersion of peaks in the 2D (^1^H/^15^N) HSQC spectra and chemical shift values \>9 ppm (due to hydrogen bond formation; [Fig. 1](#f0005){ref-type="fig"}D and E). However, there are no significant chemical shift changes in the presence of equimolar CARD9 CARD indicating that there is no interaction between the two recombinant proteins under the conditions studied ([Fig. 1](#f0005){ref-type="fig"}D and E (right panel)). Together these approaches conclusively support the view that the CARDs of NOD2 and CARD9 do not directly interact. Comparing the sequences of murine CARD9 and human CARD9 indicated that these proteins are 86% identical and 91% similar across their entire sequence, both identity and similarity increasing to 95% for the CARD (Supplementary figure 1A). Homology models of the two CARDs confirmed that four of the five substitutions were in helix 1 and the fifth in helix 6. None of the mutations affect the overall protein fold or the electrostatic properties of the CARDs (Supplementary Fig. 1B and C). This is consistent with the conserved biological function of the molecules and allows us to conclude that it is highly unlikely that the CARDs of human CARD9 and human NOD2 would interact either. 3.2. The NACHT and CARD--NACHT linker of NOD2 are important for interaction with CARD9 {#s0050} -------------------------------------------------------------------------------------- Given the inability of the respective CARDs to interact we used domain truncations of FLAG-tagged NOD2 to map the region required for interaction with CARD9. HEK293 cells were transiently transfected with NOD2 truncations and V5-tagged CARD9 and proteins immunoprecipitated after 24 h. CARD9 was immunoprecipitated by full-length, CARD--NACHT, and NACHT--LRR NOD2 constructs ([Fig. 2](#f0010){ref-type="fig"}A). Neither the CARDs nor the LRRs alone interacted with CARD9, suggesting that the NACHT domain of NOD2 has a critical role in mediating interaction with CARD9. To further investigate the importance of the NOD2 NACHT domain we tested the impact of a panel of Crohn's Disease associated single nucleotide polymorphisms (SNP) [@b0165; @b0170; @b0175; @b0180] spanning the NACHT on the interaction with CARD9. This included the widely studied R702W SNP. We also tested the hyperactive Blau Syndrome associated SNP R334W [@b0185]. All of the polymorphisms still interacted with CARD9 indicating that these residues, and potentially the surrounding regions of the NACHT, were not crucial for CARD9 interaction ([Fig. 2](#f0010){ref-type="fig"}B and C). It also demonstrates that the clinical impact of the R334W and R702W SNPs is unlikely to result from alteration of the CARD9 and NOD2 interaction. Interestingly, the W355stop SNP, which lacks any sequence downstream of this tryptophan, still immunoprecipitated Card9 suggesting that a region between the NOD2 CARDs and W355 could mediate interaction. Consequently, we generated further C-terminal NOD2 truncations ([Fig. 3](#f0015){ref-type="fig"}A) and tested their interaction with CARD9. All of these truncations, including A274stop which lacks any of the NACHT, retained the ability to interact with CARD9 ([Fig. 3](#f0015){ref-type="fig"}B). Together with the failure of the CARD only construct (residues 1--227; Figs. [2](#f0010){ref-type="fig"}A and [3](#f0015){ref-type="fig"}B) to interact with CARD9 this indicated that residues 228--274 contain a critical region for CARD9 interaction. To characterise the role of the CARD--NACHT linker in more detail we expanded our range of mutants to alter the amount of linker present ([Fig. 3](#f0015){ref-type="fig"}A). Extension of the C-terminus of the CARD only construct to residue 240 did not result in interaction with CARD9 ([Fig. 3](#f0015){ref-type="fig"}B). However, addition of a further twenty residues, to position 260, resulted in CARD9 being immunoprecipitated, albeit more weakly than with the the wild-type protein, or other truncations ([Fig. 3](#f0015){ref-type="fig"}B). All of our N-terminal truncations were able to immunoprecipitate CARD9, including a construct beginning at 274 which therefore lacks the CARD--NACHT linker ([Fig. 3](#f0015){ref-type="fig"}B). Intriguingly both the constructs 1--274 and 274--1040 interacted with Card9 thereby implying the presence of at least two binding sites in NOD2 for CARD9; one located in the CARD--NACHT linker (between residues 241 and 274) and the other in the NACHT domain. Together these results support an interaction between NOD2 and CARD9. This interaction is not mediated by the CARD domains, but instead involves two regions in NOD2 -- one linking the CARDs and NACHT, the other within the NACHT itself. 4. Discussion {#s0055} ============= The pattern recognition receptors NOD1 and NOD2 play an important role in the pro-inflammatory immune response to bacterial infections. Their ability to activate NF-κB-mediated transcription following interaction with RIP2 has been widely studied. Our understanding of how these receptors activate alternative signalling pathways, such as those utilising p38 and JNK, is limited. However, it has been previously shown in mice that engagement of the adaptor protein CARD9 is crucial in facilitating NOD2-initiated p38 and JNK signalling, with the two proteins working in synergy to mediate this response [@b0085]. How this is achieved, and how NOD2 and CARD9 interact has not previously been understood. In this work we have identified two regions of NOD2 capable of binding CARD9; one in the linker connecting the CARDs and the NACHT, and one in the NACHT itself. These binding surfaces may by spatially adjacent to one another in the tertiary structure of the protein. A random mutagenesis study of NOD2 identified two residues, A232 and V256, in the linker region that when mutated resulted in a significant loss of NF-κB signalling [@b0190]. Whilst for A232 this correlated with a loss in protein expression this was not the case for V256, suggesting that the linker may be more important for NOD2 functionality than previously assumed. Somewhat surprisingly and despite the known importance of CARDs in mediating protein--protein interaction in immune signalling complexes we saw no evidence that the CARDs of NOD2 and CARD9 could interact with one another. In this study we used murine CARD9 and human NOD2. In light of the conserved biological function [@b0075; @b0080]; the high level of sequence identity (Supplementary Fig. 1A); and the conserved nature of the electrostatic surfaces (Supplementary Fig. 1C) between the human and murine proteins we are confident that our observations and conclusions remain valid for interactions between human CARD9 and NOD2. Our observations raise interesting questions about the molecular composition and formation of the NOD2:CARD9 signalling complex. The lack of NOD2 CARD9 CARD-mediated interaction suggests that NOD2 could interact concurrently with RIP2, via its CARDs, and with CARD9 via its linker/NACHT. This would enable simultaneous activation of NF-κB, p38 and JNK signalling pathways from the same macromolecular complex. Hsu and colleagues reported that the presence of RIP2 enhances the interaction between NOD2 and CARD9 [@b0085].This could be through promotion of conformational changes due to receptor activation which expose the CARD9 binding motifs on NOD2; or by promoting protein clustering and thereby increasing the avidity of the NOD2:CARD9 interaction. CARD9 contains a predicted coiled-coil domain (CCD). CCDs are widely reported to mediate protein--protein interactions and it remains entirely feasible that the CCD mediates interaction with NOD2. However, difficulties in expressing the CCD independently of the CARD [@b0160] have made this hypothesis impractical to test to date. In conclusion, we have identified regions of NOD2 important for CARD9 interaction. These now require functional interrogation to determine precisely how the two proteins interact and whether these regions could serve as modulators of receptor signalling with the potential for the regulation of a range of signalling pathways. Appendix A. Supplementary data {#s0065} ============================== Supplementary data 1Murine CARD9 and Human CARD9 are highly similar. (A) Clustal Omega generated sequence alignment of murine CARD9 (top) and human CARD9 (bottom). The consensus sequence is shown underneath. The CARD is delineated by a solid black bar and the residues forming the hydrophobic core of the CARD are highlighted in grey. (B) Homology models of the CARD from human CARD9 (blue) and murine CARD9 (brick red). Residues that differ between the two CARDs are shown in stick representation and the models are overlayed to emphasise their similarity. (C) Electrostatic surface of the human (left) and murine (right) proteins. The orientation of the electrostatic surface is identical to that presented in panel (B). This work was funded by a Wellcome Trust Career Development Fellowship (WT085090MA) to TPM and a Medical Research Council grant (U117565398) to KR. RP and JPB were supported by BBSRC Doctoral Training Grants. None of the funders had any role in the design of the study. We thank Geoff Kelly for assistance with the acquisition of NMR spectrum. {#f0005} {#f0010} {#f0015}

*Pyrenophora graminea* \[anamorph *Drechslera graminea* (Rabenh. *ex.* Schltdl.) Ito\], the seed-borne pathogen responsible for leaf-stripe in barley (*Hordeum vulgare* L.), inflicts heavy losses on crops ([@Porta-Pugliaetal1986]; [@Arabietal2004]). Several studies on morphological, physiological and biochemical aspects have already been undertaken ([@ZribaandHarrabi1995]; [@Jawharetal2000]). Traditionally, the classification of *P. graminea* isolates, besides requiring a certain expertise in taxonomy, may be further complicated by the inherent variation in morphological features among isolates, besides being time consuming, especially in those cases where similar species may be present in one and the same field ([@Gattietal1992]). Over the years, the methods for detecting and assessing genetic diversity have extended from the analysis of discrete morphological traits to those of biochemical and molecular origin. Two classes of molecular markers which have received much attention in recent studies on genetic diversity in natural populations, are inter-retrotransposon amplified polymorphism (IRAP) ([@Kalendaretal1999]; [@Pasqualietal2007]), and restriction fragment length polymorphisms (RFLP) of PCR amplified internal transcribed spacer (ITS) regions (ITS-RFLP) ([@HsiangandWu2000]; [@Nilssonetal2008]). The usefulness of these two markers types extends to resolve genetic variation among species within a genus or among populations ([@Redeckeretal1997]; [@MartinandRygiewicz2005]; [@Brancoetal2007]). Despite the general interest, it is not clear whether these two markers have comparable power for quantifying differentiation among populations. Thus, it would be of interest to determine whether IRAP and ITS-RFLP markers are equally efficient at detecting genetic patterns existent among *P. graminea* isolates. However, differences would suggest that one marker may be more appreciated for detecting isolation, which has implications for the use of either type of marker for defining demographically independent management units ([@Mortiz1994]). The present study aimed to evaluate the usefulness of both markers in assessing and analyzing the nature and extent of genetic diversity among isolates of *P. graminea* collected from various regions in Syria. The 39 monosporic isolates of *P. graminea* used in the study were identified, cultivated, and maintained as described by [@Arabietal2002], [@Arabietal2004]). They were isolated from leaf-stripe infected barley leaves, originating from various regions in Syria, and selected from among 93 isolates, according to morphological and physiological criteria (virulence). The isolates were grown separately in 9 cm Petri dishes containing potato dextrose agar (PDA, DIFCO, Detroit, MI. USA), and incubated for 10 days at 21 ± 1 °C in the dark to facilitate mycelia growth. Genomic DNA was extracted from fungal cultures as previously described ([@ArabiandJawhar2007]). ITS regions and 5.8S rDNAs were amplified for all the isolates using the ITS1 (5\' TCCGTAGGTGAACCTGCGG 3\') and ITS4 (5\'TCCTCCGCTTATTGATATGC 3\') primers designed by [@Whiteetal1990]. The amplification protocol was as described by [@ArabiandJawhar2007]. In separate reactions, 10 μL of PCR reaction were digested for 3 h with six different endonucleases (*Alu*I, *EcoR*1, *Bsur*I, *Bam*HI, *Rsa*I and *Hind*III), according to manufacturer\'s recommendations (MBI Fermentas, York, UK). DNA fragments were size-fractionated by electrophoresis through 1.5% agarose gels. The sizes were determined by comparison with their molecular weight relative to a DNA ladder (Q.BIOgene). The IRAP method was used for retrotransposon amplification, as described by [@Kalendaretal1999]. Primer sequences, as well as retrotransposon type and orientation are shown in [Table 1](#t1){ref-type="table"}. PCR was carried out using the method described by [@JawharandArabi2009]. Amplified products were electrophoresed in a 2% agarose gel using a 1 x Tris-borate-EDTA buffer (100 mM Tris-HCl/L, pH 8.3, 83 boric acid/L, 1 mM EDTA/L) at 100 V. Subsequently, the gels were stained with ethidium bromide solution and visualized under ultraviolet illumination. The sizes of the amplified products were determined as mentioned above. ITS-RFLP and IRAP banding profiles were scored for the presence (1) or absence (0) of bands. The experiments were repeated twice for each isolate and both markers, so as to confirm repeatability and remove monomorphic bands from the analysis. The data were converted to a Jaccards similarity ([@Jaccard1908]) coefficient, which was used to construct a dendrogram by the unweighted pair-group method with arithmetic averages (UPGMA) utilising the software package Phylip 3.7 ([@Felsenstein1985]). The polymorphism information content (PIC) was calculated for each locus according to [@Andersonetal1993], which provides an estimate of the discriminating power of a locus by taking into account the number of alleles generated by each reaction unit and their frequency distribution in the population. The percent of polymorphic markers (β) was estimated by dividing the number of polymorphic markers by the number of obtained markers. The multiplex ratio (MR) is defined as the number of bands per reaction unit, and the effective multiplex ratio EMR as the product of MR with the fraction of polymorphic markers. Marker utility (MI) for genetic studies was estimated as a marker index according to [@Powelletal1996] and [@Milbourneetal1997]. As PIC values are equal to gene diversity in binary marker systems, the effective number of alleles per marker was calculated as the respective reciprocal of the PIC value. The Mantel test ([@Mantel1967]) was applied to ascertain the significance of correlations between pairwise genetic similarities in both marker systems. The probability of calculated correlation was estimated based on 1000 random permutations. These computations were carried out using the Arlequin software package ([@Excoffieretal2005]). The quality nature of data (QND) of the marker system and the effective marker index EMI as an overall criterion for the utility of molecular markers were calculated according to [@Varshneyetal2007]. Selected IRAP bands were cut with a surgical blade and purified with a QIAgene gel extraction kit according to manufacturers recommendations. Sequencing was carried out on a Genetic Analyzer (ABI 310, Perkin-Elmer, Applied Biosystems, USA). Each sequence was identified by homology search using the Basic Local Alignment Search Tool (BLAST) program ([@Altschuletal1997]) against the GenBank nonredundant public sequence database. PCR amplification with the specific primers ITS1 and ITS4 yielded single DNA fragments present in all isolates with \~ 650 bp in size, which is in agreement with the results obtained by a previous study ([@ArabiandJawhar2007]). Fingerprints generated from the five restriction digestions of the nrDNA ITS region denoted high levels of intraspecific variation within the *P. graminea* population. A total of 354 scorable DNA bands were scored, 274 of which (77%) being polymorphic, while the number of bands in isolates varied from 3 to 5. Based on IRAP patterns, 534 bands were obtained, 454 (85%) of which were polymorphic, whereas the number of DNA bands in isolates varied between 4 and 15 ([Figure 1](#fig1){ref-type="fig"}). This is sustained by the findings of [@Tayloretal2004], who found a presence of high copy numbers of *Pyggy*-like sequences in the *P. graminea* genome by using a primer derived from the LTR (long terminal repeat) of the *Pyggy* retrotransposon isolated from this fungus. However, the sequence of one IRAP fragment, when using a Sukkula primer, showed similarity 19/27 (70.4%) to the LTR of the *Pyggy* retrotransposon (AF533703.1). Similarity began from position 34, as position 4 was a G instead of an A, and position 6 an A instead of a G. Furthermore, bases were missing at positions 13 and 24. In addition, the *P.graminea* LTR sequence and a rice cDNA clone (Accession No. AK058381) were significantly similar, as attested by 117/147 bp identity (80%). Homology was also evident between *P.graminea* LTR and an *Alternaria alternata* LTR (AB025309), with 1556/2096 bp identity (74%). These results indicated the capability of LTR-specific primer to amplify in different target species. On the other hand, both markers were highly repeatable, although QND was 0.191 for ITS-RFLP markers and only 0.141 for IRAP. PIC values were 0.376 and 0.355 for IRAP and ITS-RFLP, respectively. Furthermore, IRAP markers generated a substantially higher number of markers (7.80) and a superior marker index (2.41) than ITS-RFLP ([Table 2](#t2){ref-type="table"}). The UPGMA dendrogram generated from IRAP and ITS-RFLP data demonstrated that isolates clustered into five groups for both markers by a similarity index of 0.341 for IRAP and 0.460 for ITS-RFLP ([Figure 2](#fig2){ref-type="fig"}). The correlation between IRAP and ITS-RFLP similarity matrices was moderate but significant (*r* = 0.34, p \< 0.05). The usefulness of a molecular marker technique depends upon both the polymorphic information content (PIC) of the markers and the number of markers generated by each primer ([@Varshneyetal2007]). Even though both IRAP and ITS-RFLP markers exhibited comparable PIC values, owing to the higher number of markers per assay, the MI of IRAP was 2.41 higher than for ITS-RFLP ([Table 2](#t2){ref-type="table"}). [@Bernardesetal2007] compared the performance of REMAP, a retrotransposon based maker technique, and ISSR with the fungus *Magnaporthe grisea* and has found an MI of 1.54 and 4.25, respectively. These marker systems are similar to IRAP and ITS-RFLP, in that they are PCR-based, anonymous and dominantly inherited. They also depend on repeated patterns in the genome to provide annealing sites for universal primers. The results showed that band quality could have benefited from the additional restriction step following amplification in the ITS-RFLP protocol, thereby leading to a clearly defined band pattern. The documentation capabilities of band information produced by both marker assays in gene bank systems were comparable, as they were equally evaluated. This was placed on par with the widely used AFLP markers ([@Varshneyetal2007]), which are far inferior to single locus markers such as SSRs or SNPs with unique primers for each locus. On considering both quantitative and qualitative attributes, IRAP turned out to be superior to ITS-RFLP, as depicted by a more effective marker index (0.338 and 0.255, respectively). Researchers have examined the existence of correlations between various molecular marker techniques in *Fusarium oxysporum* f. sp. *lentis*. [@Belabidetal2004] reported similar genetic relationships through RAPD and AFLP analysis. In barley, [@Russelletal1997] found that RFLP and AFLP, but not SSR, were correlated. In the present study, we found significant and moderate correlation between IRAP and ITS-RFLP in *P. graminea* pathogen according to the Mantel test, which confirmed in the partially conserved dendrogram topologies inferred from each of the similarity matrices ([Figure 2](#fig2){ref-type="fig"}). The moderate correspondence between these markers could possibly be attributed to different amplification targets in the *P. graminea* genome. To our knowledge this is the first comparative report on the two advanced IRAP and ITS-RFLP genetic marker systems. The present study emphasized that, besides their effective employment, both of these DNA markers may furnish comparable results in assays of genetic differentiation among *P. graminea* isolates. Furthermore, due to the specific advantages of each marker, the combination of both marker systems can give us greater confidence that the delineated patterns are real, through drawing on results from multiple genetic systems ([@AllendorfandSeeb2000]). {#fig1} {#fig2} The authors wish to thank the Director General of AECS for his support and Dr. N. MirAli the Head of the Biotechnology Department for his critical review of the manuscript. Associate Editor: Everaldo Gonçalves de Barros ###### Primer name, retrotransposon type, position and sequence. Name and orientation Retrotransposon type Accession Position Sequence ---------------------- ---------------------- ----------- ----------- ------------------------------------- 3\'LTR → *BARE-1* Z17327 2112-2138 TGTTTCCCATGCGACGTTCCCCAACA LTR6149 → *BARE-1* Z17327 1993-2012 CTCGCTCGCCCACACATCAACCGCGTTTATT LRT6150 ← *BARE-1* 418-439 CTGGTTCGGCCCATGTCTATGTATCCACACATGTA 5\'LRT1 ← *BARE-1* Z17327 1-26 TTGCCTCTAGGGCATATTTCCAACA 5\'LRT2 ← *BARE-1* Z17327 314-338 ATCATTCCCTCTAGGGCATAATTC Sukkula → *Sukkula* AY054376 4301-4326 GATAGGGTCGCATCTTGGGCGTGAC AY054373 Nikita → Nikita AY078074 1-22 CGCATTTGTTCAAGCCTAAACC AY078075 ###### Estimates of key statistics for evaluating the performance of IRAP and ITS-RFLP markers in 39 isolates of *P. graminea*. Component IRAP ITS --------------------------------- --------------------- --------------------- Nr. of assay units 5 (primer comb.) 6 (enzymes) Total nr. of bands 534 354 Polymorphic bands (percent) 454 (85%) 274% (0.774) Percent polymorphic loci (β) 94% 0.89% PIC^\*^ (min; average; max) 0.139; 0.376; 0.500 0.289; 0.355; 0.500 Nr. of loci PIC \> 0.3 27 14 Mean effective allele number 1.669 ± 0.357 1.639 ± 0.378 Multiplex Ratio (MR) 7.8 4.75 Effective Multiplex Ratio (EMR) 6.4 4.25 Marker index (MI) 2.41 1.50 Effective Marker Index (EMI) 0.338 0.255 Gen. simil. (min; average; max) 0.111; 0.341; 0.857 0.118; 0.460; 0.900 Quality nature of data (QND) 0.093 0.169 ^A^Value considering only polymorphic markers.

1. Introduction {#sec1} =============== Degenerative defects in articular cartilage or cartilage-like tissues, such as disc nucleus pulposus, are a significant cause of morbidity and socioeconomic burden especially in the context of an active ageing population. While cellular repopulation in replenishing and regenerating the cartilaginous matrix has been established in the literature \[[@B1]\], there has been a paradigm shift in recent years, focusing on the role of primary cells or predifferentiated cells in the absence of growth factors that can maintain their phenotype in vivo \[[@B2], [@B3]\]. For example, proposed therapies for intervertebral disc (IVD) regeneration include ADCT or autologous disc cell transplantation \[[@B4]\] and second generation NuQU using allogeneic, juvenile chondrocyte transplantation delivered in an injectable fibrin formulation \[[@B5]\]. MACI or Matrix-Induced Autologous Chondrocyte Implantation is a two-step procedure involving the isolation, culture expansion, and implantation of autologous chondrocytes on a membrane or scaffold for articular cartilage repair \[[@B6]\]. A crucial step in these approaches is cell isolation, usually obtained through mechanical and enzymatic breakdown of a tissue biopsy and subsequent laboratory expansion in cell processing facilities. In engineering appropriate constructs using primary cells, the need for large populations of viable chondrocytes has been a significant challenge. Cartilage is a relatively acellular tissue with only 5--10% of its volume consisting of chondrocytes \[[@B7]\]. In vivo, these cells reside within a pericellular matrix as chondrons \[[@B8]\], surrounded by dense extracellular matrix (ECM) consisting of collagens and proteoglycans. Cell yield from a cartilage digest is typically lower than 20% and is highly variable between donors and user competency \[[@B32]\]. Despite this, a high cell density is critical for maximising chondrogenesis \[[@B9]\] and remains a pertinent issue in cartilage regeneration. In order to reconcile the low cell yield with high cell number requirements for chondrogenesis, in vitro expansion or passaging has been employed. While costly, labour intensive, and time consuming, chondrocytes can undergo a process of dedifferentiation, increasing the relative collagen type I/collagen type II production \[[@B10]\] which may negatively impact capacity for successful cartilage regeneration \[[@B11], [@B12]\]. This poses a significant limitation in existing regenerative therapeutic strategies using culture expanded chondrocytic cell populations. Optimization of chondrocyte isolation is essential to enable further development of primary cell-based approaches. Limited work has been performed in this area and researchers have primarily investigated combinations of enzymatic regimes, multistep isolations, concentrations, and incubation times with different protocols \[[@B7]\] to improve cell yields. Previous work has investigated the role of perfusion systems in physical agitation to augment cell viability in chondrocyte isolation protocols but the role of these strategies in improving enzyme exposure is lacking \[[@B13]\]. When considering factors in combination, Oseni et al. investigated the necessity of a predigest phase in multistep approaches of chondrocyte isolation and found that it served no benefit in increasing the number of viable cells \[[@B7]\]. In manipulating the enzyme exposure in terms of concentration and incubation time, the breakdown of dense ECM which occurs gradually with time gives rise to the released chondrocytes being exposed to harsh enzymes for prolonged periods of time \[[@B14]\]. This reduces not only the final cell number, but also the viability and subsequent proliferative capacity of the cells \[[@B7]\]. While the relationship between specific digestion conditions and functional characteristics of isolated chondrocytes such as adhesion, proliferation kinetics, cell phenotype, and chondrogenic potential has been studied in rabbits, pigs, and ovine models \[[@B13]\], comprehensive characterization of matrix forming capacity is lacking in the literature. Alternative chondrocyte tissue sources have also been explored, such as those from the human ear \[[@B15], [@B16]\], nose \[[@B17]--[@B20]\], and rib cartilage \[[@B21], [@B22]\], each demonstrating varying cell yields in line with differences in cellularity of these tissues. In particular, human nasal chondrocytes have been considered as a clinically relevant source for cartilage engineering due to the high cellularity content and regenerative potential in terms of proliferative and synthetic capacities in biochemically distinct environments from their own such as joint and disc \[[@B20]--[@B26]\]. The overall objective of this study was to evaluate the effect of enzyme exposure, incubation time, and additional physical agitation cycles in optimizing chondrocyte isolation from nasal cartilage biopsies using the commonly employed collagenase enzyme. Cell yield, viability, morphology, proliferation kinetics, and subsequent matrix elaboration were evaluated for the different protocol groups. In investigating the scope for cartilage regeneration using these protocols, we focused on the effect of enzyme exposure (concentration and time) on the subsequent chondrogenic potential of nasal chondrocytes. 2. Methods {#sec2} ========== 2.1. Isolation of Nasal Septal Chondrocytes from Bovine Tissue and Monolayer Expansion {#sec2.1} -------------------------------------------------------------------------------------- Bovine nasal septa were obtained from a local abattoir within 12 h of sacrifice. Biopsies of nasal cartilage (NC) were harvested ([Figure 1(a)](#fig1){ref-type="fig"}) and washed with phosphate buffered saline (PBS) and minced ([Figure 1(b)](#fig1){ref-type="fig"}). For cell isolation, minced tissue was digested with concentrations of 750 U/ml or 3000 U/ml collagenase type II (190 U/mg, Gibco, Ireland) at a ratio of 10 ml/g of cartilage tissue for 1 h or 12 h under constant rotation at 37°C in serum-free low-glucose Dulbecco\'s modified eagles medium (LG-DMEM, 1 mg/mL D Glucose, 200 mM L-Glutamine;) containing antibiotic/antimycotic (100 U/ml penicillin, 100 mg/ml streptomycin) (all Gibco, Invitrogen) and amphotericin B (0.25 mg/ml, Sigma-Aldrich). The digest was subjected to physical agitation cycles at the start, after 30 min, and at the end of the incubation period using the Gentlemacs tissue dissociator (Miltenyi Biotech) ([Figure 1(c)](#fig1){ref-type="fig"}). Digested tissue/cell suspensions were passed through a 40 *µ*m cell strainer to remove tissue debris and washed three times by repeated centrifugation at 650*g* for 5 min. Cell yield and viability were determined with a hemocytometer and trypan blue exclusion. Cells were seeded at an initial density of 5 × 10^3^ cells/cm^2^ in T-175 flasks in LG-DMEM supplemented with 10% foetal bovine serum (FBS), penicillin (100 U/ml), streptomycin (100 mg/ml), and amphotericin B (0.25 mg/ml, Sigma-Aldrich). Cultures were expanded to passage one (P1) (7 d from initial isolation) in a humidified atmosphere at 37°C and 5% CO~2~. 2.2. Proliferation Kinetics and Cell Imaging {#sec2.2} -------------------------------------------- When subconfluent (\~80%), cells were detached by treatment with 0.05% trypsin/0.53 mM ethylenediaminetetraacetic acid (EDTA) and counted using trypan blue exclusion. The number of cell doublings during the expansion phase was determined as the logarithm (base 2) of the fold increase in the number of cells during expansion. The population doubling time was defined as the culture expansion time divided by the number of doublings during the expansion phase \[[@B27]\]. Cells from the various isolation regimes were plated in 6-well culture plates at a seeding density of 5 × 10^3^ cells/well and cultured for 7 days. Wells were subsequently washed in PBS, fixed in 4% paraformaldehyde (PFA), and stained with hematoxylin and eosin (H&E), 2% crystal violet, or DAPI/F-Actin to assess cellular morphology and cytoskeletal filament structure. 2.3. Alginate Bead Encapsulation and Culture {#sec2.3} -------------------------------------------- Monolayer expanded cells were trypsinised and counted using trypan blue staining and encapsulated in 1.5% alginate (Pronova UP LVG; FMC NovaMatrix, Sandvika, Norway) at a density of 4 × 10^6^ cells/ml. The alginate/cell suspension was passed through a 12 G needle and crosslinked in 102 mM calcium chloride (CaCl~2~) to produce beads (Ø 5 mm). Beads were allowed to ionically crosslink for 20 min and subsequently transferred to 24-well plates with one bead per well and 2 ml of chemically defined medium (CDM) at 37°C with 5% CO~2~ under low oxygen (5% O~2~) conditions. CDM consisted of LG-DMEM supplemented with penicillin (100 U/ml), streptomycin (100 *µ*g/ml), 0.25 *µ*g/ml amphotericin B, 40 *µ*g/ml L-proline, 1.5 mg/ml bovine serum albumin, 4.7 *µ*g/ml linoleic acid, 1x insulin-transferrin-selenium, 50 *µ*g/ml L-ascorbic acid-2-phosphate, and 100 nM dexamethasone (all Sigma-Aldrich) with TGF-*β*3 (10 ng/ml, PeproTech, UK) supplementation. Beads were assessed at days 0 and 21 in terms of cell viability, biochemical content, and histological analysis. 2.4. Assessment of Cell Viability {#sec2.4} --------------------------------- Cell viability was assessed using a viability/cytotoxicity assay kit (LIVE/DEAD®, Invitrogen, Ireland). Briefly, constructs were cut in half and washed in phenol-free DMEM (Sigma-Aldrich, Dublin, Ireland) followed by incubation in phenol-free DMEM containing 2 *μ*M calcein AM (live, intact cell membrane) and 4 *μ*M ethidium homodimer-1 (dead, disrupted cell membrane). Sections were again washed in phenol-free DMEM, imaged with an Olympus FV-1000 Point-Scanning Confocal Microscope at 515 and 615 nm channels, and analysed using FV10-ASW 2.0 Viewer software. Quantitative analysis of cell density (per cm^2^) was determined using ImageJ software (National Institutes of Health, Bethesda, Maryland, USA) in both peripheral and core regions of the constructs and averaged for four regions. 2.5. Quantitative Biochemical Analysis {#sec2.5} -------------------------------------- On removal from culture, wet weight of the samples was recorded and constructs were frozen at −85°C for further analysis. Samples were digested with 125 *μ*g/ml papain in 0.1 M sodium acetate, 5 mM L-cysteine-HCl, 0.05 M EDTA, and pH 6.0 (all from Sigma-Aldrich) at 60°C under constant rotation for 18 h followed by an additional incubation with 1 M sodium citrate under constant rotation for 1 h to disrupt the alginate calcium crosslinks. DNA content was determined using the Hoechst 33258 dye-based assay (DNA QF kit, Sigma-Aldrich, Ireland) with a calf thymus DNA standard. Proteoglycan (sulphated glycosaminoglycan, sGAG) content was quantified using the dimethylmethylene blue dye-binding assay (Blyscan, Biocolor Ltd., Northern Ireland), with a chondroitin sulphate standard. Total collagen was determined by measuring the hydroxyproline content. Samples were hydrolysed at 110°C for 18 h in 38% HCl and assayed using a chloramine-T assay \[[@B28]\] with a hydroxyproline:collagen ratio of 1 : 7.69 \[[@B29]\]. 2.6. Histological Analysis {#sec2.6} -------------------------- Beads were removed from culture, washed in PBS, and fixed in 4% PFA solution containing sodium cacodylate barium chloride (0.05 M) buffer overnight at 4°C. After removing the fixative and washing, samples were gradually dehydrated through 30--100% ethanol series with a final xylene change, before embedding in wax. Sections of 8 *μ*m were obtained with a microtome (Leica RM2125RT, Ireland) and affixed to microscope slides (Polylysine™, VWR, Ireland). Prior to staining, sections were dewaxed and rehydrated in 100% to 70% ethanol baths followed by distilled water. Cellular colonization and matrix deposition were observed using hematoxylin and eosin (H&E), sGAG deposition was evaluated using aldehyde fuchsin and 1% alcian blue 8GX in 0.1 M HCl, and collagen distribution was assessed using picrosirius red (all from Sigma-Aldrich, Ireland). Semiquantitative analysis of percentage (%) chondron in constructs was determined using ImageJ software (National Institutes of Health, Bethesda, Maryland, USA). 2.7. Pellet Culture Assay {#sec2.7} ------------------------- To compare freshly isolated and culture expanded chondrocytes a pellet culture model was employed. Briefly, 250,000 cells of fresh and culture expanded cells isolated using the 1 h, 3000 U/ml enzyme protocol were placed in a 1.5 ml conical microtube and centrifuged at 650*g* for 5 minutes. The pellets were cultured in low-glucose chondrogenic media without additional growth factor supplementation. For histological evaluation the pellets were embedded in paraffin, cut into 5 *µ*m thick sections, and stained with 1% alcian blue 8GX (Sigma-Aldrich, Ireland) in 0.1 M HCl to assess glycosaminoglycan (GAG) content and picrosirius red to detect collagen. Subsequent biochemical analysis was carried out to quantify GAG and collagen content as outlined above. 2.8. Statistical Analysis {#sec2.8} ------------------------- Statistical analysis was performed using GraphPad Prism (version 5) software with 3-4 samples analysed for each experimental group. One-way ANOVA was used for analysis of variance with Bonferroni\'s posttests to compare between groups. Results are displayed as mean ± standard deviation. Significance was accepted at a level of *p* \< 0.05. The entire experiment was replicated independently with tissues from two additional donors which confirmed the findings presented in this manuscript. 3. Results {#sec3} ========== 3.1. Effect of Physical Agitation in Improving Cell Yield {#sec3.1} --------------------------------------------------------- For a standard chondrocyte isolation protocol employing 750 U/ml collagenase type II, physical agitation was found to significantly increase cell yield ([Figure 1(d)](#fig1){ref-type="fig"}), with an almost fivefold increase after 1 h, compared with a twofold increase for 12 h exposure (*p* \< 0.001). Both increased enzyme incubation time and physical agitation were found to reduce viability by approximately 6% (without (w/o) physical agitation: 1 h = 95.0 ± 1.3%, 12 h = 89.2 ±1.8%, with (w/) physical agitation: 1 h = 89.5 ± 2.4%, 12 h = 82.3 ± 1.6%). 3.2. Rapid Isolation and Characterization {#sec3.2} ----------------------------------------- All further experiments utilised physical agitation to determine the effect of enzyme concentration (750 and 3000 U/ml) exposure for incubation times of 1 h and 12 h. For 3000 U/ml of enzyme, the cell yield ([Figure 2(a)](#fig2){ref-type="fig"}) at 1 h was found to be similar to the 12 h digest with 750 U/ml of enzyme (*p* \< 0.0001), with just over 1 million cells per gram of cartilage obtained. Minor changes in cell viability were observed for both increased incubation time and enzyme concentration exposure ([Figure 2(b)](#fig2){ref-type="fig"}). While there was a significant increase in cell yield at 3000 U/ml for a 12 h incubation time ([Figure 2(a)](#fig2){ref-type="fig"}), there was a concomitant reduction in cell viability ([Figure 2(b)](#fig2){ref-type="fig"}) (*p* \< 0.0001). A 750 U/ml digest for 1 h yielded half the number of cells (*p* \< 0.001) when compared with 3000 U/ml of enzyme for the same incubation time. Further, when assessing the proliferation kinetics in terms of population doubling time ([Figure 2(c)](#fig2){ref-type="fig"}), cells isolated within 1 h at 750 U/ml were found to exhibit significantly slower doubling kinetics, almost threefold slower compared with both the 1 h, 3000 U/ml and 12 h, 750 U/ml digest groups. The 12 h, 3000 U/ml group also exhibited slower proliferation kinetics (\~2-fold) (*p* \< 0.001). On evaluation of cell morphology with crystal violet, H&E, and fluorescence DAPI/F-Actin staining, diminished proliferative capacity was observed for 1 h, 750 U/ml and 12 h, 3000 U/ml groups ([Figure 2(d)](#fig2){ref-type="fig"}). 3.3. Cell Proliferation, Morphology, and Matrix Forming Capacity for Different Isolation Protocols {#sec3.3} -------------------------------------------------------------------------------------------------- The trends in proliferation kinetics observed in 2D culture were also maintained in 3D alginate constructs. DNA content increased in all groups compared to day 0, with the 1 h, 3000 U/ml group exhibiting the highest DNA content (*p* \< 0.001), almost twofold higher than the lower enzyme concentration (1 h, 750 U/ml) and increased temporal exposure (12 h, 3000 U/ml) groups (*p* \< 0.001) ([Figure 3(a)](#fig3){ref-type="fig"}). Cell viability after 21 days was found to be dependent on enzyme incubation period with more homogenous viable cell distribution observed for cells isolated after 1 h incubation for both enzyme concentrations compared to 12 h incubation ([Figure 3(b)](#fig3){ref-type="fig"}). For cells isolated after a 12 h incubation period, a higher degree of inhomogeneity was observed in the cellular density between peripheral and core regions, with higher cell densities observed in the periphery ([Figure 3(c)](#fig3){ref-type="fig"}). Enzyme concentration exposure was also observed to have an effect on cellular distribution but to a lesser extent compared to incubation period. Cells isolated in a shorter incubation time maintained a chondron morphology compared to a single cell morphology observed with higher enzyme concentration and exposure time. Higher enzyme exposure was observed to correlate with less intense eosin staining in the pericellular matrix (PCM) indicating a reduction in PCM density ([Figure 3(d)](#fig3){ref-type="fig"}). This was also observed through semiquantitative analysis, with the highest percentage chondron being retained for the 1 h incubation and 750 U/ml enzyme concentration ([Figure 3(e)](#fig3){ref-type="fig"}). 3.4. Sulphated Glycosaminoglycan (sGAG) Accumulation for Different Isolation Protocols {#sec3.4} -------------------------------------------------------------------------------------- Having assessed the effect of enzyme exposure on proliferation, the matrix forming capacity was subsequently evaluated in terms of sGAG and collagen accumulation which are key constituents of cartilaginous tissues. An important difference to note in the histology at day 0 is that the 1 h 750 U/ml exposure group exhibited more intense staining for sGAG ([Figure 4(a)](#fig4){ref-type="fig"}), reflecting higher baselines sGAG (twofold) at the start of 3D culture as corroborated by the biochemical findings (Figures [4(b)](#fig4){ref-type="fig"} and [4(c)](#fig4){ref-type="fig"}). For both enzyme concentrations, a 1 h incubation period was found to support significantly higher sGAG accumulation compared to 12 h exposure groups ([Figure 4(a)](#fig4){ref-type="fig"}). These observations were corroborated by the biochemical analysis in terms of sGAG (% w/w) ([Figure 4(b)](#fig4){ref-type="fig"}) and sGAG/DNA ([Figure 4(c)](#fig4){ref-type="fig"}), where a significant reduction in sGAG synthesis was observed with increased incubation time. 3.5. Collagen Accumulation for Different Isolation Protocols {#sec3.5} ------------------------------------------------------------ In terms of collagen accumulation, more intense histological staining was observed for shorter incubation period groups ([Figure 5(a)](#fig5){ref-type="fig"}). This was corroborated by biochemical data for both collagen (% w/w) ([Figure 5(b)](#fig5){ref-type="fig"}) and Collagen/DNA ([Figure 5(c)](#fig5){ref-type="fig"}) with significantly higher amounts of collagen, almost twofold, for shorter enzyme incubation time groups. Overall, a trend towards decreasing collagen matrix capacity is also observed with increasing enzyme exposure, with greater differences observed for increased exposure time. 3.6. Comparison of Freshly Isolated and Culture Expanded Chondrocytes {#sec3.6} --------------------------------------------------------------------- Having identified that a 1 h, 3000 U/ml isolation protocol was beneficial, we next sought to compare the proliferative and matrix forming capacity of both freshly isolated and culture expanded cells in a pellet culture model system. Freshly isolated cells were found to have a higher proliferative and matrix forming capacity ([Figure 6](#fig6){ref-type="fig"}) compared to culture expanded cells, with increased DNA content ([Figure 6(a)](#fig6){ref-type="fig"}), GAG, and collagen deposition observed (Figures [6(b)](#fig6){ref-type="fig"}--[6(d)](#fig6){ref-type="fig"}). 4. Discussion {#sec4} ============= In the treatment of cartilaginous defects, large populations of cells are needed for optimal chondrogenesis \[[@B9]\]. Optimizing cell isolation is a key challenge in terms of obtaining adequate cell numbers while maintaining a vibrant cell population capable of subsequent proliferation and matrix elaboration. In light of previous literature findings \[[@B7]\], the overall objective of this study was to optimize chondrocyte isolation by modulating collagenase enzyme exposure in terms of concentration and time combined with physical agitation cycles. The second objective was to evaluate the effects of enzyme exposure on subsequent cell viability and matrix forming capacity. Overall, increasing enzyme exposure time was found to be more detrimental than collagenase concentration for subsequent viability, proliferation, and matrix forming capacity (sGAG and collagen) of these cells resulting in nonuniform cartilaginous matrix deposition. Taken together, the results indicate that a 3000 U/ml collagenase digest for 1 h using physical agitation cycles can be applied as a clinically translatable protocol for isolation of chondrocytes to achieve adequate cell numbers. The majority of collagenase enzyme concentrations utilised for cell isolation protocols are reported in terms of mg/ml or percentage weight per volume (% w/v) with typical values quoted in literature ranging from 0.08 to 0.3 (% w/v) \[[@B7]\]. For comparison purposes, based on the batch of collagenase used in this work (190 U/mg), 750 U/ml represents 0.4% w/v and 3000 U/ml represents 1.6% w/v. While percentage weight per volume is based on physical characteristics that are easily determined, a unit of activity is a measure of the biochemical function of the enzyme. As such, a unit of activity per gram varies for different types of collagenase or different lots of the same collagenase and can easily change over time. This inconsistency in reporting enzyme concentrations and enzyme solution to tissue mass ratios could account for much of the reported heterogeneity in isolation protocols in the literature and we advocate for consistent reporting in terms of units of enzyme in this regard. Previous studies have investigated temperature modulation \[[@B30]\], human serum supplementation \[[@B31]\], and the use of ascorbic acid and NaCl in perfusion bioreactor systems to enhance cell isolation protocols \[[@B13]\]. While these approaches are highly innovative and could add significantly to advances in GMP biomanufacturing, for large scale isolation and tissue engineering approaches, short and simple protocols are desirable for clinical translation. In considering biocompatible collagenase concentrations and minimal incubation time, physical agitation and surface area of exposure become important factors in the rapid isolation of chondrocytes. Enhancing surface area through optimal mincing and tissue breakdown can dramatically improve enzymatic action to yield similar, if not superior, cell yields \[[@B32]\]. Physical agitation in a cyclical fashion was shown to improve cell yield through improved tissue exposure to enzyme and increased digestion in line with the pursuits of perfusion culture systems as proposed by Centola et al. (2015) \[[@B13]\]. In this work, for a 750 U/ml and 1 h enzyme exposure, incomplete cell release and preservation of the chondron structure were observed resulting in lower cell yields and longer population doubling times but with superior matrix forming capacity. It is clear that balancing cell yield with viability and proliferative and subsequent matrix forming capacity specific to tissue reconstitution is key to developing optimal cell isolation protocols. Furthermore, the improved cell viability with reduced enzyme exposure time is reflected in more uniform cell viability and matrix formation. When isolated cell populations were cultured in alginate beads, clear differences were observed between groups. Specifically, for cells subjected to longer incubation times (12 h) distinct differences in peripheral rim and core cell densities were observed, which were not as pronounced for the 1 h isolation protocol. Bos et al. (2002) demonstrated that with progressive breakdown of ECM, there is increased direct cellular exposure to enzyme which can be damaging \[[@B33]\]. This was observed in the changes to surrounding pericellular matrix (PCM), gradual attrition, and release of single cells with increased exposure. PCM preservation was seen to curb proliferation kinetics in the temporally less intensive enzyme regimes. We have shown here also the preservation of chondron structure in baseline constructs at day 0. PCM plays a key role in modulating the interactions of cells with the surrounding environment \[[@B34]\] and proliferative and synthetic responses in signaling \[[@B35], [@B36]\]. In this context, PCM plays a key role in signaling and regulation of matrix molecules \[[@B34], [@B37], [@B38]\]. This modulation results in lower metabolic demands \[[@B14]\] of these rapidly isolated cells and perhaps explains resulting homogenous matrix distribution. When considered in the context of tissue engineering or regeneration strategies, lower metabolic demands are perhaps more desirable due to compromised nutrition at the site of damage to be treated, thus making rapidly isolated cells with an intact PCM attractive for clinical translation. Technologies such as Carti-One™ (Orteq® Ltd., United Kingdom) are currently exploiting novel intraoperative point of care (POC) cell and tissue processing. This approach allows for single staged surgery with scope for autologous tissue combined with a carrier to be delivered arthroscopically for improved repair of cartilage defects \[[@B39]\]. While there remain limitations, this approach highlights the role of such translatable protocols in facilitating regenerative ventures using primary cells. Further, as shown in this work, minimizing duration of enzyme exposure in a rapid isolation protocol can retain subsequent matrix forming potential. The authors chose to work with nasal derived chondrocytes which have been proposed in the literature as an alternative primary cell source with the potential for low morbidity procurement, improved proliferation, and matrix forming capacity in cartilage regeneration \[[@B21], [@B22]\]. It is well established that culture expansion of chondrocytes results in changes in proliferative characteristics, matrix synthesis, and loss in expression of differentiation markers, termed "dedifferentiation" \[[@B40]--[@B42]\]. While it would have been ideal to work with fresh nasal cells for the entire study, expansion to passage 1 (P1) was necessary to obtain adequate cell numbers to demonstrate cell proliferative, live/dead characteristics, and matrix forming capacity for the various isolation protocols investigated. Having identified that a 1 h, 3000 U/ml isolation protocol was beneficial we therefore compared fresh versus culture expanded cells in a pellet culture model. In pellet culture, freshly isolated cells were found to have a higher proliferative and matrix forming capacity compared to culture expanded cells, with increased DNA content, GAG, and collagen deposition observed, further demonstrating the benefit of employing freshly isolated cells with short isolation protocols. Cell-based medicinal products (CBMPs) follow EU legislation applicable for advanced therapy medicinal products (ATMPs) \[[@B43]\] with the technical requirements defined in Directive 2009/120/EC guided by the European Medicines Agency (EMA) and committee for advanced therapies (CAT) \[[@B44]\]. The present position of CAT considers clinical application of donor cells isolated from a different anatomical site to recipient site as "nonhomologous use" (ie., the cells or tissues are not intended to be used for the same essential function or functions) and should be classified as an ATMP requiring approval and regulation by the EMA \[[@B44], [@B45]\]. Furthermore, whereby enzymatic treatment is aimed at isolating or separating cells (which typically leads to a cell suspension with altered cell structure and functionality relative to the intact native tissue), this is considered a substantial manipulation \[[@B43]\] and would also require regulation as an ATMP. The development of CBMPs for clinical translation is still in its infancy and it is evident that the legislation is complex and continuously evolving with scientific advances and understanding. As the field of regenerative medicine matures and products reach commercialisation it is envisaged that the regulatory landscape may change or adapt with experience. Future endeavours should aim at consolidating cyclical physical agitation cycles in both mincing and perfusion and modulating enzyme exposure with short incubation to yield a practical translatable protocol. Automation in a single contained unit aimed at intraoperative processing may facilitate clinically translatable strategies using chondrocytes. It should be cautioned however that further investment in these areas will be dictated by the regulatory landscape where FDA and EMA approval of point of care (POC) devices for cell isolation and intraoperative use of enzymes is necessary to apply rapid isolation of cells for use in single step approaches. 5. Conclusion {#sec5} ============= We recommend a 3000 U/ml collagenase digest for 1 h at a ratio of 10 ml/g of cartilage tissue with physical agitation cycles using a tissue dissociator device as a translatable protocol for intraoperative cell isolation (1−1.5 × 10^6^ cells per g of cartilage) applications. Subsequent culture of these rapidly isolated cell populations demonstrated superior proliferation kinetics, more robust matrix synthesis, and uniform matrix forming capacity. Automation of such a protocol in a single unit could facilitate single step, clinically translatable intraoperative regenerative strategies using chondrocytes for cartilage repair. This work was supported by the Trinity College Dublin Research Capacity Building Grant- Pathfinder and Science Foundation Ireland Career Development Award (15/CDA/3476). The authors would also like to acknowledge Servier Medical Art (<http://www.servier.com/>) for their image bank used to produce [Figure 1](#fig1){ref-type="fig"}. Image of Bovine Nasal Anatomy in [Figure 1](#fig1){ref-type="fig"} was inspired by an illustration in Bovine Anatomy: An Illustrated Text \[[@B46]\]. Competing Interests =================== The authors declare that no competing interests exist. Authors\' Contributions ======================= Srujana Vedicherla performed all experimental work and analysis, interpretation of data, and preparation of manuscript. Conor Timothy Buckley was responsible for overall experimental design, data interpretation, preparation, and editing of manuscript. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} [^1]: Academic Editor: Kibret Mequanint

Background {#Sec1} ========== Accompanied by economic development and lifestyle changes, the prevalence of childhood obesity has increased rapidly worldwide, leading to obesity-related metabolic diseases in adulthood, such as non-alcoholic fatty liver disease, type 2 diabetes, and cardiovascular disease \[[@CR1], [@CR2]\]. Indeed, the unhealthy lifestyle and academic demands of children are increasingly interfering with biological rhythms, which might contribute to childhood obesity and negative health outcomes \[[@CR3], [@CR4]\]. Therefore, it is essential to identify novel contributors to the underlying physiology of childhood obesity. Cortisol is a primary product of the hypothalamic-pituitary-adrenal (HPA) axis and acts as the terminal effector of this axis on other systems \[[@CR5]\]. In both human and animal models, cortisol has been causally demonstrated to promote fat accumulation and weight gain as well as glucose homeostasis and lipid metabolism \[[@CR6], [@CR7]\]. Considering that the production, secretion and abundance of cortisol are regulated in a robust time-of-day-dependent manner \[[@CR8]\], the diurnal cortisol rhythm is a good indicator for comprehensive evaluation of HPA axis activity. Cortisol rhythms are believed to be established between 2 and 9 months in early life \[[@CR9]\], as mediated by a combination of influences such as the light-dark cycle, pubertal development, feeding, sleep, and physical activity \[[@CR10]\]. Under non-stress conditions, the secretion and release of cortisol follows a typical circadian rhythm: cortisol rapidly increases 30 to 40 min after awakening, followed by a sharp decline during the next few hours and a gradually decline during the remainder of the day until reaching the lowest level at midnight \[[@CR11], [@CR12]\]. An interaction between the HPA axis and obesity has long been proposed. On the one hand, cortisol controls body weight via effects on both food intake and energy expenditure as well as adipogenic pathways in abdominal adipose tissue \[[@CR13]\]. On the other hand, obesity constitutes a chronic stressor and in turn alters the activity of the stress axis \[[@CR14]\]. Recently, it has been proposed that obesity is associated with circadian disruption and often accompanied by altered HPA axis rhythmicity \[[@CR5]\]. Adults with obesity usually display blunted diurnal HPA axis functioning, which manifests as decreased cortisol variability, lower morning levels, or a smaller change in cortisol throughout the day \[[@CR15]--[@CR17]\]. However, studies of diurnal cortisol patterns in childhood obesity have resulted in different findings. For example, Kjolhede reported that average salivary cortisol levels throughout the day were significantly lower in children with obesity \[[@CR18]\], whereas Hillman showed that with an increasing degree of adiposity in adolescent girls, there may be reduced serum cortisol levels during the day and increased levels at night \[[@CR14]\]. Conversely, another study reported no significant association between the HPA axis and percent body fat in pre-pubertal children with obesity \[[@CR19]\]. A potential explanation for these variable findings is with regard to methodological differences such as different measurement methods (e.g., enzyme immunoassay, radioimmunoassay, chemiluminescence immunoassay) and sampling time. Considering the characteristics of children's growth, regulation of the HPA axis in children may be affected by more complex factors than those in adults, such as age, pubertal development and stress-related activities (dietary consumption, physical activities, etc.). Therefore, exploring the factors influencing children's HPA axis may help in reaching a complete understanding of the links between dysregulation of the HPA axis and childhood obesity. In the present study, we explored the characteristics of diurnal cortisol rhythm in children and adolescents with obesity by repeated sampling of salivary cortisol over the course of a day. Moreover, we examined relationships of cortisol activity with the degree of adiposity, pubertal stage and physical activity. This information may contribute to our understanding of the associations between chronodisruption, obesity and lifestyle to provide new insight for the primary prevention of childhood obesity. Methods {#Sec2} ======= Participants {#Sec3} ------------ In this cross-sectional study, a total of 57 children and adolescents aged 6--15 years were recruited from the Department of Endocrinology and Child Health Care of Children's Hospital of Nanjing Medical University from July 2018 to June 2019. According to WHO standards \[[@CR20]\], the subjects were divided into a normal weight group (− 2 \< BMI z-score \< 1) and an obesity group (BMI z-score \> 2). The exclusion criteria were as follows: (1) a history of chronic diseases (except obesity), such as epilepsy, diabetes, hypothyroidism, tumours, mental illness, precocious puberty or short stature; (2) use of exogenous steroids in the past 3 months; (3) a history of surgery, trauma or other stress events in the past 3 months; (4) use of a medication known to affect hormones; or (5) female menstrual period. The study was approved by the Children's Hospital of Nanjing Medical University Ethical Committee. Prior to inclusion in the study, the parents provided written informed consent. Measures {#Sec4} -------- ### Anthropometric measures {#Sec5} All subjects fasted for 12 h overnight and emptied urine and stool prior to measurements. Body composition was determined using the bioelectrical impedance method (Inbody J20, Biospace, Korea), including body fat mass, fat mass percentage (FM%) and skeletal muscle mass. According to a standard protocol, height and weight were measured by experienced researchers with precisions of 0.1 cm and 0.1 kg, respectively. BMI was calculated as weight (in kilograms) divided by the square of height (in metres). Because children's BMI varies with age and sex, BMI was converted to BMI z-score according to the World Health Organization's Child Growth Standards (2006). Waist circumference (WC) was measured in centimetres to the nearest 0.1 cm. The waist-to-height ratio (WHtR) was calculated as WC (in centimetres) divided by height (in centimetres). ### Pubertal stage {#Sec6} Professional paediatricians performed visual inspection and palpation to determine pubertal stage. Females were matched for breasts and pubic hair and males for genitalia and pubic hair \[[@CR21]\]. The stage of pubertal development (I-V period) was assessed according to Tanner staging criteria, with Tanner II as the hallmark of puberty initiation. For analysis of different degrees of pubertal development, the Tanner stage was categorized into three levels: pre-pubertal (Tanner I), early pubertal (Tanner II and III) and late pubertal (Tanner IV and V) \[[@CR21]\]. ### Salivary cortisol analysis {#Sec7} Salivary cortisol reflects the levels of biologically active, non-protein-bound cortisol in serum and follows the circadian variation in serum cortisol \[[@CR22]\]. Salivary cortisol correlates strongly with plasma cortisol \[[@CR23]\] and is less prone to variability due to changes in cortisol-binding proteins \[[@CR24]\]. Due to its easy, non-invasive collection and convenient transportation and storage, salivary cortisol is widely used for paediatric research. Salivary samples were collected at 8:00, 16:00 and 23:00 h in a quiet state after a fast of 4 h. A commercial Salivette® (SARSTEDT AG &Co, Germany) tube containing a cotton wool swab was used to collect saliva. The swab was rotated in the mouth for at least 5 min and inserted back into the tube. The cortisol samples, which are stable at room temperature for a number of days,\[[@CR23]\] were centrifuged at 1500 rpm for 5 min within 24 h to obtain clear saliva with low viscosity, and 500 μL of saliva was pipetted into the EP tube with a micropipette. The saliva samples were stored at − 80 °C after being dispensed. Using an Elecsys reagent kit and a Cobas e immunoassay analyser (Roche Diagnostics GmbH, Germany), cortisol levels were determined by electrochemiluminescence immunoassay (ECLIA) with a high sensitivity of 0.054 ng/ml and intra- and inter-assay coefficients of variation below 10%. Areas under the curve relative to ground (AUCs) represent the total amount of cortisol exposure during the portions of the diurnal cortisol cycle by the trapezoidal method \[[@CR25]\]. The diurnal cortisol slope (DCS) is characterized as the decline in cortisol over the day and is calculated by the formula rise over run as the slope of the line from the first time point value to the last measured point \[[@CR26]\]. It has been proven that there is no difference between linear regression and rise over run formulas \[[@CR26]\]. Thus, we calculated HPA axis rhythm measures based on cortisol levels at 3 time points, AUC~08:00--16:00~, AUC~16:00--23:00~ and AUC~08:00--23:00~, as well as DCS. ### Assessment of glucose and lipid metabolism {#Sec8} Blood samples were taken at 8:00 after an overnight fast of 12 h to test fasting glucose (FG), fasting insulin (FI), total cholesterol (TC), triglycerides (TG) in the obesity group and part of the normal weight group. Insulin resistance was determined by the formula of the homeostasis model assessment of insulin resistance (HOMA-IR) = (\[fasting insulin (lU/mL) × fasting glucose (mmol/L)\]/ 22.5. Questionnaires for physical activities {#Sec9} -------------------------------------- Children's sleep parameters were collected by parental questionnaire. Parents reported children's bedtime and wake-up time on weekdays and weekends during the previous month. The average sleep duration was calculated by the following formula: (sleep duration on weekdays× 5 + sleep duration on weekends× 2)/7 \[[@CR27]\]. The Chinese Version of the Children's Leisure Activities Study Survey questionnaire was used to assess the physical activity of the children. The questionnaire was completed by the children with the assistance of their parents, and the reliability and validity of the Chinese version has been verified \[[@CR28]\]. A checklist of 31 physical activities and 13 sedentary behaviours was included in the questionnaire. According to the intensity of physical activity, there were 15 activities classified as vigorous-intensity physical activities (VPA, \> 6 METs) and 16 activities classified as moderate-intensity (MPA, 3--5.9 METs) \[[@CR28]\]. For data analysis, screen time consisted of 3 sedentary behaviours (SB, including watching TV or movie, playing computer games, surfing the internet or playing on the phone); the other 10 were considered non-screen sedentary behaviours. Statistical analyses {#Sec10} -------------------- IBM SPSS Statistics software (Version 24.0) was used, and the level of significance was accepted with *P* \< 0.05. The results are expressed as the means ± standard deviation or median \[interquartile range\]. The normality of data was evaluated using the Shapiro-Wilk test. Cortisol variables with a skewed distribution were logarithmically transformed for correlational analysis. Significant differences between the normal weight and obesity groups were analysed using t-tests or Mann--Whitney U-tests. Chi-square tests were applied to compare categorical variables between two groups. Differences in HPA axis measures among puberty groups were compared by analysis of variance (ANOVA) and analysis of covariance (ANCOVA). Multiple linear regressions were performed to assess the correlation of cortisol levels with different anthropometric variables and physical activities. Spearman's correlations were employed to assess the correlation of cortisol variables with testosterone, glucose or lipid metabolism in obese children . For analysis of 24-h movement, compositional data analysis was used following the guide of Chastin and colleagues \[[@CR29]\]. Four compositional linear regression models were conducted for each health indicator with each behaviour sequentially entered into the model via log--ratio transformation \[[@CR30]\]. Models examined the combined effect of the relative distribution of all movement behaviours with each health indicator \[[@CR29]\]. Model *P* values and *R*^2^ coefficients were the same across all 4 linear regression models. Next, models assessed the association between the time spent in each movement behaviour relative to the time spent in the other movement behaviours and each health indicator. The first coefficient and its *P* value for each rotated model were used to determine whether the individual movement behaviour was significantly positively or negatively associated with each health indicator relative to the time spent in the other movement behaviours \[[@CR31]\]. In summary, the compositional analysis is a multiple linear regression model where the cortisol measures were modelled as a function of sleep, screen time, non-screen time, and MVPA. Results {#Sec11} ======= Baseline characteristics {#Sec12} ------------------------ A total of 57 participants were enrolled in the study and divided into a normal weight group (*n* = 22) and an obesity group (*n* = 35) according to BMI. Demographic, anthropometric and behavioural characteristics are summarized in Table [1](#Tab1){ref-type="table"}. There were no differences between the obesity group and the normal weight group in terms of age, sex, pubertal stage, height, sleep duration or MVPA minutes. Table 1Characteristics of study participantsParameterAll (***n*** = 57)Obesity (***n*** = 35)Normal weight (***n*** = 22)***P*Age (years)**11.50 (9.35,12.85)11.80 (10.30,13.30)10.85 (8.98,12.13)0.213**Sex (boys)**41 (72%)26 (74%)15 (68%)0.618**Pubertalstage** **Pre-pubertal**19/5711/358/220.416 **Early pubertal**26/5715/3511/22 **Late pubertal**12/579/353/22**BMI z-score**2.41 (−0.01,3.48)3.21 (2.69,3.71)− 0.27(− 0.88,0.35)\< 0.001**FM (%)** **Boys**36.80 (20.10,43.50)40.15 (36.33,45.20)12.20 (9.40,21.20)\< 0.001 **Girls**28.85 (16.85,39.58)38.30 (33.35,45.95)16.70 (11.00,20.00)\< 0.001 **WHtR**0.59 (0.41,0.63)0.62 (0.59,0.66)0.41 (0.38,0.42)\< 0.001**Sleep duration (hours)**8.93 ± 0.868.78 ± 0.919.15 ± 0.750.129**MVPA (minutes)**71.43 (40.71,106.70)54.83 (34.67,103.80)77.86 (53.93,139.29)0.135**Non-screen SB (minutes)**240.79 ± 102.25210.28 ± 96.88286.56 ± 94.460.006**Screen time (minutes)**34.29 (11.40,120.00)51.40 (12.15,128.60)17.14 (3.21,45.21)0.016Data are reported as the median (interquartile range) and the Mann--Whitney U-test was used or the mean ± standard deviation and the t-test was used. *P*: obesity group vs normal weight group. *BMI z-score* body mass index z-score, *FM* fat mass, *WHtR* waist-to-height ratio, *MVPA* moderate to vigorous physical activity, *SB* sedentary behaviour Diurnal cortisol patterns {#Sec13} ------------------------- Table [2](#Tab2){ref-type="table"} reports descriptive statistics for HPA axis rhythmicity in all subjects, which showed peak cortisol levels in the morning and a nadir at midnight. Moreover, the children with obesity displayed lower cortisol levels at 08:00 (*P* = 0.030) and AUC~08:00--16:00~ (*P* = 0.027) and higher levels at 23:00 (*P* \< 0.001) than their normal weight counterparts. Figure [1](#Fig1){ref-type="fig"} depicts the variation in the diurnal cortisol curve from 08:00 to 23:00 based on BMI category, with notably flatter trajectories of circadian cortisol observed in the children with obesity. Table 2Descriptive statistics for HPA axis rhythmicityAll (***n*** = 57)Obesity (***n*** = 35)Normal weight (***n*** = 22)***P*Cortisol**~**08:00**~**(nmol/L)**6.23 (4.13,8.97)5.79 (3.42,7.73)8.44 (5.56,9.59)0.030**Cortisol**~**16:00**~ **(nmol/L)**2.28 (1.77,3.32)2.28 (1.72,3.18)2.27 (1.91,3.52)0.481**Cortisol**~**23:00**~ **(nmol/L)**0.59 (0.35,1.34)1.10 (0.48,1.46)0.40 (0.21,0.61)\< 0.001**AUC**~**08:00--16:00**~ **(nmol/L × h)**36.46 (25.35,48.15)34.84 (23.01,45.99)42.16 (32.12,52.04)0.027**AUC**~**16:00--23:00**~ **(nmol/L × h)**11.73 (8.43,14.83)11.90 (8.94,16.61)11.51 (7.81,14.07)0.263**AUC**~**08:00--23:00**~ **(nmol/L × h)**47.91 (36.31,61.56)44.93 (34.66,60.94)51.54 (42.75,66.87)0.098**DCS (nmol/L/h)**−0.35(−0.54,-0.24)−0.29(− 0.49,-0.14)− 0.52(− 0.63,-0.34)0.006Data are reported as the median (interquartile range), and the Mann--Whitney U-test was used. *P*: obesity group vs normal weight groupFig. 1Diurnal cortisol patterns in children with obesity and normal weight. Data are expressed as the mean ± SEM, and error bars show the standard error of the mean. Cortisol variables were logarithmically transformed Measures of the HPA axis and pubertal stage {#Sec14} ------------------------------------------- There were no significant correlations between HPA axis measures and sex or age. We then tested the hypothesis that cortisol AUC may be influenced by puberty, which was proposed in other studies \[[@CR32], [@CR33]\]. The AUC increased with pubertal development (AUC~08:00--16:00~:*P* = 0.008; AUC~08:00--23:00~: *P* = 0.005; ANOVA). After adjustments for BMI, the above relationships remained (AUC~08:00--16:00~: *P* = 0.002; AUC~08:00--23:00~: *P* = 0.002; ANCOVA), as shown in Fig. [2](#Fig2){ref-type="fig"}. Moreover, testosterone was positively related to AUC~08:00--16:00~ (*r* = 0.407, *P* = 0.023) and AUC~08:00--23:00~ (*r* = 0.443, *P* = 0.014) in children with obesity. Fig. 2Comparison of HPA axis measures among different puberty groups. Cortisol variables were logarithmically transformed. Box plots represent interquartile range with the symbol +, inside the box plot representing the mean score. a: *P*\<0.05 Pre-pubertal vs Late pubertal; b: *P*\<0.05 Early pubertal vs Late pubertal Measures of the HPA axis and anthropometry {#Sec15} ------------------------------------------ The results of multiple regression for associations between HPA axis measures and anthropometry in all participants are shown in Table [3](#Tab3){ref-type="table"}. Cortisol~08:00~ was inversely associated with BMI z-score (β = − 0.247, *P* = 0.036) and WHtR (β = − 0.295, *P* = 0.027). Cortisol~23:00~ was positively associated with BMI z-score (β = 0.490, *P*\<0.001), WHtR (β = 0.485, *P*\<0.001) and FM% (β = 0.464, *P*\<0.001), and AUC~08:00--16:00~ was inversely associated with BMI z-score (β = − 0.288, *P* = 0.033) and WHtR (β = − 0.316, *P* = 0.020). Absolute values of DCS were inversely associated with BMI z-score (β = − 0.350, *P* = 0.009), WHtR (β = − 0.384, *P* = 0.004) and FM% (β = − 0.322, *P* = 0.019). After adjustments for puberty, cortisol~08:00~ was inversely associated with BMI z-score (β = − 0.247, *P* = 0.048) and WHtR (β = − 0.271, *P* = 0.030). Cortisol~23:00~ was positively associated with BMI z-score (β = 0.454, *P*\<0.001), WHtR (β = 0.484, *P*\<0.001) and FM% (β = 0.451, *P*\<0.001), and absolute values of DCS were inversely associated with BMI z-score (β = − 0.327, *P* = 0.013), WHtR (β = − 0.366, *P* = 0.005) and FM% (β = − 0.313, *P* = 0.017). Table 3Associations between HPA axis rhythm index and anthropometryBMI z-scoreWHtRFM (%)β***P***β\****P\****β***P***β\****P\****β***P***β\****P\**Cortisol**~**08:00**~− 0.247**0.036**− 0.247**0.048**− 0.295**0.027**− 0.271**0.030**− 0.2380.080− 0.2210.078**Cortisol**~**16:00**~− 0.0660.631− 0 .0550.692− 0.0770.580− 0.0660.636− 0.0300.831−0.0160.907**Cortisol**~**23:00**~0.490**\<0.001**0.454**\<0.001**0.485**\<0.001**0.484**\<0.001**0.464**\<0.001**0.451**0.001AUC**~**08:00--16:00**~−0.288**0.033**−0.2140.092−0.316**0.020**−0.2460.052−0.2560.065−0.1880.161**AUC**~**16:00--23:00**~0.1570.2650.1920.1610.1880.2660.1920.1620.1810.2050.2290.092**AUC**~**08:00--23:00**~− 0.1970.161−0.1230.349− 0.2330.096−0.1620.215−0.1700.234−0.0920.483**\|DCS\|**−0.350**0.009**−0.327**0.013**−0.384**0.004**−0.366**0.005**−0.322**0.019**−0.313**0.017**\* denotes adjusted values for pubertal stage using multiple linear regression. Cortisol variables were logarithmically transformed. *BMI z-score* body mass index z-score, *FM* fat mass, *WHtR* waist-to-height ratio, *\|DCS\|* absolute values of diurnal cortisol slope HPA axis measures and 24-h physical activity {#Sec16} -------------------------------------------- For the entire sample, correlations of each movement behaviour with HPA axis measures relative to the other movement behaviours are displayed in Table [4](#Tab4){ref-type="table"}. After adjustments for pubertal stage and BMI z-score, inverse associations between cortisol~08:00~ (γ~MVPA~ = − 0.107; *P* = 0.018), AUC~08:00--16:00~ (γ~MVPA~ = − 0.081; *P =* 0.038), and absolute values of DCS (γ~MVPA~ = − 0.150; *P* = 0.007) with the time spent in MVPA relative to other movement behaviours were detected. Moreover, AUC~16:00--23:00~ correlated positively with time spent in non-screen sedentary behaviours (γ~non-screen\ SB~ = 0.169; *P* = 0.009) and negatively with the relative time spent in sleeping (γ~sleep~ = − 0.212; *P* = 0.018). Table 4Compositional behavior model for the associations between HPA axis measures and the proportion of the day spent in screen time, non-screen sedentary behaviours, MVPA, and sleep durationCortisol variablesModel\ ***P***Model\ R^**2**^γ~**Screentime**~***P***γ~**non-screen/SB**~***P***γ~**MVPA**~***P***γ~**sleep**~***P*Cortisol**~**08:00**~0.0020.350−0.0040.895−0.0300.686−0.107**0.018**0.1420.183**Cortisol**~**16:00**~0.2860.1360.0160.6180.1470.045−0.0090.377−0.1530.129**Cortisol**~**23:00**~0.0040.3320.0040.9410.2240.0800.0860.246−0.3150.080**AUC**~**08:00--16:00**~0.0050.3240.0020.9360.0300.653−0.081**0.038**0.0480.603**AUC**~**16:00--23:00**~0.0350.2520.0240.3850.169**0.009**0.0190.594−0.212**0.018AUC**~**08:00--23:00**~0.0090.3090.0100.7110.0640.292−0.0590.091−0.0150.860**\|DCS\|**0.0010.382−0.0030.936−0.0690.445−0.150**0.007**0.2220.085All models were adjusted for pubertal stage and BMI z-score using multiple linear regression. Cortisol variables were logarithmically transformed. Regression coefficients correspond to change in the log-ratio of the given behaviour relative to other behaviours. *MVPA* moderate to vigorous physical activity, *SB* sedentary behaviours, *\|DCS\|* absolute values of diurnal cortisol slope Measures of the HPA axis and glucose or lipid metabolism {#Sec17} -------------------------------------------------------- There were no significant correlations between HPA axis measures and serum glucose or lipid levels, as shown in Table [5](#Tab5){ref-type="table"}. Table 5Associations between HPA axis rhythm index and glucose or lipid metabolismCortisol variablesFGFIHOMA-IRTGTCr***P***r***P***r***P***r***P***r***P*Cortisol**~**08:00**~0.1440.3450.0090.9550.0110.944−0.1150.450−0.1440.346**Cortisol**~**16:00**~0.2230.1500.1760.2720.1910.232−0.2500.106−0.0720.646**Cortisol**~**23:00**~0.1120.4790.2560.1110.2260.1610.0800.615−0.1980.209**AUC**~**08:00--16:00**~0.2230.152−0.0100.952−0.0050.975−0.1580.312−0.1240.429**AUC**~**16:00--23:00**~0.1010.5340.2710.0950.2590.111−0.1960.225−0.1570.335**AUC**~**08:00--23:00**~0.2160.1800.1320.4240.1200.467−0.2270.159−0.2640.100**DCS**−0.0120.9390.0740.6500.0700.6690.1730.2730.1540.330Spearman's correlations and *P* values were reported. FM: fat mass, *WHtR* waist-to-height ratio, *\|DCS\|* absolute values of diurnal cortisol slope, *FG* fasting glucose, *FI* fasting insulin, *HOMA-IR* the homeostasis model assessment of insulin resistance, *TC* total cholesterol, *TG* triglycerides Discussion {#Sec18} ========== In this cross-sectional study, we report the influences of obesity, puberty and physical activity on diurnal cortisol rhythm in children and adolescents. We found a dampened circadian cortisol rhythm in children with obesity, and flatter and less sharply declining slopes correlated with degrees of adiposity. The altered dynamics of the HPA axis also appeared to be influenced by puberty and the distribution of 24-h movement. Therefore, stabilizing circadian cortisol rhythms through circadian regulation strategies may be an important approach for preventing childhood obesity. HPA axis dysfunction is a risk factor for metabolic diseases such as obesity and is closely related to negative health outcomes. It has been proven that individuals with obesity may display blunted diurnal HPA axis functioning, which mainly manifests as decreased cortisol variability, lower morning levels, or elevated evening levels \[[@CR15]--[@CR17]\]. As previously reported, obese Zucker rats lack a circadian rhythm of 11β-HSD1 gene expression in the hippocampus, which may contribute to dampened diurnal variation of circulating corticosterone levels \[[@CR34]\]. One paediatric study demonstrated that daytime cortisol levels are inversely associated but that night-time levels are positively associated with BMI z-score and central adiposity \[[@CR14]\]. In adults, higher BMI or WHtR correlates with a flatter diurnal cortisol slope, suggesting a shallower decline throughout the day \[[@CR25], [@CR35]\]. These studies incorporated multiple sampling time points, allowing more precise slope measurement and more reliable results for associations between cortisol rhythms and obesity. Similar to the above studies, we show that salivary cortisol levels were lower in the morning and higher at night with flatter and less sharply declining cortisol slopes in children with obesity than in those with a normal weight. Moreover, we found salivary cortisol slopes and night-time cortisol to be positively related to weight gain, abdominal fat distribution (WHtR) and body fat percentage in all participants. Such findings are supported by a large cross-sectional study of adults, which showed that bedtime salivary cortisol output tended to increase with BMI, indicating that individuals with obesity display abnormal HPA hyperactivity at night \[[@CR36]\]. However, other paediatric studies have reported different findings. Based on the HPLC-MS/MS method, Chu showed higher morning salivary cortisol and morning urinary cortisol in children with obesity aged 4--5 years \[[@CR37]\], and Kjolhede presented an inverse association between obesity and morning or evening salivary cortisol levels in children aged 6--12 years by EIA \[[@CR18]\]. Such inconsistent findings might result from single sampling time points or different sampling times, cortisol measurements or age distributions. Recent human studies have shown that cortisol concentrations increase significantly throughout puberty and adolescence \[[@CR38], [@CR39]\], which is consistent with our findings. The increased salivary cortisol AUC might reflect higher overall activity of the adrenal gland throughout puberty. In fact, the developmental process of puberty, along with endocrine changes, has been suggested to influence HPA axis functioning \[[@CR40]\]. We also found that testosterone in children with obesity correlated positively with AUC~08:00--16:00~ and AUC~08:00--23:00~, consistent with the phenomenon of co-activation, where cortisol and testosterone (and dehydroepiandrosterone) are positively linked within an individual \[[@CR41]\]. Accordingly, these findings highlight the important role of gonadal hormones in the development of the circadian cortisol cycle during puberty, indicating that puberty is a highly interrelated variable and should be included as a covariate in studies seeking to explore the relationship between cortisol rhythms and adolescent obesity. Compositional analyses provide an appropriate statistical means for understanding the collective health implications of finite, co-dependent, 24-h movement behaviours \[[@CR29]\]. In our results, the relative time spent in MVPA was related to lower morning cortisol concentrations, daytime cortisol output (AUC~08:00--16:00~) and flatter DCS, independent of puberty and BMI z-score. Labsy pointed out that acute exercise does not significantly affect steroid circadian rhythms but that medium-to-long term training, intended as chronic exercise, appeared to play a key role as a synchronizer for the whole circadian system \[[@CR10]\]. Thus far, chronic physical activity has been reported to lower diurnal HPA activity and reduce HPA reactivity to acute stress in pre-pubertal children \[[@CR42]\]. In cancer patients, moderate chronic PA positively influences sleep behaviour and the activity--wake circadian rhythm \[[@CR43]\]. As lower cortisol secetion in daytime may act as a protective factor due to prior over-stimulated HPA axis in obesity \[[@CR25]\], a reduction in morning cortisol concentrations and daytime cortisol output may also contribute to the role of MVPA as a protective factor in response to chronic stress. In both adults and children, traditional research has mainly focused on a single exercise or unclassified physical activity, without consideration of the combined effects of the composition of the rest of the day. In this study, we eliminated such drawbacks and emphasized the important role of the proportion of time spent in MVPA in the circadian system. Our findings also suggest that increased non-screen sedentary behaviours and inadequate sleep duration are associated with higher night-time cortisol output. With sleep loss, cortisol may exert its deleterious metabolic effects by maintaining high night-time concentrations, which are associated with insulin resistance (IR), suppressed immunity and increased inflammation \[[@CR44]\]. Thus, the findings of this integrated approach indicate that the relative distribution of time spent in different physical activities within a 24-h period is important for health promotion and maintenance of diurnal cortisol rhythm in the paediatric population. Flat slopes with lower amplitude, i.e., those exhibiting suppressed peak levels or failing to reach sufficiently low levels by evening, are indicative of HPA dysregulation \[[@CR45]\] and associated with a higher risk of obesity, cardiovascular disease and type 2 diabetes \[[@CR17], [@CR25]\]. As the circadian system also plays a role in modulating appetite with self-reported hunger peaks at night \[[@CR46]\], elevated nadir cortisol may further increase appetite and promote the consumption of foods enriched in fat and sugar at night \[[@CR13]\]. Moreover, fasting glucose is supposed to be lowest at night, and glucose elevation at night has been demonstrated to be temporally and quantitatively correlated with cortisol rise \[[@CR47]\]. Human explant visceral and subcutaneous adipose tissue clock gene expression rhythms can be altered by dexamethasone administration \[[@CR48]\]. In light of this, an interaction pathway with the HPA axis to mediate food intake and body weight via the circadian output of adipocytes is postulated \[[@CR49]\]. In the present study, there were no associations between cortisol rhythms and glucose or lipid metabolism due to the non-corresponding sampling time to verify the above hypothesis. Nonetheless, metabolic disorders in children with obesity were increased compared with those in normal weight children (data not shown). Further analysis is warranted to verify this pattern and assess the relationship between obesity complications and cortisol slope. Here, we present a preliminary study that examined the relationship between indices of salivary cortisol and the clinical characteristics of children and adolescents. Nonetheless, some limitations should be noted. Socioeconomic status index, dietary data and growth hormone levels were not included, and these factors may be related to cortisol rhythms. Moreover, the small sample size and sampling time limited additional findings, especially regarding verification of the correlations of cortisol rhythms with lipid and glucose metabolism. Finally, the level and intensity of PA, SB and sleep duration were parent- or self-reported, and these subjective measurements might confound the results. Conclusion {#Sec19} ========== This study offers initial insight into the complex and interrelated associations of diurnal cortisol rhythm and obesity during childhood and adolescence. We demonstrated reduced cortisol levels in the morning and increased levels at night in childhood obesity. Flatter and less sharply declining slopes correlated with adiposity, indicating an alteration in the circadian rhythm of cortisol with adiposity. Our findings also support the importance of an appropriate distribution of 24-h movement for optimal health and the circadian system in children and young people. Synchronizing exercise and nutrient interventions to the circadian clock might maximize the health-promoting benefits of interventions to prevent and treat metabolic disease \[[@CR50]\]. Thus, chronotherapeutic approaches targeting the maintenance of normal rhythms via a healthy lifestyle may be effective in counteracting obesity and other metabolic diseases in children and adolescents. ANOVA : Analysis of variance ANCOVA : Analysis of covariance AUC : Areas under the curve DCS : Diurnal cortisol slope ECLIA : Electrochemiluminescence immunoassay FM% : Fat mass percentage FG : Fasting glucose FI : Fasting insulin HOMA-IR : The homeostasis model assessment of insulin resistance MPA : Moderate-intensity physical activities HPA axis : The hypothalamic-pituitary-adrenal axis IR : Insulin resistance SB : Sedentary behaviours TC : Total cholesterol TG : Triglycerides VPA : Vigorous-intensity physical activities WC : Waist circumference WHtR : The waist-to-height ratio **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We thank the Clinical Laboratory of Children's Hospital of Nanjing Medical University for their technical assistance. TY and XL conceived and carried out the experiments. TY and SW performed the data collection. TY performed the analyses. TY and XL wrote the paper. WZ, QL and XL reviewed the manuscript. All authors had final approval of the submitted and published versions. This work was financially supported by the Natural Science Foundation of China (81773421), Jiangsu Province Key Research and Social Development Program (BE2015607), and the Innovation Team of Jiangsu Health (CXTDA 2017035). The funders had no role in the study design or collection, analysis, or interpretation of data or in writing this manuscript. The data used to support the findings of this study are available from the corresponding author upon request. The study was approved by the Children's Hospital of Nanjing Medical University Ethical Committee (NO.201603004--1). The parents provided written informed consent prior to inclusion in the study. Not applicable. The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of the research reported.

T[he]{.smallcaps}regulation of lymphocyte extravasation from the circulation into sites of inflammation is critical in coordinating an appropriate and effective immune response. There is careful control not only of the particular tissues into which migration occurs, but also of the various subtypes of leukocytes involved ([@B7]). Under flow conditions, circulating lymphocytes can attach and roll on vascular endothelium using the selectins and α4 integrins ([@B1]; [@B4]; [@B18]). These adhesion receptors are able to slow the transit of leukocytes and expose them to stimuli causing activation-dependent firm adhesion. The integrins α4β1, α4β7, and leukocyte function--associated antigen-1 (LFA-1)^1^ have been implicated in activation-dependent stable arrest of lymphocytes under flow ([@B2]). LFA-1, however, unlike the α4 integrins, cannot initiate adhesion under these conditions without L-selectin and/or α4β7 first tethering the lymphocyte to the vessel wall ([@B2]). Subsequently, LFA-1 is the principal integrin involved in transendothelial migration ([@B32]; [@B30]; [@B24]), although the stimulus that induces this β2 integrin--dependent movement across the endothelial layer is unclear. Interaction with and migration across fibronectin and other extracellular matrix components are then necessary for the successful completion of migration into the tissues. Therefore, transmigration of any one T cell is a multistep process dependent on the tight regulation of the sequential and often overlapping activities of the expressed integrins ([@B2]). There is increasing evidence that, on a given cell, one subset of integrins may be negatively regulated by ligation of another. Transfection of αvβ3 into K562 cells that endogenously express α5β1 provides a system in which ligation of β3 inhibits the phagocytic but not adhesive function of α5β1 ([@B5]). Similarly, ligation of transfected αIIbβ3 will inhibit the function of cotransfected α2β1 or endogenous α5β1 in CHO cells expressing these integrins (Diáz-González et al, 1996). Such effects are dependent on an intact β3 cytoplasmic tail, and are considered to involve signal transduction ([@B6]; [@B8]). In a further example, ligand-binding by α4β1 on fibroblasts was able to suppress the ability of α5β1 to induce metalloproteinase expression ([@B14]). There is also indirect evidence of similar integrin regulation on T cells. For example, resting lymphocytes use both LFA-1 and α4β1 to bind endothelial cells, but when T cells become activated, adhesion is mediated through LFA-1 with little or no contribution from α4β1 ([@B33]). Additionally, in some leukemic T cell lines, functional α4β1 is found only when LFA-1 is either not expressed or inactive ([@B33]). Therefore, there appears to be a T cell integrin hierarchy in which α4β1 is inactive if LFA-1 is active. Here we provide direct evidence of cross talk on T cells between β2 and β1 integrins. The avid state of the β2 integrin, LFA-1, can be maintained by the ligand intercellular adhesion molecule-1 (ICAM-1) or by activation mAbs. Such sustained activation of LFA-1 is shown to downregulate the adhesion through α4β1 and, to a lesser extent, through α5β1 to the ligands fibronectin and vascular cell adhesion molecule-1 (VCAM-1) (α4β1 alone). The result, phenotypically, is a less adhesive, more migratory T cell. Therefore, we have demonstrated another way in which integrin activities may be regulated. Materials and Methods ===================== Preparation of T Lymphoblasts ----------------------------- Peripheral blood mononuclear cells were prepared from single donor leukocyte buffy coats by centrifugation through Lymphoprep^®^ (Pharmacia Diagnostics AB, Uppsala, Sweden). T cells were expanded from this population by culturing in RPMI 1640 plus 10% FCS (GIBCO BRL, Paisley, UK) in the presence of phytohaemagglutinin (Murex Diagnostics, Dartford, UK) at 1 μg/ml for 72 h as previously described ([@B10]). Cells were then washed and maintained for 1--2 wk in medium supplemented with 20 ng/ml recombinant IL-2 (Euro Cetus UK Ltd., Harefield, UK). The cells, which were used between days 10 and 14, were a 99% CD3^+^ population, containing 65% CD8^+^ and 35% CD4^+^ cells. The population was negative for the natural killer cell marker CD56. mAbs and Other Reagents ----------------------- mAb 925.2 (CD11a; LFA-1α subunit, nonblocking) was purchased from Becton Dickinson (Oxford, UK). mAbs 38 (CD11a; LFA-1 α subunit function--blocking), and 24 (CD11/CD18; β2 integrin activation reporter) ([@B11]; [@B9]) and 52U (control antibody) were prepared in this laboratory, and purified from ascites or tissue culture supernatant by protein A--Sepharose chromatography ([@B12]). mAbs HP1/2 (CD49d; α4 subunit function--blocking) and TS2/ 16 (CD29; β1 subunit-activating) were gifts from R. Lobb (Biogen, Inc., Cambridge, MA). mAb SAM-1 (CD49e; α5 subunit--blocking) was purchased from Eurogenetics (Hampton, UK). mAb 7.2 (CD49d; α4 subunit, nonblocking) was a gift from J. Marshall (Imperial Cancer Research Fund, London, UK). mAb UCHT2 (CD5) was a gift from P. Beverley (University College, London, UK). mAb G19.4 was a gift from Bristol Myers-Squibb Pharmaceuticals (Princeton, NJ). Rabbit anti--mouse IgG was purchased from Sigma Chemical Co. (Poole, UK). ICAM-1Fc was produced as a chimeric protein, consisting of the five extracellular domains of ICAM-1 fused to the Fc fragment of human IgG1 ([@B3]). VCAM-1Fc, produced as a chimera consisting of the two amino-terminal domains of human VCAM-1 fused to the Fc fragment of human IgG1 ([@B16]), was a gift from R. Lobb. Fibronectin (0.1% solution from human plasma) was purchased from Sigma Chemical Co., and cytochalasin D from GIBCO BRL. The fluorescent cell label 2′,7′-bis-(\[carboxyethyl\]-5\[6′\]-carboxyfluorescein) (BCECF-AM) was purchased from Calbiochem Corp. (Nottingham, UK). All other chemicals and reagents were purchased from Sigma Chemical Co. Ligand-coated Beads ------------------- A modified protocol ([@B27]) was developed in which 200 μl (10^8^) of 3.2-μm polystyrene beads (Sigma Chemical Co.) were washed twice in distilled water, followed by two further washes and resuspension in 0.1 M bicarbonate buffer, pH 9. Fibronectin, ICAM-1, VCAM-1, or BSA as control were added to these beads to a final concentration of 10 μg/ml. The beads were rotated at room temperature for 1 h, washed once in PBS, and blocked with 0.1% denatured BSA for 2 h at room temperature while being rotated. The beads were then washed twice in 20 mM Hepes, 140 mM NaCl, 2 mg/ml glucose, pH 7.4 (assay buffer), containing 3 mM Mg^2+^/2 mM EGTA, for use in the adhesion experiments as described. Cell Bead Attachment Assay -------------------------- Multiwell Lab-Tek^®^ Chamber Slides (Nunc, Inc., Naperville, IL) were left uncoated as controls or coated with either ICAM-1Fc (10 μg/ml in PBS) or rabbit anti--mouse Ig (1:100 dilution in PBS) overnight at 4°C. The next day, mAb tissue culture supernatant was added to wells precoated with anti--mouse Ig and left on ice for 1 h. Wells were washed twice with PBS, and nonspecific binding sites were blocked with 0.1% denatured BSA for 2 h at room temperature. Cells (150 μl of 2 × 10^6^/ml) in assay buffer (see above) containing 3 mM Mg^2+^/2 mM EGTA were added to the wells and allowed to settle on ice for 15 min. Freshly prepared ligand-coated beads (see above) were added to the wells at 100:1 bead-to-cell ratio in 50 μl. After 30 min at 37°C, the unbound beads and cells were removed with four washes in warmed assay buffer. Bound cells were fixed with 1% formaldehyde in PBS for 20 min at room temperature. Cells were then stained with haematoxylin for 10 min. Beads and cells were counted per high power field (×40 oil immersion objective; Carl Zeiss, Inc., Thornwood, NY), and the number of beads per 100 cells was determined (attachment index). Cell Attachment Assays ---------------------- Flat-bottomed Immulon-1^®^ 96-well plates (Dynatech Labs., Inc., Chantilly, VA) were precoated with 50 μl fibronectin (20 μg/ml), VCAM-1Fc (7 μg/ml), or ICAM-1Fc (2.4 μg/ml) in PBS overnight at 4°C. The plates were blocked with 2.5% BSA in PBS for 2 h at room temperature and then washed four times in assay buffer (see above) at 4°C. T cells were washed three times in assay buffer and labeled with 2.5 μM BCECF-AM in the same buffer for 30 min at 37°C, followed by two further washes. T cells (2 × 10^5^ cells) were treated with 3 mM Mg^2+^/2 mM EGTA, 50 nM phorbol-12,13-dibutyrate (PdBu), or CD3 mAb at indicated levels, as well as inhibitors and mAbs in 100 μl assay buffer. Ca^2+^ and Mg^2+^ were included at 0.4 mM for experiments involving PdBu or T cell receptor/CD3 complex (TCR/CD3) cross-linking with mAb G19.4. Blocking mAbs were titrated on T cells by FACS^®^ analysis (Becton Dickinson, Mountain View, CA) and used at saturating concentrations to block T cell function. For fibronectin- and VCAM-1--binding assays, all wells contained anti--LFA-1 mAb 38 at function-blocking concentrations of 10 μg/ml. This prevents cells aggregating via LFA-1/ICAM-1 interactions, which would cause spuriously high binding to β1 ligands through the piggy-back interaction of nonadherent cells with truly adherent cells. Plates were incubated for 15 min on ice, followed by centrifugation at 40 *g* for 1 min, before 40-min incubation at 37°C. Nonadherent cells were removed by washing four times in warmed assay buffer (150 μl/well). Adhesion was quantified by recording emission at 530 nm, after excitation at 485 nm, using a Fluoroskan^®^ II (Labsystems, Inc., Basingstoke, UK), and by expressing the reading for each well as a percentage of the total emission before incubation. Transmigration Assays --------------------- Assays were performed in 20 mM Hepes, 140 mM NaCl, 2 mg/ml glucose, pH 7.4, 0.25% BSA, 3 mM Mg^2+^/2 mM EGTA (transmigration buffer) using 6.5-mm-diam Transwell^®^ plates (Costar Corp., Cambridge, MA). The upper and lower surfaces of the inserts were coated with fibronectin at concentrations ranging from 0 to 50 μg/ml in PBS overnight at 4°C. The inserts were positioned in wells containing 600 μl transmigration buffer. Cells were then plated in the insert at a concentration of 5 × 10^5^ cells in 100 μl transmigration buffer with appropriate mAbs. The mAbs were used at the following final concentrations: mAb HP1/2 at 0.7 μg/ml, mAb 7.2 at 5 μg/ml, mAb SAM-1 at 5 μg/ml, mAb 24 at 5 μg/ml, and mAb 52U at 5 μg/ml. The anti--LFA-1 mAb 38 was added to all inserts at function-blocking concentrations of 10 μg/ml to prevent a spurious decrease in migration due to cell aggregation when LFA-1 is activated. The plates were then incubated for 6 h at 37°C. The bottom surface of the insert was then scraped to release migrated but adherent cells into the bottom well, and the migrated cells were counted in a hemocytometer. Nine grids (0.1 mm^3^ per grid) were counted per well, and readings were averaged from duplicate samples. All assays were performed in duplicate, and each experiment was repeated a minimum of four times. Results ======= In this study, we have investigated cross influences on function between the β1 integrins and LFA-1 on T cells. As one method of activating these leukocyte integrins, we treated T cells with 3 mM Mg^2+^/2 mM EGTA ([@B10]). For the β2 integrin, LFA-1, the advantage of such treatment is that it directly alters the integrin ectodomain, bypassing the requirement for an intracellular stimulus ([@B31]). This form of LFA-1 is considered to be of high affinity because it is able to bind soluble ICAM-1 ([@B31]). There have been both positive and negative reports of the ability of Mg^2+^/ EGTA to induce fibronectin receptor--mediated adhesion ([@B29]; [@B17]). In this study, we show that T cells do bind fibronectin, immobilized either on plates or on beads, in an Mg^2+^-dependent manner. To examine further the generality of cross-influences between these integrins, we also investigated T cells stimulated with phorbol ester or by TCR/CD3 cross-linking. Both of these stimuli act from within the cell to activate integrins, so called inside-out signaling, and may be considered more representative of the in vivo situation. Adhesion of T Cells to ICAM-1 Will Decrease Binding of Fibronectin-coated Beads ------------------------------------------------------------------------------- To determine the effect of LFA-1 ligation on the function of the β1 integrins on the same T cell, we developed a ligand-coated bead--binding system. T cells were adhered to immobilized ICAM-1 via LFA-1, or to a control substrate, and their ability to bind beads coated with ligand for the β1 integrins, α4β1 and α5β1, was then investigated. Fibronectin-coated beads were bound by T cells adherent to the control substrate, anti-CD5 mAb, immobilized on plastic (Fig. [1](#F1){ref-type="fig"} *A*), and the specificity of adhesion was demonstrated by blocking bead-binding with a combination of α4 and α5 mAbs (Fig. [1](#F1){ref-type="fig"} *B*). However, when T cells were adherent to ICAM-1 as substrate, they bound fewer fibronectin-coated beads (Fig. [1](#F1){ref-type="fig"} *C*). When binding of the fibronectin-coated beads was quantified, there was a decreased level of fibronectin bead-binding when T cells were adherent to ICAM-1 (Fig. [2](#F2){ref-type="fig"} *A*) (inhibition: 65.0 ± 23.4% = mean ± SD; *n* = 6). This result demonstrated that, on human T cells, the interaction of LFA-1 with its ligand ICAM-1 could downregulate the function of the β1 integrins. In contrast, T cells adhered to anti--LFA-1 mAb bound beads at a similar level as T cells adherent to control mAb. This indicated that the LFA-1 inhibitory effect could not be mimicked by cross-linking LFA-1 with immobilized CD11a mAb 38 (Fig. [2](#F2){ref-type="fig"} *A*). Conversely, there was no difference between the ability of T cells adherent to anti-CD5 mAb, fibronectin, or ICAM-1 to bind ICAM-1-- coated beads (Fig. [2](#F2){ref-type="fig"} *B*), indicating that adhesion to immobilized fibronectin did not alter the extent of ICAM-1 bead binding by LFA-1. This is the first evidence that LFA-1 could dominate the activity of the fibronectin-binding receptors and that the reverse situation did not hold. Adhesion of T Cells to ICAM-1 Decreases Binding of Fibronectin- and VCAM-1--coated Beads by Downregulating α4β1 Activity ------------------------------------------------------------------------------------------------------------------------ We then looked at the effects of LFA-1 ligand--binding on each of the T cell fibronectin receptors, α4β1 and α5β1. Both of these integrins can be involved in T cell binding to fibronectin, and, if one is blocked, the other can partially compensate, as previously described ([@B34]). Fig. [3](#F3){ref-type="fig"} *A* shows the binding of fibronectin-coated beads by T cells adherent to anti-CD5 mAb. This binding can be partially blocked with either an α4-blocking mAb, an α5-blocking mAb, or completely with a combination of both blocking mAbs, showing that T cells bind these beads through a mixture of α4β1 and α5β1 integrins. However, when T cells are adherent to ICAM-1, the binding index for fibronectin-coated beads is lower, and is reduced only by an α5-blocking mAb (Fig. [3](#F3){ref-type="fig"} *B*). Therefore, binding in this situation is mediated mainly through α5β1 with little contribution from α4β1, demonstrating that the binding activity of α4β1 has been compromised. The binding by T cells of VCAM-1--coated beads, which is mediated exclusively by α4β1, reveals a similar downregulation when T cells are adherent to ICAM-1 as compared to anti--LFA-1 mAb (Fig. [4](#F4){ref-type="fig"}). Activation of the β2 Integrin LFA-1 on T Cells Inhibits Their β1-Mediated Binding to Fibronectin ------------------------------------------------------------------------------------------------ The inhibitory effect of LFA-1 on β1-mediated ligand-binding could not be demonstrated by cross-linking LFA-1 with an anti--LFA-1 mAb, but required binding to ligand ICAM-1. This suggested that high affinity LFA-1 rather than receptor cross-linking was necessary for cross talk. This led to the development of an assay in which T cells were first stimulated with Mg^2+^/EGTA, TCR/CD3 cross-linking, or PdBu and then exposed to mAb 24, which holds LFA-1 in an active conformation as if occupied by ligand ([@B11]). mAb 24 caused increased T cell binding to ICAM-1 after titration of Mg^2+^ (Fig. [5](#F5){ref-type="fig"} *A*), CD3 mAb G19.4 (Fig. [5](#F5){ref-type="fig"} *B*), and PdBu (not shown). In contrast, mAb 24 caused inhibition of T cell binding to fibronectin after TCR/CD3 cross-linking (Fig. [6](#F6){ref-type="fig"} *A*) or Mg^2+^/EGTA (data not shown). Monovalent Fab′ fragments of mAb 24 produced the same degree of inhibition as bivalent mAb 24 (data not shown). This demonstrated that activation or ligand occupancy of LFA-1 in the absence of clustering is sufficient to alter fibronectin-mediated adhesion. mAb KIM185 (CD18; β2-activating) behaved similarly to mAb 24, depressing binding of T cells to fibronectin while enhancing adhesion to ICAM-1 (data not shown). The Effect of LFA-1 Activation on T Cells Is Mediated Predominantly through α4β1 -------------------------------------------------------------------------------- Because T cells bind to fibronectin through both α4β1 and α5β1, we analyzed the effects of LFA-1 activation individually on these integrins using function-blocking mAbs and either TCR/CD3 cross-linking (Fig. [6](#F6){ref-type="fig"}, *B* and *C*) or Mg^2+^/ EGTA (data not shown) to stimulate adhesion. Prolonged activation of LFA-1 with mAb 24 had only a small effect on total fibronectin-binding (Fig. [6](#F6){ref-type="fig"} *A*) and on α5β1-mediated adhesion (Fig. [6](#F6){ref-type="fig"} *B*), but had a much greater effect on α4β1-mediated adhesion (Fig. [6](#F6){ref-type="fig"} *C*). Similar levels of α4β1 inhibition by mAb 24 were seen when the integrins were activated with PdBu or Mg^2+^/EGTA (Fig. [7](#F7){ref-type="fig"}). In addition, there was no effect of the nonfunction-altering anti--LFA-1 mAb G25.5, which again emphasized the requirement for LFA-1 activation (Fig. [7](#F7){ref-type="fig"}). Under equivalent activating conditions, mAb 24 and the β2 integrin--activating mAb KIM185 decreased α4β1-mediated adhesion to VCAM-1 to the same extent as to fibronectin (data not shown). Together, these results reinforced the findings that α4β1 function is particularly sensitive to the state of LFA-1 activation. Activation of β1 Integrins on T Cells Has No Effect on LFA-1 Binding to ICAM-1 ------------------------------------------------------------------------------ We then reversed the situation to investigate the effect on β2 integrin--mediated adhesion of maintaining β1 integrins in an active state, using the β1 integrin--stimulating mAb TS2/16. This mAb increased binding to fibronectin after the three activating treatments (Fig. [8](#F8){ref-type="fig"} *A*), but had no effect on β2-mediated binding to ICAM-1 (Fig. [8](#F8){ref-type="fig"} *B*), confirming that the β1 integrins were unable to influence the ligand-binding activity of LFA-1. Inhibition of α4β1 with Blocking mAbs or by LFA-1 Activation Increases α5β1-Mediated Migration ---------------------------------------------------------------------------------------------- The effects of LFA-1--mediated cross talk on α4β1- and α5β1-mediated T cell migration on fibronectin were then investigated. Using the Transwell^®^ system, we established that T cells undergo random migration on fibronectin using α5β1 exclusively, and that this migration was enhanced by an α4-blocking mAb HP1/2, but not affected by an α4-- nonblocking mAb 7.2 (Fig. [9](#F9){ref-type="fig"} *A*). In addition, migration at different concentrations of fibronectin remained solely dependent on α5β1 (Fig. [9](#F9){ref-type="fig"} *B*), and could be enhanced either with α4-blocking mAb HP1/2 (Fig. [9](#F9){ref-type="fig"} *C*), or by maintaining LFA-1 activation with mAb 24 (Fig. [9](#F9){ref-type="fig"} *D*). Such increased migration was blocked with an α5-blocking mAb. Therefore, by decreasing the α4β1 contribution to fibronectin adhesion with either an α4-blocking mAb or an LFA-1-- activation mAb, the ability of α5β1 to mediate migration was increased. Investigation of the Mechanism of LFA-1 Cross Talk -------------------------------------------------- Treatment of T cell LFA-1 with Mg^2+^/EGTA directly induces a high affinity form of the integrin that is able to bind soluble ICAM-1 ([@B31]). Therefore, we looked at the ability of β1 integrins to adopt a high affinity state. However, treatment with Mg^2+^/EGTA yielded no detectable binding of soluble fibronectin or VCAM-1, even when these ligands were used at concentrations up to 1.2 mg/ml. In contrast, 0.5 mM Mn^2+^ was able to induce fibronectin (α4β1 and α5β1)-- and VCAM-1 (α4β1)--binding to both T cells and Jurkat cells (data not shown), as has been reported by others ([@B16]; [@B13]). Similarly, while 0.5 mM Mn^2+^ was able to induce expression of the β1 activation reporter epitopes recognized by mAb 15/7 ([@B35]) or mAb HUTS-21 ([@B17]), no expression of these epitopes was observed with Mg^2+^/EGTA treatment (data not shown). These findings imply that the fibronectin-binding integrins are in a low affinity state after all three methods of stimulation, and that LFA-1 cross talk causes inhibition of postreceptor occupancy events, rather than direct modulation of receptor affinity. We next tested the possibility that LFA-1 activation might be influencing a cytoskeletal event. Although mAb 24 decreased the overall level of α4β1-mediated adhesion to fibronectin, there was no change in the sensitivity of binding to cytochalasin D (Fig. [10](#F10){ref-type="fig"}). Therefore, LFA-1 cross talk affects an event in cell adhesion after receptor occupancy but before changes in the actin cytoskeleton, and is independent of both. Discussion ========== This study was undertaken to examine the functional interaction on T cells between LFA-1 and the β1 integrin fibronectin receptors α4β1 and α5β1. The main findings are that (*a*) the occupation of T cell LFA-1 by its ligand ICAM-1 decreases the binding of α4β1 to ligands fibronectin and VCAM-1; (*b*) this inhibitory cross talk also results from the prolonged activation of LFA-1 induced by the activation reporter mAb 24 in combination with several T cell adhesion--inducing protocols; (*c*) the adhesive activity of α5β1 is affected to a lesser extent; (*d*) while active LFA-1 downregulates the avidity of α4β1, the reverse does not occur, as neither β1 integrin--activating mAb TS2/16 nor β1-mediated binding to fibronectin affected the avidity of LFA-1; and (*e*) downregulation of α4β1 activity increases the efficiency of α5β1-mediated migration on fibronectin. Therefore, we have demonstrated differential regulation of two integrin subclasses and a hierarchy of integrin usage in which the β2 integrin LFA-1 will suppress the function of β1 integrins, particularly α4β1. Previous studies have demonstrated the involvement of α4β1 in leukocyte adhesion to but not migration across endothelium, and of LFA-1 as the chief integrin in transendothelial migration ([@B32]; [@B24]; [@B23]). Furthermore, in vitro experiments during flow have emphasized the requirement that an integrin hierarchy allow coordinated migration of lymphocytes across the endothelium into the tissues ([@B7]). Our finding that active LFA-1 is able to decrease the ligand-binding activity of α4β1 has direct implications for the sequential activity of these integrins in such an adhesion cascade; LFA-1 may function optimally in the absence of α4β1 adhesion, allowing the T cell to deadhere from the apical surface of the endothelium and transmigrate. Our findings also argue against a redundancy among integrin--ligand pairs in leukocyte transmigration, and imply specific roles for each integrin. In this study, we have demonstrated that, in contrast to adhesion, the migration of activated T cells on fibronectin is mediated by α5β1 with no contribution from α4β1. In addition, suppressing α4β1 activity on T cells either by mAb 24 or α4 function--blocking mAbs enhanced the level of α5β1 migration, particularly at low fibronectin levels. This may reflect the compensatory increase in α5β1 adhesion, with its migratory potential, when binding through the nonmigratory α4β1 is blocked. Another possibility is that the enhanced migration by α5β1 is due to removal of a restraint imposed by α4β1. The importance of strength of adhesion in regulating cell migration is well documented ([@B15]; [@B26]), suggesting that firm adhesion by both α4β1 and α5β1 may make conditions suboptimal for migration. Alternatively, α4β1 may be involved in a more specific inhibition of α5β1 function, as has been described in the control of metalloproteinase expression in fibroblasts ([@B14]). The promotion of migratory behavior by α5β1 through loss of α4β1-binding activity is in keeping with its more prominent role within the tissues after successful negotiation of T cells across the endothelium ([@B19]). Therefore, a hierarchy of integrin activity may feature at this later stage of the adhesion cascade, with LFA-1 providing a link between α4β1 and α5β1. The mechanism for LFA-1 downregulation of α4β1 was explored in several ways. We first established that there was no alteration in expression of either α4β1 or α5β1 during the experimental period (data not shown). Furthermore, confocal microscopy using mAbs specific for α4β1, α5β1, and the β1-activation reporter mAb 15/7 indicated that avid LFA-1 did not cause β1 integrin redistribution on the T cell membrane (data not shown). In addition, although stimulation of T cells with Mg^2+^/EGTA induces high affinity LFA-1 ([@B31]), none of the three stimulating protocols induced high affinity α4β1 or α5β1. This implied that cross talk was not affecting high affinity β1 integrins. Together, these findings suggested that the β1 integrins had not undergone a detectable alteration in affinity nor been redistributed or shed from the cell surface, and that LFA-1 cross talk was targeting events after ligand-binding. This result is in keeping with other studies in which cross talk is ultimately dependent on the presence of the β subunit cytoplasmic tail and steps subsequent to modulation of integrin affinity ([@B5]; [@B8]). It seemed possible that the cytoskeleton was a target of LFA-1--mediated cross talk because both α4β1- and α5β1-mediated adhesion were more sensitive to changes in actin than was adhesion through LFA-1 (data not shown). However, for β1 integrin--mediated adhesion, the similarity in cytochalasin D sensitivity of mAb 24--treated and untreated cells and the synergism between suboptimal doses of cytochalasin D and mAb 24 in the inhibition of α4β1-mediated binding to fibronectin (data not shown) supported the evidence that inhibition occurs upstream of cytoskeletal changes. These results implied that cross talk affects an event in cell adhesion occurring after receptor occupancy but before actin-mediated cytoskeletal changes, and independent of both. In addition, protein kinase A, associated with LFA-1 signaling and deadhesion ([@B28]), and protein kinase C, implicated in some previous cross talk studies ([@B5]; [@B25]), were not involved in this phenomenon (data not shown). LFA-1 cross talk was evident after several different adhesion-inducing protocols, showing that the phenomenon was not stimulus specific. The fact that cross talk was dependent on ICAM-1 or mAb 24 indicated that occupancy of LFA-1 was a prerequisite. Although the signaling pathways activated upon engagement of the β2 integrins are not well understood, certain observations suggested that cross talk did activate specific intracellular signaling pathways. Cross talk was not observed using the Jurkat T cell line, which is known to have a defect in LFA-1 signaling ([@B22]). In addition, cross talk was induced by mAb 24 Fab′ fragments but not by immobilized anti-- LFA-1 mAb, emphasizing the requirement for a mechanism beyond LFA-1 clustering. For α5β1 on human fibroblasts, although clustering by mAbs of integrin on beads induced phosphorylation and accumulation of p125 focal adhesion kinase and tensin, ligand occupancy recruited further cytoskeletal proteins to the signaling complex ([@B20],*b*). One speculation is that the targets of LFA-1 cross talk may be the proteins providing the link between integrins and actin. However, several observations suggested that cross talk does not represent a simple sequestering of such intracellular proteins. First, integrin activity operates in one direction only, so prolonged activation of the β1 integrins using mAb TS2/16 does not alter LFA-1 binding to ICAM-1. Second, LFA-1 predominantly affects the activity of α4β1, despite a sixfold abundance of α4β1 over α5β1 (data not shown). Future work will address the role of potential integrator molecules in the cross talk phenomenon. In summary, we describe inhibition of α4β1-binding activity in T cells as a consequence of LFA-1 activation. A speculation is that deadhesion of α4β1 from the apical surface of the endothelium is required for LFA-1--mediated migration across endothelium to proceed. Another observation is that T cell migration on fibronectin is mediated by α5β1, and that this migration is enhanced by interfering with α4β1 adhesion. LFA-1 might provide a link between α4β1 and α5β1 by uncoupling the former in order to enhance migration by the latter. While the actual mechanism by which cross talk is achieved is unclear, our findings implicate a downstream signaling event brought about by maintaining LFA-1 in a highly avid state. We gratefully acknowledge Dr. Roy Lobb for supplies of VCAM-1Fc and mAbs, Dr. Fumio Takei (Vancouver, Canada) for invaluable assistance with the bead-binding assays, Dr. Martyn Robinson, Dr. Carlos Cabañas, Dr. John Marshall, Dr. Peter Beverley, and Dr. Ted Yednock for mAbs, Alison McDowall for preparation of mAb 24 Fab′ fragments, and all our colleagues in the Leucocyte Adhesion Laboratory for their helpful comments and critical reading of the manuscript. This work was supported by the Imperial Cancer Research Fund, London, UK. J.C. Porter is a Medical Research Council (UK) Clinical Training Fellow. Address all correspondence to J.C. Porter, Leukocyte Adhesion Laboratory, Imperial Cancer Research Fund, Lincoln\'s Inn Fields, London WC2A 3PX, United Kingdom. Tel.: 44-171-269-3569; Fax: 44-171-269-3093; e-mail: <porterj@europa.lif.icnet.uk> ICAM-1 : intercellular adhesion molecule-1 LFA-1 : lymphocyte function--associated antigen-1 PdBu : phorbol-12,13-dibutyrate TCR/CD3 : T cell receptor/CD3 complex VCAM-1 : vascular cell adhesion molecule-1 {#F1} {#F2} {#F3} {#F4} {#F5} {#F6} {#F7} {#F8} {#F9} {#F10}

Abbreviations used in this paper: ArfGAP, Arf GTPase activating protein; Arl, Arf-like; FG, phenylalanine-glycine; GAP, GTPase-activating protein; GEF, guanine nucleotide exchange factor. Introduction ============ Regulatory GTPases act as molecular switches that can rapidly interconvert between two conformational states, depending on whether GTP or GDP is bound. The cellular actions of GTPases are typically initiated by GTP binding, promoted by guanine nucleotide exchange factors (GEFs), and terminated by GTP hydrolysis that is facilitated by GTPase-activating proteins (GAPs). Within the Ras superfamily ([@bib26]), comprised of \>150 GTPases, each family (Ras, Rho, Rab, Arf, and Ran) uses distinct sets of GEFs and GAPs to regulate signaling. Though originally envisioned as simple "off switches" in Arf signaling, the ArfGAPs have also emerged as effectors and key components in the assembly of nanomachines with complex signaling potential ([@bib5]; [@bib9]). ArfGAPs are a family of proteins containing a characteristic module, the ArfGAP domain, which was first identified in rat ArfGAP1 as the domain responsible for stimulation of GTP hydrolysis on Arf1 ([@bib3]). ArfGAP domains are ancient and highly conserved since the earliest eukaryotes. Five ArfGAPs have been identified in the yeast *Saccharomyces cerevisiae* and shown to display a combination of redundant and unique functions. Mammalian cells express an array of ArfGAPs ranging from relatively small proteins resembling those found in yeast to the large, multi-domain ArfGAPs that are proposed to function as scaffolds for cell signaling ([Fig. 1 A](#fig1){ref-type="fig"}). Structure, mechanism, and specificity ===================================== ArfGAP domains are ∼130 amino acids in length and were originally defined as the minimal fragment possessing ArfGAP activity ([@bib3]). They contain a characteristic C~4~-type zinc finger motif and a conserved arginine that is required for activity, within a particular spacing (CX~2~CX~16~CX~2~CX~4~R). The zinc finger has an architectural rather than catalytic role ([Fig. 1 B](#fig1){ref-type="fig"}) ([@bib6]). The invariant arginine was proposed to serve in a catalytic "arginine finger" mechanism, similar to that found in GAPs for other GTPases, including Ras and Rho ([@bib20]), and is highly exposed to solvent in the crystal structure ([Fig. 1 B](#fig1){ref-type="fig"}). However, the potential for other binding partners (e.g., coatomer; [@bib6]) or other domains within some ArfGAPs (e.g., PH domains) serving supportive or regulatory roles in GAP-stimulated hydrolysis has also been demonstrated. For example, the ArfGAP activity of ASAP1 is dependent on the PH domain and is sensitive to PI(4,5)P~2~, and that of GITs is stimulated by PIP~3~. {#fig1} ArfGAPs display various degrees of specificity for individual members of the Arf family both in vitro and in live cells. However, these data are not easy to interpret due to uncertainties in colocalization of the ArfGAP and Arfs in cells, incomplete knowledge of the importance of coregulators (lipids or other proteins), and because some ArfGAPs use their GAP domain to bind Arf without promoting GTP hydrolysis. For the most part (see below) ArfGAPs are active on one or more of the "true" Arfs (Arf1-6) but not on the Arf-like (Arl) or Sar proteins, which use distinct families of GAPs. Arf GAP activity has been demonstrated in vitro for at least one member of each subfamily, with the exception of the ADAPs, which appear to lack in vitro GAP activity. However, overexpression of ADAP1 reduces activated Arf6 (but not Arf1) levels, and diminishes cortical actin and stress fibers, consistent with function as an Arf6 GAP. Although ArfGAPs had been thought to distinguish between Arfs and Arls, Gcs1p (a yeast orthologue of mammalian ArfGAP1) was reported to function as GAP for both yeast Arf1p and Arl1p ([@bib12]). Conversely, Arl2 GAP isolated from bovine testis lacks the ArfGAP domain but exhibits GAP activity toward Arfs ([@bib2]). The latter finding suggests that the "canonical" ArfGAPs, those containing the ArfGAP domain, may not represent the entire repertoire of proteins capable of stimulating GTP hydrolysis by Arf proteins. Consensus nomenclature of the ArfGAP family of human genes and encoded proteins =============================================================================== Like most gene families today, gene/protein discovery and the accompanying nomenclature of ArfGAPs has come from a number of different laboratories and techniques and over a span of more than a decade. We have developed a consensus nomenclature for the human ArfGAPs that is based upon the phylogeny of the ArfGAP domains and the domain organization of the proteins, and updated or clarified the acronyms in use to more accurately describe the current understanding of each protein. [Table I](#tbl1){ref-type="table"} lists the consensus nomenclature for the human genes, previously published gene symbols and names, database identifiers, chromosome locations, and accession numbers. These gene names should also be used for the encoded proteins, with appropriate indications when more than one splice variant is known to exist. The 31 predicted human ArfGAPs have been classified into 10 subfamilies, based on sequence similarities of their ArfGAP domains, and supported by the conservation of the domain architecture within each subfamily (see [Fig. 1 A](#fig1){ref-type="fig"}). The acquisition of additional domains throughout eukaryotic evolution has contributed to the acquisition of new functions by different ArfGAPs. A brief summary of each subfamily is presented below. ###### Consensus names for human ArfGAPs, with additional information --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Subfamily New\ Previous\ Literature or database aliases/synonyms consensus\ HGNC\ gene symbols gene symbols ------------- -------------- -------------- ----------------------------------------- ------------------------------- ---------------- ------------- ---------- -------- ---------------- --------------------------------------------------------------------- **ARFGAP1** ARFGAP1 ARFGAP1 ArfGAP1 Arf1GAP Arf GAP MGC39924 55738 20q13.33 [NM_018209](NM_018209); [NP_060679](NP_060679) (406aa) **ARFGAP2** ARFGAP2 ARFGAP2 ArfGAP2 ZNF289/Zfp289 FLJ14576 FLJ26000 84364 11p11.2-p11.12 [NM_032389](NM_032389); [NP_115765](NP_115765) (521aa) ARFGAP3 ARFGAP3 ArfGAP3 ArfGAP1 26286 22q13.2-q13.3 [NM_014570](NM_014570); [NP_055385](NP_055385) (516aa) **ADAP** ADAP1 CENTA1 Centaurin\ PIPBP p42IP4 GCS1L 11033 7p22.3 [NM_006869](NM_006869); [NP_006860](NP_006860) (374aa) (alpha1) ADAP2 CENTA2 Centaurin\ Cent. Alpha2 cent-b HSA272195 55803 17q11.2 [NM_018404](NM_018404); [NP_060874](NP_060874) (381aa) (beta) **SMAP** SMAP1 SMAP1 SMAP-1 FLJ13159 FLJ42245 60682 6q13 [NM_001044305](NM_001044305); [NP_068759](NP_068759) (440aa) SMAP2 SMAP2 SMAP1L RP1-228H13.3 SMAP2 64744 1p35.3-p34.1 [NM_022733](NM_022733); [NP_073570](NP_073570) (429aa) **AGFG** AGFG1 HRB HRB1 RIP HIV-1 Rev BP RAB 3267 2q36.3 [NM_004504](NM_004504); [NP_004495](NP_004495) (562aa) AGFG2 HRBL HRB2 RAB-R/HRB1 HIV-1 Rev BP-L 3268 7q22.1 [NM_006076](NM_006076); [NP_006067](NP_006067) (481aa) **GIT** GIT1 GIT1 Git1 Cat1 p95APP1 28964 17p11.2 [NM_014030](NM_014030); [NP_054749](NP_054749) (761aa) GIT2 GIT2 Git2 Cat2 p95APP2 p95PKL KIAA0148 9815 12q24.1 [NM_057169](NM_057169); [NP_476510](NP_476510) (759aa) **ASAP** ASAP1 DDEF1 AMAP1 DEF1/DDEF1 PAG2/SHAG1 Cent. Beta4 KIAA1249 50807 8q24.1-q24.2 [NM_018482](NM_018482); [NP_060952](NP_060952) (1129aa) ASAP2 DDEF2 AMAP2 PAP/DDEF2 SHAG2/PAG3 Cent. Beta3 KIAA0400 8853 2p24 [NM_003887](NM_003887): [NP_003878](NP_003878) (1006aa) ASAP3 DDEFL1 DDEFL1 UPLC1 ACAP4 Cent. Beta6 55616 1p36.12 [NM_017707](NM_017707); [NP_060177](NP_060177) (903aa) **ACAP** ACAP1 CENTB1 ACAP1 Cent. Beta1 KIAA0050 9744 17p13.1 [NM_014716](NM_014716); [NP_055531](NP_055531) (740aa) ACAP2 CENTB2 ACAP2 Cent. Beta2 KIAA0041 23527 3q29 [NM_012287](NM_012287); [NP_036419](NP_036419) (778aa) ACAP3 CENTB5 ACAP3 Cent. Beta5 KIAA1716 116983 1p36 [NM_030649](NM_030649); [NP_085152](NP_085152) (759aa) **AGAP** AGAP1 CENTG2 AGAP1 Cent. Gamma2 GGAP1 MGC71657 KIAA1099 116987 2q37 [NM_014914](NM_014914); [NP_055729](NP_055729) (804aa) AGAP2 CENTG1 AGAP2 Cent. Gamma1 GGAP2/PIKE FLJ16430 KIAA0167 116986 12q14.1 [NM_014770](NM_014770); [NP_055585](NP_055585) (836aa);\ [AAM97540](AAM97540) (1192aa) AGAP3 CENTG3 AGAP3 Cent. Gamma3 MRIP1/CRAG FLJ16146 116988 7q36.1 [NM_031946](NM_031946); [NP_114152](NP_114152) (911aa) AGAP4 CTGLF1 CTGLF1 Centaurin gamma-like family 1 MRIP2 119016 10q11.21 [NM_133446](NM_133446); [NP_597703](NP_597703) (663aa) AGAP5 CTGLF2 CTGLF2 Centaurin gamma-like family 2 729092 10q22.2 XM_001132588; XP_001132588 (686aa) AGAP6 CTGLF3 CTGLF3 Centaurin gamma-like family 3 414189 10q11.23 [NM_001077665](NM_001077665); [NP_001071133](NP_001071133) (686aa) AGAP7 CTGLF4 CTGLF4 Centaurin gamma-like family 4 653268 10q11.23 [NM_001077685](NM_001077685); [NP_001071153](NP_001071153) (663aa) AGAP8 CTGLF5 CTGLF5 Centaurin gamma-like family 5 728404 10q11.23 [NM_001077686](NM_001077686); [NP_001071154](NP_001071154) (663aa) AGAP9 CTGLF6 CTGLF6 Centaurin gamma-like family 6 FLJ00312 642517 10q11.22 [NM_001077686](NM_001077686) (663aa) AGAP10 CTGLF7 CTGLF7 Centaurin gamma-like family 7 728127 10q11.22 XR_015281 (655aa) AGAP11 KIAA1975 Similar to MRIP2 KIAA1975 119385 10q23.2 [NP_597704](NP_597704) (550aa) **ARAP** ARAP1 CENTD2 ARAP1 Cent. Delta2 KIAA0782 116985 11q13.4 [NM_001040118](NM_001040118); [NP_001035207](NP_001035207) (1450aa) ARAP2 CENTD1 ARAP2 Cent. Delta1 FLJ13675 PARX KIAA0580 116984 4p14 [NM_015230](NM_015230); [NP_056045](NP_056045) (1704aa) ARAP3 CENTD3 ARAP3 Cent. Delta3 FLJ21065 DRAG1 64411 5q31.3 [NM_022481](NM_022481); [NP_071926](NP_071926) (1544aa) --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- List of consensus acronyms: ArfGAP = ADP-ribosylation factor GTPase activating proteins. ACAP = ArfGAP with coiled-coil, ankyrin repeat, and PH domains. ADAP = ArfGAP with dual PH domains. AGAP = ArfGAP with GTPase domain, ankyrin repeat, and PH domain. AGFG = ArfGAP with FG repeats. ARAP = ArfGAP with Rho GAP domain, ankyrin repeats and PH domain. ASAP = ArfGAP with SH3 domain, ankyrin repeat, and PH domain. GIT = G protein receptor kinase (GRK) interacting ArfGAP. SMAP = Small ArfGAP. List of proposed discontinued acronyms: APP = ArfGAP putative, PIX1-interacting, paxillin binding protein. CAT = Cool-associated tyrosine phosphorylated protein. CTGLF = Centaurin gamma-like family. DEF = differentiation-enhancing factor. HRB = HIV Rev binding protein. PAG = paxillin associated Arf GAP. PAP = Pyk2 associated protein. PKL = paxillin-kinase linker. RAB = HIV Rev-associated binding protein. RIP = HIV Rev-interacting protein. ArfGAP1 subfamily ================= This founder of the ArfGAP family is also the smallest member, at ∼45 kD. ArfGAP1 shuttles between cytosol and the Golgi, where it is involved in regulating the COPI mechanism of membrane traffic. The region of ArfGAP1 C-terminal to the ArfGAP domain is predicted to be largely unstructured. This region contains two stretches, termed ALPS motifs, that have the propensity in vitro to fold into amphipathic α-helices upon interaction with membranes that are highly curved or contain loosely packed lipids ([@bib1]; [@bib13]), and are required for Golgi targeting of the protein in vivo ([@bib10]). The ALPS motifs are predicted to orient the protein on the membrane and to contribute substantially to regulation of its activity in cells. ArfGAP activity is also regulated by the COPI coat complex ([@bib6]). While most studies have focused on the role of ArfGAP1 in the COPI system, ArfGAP1 also interacts with components of clathrin coated carriers (including clathrin, AP-1, and AP-2), although the functional consequences of these interactions remains to be established. ArfGAP2 subfamily ================= ArfGAPs 1--3 are predicted to have arisen from a common ancestor (e.g., a single gene is present in *G. lamblia*) with an early split into distinctive ArfGAP1 and ArfGAP2 subfamilies and later duplications leading to the ArfGAP2/ArfGAP3 divergence. Human ArfGAP2 and 3 are closely related proteins (58% identity) with little similarity to ArfGAP1 outside the catalytic domain. ArfGAP1 and ArfGAP2/3 display functional interplay, as indicated by synthetic lethality observed between these ArfGAPs in HeLa cells ([@bib4]), and in *S. cerevisiae* (Gcs1 and Glo3; [@bib16]). ArfGAP2/3 lack ALPS motifs but are found on Golgi membranes as a result of strong interactions with the COPI coat ([@bib25]; [@bib4]). Like ArfGAP1, the GAP activities of ArfGAP2/3 are stimulated by coatomer (unpublished data). How each of these ArfGAPs contributes to the COPI mechanism remains unclear. ADAP subfamily ============== ADAP1 (ArfGAP with dual PH domains; previously centaurin α1, p42(IP~4~), or PIP~3~BP) was identified as a high affinity PI(3,4,5)P~3~ and Ins(1,3,4,5)P~4~ binding protein. ADAP1 is proposed to function as an Arf6 GAP that regulates the actin cytoskeleton, membrane traffic, and neuronal differentiation ([@bib23]; [@bib24]). ADAP1 localizes to dendrites, spines, and synapses of developing and adult neurons and can impact traffic of regulated secretory vesicles in neuronal cells. Studies of ADAP2 have not been reported. SMAP subfamily ============== SMAPs have been implicated as regulators of endocytosis and oncogenesis ([@bib22]). Human SMAPs are ∼50 kD and lack other defined domains, thus the acronym small ArfGAP protein. The two human SMAP proteins share 47% identity overall, which rises to 83% identity in the ArfGAP domain. SMAPs bind to clathrin heavy chain via the clathrin binding motif (LLGLD) as well as the clathrin assembly protein, CALM ([@bib14]). SMAP1 is cytosolic but is recruited to membranes where it regulates constitutive endocytosis. SMAP2 is more stably bound to endosomes and is involved in the retrograde transport of TGN46 from early endosomes to the TGN ([@bib14]). AGFG subfamily ============== AGFG1 has been reported to be an essential HIV Rev cofactor, based on experiments that suggest that the ArfGAP domain mediates the release of Rev-directed HIV-1 RNAs from the perinuclear region ([@bib19]). The Rev--AGFG1 interaction is indirect and possibly bridged by the nuclear export receptor CRM1. The AGFG subfamily arose very early in eukaryotes, with a predicted progenitor in *G. lamblia* and representatives in molds, plants, fish, and mammals. AGFG1 contains 10 phenylalanine-glycine (FG) repeats, reminiscent of those found in nucleoporins. Thus, the AGFGs are ArfGAPs with FG repeats. Much less information is available on AGFG2. GIT subfamily ============= GIT genes arose in animals and were duplicated in vertebrates. GIT2 expression is nearly ubiquitous, whereas GIT1 appears absent from many major cell types (muscle, hepatoctyes, pneumocytes, adipocytes) but is especially prominent in endothelial cells ([@bib21]). Unlike other ArfGAPs, GITs tightly associate with a specific partner, the PIX/Cool proteins, to form oligomeric complexes ([@bib17]). PIX/Cool proteins are GEFs for Rac1 and Cdc42 GTPases. GIT/PIX complexes function as scaffolds for a variety of signaling enzymes (including G protein receptor kinases, p21 activated kinases, focal adhesion kinases, MEK/Erk, and phospholipase Cγ) and are recruited to distinct cellular locations via specific partners (e.g., focal adhesions via paxillin or integrinα4, synapses via piccolo or liprin-α, and plasma membranes via scribble) ([@bib8]). GIT/PIX complexes function as sites of signal integration from multiple GTPase inputs, a feature shared by the ARAPs (see below). ASAP subfamily ============== The ASAPs are associated with plasma membrane specializations (e.g., focal adhesions and invadopodia), and regulate aspects of endocytic traffic and actin remodeling ([@bib15]; [@bib9]; [@bib18]). ASAPs are found exclusively in animals, and have been duplicated in vertebrates, with three genes in humans that share multiple domains (see [Fig. 1 A](#fig1){ref-type="fig"}) but differ in their C termini. ASAP1 has tandem repeats of \[D/ELPPKP\] and an SH3 domain, ASAP2 has an SH3 domain but no E/DLPPKP repeats, and ASAP3 has neither of these motifs. These differences at the C termini have led to confusion in nomenclature, but phylogenetic analysis of the ArfGAP domains supports the conclusion that these three ASAPs form a distinct group. ASAP1 was identified as a Src-binding protein but also binds CrkL, focal adhesion kinase, CD2AP, and CIN85 ([@bib9]). ASAP2 binds pyk2, a focal adhesion kinase, whereas ASAP3 associates with focal adhesions and regulates stress fibers ([@bib7]). AGAP subfamily ============== AGAPs are present in animals only and there are 11 human genes predicted to encode AGAP-type proteins, arising from amplifications in regions of human chromosome 10q, with a number of pseudogenes predicted in this same region. AGAPs possess a GTP-binding domain, reported to directly bind and activate Akt and other Ras effectors ([@bib27]). AGAP2 has multiple splice variants with one possessing an SH3 binding motif N-terminal to the GTP-binding protein-like domain. AGAP1 and AGAP2 have been the most extensively characterized members of this group. They function in the endocytic system; AGAP1 working with AP-3 and AGAP2 with AP-1 ([@bib15]). ACAP subfamily ============== ACAPs regulate Arf6-dependent actin remodeling and endocytosis and receptor tyrosine kinase-dependent cell movement ([@bib9]). ACAP1 functions as part of an Arf6-regulated clathrin coat ([@bib11]). ACAPs are found in *Dictyostelium* and metazoans, with vertebrate duplications resulting in three human genes. ACAP is an acronym for ArfGAP with coiled coil, ankyrin repeat, and PH domains, though the coiled coil domain was later identified as a BAR domain. ARAP subfamily ============== The presence of ArfGAP, RhoGAP, ankyrin repeats, and Ras association domains in ARAPs intimates that they are important coordinators of two or more GTPase signaling pathways. Although the domain structures of the three ARAPs are similar, the proteins are distinct in functions and cellular locations with individual ARAPs showing different Arf, Rho, and Ras binding specificities. Pathways affected by different family members include signaling through EGF receptor, focal adhesion dynamics, and lamellipodia formation ([@bib9]; [@bib18]). ARAPs are specific to chordates, with three human genes. Summary ======= ArfGAPs play critical roles in secretory and endocytic membrane traffic as well as actin remodeling. Other ArfGAPs serve as scaffolds for cell signaling, allowing them to mediate cross-talk between members of different GTPase families. Future studies aimed at dissecting ArfGAP functions and at identifying sources of specificity and regulation in their actions are certain to provide important insights into the role of ArfGAPs in cell physiology and in human pathologies. These outcomes should be facilitated by rapid adoption of the consensus nomenclature described herein. The authors thank Jonathan Goldberg (Memorial Sloan-Kettering Cancer Center) for providing coordinates of his ArfGAP structure and Eric Ortlund (Emory University) for help in the preparation of [Fig. 1 B](#fig1){ref-type="fig"}. We also acknowledge the many contributions to the field that we were not able to specifically cite due to limits on the number of citations. This work was performed with support from extramural grants (GM67226, GM68029, R.A. Kahn; National Institutes of Health GM59989, DA016347, R.T. Premont; AI-043208, M.L. Zapp) and that from the Intramural Program (P.A. Randazzo, H. Inoue) of the National Institutes of Health, Department of Health and Human Services as well as the American Heart Association (0655454U, R.T. Premont), and a grant from the Israel Science Foundation 448/04 (D. Cassel). [^1]: Correspondence to Richard A. Kahn: <rkahn@emory.edu>

Introduction ============ Lung cancer is one of the most prevalent cancers worldwide, causing 1.76 million deaths per year \[[@ref1],[@ref2]\]. Chest computed tomography (CT) scans play an essential role in the screening for \[[@ref3]\] and diagnosis of lung cancer \[[@ref4]\]. A randomized controlled trial demonstrated that low-dose CT screening reduced mortality from lung cancer among high-risk patients \[[@ref3]\], and recent studies showed the benefit of CT screening in community settings \[[@ref5]\]. The wide adoption of lung cancer screening is expected to benefit millions of people \[[@ref6]\]. However, millions of CT scan images obtained from patients constitute a heavy workload for radiologists \[[@ref7]\]. In addition, interrater disagreement has been documented \[[@ref8]\]. Previous studies suggested that computer-aided diagnostic systems could improve the detection of pulmonary nodules in CT examination \[[@ref9]-[@ref12]\]. To stimulate the development of machine learning models for automated CT diagnosis, the Kaggle Data Science Bowl provided labeled chest CT images from 1397 patients and awarded \$1 million in prizes to the best algorithms for automated lung cancer diagnosis, which is the largest machine learning challenge on medical imaging to date. In response, 1972 teams worldwide have participated and 394 teams have completed all phases of the competition \[[@ref13]\], making it the largest health care--related Kaggle contest \[[@ref14]\]. This provides a unique opportunity to study the robustness of medical machine learning models and compare the performance of various strategies for processing and classifying chest CT images at scale. Due to the improved performance of machine learning algorithms for radiology diagnosis, some developers have sought commercialization of their models. However, given the divergent software platforms, packages, and patches employed by different teams, their results were not easily reproducible. The difficulty in reusing the state-of-the-art models and reproducing the diagnostic performance markedly hindered further validation and applications. To address this gap, we reimplemented, examined, and systematically compared the algorithms and software codes developed by the best-performing teams of the Kaggle Data Science Bowl. Specifically, we investigated all modules developed by the 10 award-winning teams, including their image preprocessing, segmentation, and classification algorithms. To ensure the reproducibility of results and the reusability of the developed modules, we generated a Docker image for each solution using the Docker Community Edition, a popular open-source software development platform that allows users to create self-contained systems with the desired version of software packages, patches, and environmental settings. According to Docker, there are over 6 million Dockerized applications, with 130 billion total downloads \[[@ref15]\]. The Docker images are easily transferrable from one server to another, which ensures the reproducibility of scientific computing \[[@ref16]\]. Our Dockerized modules will facilitate further development of computer-aided diagnostic algorithms for chest CT images and contribute to precision oncology. Methods ======= Data and Classification Models ------------------------------ We obtained the low-dose chest CT datasets in Digital Imaging and Communications in Medicine format from the Kaggle Data Science Bowl website \[[@ref13]\]. The dataset was acquired from patients with high risks for developing lung cancers. In this Kaggle challenge, a training set (n=1397) with ground truth labels (362 with lung cancer; 1035 without) and a public test set (n=198) without labels were provided to the participants. The ground truth label is 1 if the patient developed lung cancer within 1 year of the date the CT scan was performed and 0 otherwise. The diagnosis was confirmed by pathology evaluation as a part of the National Lung Screening Trial \[[@ref3],[@ref17]\]. Once participants submitted the prediction results for the public test set, the Kaggle competition platform reported their models' performance on the public leaderboard instantaneously. The final test set (n=506, ground truth labels were not disclosed to participants) was only available to participants after the model submission deadline, thus serving as an independent validation set that decided the final winners. The chest CT images in the training set, public test set, and final test set all came from multiple hospitals and had different qualities. In particular, the final test set contained more recent and higher quality data with thinner slice thickness than those in the two other sets \[[@ref18]\]. To systematically compare the solutions developed by the award-winning teams, we acquired the source codes of the winning solutions and their documentation from the Kaggle news release after the conclusion of the competition. Per the rules of this Kaggle challenge, the source codes of these award-winning solutions were required to be released under open-source licenses approved by the Open Source Initiative \[[@ref19]\] in order to facilitate free distribution and derivation of the solution codes \[[@ref20]\]. The default license is the MIT license \[[@ref20]\]. Under the open-source licenses approved by Open Source Initiative, the software can be freely used, modified, and shared \[[@ref19]\]. Comparison of the Approaches and Their Performance -------------------------------------------------- We compared the workflows of the top 10 solutions by examining and rerunning their source codes. For each solution, we inspected all steps taken from inputting the CT images to outputting the prediction. We documented the versions of the software package and platform dependencies of each solution. The Kaggle Data Science Bowl used the log-loss function to evaluate the performance of the models \[[@ref13]\]. The log-loss function where n is the number of patients in the test set, *y~i~* is 1 if patient i has lung cancer, 0 otherwise, and *ŷ~i~* is the predicted probability that patient i has lung cancer \[[@ref13]\]. If the predicted outcome is set as 0.5 for all patients, the log-loss value would be ln (2) ≈ 0.69. To investigate whether models with high performance in the public test set generalize to the images in the final test set, we computed the Spearman correlation coefficient of the log-loss in the two test sets. All analyses were conducted using R version 3.6 (R Foundation for Statistical Computing). Docker Image Generation for the Top Ten Solutions ------------------------------------------------- We reproduced the results by recompiling the source codes and dependencies of each of the top ten solutions. Since the solutions used various platforms and different versions of custom software packages, many of which were not compatible with the most updated packages or mainstream release, we generated Docker images \[[@ref16]\] to manage the software dependencies and patches required by each solution to enhance the reusability and reproducibility of the developed algorithms. Results ======= Performance Comparison ---------------------- [Figure 1](#figure1){ref-type="fig"} summarizes the public and private test set scores of the top 250 teams that participated in the Kaggle Data Science Bowl. Results showed that the top 20 teams achieved a log-loss less than 0.5 in the final test set, and more than 80 submissions reached a log-loss less than 0.6 in the same set. However, these models had varying performances in the public test set. Surprisingly, 11 out of the top 50 teams had a public test set log-loss greater than 0.69, which was worse than blindly submitting "0.5" as the cancer probability for every patient. The correlation between the public test set scores and the final test set scores was weak among all teams that completed the contest (Spearman correlation coefficient = .23; [Figure 2](#figure2){ref-type="fig"}A). In the top 10 teams, the correlation is moderate (Spearman correlation coefficient = .39; [Figure 2](#figure2){ref-type="fig"}B). {#figure1} {#figure2} Data Workflow Comparison ------------------------ [Figure 3](#figure3){ref-type="fig"} summarizes the most frequently used strategy by the winning teams. Most solutions used additional publicly available datasets, generated lung segmentation, rescaled the voxels, and performed nodule segmentations before fitting the classification models. [Table 1](#table1){ref-type="table"} compares the additional datasets, data preprocessing, segmentation, classification, implementation, and final test set scores of the top 10 solutions. In addition to the training dataset provided by the Kaggle challenge, most teams used CT images and nodule annotations from other publicly available resources. [Table 2](#table2){ref-type="table"} summarizes the sample size, availability of nodule locations, nodule segmentation, diagnoses, other characteristics of the Kaggle dataset, and additional datasets employed by the participants. Most of the top solutions used images and nodule segmentations from the Lung Nodule Analysis 2016 (LUNA16) challenge to develop their segmentation algorithms. LUNA16 is a closely related competition organized in 2016 with an aim to detect lung nodules in chest CT images \[[@ref21],[@ref22]\]. Two teams also reported using the lung CT images, diagnostic annotations, and nodule location data from the International Society for Optics and Photonics (SPIE)--American Association of Physicists in Medicine (AAPM) Lung CT Challenge \[[@ref23]\], but one of them did not incorporate this relatively small dataset (n=70) when building the final models. Only one of the top 10 teams did not use any additional datasets outside of the competition. Frequently used image preprocessing steps include lung segmentation and voxel scaling. Voxel scaling ensures that the voxels of images from various CT scan protocols correspond to similar sizes of physical space. Variants of U-Net \[[@ref24]\], VGGNet \[[@ref25]\], and residual net (ResNet) \[[@ref26]\] were commonly used as the nodule segmentation algorithms, and the nodule segmentation models trained on the LUNA16 dataset were often applied to the Data Science Bowl dataset. After lung nodule segmentation, classification algorithms were employed to generate final cancer versus noncancer predictions. Most of the solutions leveraged existing ImageNet-based architecture and transfer learning \[[@ref12],[@ref27]\]. All teams employed 2D or 3D convolutional neural networks (CNN). A few teams employed CNNs as feature extractors and used tree-based classifiers for this classification task. {#figure3} ###### Comparisons of the top-performing solutions of the Kaggle Data Science Bowl. Rank Team name Additional datasets used Data preprocessing Nodule segmentation Classification algorithms Implementation Final test set score ------ ---------------------------------- -------------------------- ----------------------------------------------------------------------------------------- ----------------------------------------------------------- --------------------------------------------------------------------------------------------------------------- ----------------------------------- ---------------------- 1 Grt123 LUNA16^a^ Lung segmentation, intensity normalization Variant of U-Net Neural network with a max-pooling layer and two fully connected layers Pytorch 0.39975 2 Julian de Wit and Daniel Hammack LUNA16, LIDC^b^ Rescale to 1×1×1 C3D^c^, ResNet-like CNN^d^ C3D, ResNet-like CNN Keras, Tensorflow, Theano 0.40117 3 Aidence LUNA16 Rescale to 2.5×0.512×0.512 (for nodule detection) and 1.25×0.5×0.5 (for classification) ResNet^e^ 3D DenseNet^f^ multitask model (different loss functions depending on the input source) Tensorflow 0.40127 4 qfpxfd LUNA16, SPIE-AAPM^g^ Lung segmentation Faster R-CNN^h^, with 3D CNN for false positive reduction 3D CNN inspired by VGGNet Keras, Tensorflow, Caffe 0.40183 5 Pierre Fillard (Therapixel) LUNA16 Rescale to 0.625×0.625×0.625, lung segmentation 3D CNN inspired by VGGNet 3D CNN inspired by VGGNet Tensorflow 0.40409 6 MDai None Rescale to 1×1×1, normalize HU^i^ 2D and 3D ResNet 3D ResNet + a Xgboost classifier incorporating CNN output, patient sex, \# nodules, and other nodule features Keras, Tensorflow, Xgboost 0.41629 7 DL Munich LUNA16 Rescale to 1×1×1, lung segmentation U-Net 2D and 3D residual neural network Tensorflow 0.42751 8 Alex, Andre, Gilberto, and Shize LUNA16 Rescale to 2×2×2 Variant of U-Net CNN, tree-based classifiers (with better performance) Keras, Theano, xgboost, extraTree 0.43019 9 Deep Breath LUNA16, SPIE-AAPM^j^ Lung mask Variant of SegNet Inception-ResNet v2 Theano and Lasagne 0.43872 10 Owkin Team LUNA16 Lung segmentation U-Net, 3D VGGNet Gradient boosting Keras, Tensorflow, xgboost 0.44068 ^a^LUNA16: Lung Nodule Analysis 2016. ^b^LIDC: Lung Image Database Consortium. ^c^C3D: convolutional 3D. ^d^ResNet-like CNN: residual net--like convolutional neural network. ^e^ResNet: residual net. ^f^DenseNet: dense convolutional network. ^g^SPIE-AAPM: International Society for Optics and Photonics--American Association of Physicists in Medicine Lung CT Challenge. ^h^R-CNN: region-based convolutional neural networks. ^i^HU: Hounsfield unit. ^j^Dataset has been evaluated but not used in building the final model. ###### A summary of the chest computed tomography datasets employed by the participants. Datasets Number of CT^a^ scan series Data originated from multiple sites Availability of nodule locations Availability of nodule segmentations Availability of patients' diagnoses (benign versus malignant) --------------------------------------------- ----------------------------------------------------------- ------------------------------------- ---------------------------------- -------------------------------------- --------------------------------------------------------------- Kaggle Data Science Bowl (this competition) Training: 1397; public test set: 198; final test set: 506 Yes No No Yes Lung nodule analysis 888 Yes Yes Yes Yes SPIE-AAPM^b^ Lung CT Challenge 70 No Yes No Yes Lung Image Database Consortium 1398 Yes Yes Yes Yes ^a^CT: computed tomography. ^b^SPIE-AAPM: International Society for Optics and Photonics--American Association of Physicists in Medicine. Comparison of the Implementation Platforms and Software Dependencies -------------------------------------------------------------------- Most of the winning teams developed their modules with Keras and Tensorflow. Only one team used Pytorch (the top-performing team), Caffe, or Lasagne. All of the top 10 teams employed a number of python packages for scientific computing and image processing, including NumPy, SciPy, and Scikit-image (skimage). A summary of package dependencies is shown in [Figure 4](#figure4){ref-type="fig"}. This reflected the popularity of the tools for processing chest CT images, building neural networks, and scientific computing among the top contestants of this contest. {#figure4} Docker Images of the Top Solutions ---------------------------------- To facilitate reusing the code developed by the top teams, we generated a Docker image for each of the available solutions. Our developed Docker images are redistributed under the open-source licenses chosen by the original developers \[[@ref28]\]. Detailed instructions on accessing the Docker images can be found on GitHub \[[@ref29]\]. Discussion ========== Principal Findings ------------------ This is the first study that systematically compared the algorithms and implementations of award-winning pulmonary nodule classifiers. Results showed that the majority of the best-performing solutions used additional datasets to train the pulmonary nodule segmentation models. The top solutions used different data preprocessing, segmentation, and classification algorithms. Nonetheless, they only differ slightly in their final test set scores. The most commonly used data preprocessing steps were lung segmentation and voxel scaling \[[@ref30]\]. For nodule classification, many solutions used CNNs. However, 2 of the top 10 teams employed tree-based methods for cancer versus noncancer classification. Tree-based approaches require a predefined set of image features, whereas CNNs allow data to refine the definition of features \[[@ref31]\]. Given sufficient sample size, CNNs outperformed tree-based methods in many image-related tasks \[[@ref12],[@ref32]\], whereas tree-based methods could reach satisfactory performance when the sample size was small, and they provided better model interpretability \[[@ref33]-[@ref35]\]. Since the conclusion of the contest, additional works on machine learning for CT evaluation have been published \[[@ref36]-[@ref40]\]. Nonetheless, these works reported similar strategies for data processing and classification overall. To enhance the reproducibility of the developed modules, we generated a Docker image for each of the award-winning solutions. The Docker images contain all software dependencies and patches required by the source codes and are portable to various computing environments \[[@ref16]\], which will expedite the application and improvement of the state-of-the-art CT analytical modules implemented by the contest winners. Limitations ----------- Since it was difficult to compile and release a large deidentified chest CT dataset to the public, the public test set only contains images from 198 patients. Leveraging the 5-digit precision of the log-loss value shown on the leaderboard, one participant implemented and shared a method for identifying all ground truth labels in the public test set during the competition \[[@ref41]\]. Several participants successfully replicated this approach and got perfect scores on the public leaderboard. Thus, solutions with very low log-loss in the public test set may result from information leakage. Interestingly, among the top-10 models defined by the final test set, 2 performed worse than random guessing in the public test set, which raised concerns on their generalizability \[[@ref42]\]. There are several approaches future contest organizers can take to ensure the generalizability of the developed models. First, a multistage competition can filter out the overfitted models using the first private test set and only allow reasonable models to advance to the final evaluation. In addition, organizers can discourage leaderboard probing by only showing the performance of a random subset of the public test data or limiting the number of submissions allowed per day. Finally, curating a larger test set can better evaluate the true model performance and reduce random variability \[[@ref43]\]. If data deidentification is difficult, requiring contestants to submit their models to a secure computing environment rather than distributing the test data to the participants can minimize the risk of leaking identifiable medical information. Conclusion ---------- In summary, we compared, reproduced, and Dockerized state-of-the-art pulmonary nodule segmentation and classification modules. Results showed that many transfer learning approaches achieved reasonable accuracy in diagnosing chest CT images. Future works on additional data collections and validation will further enhance the generalizability of the current methods. The authors express their appreciation to Dr Steven Seltzer for his feedback on the manuscript; Drs Shann-Ching Chen, Albert Tsung-Ying Ho, and Luke Kung for identifying the data resources; Dr Mu-Hung Tsai for pointing out the computing resources; and Ms Samantha Lemos and Nichole Parker for their administrative support. K-HY is a Harvard Data Science Fellow. This work was supported in part by the Blavatnik Center for Computational Biomedicine Award and grants from the Office of the Director, National Institutes of Health (grant number OT3OD025466), and the Ministry of Science and Technology Research Grant, Taiwan (grant numbers MOST 103-2221-E-006-254-MY2 and MOST 103-2221-E-168-019). The authors thank the Amazon Web Services Cloud Credits for Research, Microsoft Azure Research Award, and the NVIDIA Corporation for their support on the computational infrastructure. This work used the Extreme Science and Engineering Discovery Environment Bridges Pylon at the Pittsburgh Supercomputing Center (through allocation TG-BCS180016), which is supported by the National Science Foundation (grant number ACI-1548562). Conflicts of Interest: None declared. AAPM : American Association of Physicists in Medicine CNN : convolutional neural network CT : computed tomography LUNA16 : Lung Nodule Analysis 2016 ResNet : residual net skimage : Scikit-image SPIE : International Society for Optics and Photonics

{.117} {.118}

Background {#Sec1} ========== Pepper (*Capsicum* spp.) is one of the major vegetable and spice crops grown worldwide, and is rich in bioactive compounds, such as capsaicinoids and carotenoids, which contribute to the improvement of human health \[[@CR1], [@CR2]\]. Because of its economic and nutritional importance, breeders have improved agronomic traits of pepper, such as pungency, fruit shape, abiotic stress tolerance, and disease resistance. Meanwhile, genetic diversity of breeding lines has become smaller and some useful genes in the landraces are lost due to the breeding activities \[[@CR3], [@CR4]\]. Therefore, conservation and sustainable utilization of genetic resources are keys to continuous improvement of peppers \[[@CR5]\]. During the last several decades, there has been remarkable progress in germplasm collection and conservation of various plants. Although a large number of germplasms have been collected, their management has become more and more complicated due to their huge sizes. Furthermore, little is known about the genetic diversity and structure of such collections at the interspecific and intraspecific levels \[[@CR6]\]. To make efficient use of large germplasm collections, the concept of core collections has been proposed. A core collection is a subset of a germplasm collection of a species that represents the genetic diversity of the entire collection \[[@CR7]\]. A good core collection is one that has no redundant accessions, is small enough to be easily managed, and represents the total genetic diversity \[[@CR8]\]. Various types of data including passport data, geographic origin \[[@CR9], [@CR10]\], agronomic traits \[[@CR11]--[@CR13]\], and molecular markers \[[@CR14]\] can be used for selecting a core set. Although the major reason for establishing a core set is to reduce the number of representative accessions up to 10 % while maintaining the diversity of the entire collection, there are a number of possible methods for selection of a core set depending on the research goals. In the early 2000s, most researchers performed random sampling using various assignment methods \[[@CR9], [@CR11]\]. Later, the M (maximization) strategy was proposed as a more effective method to select a core set representing the maximum genetic diversity without redundancy \[[@CR12], [@CR15]\]. Several research institutions have collected and conserved thousands of *Capsicum* accessions, ranging from 1,000 in the Centre for Genetic Resources (CGN), the Netherlands \[[@CR16]\] to almost 8,000 in the Asian Vegetable Research and Development Center (AVRDC), Taiwan \[[@CR17]\]. Researchers and institutions have attempted to construct core collections of *Capscicum* spp. for various purposes. Fan et al. \[[@CR13]\], Nicolai et al. \[[@CR14]\], and Zewdie et al. \[[@CR12]\] established core collections to reveal phenotypic and genetic variation. Thies and Fery \[[@CR9]\], and Quenouille et al. \[[@CR10]\] constructed a core collection for disease resistance against northern root-knot nematode and *Potato virus Y* (PVY), respectively. Hanson et al. \[[@CR11]\] developed a core collection to analyze antioxidant activities. However, most studies involved a relatively small number of accessions, using fewer than 1,000 accessions with limited numbers of morphological traits and molecular markers \[[@CR11], [@CR12], [@CR14]\]. The limited number of morphological traits and markers allow us to survey only a small portion of the genetic diversity of the entire germplasm, and the resulting data cannot be used for genome-wide variation studies. In this study, we performed population structure analysis in a large *Capsicum* germplasm collection consisting of 3,821 accessions by applying 48 genome-wide SNPs, and selected a core set using the SNP data together with data for 32 morphological traits. This allowed us to 1) examine the level of genetic diversity and the population structure within the worldwide *Capsicum* germplasm collection; 2) optimize selection methods by comparing different core sets, which were selected using a stepwise selection strategy based on various combinations of data and clustering methods; and 3) ultimately construct a *Capsicum* core collection that represents the entire germplasm collection without redundancy. Finally, we validated the core collection by evaluating the diversity of a range of traits and genotyping additional molecular markers. This core collection will be a valuable data set for both pepper breeding and genome-wide association studies. Methods {#Sec2} ======= Plant materials {#Sec3} --------------- A total of 4,652 *Capsicum* accessions used in this study originated from 97 countries and included 11 species: *C. annuum*, *C. baccatum*, *C. cardenasii*, *C. chacoense*, *C. chinense*, *C. eximium*, *C. frutescens*, *C. galapagoense*, *C. praetermissum*, *C. pubescens*, and *C. tovarii*. The geographic origin and passport data of the germplasm accessions were obtained from the Rural Development Administration (RDA, Jeonju, Korea) and Seoul National University (SNU, Seoul, Korea). Among the germplasm accessions, 3,599 were obtained from the RDA, and 1,053 were obtained from SNU. Most of the accessions were *C. annuum*, accounting for 4,163 accessions. Four other domesticated species, *C. baccatum*, *C. chinense*, *C. frutescens*, and *C. pubescens* accounted for 163, 122, 152, and 11 accessions, respectively. Among the wild *Capsicum* species, *C. cardenasii*, *C. chacoense, C. eximium*, *C. galapagoense*, *C. praetermissum* and *C. tovarii* accounted for 1, 28, 4, 2, 5, and 1 accessions, respectively. DNA extraction and SNP genotyping {#Sec4} --------------------------------- Two young leaves from each accession were used for DNA extraction. DNA was extracted using the cetyl trimethylammonium bromide (CTAB) method as described previously \[[@CR18]\]. The concentration and purity of DNA samples were determined with a NanoDrop 1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA). DNA samples showing absorbance ratios above 1.8 at 260/280 nm were used for marker analysis. A set of 48 SNP markers evenly distributed in 12 pepper chromosomes were used in this study \[[@CR19]\] (Additional file [1](#MOESM1){ref-type="media"}: Table S1). In a preliminary study a total of 282 accessions were randomly selected from entire germplasm collection for genetic diversity study with 412 SNP markers developed by Kang et al. \[[@CR19]\]. Based on this analysis, highly polymorphic SNP markers (PIC \> 0.45) were selected. Genotyping was performed using the BioMark™ HD system (Fluidigm, San Francisco, CA, USA), EP1™ system (Fluidigm, San Francisco, CA, USA), and 48 × 48 Dynamic Array IFCs (Fluidigm, San Francisco, CA, USA) according to the manufacturer's protocol \[[@CR20]\]. Specific target amplification (STA) was performed prior to SNP genotyping analysis. PCR was performed in a 5-μL reaction containing 60 ng of the DNA sample according to the manufacturer's protocol. Thermal cycling conditions were 15 min at 95 °C, followed by 14 cycles of a 2-step amplification profile of 15 s at 95 °C and 2 min at 60 °C. For genotyping, SNPtype assays were performed using STA products following manufacturer's protocol. Thermal cycling was carried out at 95 °C for 15 s, 64 °C for 45 s and 72 °C for 15 s with a touchdown of −1 °C per cycle from 64 to 61 °C, followed by 34 cycles of 95 °C for 15 s, 60 °C for 45 s and 72 °C for 15 s. For the species verification and/or identification of pepper accessions with missing species information, SNP markers C2_At5g04590, C2_At1g50020, and C2_At2g19560 were used based on high resolution melting (HRM) analysis \[[@CR21]\]. Genotyping analysis was performed using a Rotor Gene 6000 (Qiagen, Valencia, CA, USA). Population structure analysis {#Sec5} ----------------------------- To analyze the population structure of the entire germplasm collection used in this study, we used a model based genetic clustering algorithm \[[@CR22]\] as implemented in the STRUCTURE program ver. 2.3.4 \[[@CR23]\]. The number of sub-populations (ΔK) was determined using the *ad-hoc* statistical method, based on the rate change in the log probability of data between successive K values \[[@CR24]\]. Fifty independent runs for K values ranging from 1 to 20 were performed with a burn-in length of 50,000 followed by 1,000,000 iterations. Phylogenetic and principal coordinate analyses {#Sec6} ---------------------------------------------- Phylogenetic trees were produced using genotyping data with 48 SNP markers using both the unweighted neighbor-joining method and the hierarchical clustering method based on the dissimilarity matrix calculated with Manhattan index, as implemented in the DARwin software (version 6.0.9). Principal coordinate analyses were also performed with DARwin 6.0.9 \[[@CR25]\]. Statistical analysis of genetic diversity {#Sec7} ----------------------------------------- Different indices were used for analysis and comparison of diversity among the *Capsicum* collections. These include levels of observed heterozygosity (H~O~), expected heterozygosity (H~E~), polymorphic information content (PIC), genetic differentiation (F~ST~), Shannon's information index of diversity (I), and genetic evenness (J'). Indices Ho, H~E~, PIC, and F~ST~ were calculated using Power Marker 3.25 \[[@CR26]\]. For analysis of genetic diversity of core collections, I and J' were calculated following Hennink and Zeven \[[@CR27]\] and Pielou \[[@CR28]\], respectively. Analysis of molecular variance (AMOVA) was conducted to detect the genetic variance within and among population using GenAlEx ver 6.502 \[[@CR29]\]. Establishment of the core collection {#Sec8} ------------------------------------ To establish a core collection, five different methods were used. Specifically, core sets were selected based on 1) genotype analysis of the entire collection, 2) genotype analysis of each cluster after grouping based on genotype dissimilarity, 3) phenotype analysis of the entire collection, 4) a combination of genotype and phenotype analysis of entire collection, and 5) a combination of phenotype and genotype analysis of each cluster after grouping based on genotype dissimilarity. Representative accessions were selected based on the advanced M strategy using a modified heuristic algorithm implemented in PowerCore software \[[@CR30]\]. Categorical variables, such as genotype and qualitative phenotype were applied in several classes (3 to 12 classes) based on distinct characters. Continuous variables (quantitative phenotypes, 7 to 12 classes) were automatically classified into different categories in the software based on Sturges' rule \[[@CR31]\]. Therefore, a total of 264 phenotypic alleles were used to select the core entries (Additional file [1](#MOESM1){ref-type="media"}: Table S2). Evaluation of the core collections {#Sec9} ---------------------------------- To evaluate each core collection, diverse statistical indicators were calculated for two types of variables, continuous and categorical variables. For continuous variables, the percentage of significant difference between core collections and the entire germplasm collection was calculated based on the mean difference (MD) percentage, the coincidence rate (CR) of range, the variance difference (VD) percentage, and variable rate (VR) of coefficient of variation. Among the candidate core sets selected from each different data set, a core set with MD less than 20 % and CR more than 80 % was considered as a representative collection. In addition, a lower value in VD and higher value in VR was considered to indicate a more effective core collection \[[@CR32]\]. For categorical variables, the I and J' values were calculated and compared between the five core collections and the entire germplasm collection. The maximum value of I (I max) is calculated based on the log of the number of classes used in the entire collection; the value for a core collection should be comparable to that of the entire collection \[[@CR8]\]. Three additional markers having multiple alleles, COS643, COS111, and L4RP-3 F, which were selected from the Sol Genomics Network \[[@CR33]\] and Yang et al. \[[@CR34]\], were used for validation of the core set. Melting curve patterns were identified by HRM analysis using a Rotor Gene 6000 (Qiagen, Valencia, CA, USA). Thermal cycling conditions were 10 min at 95 °C, 50 cycles of 3-step amplification profile of 20 s at 94 °C, 20 s at 55 °C, and 40 s at 72 °C, followed by final extension 60 s at 95 °C and 60 s at 40 °C. HRM analysis was performed increasing 0.1 °C for every two seconds from 70 to 90 °C. Finally, the core collection (CC240) with the highest genetic diversity and evenness was planted in 2014 in a research farm (Suwon, Korea) to monitor the variation of the diverse traits. Morphological data were obtained for the same accessions that were genotyped. Thirty-two different traits related to plant habit (9), leaf (4), flower (6), fruit (10), and seed (3) were analyzed. Phenotype data were presented as the mean ± SE. The differences between the mean values of individual clusters were assessed using one-way ANOVA and Duncan's multiple range tests. *P* \< 0.05 was considered to indicate a statistically significant difference. The IBM SPSS Statistics v23 software (IBM Corp., Armonk, NY, USA) was used for analysis. Results {#Sec10} ======= Genetic diversity of the *Capsicum* germplasm {#Sec11} --------------------------------------------- In our preliminary studies, a total of 4,652 non-redundant accessions from 11 species were screened using SNP markers to reveal the genetic diversity (Additional file [1](#MOESM1){ref-type="media"}: Table S3). Based on the H~O~ values, 673 accessions mostly from *C. annuum* with Ho value more than 0.3 were considered as F1 hybrids (Additional file [2](#MOESM2){ref-type="media"}) and excluded from analysis. In addition, 158 accessions with more than seven missing genotype data points were also excluded. Ultimately, a total of 3,821 accessions were used for further experiments (Table [1](#Tab1){ref-type="table"}).Table 1Genetic diversity analysis of the 3,821 pepper accessionsSpeciesNumberH~O~H~E~I*C. annuum*3,3830.120.440.69*C. baccatum*1500.120.260.51*C. cardenasii*10.210.10.14*C. chacoense*240.170.280.54*C. chinense*1050.110.380.56*C. eximium*30.140.230.45*C. frutescens*1370.090.370.55*C. galapagoense*10.130.060.07*C. praetermissum*50.210.180.31*C. pubescens*110.160.120.29*C. tovarii*10.150.070.12Total3,8210.150.230.38*Ho* observed heterozygosity, *H* ~*E*~ expected heterozygosity, *I* Shannon's information index of diversity Using the SNP genotyping results, the H~E~, H~O~, and I were calculated for 3,821 pepper accessions (Table [1](#Tab1){ref-type="table"}). The H~E~ values ranged from 0.10 to 0.44, and I values ranged from 0.07 to a maximum of 0.69. The highest diversity values in *C. annuum* accessions (H~E~ = 0.44, I = 0.69) suggests that there is extensive genetic variation within this species. With the exceptions of *C. baccatum* and *C. pubescens*, the other domesticated species showed relatively high H~E~ values, above 0.37. The H~O~ value of *C. annuum* was 0.12, whereas those of the other species varied from 0.09 to 0.21. Four domesticated species *C. annuum*, *C. baccatum*, *C. chinense*, and *C. frutescens* and two wild species *C. chacoense*, and *C. eximium* had lower values for H~O~ compared to H~E,~ (Table [1](#Tab1){ref-type="table"}) whereas *C. cardenasii*, *C. galapagoense, C. pratermissum*, *C. pubescens*, and *C. tovarii* had relatively higher values of H~O~ compared to H~E~. This pattern suggests that the first six species have experienced inbreeding for a long time which could be attributed to the interplay of many factors such as artificial selection, non-random mating between individuals, population structure and size, and Wahlund effect (mixing of individuals from different genetic sources) \[[@CR35], [@CR36]\]. By contrast, accessions of the latter five species were collected in different isolated locations where each accession had evolved independently. Population structure of the germplasm collection {#Sec12} ------------------------------------------------ The SNP genotyping results were used to perform population structure analysis for the 3,821 accessions under an admixed model using the STRUCTURE program \[[@CR23]\]. Estimated likelihood (LnP (D)) was found to be greatest when K = 10, suggesting that the population used in this study can be divided into ten clusters (Fig. [1](#Fig1){ref-type="fig"}). The clusters 3, 8, 9, and 10 were rather well separated from others whereas the cluster 1, 2, 4, 5, 6, and 7 were admixtures. Each of the 10 clusters included different numbers of accessions, ranging from 85 to 806 (Table [2](#Tab2){ref-type="table"}). The average distance (H~E~) between individuals in each cluster was 0.32. The highest H~E~ value of 0.43 was observed in cluster 5, indicating greater genetic diversity within this cluster, whereas cluster 9 showed the lowest H~E~ value of 0.11. Genetic differentiation (F~ST~) values varied from 0.08 to 0.78 with an average of 0.33. The smallest F~ST~ value (0.08) was observed in cluster 5, whereas cluster 9 had the highest F~ST~ value (0.78), indicating that accessions in this cluster have several different genotype patterns.Fig. 1Population structure of the *Capsicum* germplasm collection. **a** ΔK reached its maximum value when K = 10 following the *ad-hoc* method. **b** Ten subpopulation clusters inferred by STRUCTURE are represented by different colors Table 2Diversity-related summary statistics for all clusters inferred by STRUCTURE analysisClusterNumberH~E~IF~ST~13410.410.610.1328060.340.530.2734870.370.570.2245350.400.600.1654260.430.660.0863410.350.540.2774610.330.510.318850.200.340.5791960.110.190.78101430.240.400.49Total3,8210.320.490.33*H* ~*E*~ expected heterozygosity, *I* Shannon's information index of diversity, *F* ~*ST*~ genetic differentiation Most of the *C. annuum* accessions were found in clusters 1 to 7. *C. chinense* was mostly distributed in clusters 8 and 9, whereas *C. frutescens* was mostly found in clusters 4, 9, and 10. By contrast, *C. baccatum* was distributed in clusters 8, 9, and 10. *C. pubescens* was placed in cluster 10. Wild species *C. chacoense*, *C. cardenasii*, *C. eximium*, *C. praetermissum*, and *C. tovarii* were distributed in cluster 10 along with *C. baccatum* accessions (Additional file [1](#MOESM1){ref-type="media"}: Table S4). Although not fully distinct, the ten clusters were roughly separated according to geographic distribution. Clusters 1 to 3 were composed of an admixture of accessions from East Europe countries (Additional file [3](#MOESM3){ref-type="media"}). Cluster 4 to 7 were mostly composed of collections from East Asia. Interestingly, the Korean landraces belonged to clusters 6 and 7. The accessions of clusters 8 to 10 were mostly from South America. Molecular phylogenetic analysis of the germplasm collection {#Sec13} ----------------------------------------------------------- Using the genotyping data, an unrooted phylogenetic tree of the 3,821 pepper accessions was generated using the unweighted neighbor joining method based on genetic dissimilarity calculated with the Manhattan index. The tree showed five large clades (A-E) in which accessions of *C. annuum* were grouped separately from the other species (Fig. [2](#Fig2){ref-type="fig"}). The *C. annuum* accessions were found in four large clades. *C. annuum* accessions collected (or originated) in European countries were distributed among upper branches including the clades A and B. *C. annuum* accessions from Asian countries were distributed among lower branches (clades C and D). The accessions belonging to other species were clustered together in clade E. Within clade E, most of *C. chinense* accessions were clearly distinguished from those of other species and were placed next to clade D. When the unrooted phylogenetic tree was compared with the clusters obtained from the STRUCTURE analysis, the phylogenic tree matched well with the cluster separation in the STRUCTURE analysis. Accessions in cluster 3 belonged to clade A, accessions in clusters 1 and 2 to clade B, and accessions in clusters 4 to 7 to clade C; some of the accessions in cluster 5 and admixtures belonged to clade D and the accessions in clusters 8 to 10 were in clade E (Figs. [1](#Fig1){ref-type="fig"} and [2](#Fig2){ref-type="fig"}). The unrooted phylogenetic tree more clearly differentiated groups according to their geographic origin.Fig. 2Unrooted phylogenetic tree of the *Capsicum* germplasm. The dendrogram was produced using the unweighted neighbor-joining method based on genetic dissimilarity among the 3,821 germplasm accessions. The colors of branches indicate accessions corresponding to the clusters (cluster 1 to 10) from population structure analysis as in Fig. [1](#Fig1){ref-type="fig"}. Five clades (**a**-**e**) were distinguished by distance between branches; **a** and **b** included *C. annuum* species mostly from Europe; **c** and **d** included *C. annuum* species mostly from Asia; **e** comprised other *Capsicum* species except *C. annuum*. **a** to **e** were used for a clustering range to select entries to construct the core collection Optimization of core set selection methods and construction of the core collection {#Sec14} ---------------------------------------------------------------------------------- To determine the best possible method for selection of a core collection, five different sampling strategies were tested (Table [3](#Tab3){ref-type="table"}). The first three methods were carried out using the entire collection without considering clustering analysis. These methods included selection of core entries based on SNP genotype data only (Gcc), phenotype data only (Pcc), or the combination of genotype and phenotype data (G + Pcc). In the other two methods, core entries were selected from each cluster of the unrooted phylogenetic tree. In these methods, after analysis of the entire collection, core entries in each cluster were selected based on genotype data only (Ggcc), or the combination of genotype and phenotype data (Gg + Pcc).Table 3Comparisons of distribution frequency and representativeness among five different core collections constructed based on diverse sampling strategiesEvaluated parameterData combinationEntire germplasm collectionGccGgccPccG + PccGg + PccNumber of entries (%)7 (0.2)53 (1.4)76 (2.0)76 (2.0)240 (6.3)3,821Number of alleles9696264360360360Continuous variablesMD %38.2818.8915.8515.359.45-CR %13.9550.8996.9997.0498.40-VD %5757.7080.0249.3746.5532.46-VR %52.15112.37126.99125.95115.65-Categorical variablesI0.880.910.940.970.950.85I max1.251.251.251.251.251.25J'0.770.790.780.810.800.73Gcc: core collection constructed based on genotype with nongroup based strategy, Ggcc: core collection constructed from entries in each cluster grouped by genotype, Pcc: core collection constructed based on phenotype, G + Pcc: core collection constructed based on combination of genotype and phenotype, Gg + Pcc: core collection constructed based on genotype and phenotype combination from each cluster grouped by genotype. Distributional aspects to validate a representativeness of core collection; MD%: the mean difference percentage, CR%: the coincidence rate, VD%: the variance difference percentage, VR%: variable rate. Genetic diversity indices to validate categorical variables; I: Shannon's information index of diversity, I max: logarithmic number of classes in entire collection, J': genetics evenness When only genotype data were used, 7 and 53 core entries were selected for Gcc and Ggcc, respectively. Ggcc showed a MD of less than 20 %, which is close to the mean value of the entire collection, whereas Gcc showed a MD of more than 20 %, poorly representing the entire collection. Both of them showed a CR of less than 80 %, demonstrating insufficient coverage of the phenotype variation of the entire collection. However, Ggcc exhibited a smaller percentage of VD and larger percentage of VR, which indicated that selection of a core set after clustering analysis (Ggcc) better represented the entire collection. Furthermore, the comparison of categorical variables including 48 SNPs markers and 15 qualitative traits produced a higher value in I for Ggcc (0.91) than for Gcc (0.88), but a similar value in J' for Ggcc (0.79) and Gcc (0.77). Therefore, selecting the core entries after clustering analysis is more effective to represent the entire collection in terms of both phenotype and genotype data even using same number of alleles. Since the core sets selected using only genotype data could not represent the diversity of the entire collection presumably due to limitations of number of SNP markers used, the available phenotype data for 32 traits were included for selection of core sets. Each trait included 3 to 12 phenotype classes providing at least 264 variations (Additional file [1](#MOESM1){ref-type="media"}: Table S2). A total of 76 entries were selected based on only phenotype data (Pcc) and produced 15.85 % in MD, 96.99 % in CR, 0.94 in I and 0.78 in J', which reflects slightly better representation of the entire collection than that of Gcc. When both genotype and phenotype data were used (G + Pcc), the same number of entries (76), but slightly better representation of the entire collection was achieved compared to that with Pcc. As we found that selection of a core set after clustering analysis more effectively represented the entire collection, the final core collection was built using a combination of genotype and phenotype data after cluster analysis (Gg + Pcc). A total of 240 accessions representing six species, *C. annuum* (176), *C. baccatum* (21), *C. chinense* (22), *C. eximium* (2), *C. frutescens* (18), and *C. praetermissum* (1) were ultimately selected as a core collection (CC240) (Additional file [1](#MOESM1){ref-type="media"}: Table S5). Compared with the entire germplasm collection, CC240 showed 9.45 % in MD and 98.40 % in CR, which provided good coverage of most of the range of continuous phenotypes in the entire collection. Furthermore, CC240 showed the lowest MD and the highest CR of all tested core collections. In addition, the values of I and J' were 0.95 and 0.80, respectively, which represents increased genetic diversity compared to the entire germplasm collection. To validate and confirm the distribution of core entries, core collections selected from five methods PCA was performed. The distribution of the entire germplasm collection and core collection entries on the basis of genotyping was explained by the first two principal components, where the first and second axes explained 8.13 and 6.19 % of the total variation among the accessions, respectively, and showed a clear separation of *C. annuum* from other species. PCA analysis based on phenotyping included 589 accessions with no more than 10 % of missing data points, whereas 2,006 accessions were plotted in the genotype plus phenotype background with no more than 20 % missing data to reach the least condition of unit pairing. In contrast to the phenotype (21.87 %, 11.44 %), the genotype (8.13 %, 6.19 %) and genotype plus phenotype (6.44 %, 4.91 %) revealed lower variation in each axis (Fig. [3](#Fig3){ref-type="fig"}). Overall, regardless of the selection method, PCA analysis showed that core entries were distributed evenly without obvious grouping, covering the variation of the entire germplasm collection.Fig. 3Principal coordinate analysis of distributions of diverse core entries selected from different sampling strategies. PCA was performed with DARwin 6.0.9 software based on distance matrices. **a** distribution of pepper accessions based on genotype included 3,821 accessions from entire germplasm collection, and 7, 53, 76, 76, and 240 accessions from Gcc, Ggcc, Pcc, G + Pcc, and Gg + Pcc sampling strategies, respectively; **b** distribution of pepper accessions based on phenotype included 589 accessions from entire germplasm collection, and 1,9, 42, 43, and 115 accessions from Gcc, Ggcc, Pcc, G + Pcc, and Gg + Pcc sampling strategies, respectively; **c** distribution of pepper accessions based on genotype and phenotype combination included 2,006 accessions from entire germplasm collection, and 3, 27, 76, 76, and 235 accessions from Gcc, Ggcc, Pcc, G + Pcc, and Gg + Pcc sampling strategies, respectively. Matrices surrounded with blue boxes indicate distributions of the pepper accessions from entire collection Evaluation of the core set using markers with multiple alleles {#Sec15} -------------------------------------------------------------- Evaluation of the quality of core collections should be based on data that were not used in the selection of the core set \[[@CR37]\]. Accordingly, three additional multiple allelic markers, COS643, COS111, and L4RP-3 F, were used to evaluate the core set (Additional file [1](#MOESM1){ref-type="media"}: Table S6); the markers had 9, 16, and 9 alleles, respectively, in the entire germplasm collection (Table [4](#Tab4){ref-type="table"}). The numbers of alleles in CC240 were the same as those in entire collection except for COS111 (11 instead of 16). The genetic diversity of CC240 revealed by these markers was compared with that of the entire collection. The average value for I in CC240 was higher (1.65) than that of the entire collection (1.50). Furthermore, the average genetic evenness was more stable in CC240 (0.74) compared to the entire collection (0.65). However, L4RP-3 F did not show a difference in genetic diversity or evenness because the I value of the entire collection (2.02) for that marker was already close to maximum value (I max = 2.20). In summary, the high genetic diversity and evenness of CC240 evaluated by three additional markers demonstrated that the core accessions in CC240 well represent the entire collection.Table 4Comparison of genetic diversity between the 3,821 accession collection and different core collections using an additional set of multiplex markersCriteria3,821 germplasm collectionCC240COS643COS111L4RP-3 FAvg.COS643COS111L4RP-3 FAvg.Genotype patterns9169-9119-I max2.202.772.202.392.202.302.202.23I1.540.942.021.501.791.221.961.65J'0.700.340.920.650.810.530.890.74I max: logarithmic number of classes in entire collection, I: Shannon's information index of diversity, J': genetic evenness Morphological variations of CC240 {#Sec16} --------------------------------- Accessions of CC240 were planted in an experimental farm and grown for 1 year to evaluate various traits. Four accessions, namely Javitott bogyiszloi (*C. annuum*), 9146 (*C. annuum*), Tabasco (*C. frutescens*), and 9148 (*C. frutescens*) were excluded from the phenotype analysis due to poor growth. Thus, phenotype evaluation was performed for 236 accessions for 32 different traits (Additional file [1](#MOESM1){ref-type="media"}: Table S2). Overall, CC240 showed a similar range of diversity in morphological traits as that of entire collection. For plant architecture, about one half of the accessions (105) showed the half-spreading phenotype. Plant height was varied between 40 cm to 200 cm, and plant width ranged between 25 cm to 130 cm. Leaf color varied from light green to dark green except for one accession having purple leaves; leaf length was 4.83 cm to 15.43 cm, and leaf width was 2.13 cm to 8.77 cm. Flower color of most accessions (188) was white, whereas 23 had light green flowers, and 21 had white flowers with yellow spots. Among all accessions, the earliest flowering date was 63 days from planting in *C. annuum* 'Swedish' and 'A9E0211', whereas the latest date was 103 days in *C. baccatum* 'C01543' and *C. annuum* 'ACC160'. Length of fruit was distributed between 4.8 mm and 249 mm with an average of 72.66 mm, fruit width varied between 4.8 mm and 84.12 mm with an average of 24.82 mm, and thickness of pericarp was 0.2 mm to 7.2 mm. Fruit weight was distributed between 0.06 g and 177.32 g. There were five to 256 seeds in each fruit (Fig. [4](#Fig4){ref-type="fig"}).Fig. 4Comparison of phenotypic measurements among five clusters in CC240. Dendrogram was generated by hierarchical clustering (UPGMA) based on genetic dissimilarity. Average values of 10 different phenotypic characters (plant height, plant width, leaf length, leaf width, flowering date, fruit length, fruit width, fruit pericarp thickness, fruit weight, and number of seeds per fruit) were recorded to compare among the five clusters. Data are presented as the mean ± SE. *P* \< 0.05 was considered to indicate a statistically significant difference, indicated by different lowercase letters Using genotype, accessions of CC240 were divided into five clusters (I-V) by hierarchical clustering (UPGMA) based on genetic dissimilarity of 48 SNP markers. Clusters I, II, and III included *C. annuum* species whereas clusters IV and V included other species, such as *C. baccatum*, *C. chinense*, *C. eximium*, *C. frutescens*, and *C. praetermissum*. Among the 32 morphological traits, 10 different quantitative traits including plant height, plant width, leaf length, leaf width, flowering date, fruit length, fruit width, fruit weight, number of seeds per fruit, and fruit pericarp thickness were significantly different between the clusters (Fig. [4](#Fig4){ref-type="fig"}). Cluster I was characterized by large-fruited peppers with thick pericarp. Leaves were large, plant height and width are slightly shorter, and flowering date was relatively earlier than that of *C. annuum* accessions in cluster III. Accessions in cluster II were characterized by small and short-fruited peppers. The pericarp of accessions in cluster II was thinner than that of those in cluster I, whereas plant height and width were slightly larger than those of cluster I. The flowering date of cluster II was similar to that of accessions of cluster I and much earlier than that of cluster III. Accessions in cluster III were characterized by elongated fruits. Plant height and width were larger and flowering date was later when compared with those of accessions in clusters I and II. The species included in clusters IV and V also exhibited differences in fruit shape, where slightly smaller fruit and thicker pericarp were observed for cluster IV. Plant height and width of cluster IV were smaller than those of accessions of cluster V. Slightly wider fruits with thinner pericarp were observed for cluster V. Leaf size was much larger, and the flowering date was slightly later than those of accessions of cluster IV. Overall, accessions in clusters IV and V exhibited small fruits with slightly higher plant height and late flowering. Discussion {#Sec17} ========== Despite numerous *Capsicum* germplasm accessions having been documented worldwide, little is known about their population structure or genetic diversity in contrast to other crops. Previously, *Capsicum* germplasm collections have been examined for genetic diversity using accessions from Mesoamerica (Central Mexico to northwestern Costa Rica) to survey geographic origin and understand the domestication process \[[@CR6], [@CR38]\]. Recently, STRUCTURE analysis was performed in a *Capsicum* germplasm collection with 1,352 accessions, which was grouped into six distinct clusters based on genetic analysis with six SSR markers \[[@CR14]\]. In the present study, a *Capsicum* germplasm collection consisting of 3,821 accessions was divided into ten clusters by STRUCTURE analysis and five distinct groups by phylogenetic analysis (Figs. [1](#Fig1){ref-type="fig"} and [2](#Fig2){ref-type="fig"}). The AMOVA analysis revealed that the genetic variance among and within the populations was significant (*p* ≤ 0.01). Variance among populations and within a population of five phylogenetic groups were seven and 93 %, respectively and the variance among and within populations of ten STRUCTURE clusters were 31 and 69 %, respectively (Additional file [1](#MOESM1){ref-type="media"}: Table S7). Both STRUCTURE and phylogenetic analyses showed that *C. annuum* accessions were separated from other species and sub-clustered into two large groups, one from European countries and the other from Asian countries. In comparison to the STRUCTURE analysis, the unrooted phylogenetic tree showed rather clear separation according to geographic origin and species classification. Accessions collected from Korea were spread in two clusters (clusters 6 and 7) as per the population structure analysis, whereas in the unrooted phylogenetic tree accessions corresponding to those two clusters were placed in a same node. Clade E included species other than *C. annuum* and showed distinct grouping. *C. chinense* accessions were separated out from other species and closely placed next to *C. annuum*. In pepper breeding, agriculturally useful traits such as disease resistance, fragrance, yield, and pungency have been introgressed from wild species by interspecific hybridizations. Among the domesticated species, *C. chinense* has better crossability with *C. annuum* and is used as a bridge species between *C. annuum* and other species \[[@CR39], [@CR40]\]. The location of *C. chinense* in this tree, between *C. annuum* and other species, may explain why *C. chinense* has played a role as an interspecific cross-bridge. Based the topology of the phylogenic tree, *C. annuum* accessions in clade D (Fig. [2](#Fig2){ref-type="fig"}) are candidate to be used as interspecific bridges to introgress genes from other species. In this study, we confirmed that *C. galapagoense* was located more closely with *C. annuum* than other species. However, other species such as *C. baccatum*, *C. frutescens*, *C. pubescens*, and *C. chacoense* were not clearly separated from each other (Fig. [2](#Fig2){ref-type="fig"}). In previous work \[[@CR14]\], classification with SSR markers showed rather clear distinction of species. It may indicate that SSR markers are more prone to be affected by speciation and evolution processes, whereas SNP markers are more appropriate for the analysis of genetic variation in various aspects of agronomic and morphological traits \[[@CR41]\]. It is also possible that we did not use enough SNP markers to allow clear differentiation among species. In this study, we used 96 alleles to survey genetic diversity. The cost of genotyping of large germplasm collection is relatively expensive, therefore based on our preliminary studies with 412 SNPs \[[@CR19]\], 48 SNPs with high PIC values were used for diversity study. Even though most of the SNP makers used in this study had high PIC values close to 0.5, SNP markers are less powerful than SSR markers in terms of relative kinship estimation and population structure analysis \[[@CR42]--[@CR44]\] because SSR markers have higher allelic diversity than SNP markers. To compensate for the small number of SNP markers, we also used 32 different traits which account for 264 phenotypic variations to build a core collection. Core collection built by more variations showed higher genetic diversity, evenness and representation. These results indicated that even with a small number of SNP markers used combination with diverse phenotypic data can be also effective to construct a core collection with the aim to conserve the phenotypic and genetic variability within species. Representative core accessions have been selected in diverse crops using various sampling strategies combined with different clustering methods \[[@CR15], [@CR32], [@CR45]--[@CR47]\]. Among the strategies, the M strategy was reported to be a useful method in selecting a core set conserving high genetic diversity with a reasonable size \[[@CR45]\]. There are two representative core selection methods implementing the M strategy, namely the MSTRAT algorithm \[[@CR48]\] and PowerCore software \[[@CR30]\]. Here, we used the advanced M strategy as implemented in PowerCore 1.0 software and successfully established a representative core collection with high genetic diversity. The advanced M strategy is based on the M strategy with heuristic searching that enables retention of all variations of the entire collection in the core collection with a minimum number of accessions. This strategy is more effective when using continuous variables in the dataset to capture a maximum of alleles with a minimum redundancy \[[@CR30], [@CR49]\]. Use of either genotype or phenotype information only for selection of core collection entries may not be efficient for capturing genetic diversity of the entire germplasm of a species. Therefore, we used both genotype and phenotype information along with clustering to select core collection entries. To determine the optimal core set selection methods, we compared five different methods and found that selection of the core set using genotype and phenotype data after clustering analysis (Gg + Pcc) is the best method (Table [3](#Tab3){ref-type="table"}). Moreover, we investigated the relationship between the number of clusters and genetic diversity among different core sets in clustering analysis. Different combinations of clades A, B, C, D, and E from the unrooted phylogenetic tree (Fig. [2](#Fig2){ref-type="fig"}) and the 10 clusters from the population structure analysis (Fig. [1](#Fig1){ref-type="fig"}) were considered to select a core set. Core sets selected from the cluster combinations in tree were named CG3, CG4, and CG5, respectively and CST10 were from the 10 clusters in STRUCTURE analysis (Additional file [1](#MOESM1){ref-type="media"}: Table S8). From those core collections, 174 to 420 entries were selected. Every collection showed higher values of genetic diversity and evenness than the 3,821 germplasm collection; however, the core collections did not show statistically significant difference from each other (*P* \> 0.05). Therefore, the number of clusters in the collection is not a critical factor to select highly diverse core entries. To reveal the phenotype variation in CC240, core accessions were clustered into five distinct subclusters based on genotype relationship (Fig. [4](#Fig4){ref-type="fig"}). Among the five clusters, three of them (I, II, and III) represented *C. annuum* and other two (IV and V) included other species, such as *C. baccatum*, *C. chinense*, *C. eximium*, *C. frutescens*, and *C. praetermissum* without clear species distinction. In a previous study, *Capsicum* germplasm was divided into six clusters \[[@CR14]\], in which three of them (1, 2, and 3) were composed of *C. annuum* accessions. Those three clusters were clearly distinguished mainly by fruit shape, such that cluster 1 was characterized by elongated fruited peppers, thin pericarp and late flowering, whereas cluster 2 exhibited conical fruit and rather thick pericarp, and cluster 3 had large-fruited peppers with thick pericarp and elongated-fruited peppers. Consistent with Nicolai's work \[[@CR14]\], three clusters in CC240 found in this study were characterized by large fruit with thick pericarp, large leaves, and early flowering date in cluster I, small, short fruit, small leaves, and early flowering date in cluster II, and elongated fruit with thin pericarp, late flowering date in cluster III. Thus, it appears that *C. annuum* accessions, which are mainly used as fundamental breeding materials, were clustered by breeding features based on food culture. Though, M strategy is the most powerful option for the selection of accessions with rich allelic diversity and for eliminating redundancies from noninformative alleles, it does not consider species composition while selecting core entries, which is a one of the disadvantage of the model based M strategy and therefore, future works should consider other measures of model fit including a rarefaction analysis, which corrects for sample sizes and manual inclusion of some representative wild species depending on the purpose of the core collection. Conclusions {#Sec18} =========== Establishing a core collection of *Capsicum* will enhance the proper utilization of *Capsicum* genetic resources. In the present study, based on population structure, a core collection (CC240) of *Capsicum* was constructed using 48 SNP markers and 32 different traits. The core collection 'CC240' is composed of six *Capsicum* species from 44 geographic locations and was found to represent the diversity of the entire germplasm collection. This core collection will serve as a primary source for SNP mining and further genetic association and functional analyses for novel genes in *Capsicum*. Additional files {#Sec19} ================ Additional file 1: Table S1.Summary of SNP markers used in the genetic analysis of pepper germplasm collection. **Table S2.** Description of various traits used in this study. **Table S3.** Genetic diversity analysis of the 4,652 pepper accessions. **Table S4.** Distribution of *Capsicum* species in STRUCTURE clusters within 3,821 germplasm accessions. **Table S5.** Accession in the 'CC240' core collection and group position of every accession in 'CC240' and the 3,821 germplasm collection. **Table S6.** Multiple allelic markers used to evaluate the core collection in *Capsicum.* **Table S7.** Analysis of molecular variance (AMOVA) among various subpopulations based on different clustering methods within 3,821 *Capsicum* accessions. **Table S8.** Comparison among different core collections established by diverse genetic clustering methods. CG3-CG5: Core collections based on unrooted phylogenetic tree clusters (A to E) grouped by 3 (A + B, C, D + E), 4 (A + B, C, D, E), and 5 (A, B, C, D, E) in respectively. CST10: Core collection based on 10 clustered population structure as in Fig. [1](#Fig1){ref-type="fig"}. Evaluated parameters; MD %: the mean difference percentage, CR %: the coincidence rate, VD %: the variance difference percentage, VR %: variable rate, I: Shannon's information index of diversity, J\': Genetics evenness, I max: logarithmic number of classes in entire collection. (XLSX 51 kb) Additional file 2:Distribution of 4,652 *Capsicum* germplasm accessions based on H~O~ (observed heterozygosity). Accessions with an H~O~ value of more than 0.3 were considered as F1 hybrids. A total of 673 accessions were excluded from the fundamental germplasm collection to construct a core collection. (TIF 360 kb) Additional file 3:Distribution of 3,821 germplasm accessions in population structure clusters according to their origin and geographic location. The colors of pie graph correspond to the clusters from STRUCTURE analysis as in Fig. [1](#Fig1){ref-type="fig"}. The area of each pie graph indicates the proportion of included accessions. (TIF 1168 kb) CR : Coincidence rate of range CTAB : Cetyl trimethylammonium bromide F~ST~ : Genetic differentiation G + Pcc : Core entries based on the combination of genotype and phenotype data Gcc : Core entries based on SNP genotype data only Gg + Pcc : Core entries based on the combination of genotype and phenotype data Ggcc : Core entries in each cluster were selected based on genotype data only H~E~ : Expected heterozygosity H~O~ : Observed heterozygosity I : Shannon's information index of diversity J' : Genetic evenness M strategy : Maximization strategy MD : Mean difference percentage Pcc : Core entries based on phenotype data only PIC : Polymorphic information content STA : Specific target amplification VD : Variance difference percentage VR : Variable rate of coefficient of variation We are grateful to RDA-Genebank for kindly providing the plant materials with phenotype data and managing plants, and to Hyeon-Seok Jeong and Ho-Hyun Kim for helping with DNA extraction and SNP genotyping, and to Muhammad Irfan Siddique, Young-Shim Park, and Mira Lee for critical review of the manuscript. Funding {#FPar1} ======= This work was carried out with the support of "Cooperative Research Program for Agriculture Science & Technology Development (PJ01120401)" Rural Development Administration, Republic of Korea and grant (710001--07) from the Vegetable Breeding Research Center through Agriculture, Food and Rural Affairs Research Center Support Program, Ministry of Agriculture, Food and Rural Affairs. Availability of data and materials {#FPar2} ================================== The datasets supporting the conclusions of this article are included within the article and its additional files. Authors' contributions {#FPar3} ====================== HYL participated in the design of the study, performed the DNA extractions, SNP genotyping, carried out genetic diversity analysis, population structure and phylogenetic analysis, constructed core collections, and drafted the manuscript. NYR participated in phenotyping and managed the phenotype dataset. HJJ participated in the conception of the study, SNP selection, data analysis, and helped in the discussion of results. JKK managed the project and revision of the manuscript. JKJ, YH, and AJ participated in extraction of DNA, genotyping for core collection evaluation, and made part of the figure. JWH participated in phenotyping. JV participated in revision of the manuscript. BCK participated in the conception of the study, discussion and revision of the manuscript. All authors have read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== Not applicable.

Introduction ============ We have shown previously that Decorin, by antagonizing TGF-β--mediated subarachnoid fibrosis, prevents ventriculomegaly in experimental juvenile hydrocephalus. To focus on white matter alterations, we sought to correlate cytopathological changes induced by hydrocephalus with diffusion tensor imaging (DTI) parameters and determine if Decorin could prevent these changes. Methods ======= Communicating hydrocephalus was induced in 3-week-old rats with basal cistern injections of kaolin; age-matched controls were intact (n=4) and kaolin-no treatment (n=4) animals. Immediately following kaolin injections, animals received a 14-day continuous intraventricular infusion of phosphate-buffered saline (n=6) or human recombinant Decorin (n=5) via osmotic minipumps. At 14 days post-kaolin, all rats underwent MRI/DTI scanning followed immediately by sacrifice and brain fixation. DTI voxel-based analysis was performed on 4 serial rostral-to-caudal slices to quantify mean fractional anisotropy (FA), diffusivity (MD), axial diffusivity (AD) and radial diffusivity (RD) of the corpus callosum (CC) and periventricular white matter (PVWM). Immunohistochemistry and stereology were employed to quantify astrogliosis (GFAP) and aquaporin-4 (AQP4) levels in the CC and PVWM at caudal levels. Results ======= Compared to intact animals (rostral 1.3±0.1 and caudal 0.9±0.1 ventricular volume), the caudal lateral ventricles were significantly larger in kaolin-only (16.2±2.8, p=0.005) and kaolin-PBS (21.0±5.4,p\<0.001) animals than rostral portions (8.0±1.7 and 10.1±3.8, respectively). Following this gradient, untreated hydrocephalic rats exhibited significantly (p\<0.01) decreased FA and increased RD in the caudal-most CC and increased MD and AD in the caudal PVWM compared to intact controls. Decorin significantly (p\<0.05) reversed the RD and MD changes in the caudal CC and PVWM MD (p\<0.05). Such DTI reversals were not discovered in the rostral CC and PVWM. A significant increase in GFAP immunostaining resulted in a positive correlation (p\<0.05) between CC GFAP levels and the caudal-most CC RD. In the caudal PVWM, MD and AQP4 levels and AD and GFAP presence were positively correlated (p\<0.01). Conclusions =========== These results indicate that regional differences exist in ventricular and DTI parameters, and that Decorin has the therapeutic potential to decrease microstructural damage in juvenile hydrocephalus.

Background ========== Cysteine proteinases (CP) represent a large group of proteins in plants, with over 140 annotated gene sequences identified to date in the Arabidopsis genome \[[@B1]-[@B3]\]. As expected for such a large family, the functions of these proteins are diverse, ranging from involvement in programmed cell death (PCD) \[[@B4],[@B5]\] to influencing tissue development \[[@B6],[@B7]\] and pathogen response signalling \[[@B8],[@B9]\]. During seed development, cysteine proteinases have been found to participate in PCD events associated with embryogenesis \[[@B10]\] and seed coat formation \[[@B11]\], as well as playing a role in the processing of proteins, particularly the seed storage proteins found in protein storage vacuoles \[[@B12]\]. Different cysteine proteinases are also thought to make a major contribution to the mobilization of the stored seed protein reserves as germination progresses \[[@B13],[@B14]\]. In germinating mung bean seeds, it has been shown that at least two cysteine proteinases are induced soon after germination has started \[[@B15]\], and these authors proposed that vacuolar receptors (VCRs) transport these newly made proteinases to the protein storage vesicles (PSVs) thereby enabling them to participate in the mobilization of the seed protein reserves. In plants, protein hydrolysis via cysteine proteinases is thought to be modulated, at least in part, by a group of proteins called the cysteine proteinase inhibitors. These polypeptides, also called phytocystatins, are a group of plant polypeptides that inhibit C1A and C13 type plant cysteine proteinases by acting as pseudosubstrates \[[@B16],[@B17]\]. While it is believed that the key biological function of the plant cysteine proteinase inhibitors (CPI) is to modulate the function of target proteinases *in-vivo*, to date, only a limited number of CPI have been tested with plant cysteine proteinases. In one such study \[[@B14]\], the inhibitory effects of a series of recombinant barley CPI were tested against multiple barley cathepsin L-like cysteine proteinases. These authors showed that most of the barley CPIs showed activity against all the CP\'s tested, although a few CPI did show increased inhibition effects towards one or two specific barley cysteine proteinases. CPIs have attracted particular attention due to their capability to inhibit cysteine proteinases found in the digestive tracts of herbivorous insects, an effect that can significantly reduce the destructive effects of these insects \[[@B18],[@B19]\]. For example, Urwin et al. \[[@B20]\] showed that over-expression of sunflower or rice CPI polypeptides in potato increased its resistance to *Globodera*root nematodes, and it has been demonstrated that simultaneously over-expressing a CPI with a second protease inhibitor acting on another protease family (carboxypeptidases) allowed tomato plants to have protection for a longer duration from two different tomato pathogens due to a reduced build-up of insect tolerance \[[@B21]\]. Plant CPIs have been also been demonstrated to increase tolerance to fungal and bacterial pathogens in transgenic plants \[[@B22]\]. Coffee is one of the most important agricultural commodities traded worldwide, however, there continues to be a lack of fundamental knowledge on many aspects of this crop. To date, for example, there is little information on the proteinase and proteinase inhibitor genes of coffee. As shown above, the cysteine proteinases and their inhibitors play important roles in plant seeds. Thus, we decided to begin an investigation of the CP/CPI genes expressed in the semi-recalcitrant coffee grain. In addition, because amino acids and peptides are an important group of coffee flavour/aroma precursors in coffee \[[@B23],[@B24]\], such a study could also yield some clues regarding the potential role of CP/CPI gene products on coffee quality. In this work, we describe cDNA representing several coffee CP and CPI genes, and we present the expression of these genes in developing and germinating grain. To begin studying the functional properties of two highly expressed CP proteins, we have also expressed these proteins in *E. coli*and tested the recombinant polypeptides for protease activity. The results obtained are discussed in relation to the potential roles of the gene products in the development and germination of the coffee grain. Methods ======= Plant material -------------- ### Robusta samples The *Coffea canephora*(BP409) \"maturation\" tissues (roots, branches, leaves and cherries at different stages of development) were harvested in 2007 from field grown trees (Equator), immediately put into liquid nitrogen, then held at -20°C before being sent frozen to Tours, France. Once at Tours, these samples were kept at -80°C until use. Coffee cherries of *Coffea canephora*(BP409) used to obtain the \"germination\" tissues were harvested at mature stage from field grown trees in Equator in 2008, and sent to Tours at room temperature. On arrival, they were manually depulped, washed, and the light grain removed by floating. The remaining grain were dried and the tegument were manually removed. Subsequently, the grain were sterilized by a 1 h treatment in calcium hypochlorite (50 g/l), followed by three washes using sterile water. The grain were then incubated *in vitro*on Heller medium without added sugar or hormone (Agar 7 g/l), at room temperature (25°C). Then five grain were harvested at various times (DAI = Days After Imbibition), and frozen in liquid nitrogen. In the experiment presented, 14DAI (T4) corresponds to the first sign of radical emergence. *Coffea arabica*(T2308) leaves were harvested at different stages of development, in 2006, from trees grown under greenhouse conditions at Tours, France and kept at -80°C before use. Two independent sets of leaves were harvested. The development stages of the leaves are defined as follows: Very Young Leaves (VYL), Young Leaves (YL), Mature leaves (ML), Old Leaves (OL). The sizes of the leaves collected were: approximately 2-3 cm for VYL stage; 6-9 cm for YL stage and 12-15 cm for ML and OL stages. RNA preparation --------------- The samples from the various tissues were reduced to a powder in a SPEX CertiPrep 6800 Freezer Mill with liquid nitrogen and the powders were then stored at -80°C until total RNA was extracted. In the case of the coffee cherries at different stages of development, these were first separated into pericarp and grain tissues and then each was very rapidly reduced to a powder and stored as described above. RNA were extracted and purified from the stored powders using the RNeasy Plant mini kit (QIAGEN) that included a DNase treatment using the manufacturer\'s instructions. The quality of the final RNA samples obtained were checked by agarose gel electrophoresis and ethidium bromide staining. cDNA synthesis -------------- The method used to make the cDNA was very similar to the protocol described in the Transcriptor Reverse Transcriptase kit (Roche) using around 1 μg total RNA sample and 870 ng of oligo dT(18) (Proligo), with reactions performed at 55°C for 30 min. The cDNA samples generated were then diluted one hundred fold in sterilized water and aliquots were stored at -20°C for later use in QPCR experiments. DNA sequence analysis --------------------- Plasmid DNA was purified using Qiagen kits according to the instructions given by the manufacturer. Prepared plasmid DNA was then sequenced by the dideoxy termination method \[[@B25]\]. Computer analyses were performed using the Laser Gene software package (DNASTAR, version 7.1.0). Real time QRT-PCR experiments ----------------------------- cDNA prepared as described above was employed for the quantitative RT-PCR experiments using TaqMan probes, as described by Simkin et al. \[[@B26]\], with an Applied 7500 instrument; except the cDNA dilutions and the Taqman primers/probes were different. The QRT-PCR primers and TaqMan probes were designed using Primer Express^®^software v2.0 from Applied Biosystems and are listed in Table [1](#T1){ref-type="table"}, below. ###### Primer and TAQMAN probes used for the quantitative PCR experiments. Primers and Probes Sequences -------------------- -------------------------------------- rpl39-F1 ^5\'^GAACAGGCCCATCCCTTATTG^3\'^ rpl39-R1 ^5\'^CGGCGCTTGGCATTGTA^3\'^ **rpl39-MGB1** **^5\'^ATGCGCACTGACAACA^3\'^** CP1-F1 ^5\'^ACACAGACCTCTTGATACCAAAACAT^3\'^ CP1-R1 ^5\'^TCTTCCAAGAGCAAACCACCTT^3\'^ **CP1-MGB1** **^5\'^TCTGCTCTTCAGAGGTTGTA^3\'^** CP4-F1 ^5\'^CAGGATGCAACGTGGTGTTG^3\'^ CP4-R1 ^5\'^CCTCCATTGCTATCCCACAAA^3\'^ **CP4-MGB1** **^5\'^TGCTGCTGAAGGCG^3\'^** CPI1-F1 ^5\'^TGTTTGGGAGATCTAATCTGATGATT^3\'^ CPI1-R1 ^5\'^AAACCGAACACTTAACAGCAAAGA^3\'^ **CPI1-MGB1** **^5\'^TTAGTACCTTTCAGTGCAAAT^3\'^** CPI2-F1 ^5\'^CGCTATTGCCTATTTGCCTAGTAGA^3\'^ CPI2-R1 ^5\'^GAAACTCCAATCTTTCCAACTGAAA^3\'^ **CPI2-MGB1** **^5\'^TAAAGCTAACGCGTAAATG^3\'^** CPI3-F1 ^5\'^AACCGACGCTGCAAGAATG^3\'^ CPI3-R1 ^5\'^CAGGGTGGTGAGTAGGAGGAGAT^3\'^ **CPI3-MGB1** **^5\'^CTTCTGCCTTTCCC^3\'^** CPI4-F1 ^5\'^TGTTTATGGTGTGGCTTTCAGTTT^3\'^ CPI4-R1 ^5\'^CGTAGGGAGACGTATGCATGAC^3\'^ **CPI4-MGB1** **^5\'^TGCATGGATGATGTACTG^3\'^** All MGB Probes were labeled at the 5\' end with the fluorescent reporter dye 6-carboxyfluorescein (FAM) and at the 3\' end with the quencher dye 6-carboxy-tetramethyl-rhodamine (TAMRA), except rpl39 probe which was labeled at the 5\' end with the fluorescent reporter dye VIC and at the 3\' end with quencher TAMRA Quantification was carried out by the method of relative quantification, using the constitutively expressed ribosomal protein rpl39 as the reference. In order to use the method of relative quantification, it was necessary to show that the amplification efficiency for the different gene sequences were roughly equivalent to the amplification efficiency of the reference sequence (rpl39 cDNA sequence) using each specifically defined primer and probe sets. To determine this relative equivalence, plasmid DNA containing the appropriate cDNA sequences were diluted 1/1000, 1/10,000, 1/100,000, and 1/1,000,000 fold, and using the QPCR conditions described above, the efficiencies of amplification were calculated. All the primer/probe sets showed acceptable efficiencies. Production of recombinant *Coffea canephora*CcCP1 and CcCP4 in *E. coli* ------------------------------------------------------------------------ Expression vectors were generated using the \"Champion™ pET SUMO Protein Expression System\" (Invitrogen). The CcCP1 sequence minus its N-terminal 28 amino acids was amplified by PCR as follows: 50 μl reactions contained the plasmid pA4-43, 5 μL of TaKaRa^®^DNA Polymerase 10X LA PCR^®^Buffer, 600 μM of each CcCP1 specific primers (CP1-FP ^5\'^ATGTTCCAACATGAAATTCAGTATC^3\'^and CP1-RP ^5\'^TCAAGAGGTCTGTGTCACCA^3\'^), 200 μM each dNTP, and 0.5 U of TaKaRa DNA Polymerase (Takara Bio Inc). The PCR cycling conditions were as follows: 94°C for 2 min; then 35 cycles of 94°C 1 min, 55°C 1.5 min, and 72°C 1.5 min followed by a final step at 72°C 7 min. The PCR product was then gel purified. The CcCP4 sequence minus its N-terminal 22 amino acids was produced as described for the CcCP1 insert except the initial DNA substrate was plasmid pcccs46w7n5, and the specific primers were (CP4-FP ^5\'^ATGGAGATCACAGAAAGAGATT^3\'^and CP4-RP ^5\'^CTAGAGGTCGTCCTTAGGT^3\'^). The gel purified fragments were then cloned into the TA cloning site of the pET-SUMO vector, as recommended by the vector manufacturer. Ligated plasmids were transformed into One Shot^®^Mach1™-T1^R^Chemically Competent Cells (Invitrogen). Clones with the inserts in the correct orientation were selected by PCR screening and the plasmid containing CcCP1 was named pNM17 and the plasmid containing CcCP4 was named pNM6. For protein expression, BL21(DE3) One shot^®^Chemically Competent Cells (Invitrogen) were transformed with pNM17 and pNM6 plasmids as recommended in the manufacturer\'s protocol. Five ml overnight cultures of the selected transformants were used to inoculate 100 ml cultures of LB medium containing 50 μg/ml kanamycin (except control, i.e.: untransformed BL21(DE3) cells). The cells were grown at 37° and 200 rpm shaking to an OD~600~of 0.4-0.6. Then, 90 ml was taken and \"induced\" by addition of IPTG (1 mM final). Both \"Induced\" and \"Not Induced\" cultures were further incubated at 37°C (200 rpm shaking) for 5.5 h, followed by centrifugation at 6000 g for 10 min at 4°C. Cell pellets were resuspended at room temperature in BugBuster^®^Protein Extraction Reagent ((Novagen)) using 5 ml reagent per gram of wet cell paste. Then, 25 U benzonase nuclease (Novagen) and 1 KU rLysozyme solution (Novagen) were added per 1 mL Bugbuster and incubated 25 min at 70 rpm, at room temperature, followed by centrifugation at 6000 g for 30 min at 4°C. The pellets obtained from the induced cultures, which contained the inclusion bodies, were again resuspended in BugBuster^®^solution using the same volume that was used to resuspend the initial cell pellet (5 mL per gram of initial wet cell pellet) and 1 KU rLysozyme solution was added per 1 mL BugBuster^®^and the mixture was incubated at room temperature for 5 min. Then, 6 volumes of 1/10 diluted BugBuster^®^solution was added and the tubes vortexed for 1 min. The resulting suspensions were centrifuged at 5000 g for 15 min at 4°C. The \"washed\" inclusion bodies collected were resuspended in 7 volumes of 1/10 diluted BugBuster^®^solution and centrifuged as previously. This wash step was repeated three more times to remove non-specific material associated with the inclusion bodies. The final pellets of the purified inclusion bodies (IBS) obtained were resuspended in 2 volumes of denaturing buffer A (8 M urea, 50 mM NaH~2~PO~4~, 10 mM Tris-HCl adjusted to pH8 with NaOH) and incubated at 28°C for 1 h as described by Zhang et al. \[[@B4]\], 2 volumes of buffer B (8 M urea, 50 mM NaH~2~PO~4~, 10 mM Tris-HCl adjusted to pH6.3 with HCl) were then added and purification was carried out with Ni-NTA Superflow Columns (QIAGEN). Briefly, the Ni-NTA slurry was mixed with the denatured proteins by shaking on a rotary shaker for 1 h at 70 rpm at room temperature, followed by a loading of the slurry on an empty column and collecting the flow-through. Two successive column washes were carried out with 1.2 volumes Buffer B, followed by elution of the recombinant proteins with 0.6 volumes of buffer C (8 M urea, 50 mM NaH~2~PO~4~, 10 mM Tris-HCl adjusted to pH5.9 with HCl), giving a fraction called El1. This was followed by 2 elutions of 0.6 volumes with buffer D (8 M urea, 50 mM NaH~2~PO~4~, 10 mM Tris-HCl adjusted to pH4.5 with HCl) giving fractions El2 and El3. 20 μl samples of the different fractions were analysed by SDS-PAGE gel electrophoresis and those containing recombinant proteins were pooled. Dialysis/Refolding method ------------------------- Pooled fractions containing purified recombinant proteins were dialyzed in a similar fashion to that described by Zhang et al. \[[@B4]\]; briefly, the purified protein fractions were introduced into Slide-A-lyser^®^dialysis cassettes (10 kD MWCO Pierce). Then, 4 successive dialysis steps were carried out for 3 h at 4°C with stirring using buffers containing decreasing levels of urea (6 M, then 4 M, 2 M and 0 M urea, see Table [2](#T2){ref-type="table"} below). The dialysis buffer volumes were approximately 50 times higher than the volumes of extracts. ###### Names and composition of dialysis buffers Names of dialysis buffers Composition --------------------------- ----------------------------------------------------------------------------------------------------------------- Buffer 6 M 50 mM potassium phosphate pH10.7, 5 mM EDTA, 1 mM reduced glutathione, 0.1 mM oxidized glutathione, **6 M**urea Buffer 4 M 50 mM potassium phosphate pH10.7, 5 mM EDTA, 1 mM reduced glutathione, 0.1 mM oxidized glutathione, **4 M**urea Buffer 2 M 50 mM potassium phosphate pH10.7, 5 mM EDTA, 1 mM reduced glutathione, 0.1 mM oxidized glutathione, **2 M**urea Buffer 0 M 50 mM potassium phosphate pH10.7, 5 mM EDTA, 1 mM reduced glutathione, 0.1 mM oxidized glutathione, **0 M**urea Assay for cysteine protease activity ------------------------------------ The assay for cysteine protease activity used here is a slight modification of the one developed by Zhang et al. \[[@B4]\] and Troen et al. \[[@B27]\]. Protein samples made up to 10 μl with water were mixed with 20 μL 50 mM sodium formate buffer (pH3), then incubated 30 sec at 37°C for activation, with parallel \"non-activated\" control reactions set up with samples in which 20 μL milliQ pure water replaced the 20 μL sodium formate buffer. This was immediately followed by the addition of 6.7 μl of the \"reaction mix\" (1% BSA, 1xPBS and 6 mM L-cysteine, pH7.5). The enzyme reactions were subsequently incubated at 37°C and 3 μL aliquots were taken at different times and added to sample loading buffer, heated for 7 min at 95°C, then run on SDS-PAGE gels and stained with coomassie. Results ======= Identification of cysteine proteinase sequences expressed during coffee grain maturation ---------------------------------------------------------------------------------------- Coffee cDNA encoding cysteine proteinases were found by carrying out BLAST searches against the Nestlé/Boyce Thompson Institute coffee EST database (*Coffea canephora*built \#3, located at <http://solgenomics.net/>) \[[@B28]\] using the protein sequences of two biochemically characterized cysteine proteinases: NtCP56-KDEL from *Nicotiana tabacum*(genbank accession number [ACB70409](ACB70409)\[[@B4]\]) which is a peptidase from the C1A subfamily (MEROPS database nomenclature; <http://merops.sanger.ac.uk>\[[@B2]\]), and *SlCP*from *Solanum lycopersicum*(genbank accession number [CAH56498](CAH56498)\[[@B5]\]) which is from the peptidase C13 family (asparaginyl endopeptidase, cysteine catalytic type). This analysis yielded 15 candidate unigene sequences (data not shown). As our main objective was to study genes highly and specifically expressed in the maturing grain, we examined the \"*in-silico\"*expression profiles associated with these unigenes. Three unigenes (SGN-U613831, SGN-U613447 and SGN-U620235) were found that exhibited multiple ESTs and all were found in either grain or cherry EST libraries (data not shown). Further analysis indicated that two of the unigenes (SGN-U613447 and SGN-U620235) were probably different alleles of the same gene, thus giving only two, clearly different unigenes with high expression in the grain for further study. Plasmids potentially containing the longest sequences for each unigene were then selected from our available EST libraries and fully sequenced to confirm the \"*in-silico*\" unigene sequences. In the case of Unigene SGN-U613831, the plasmid pA4-43 was selected to characterize the first *Coffea canephora*CP cDNA, which we named CcCP1; the cDNA has a 1511 bp long insert, with a 1194 bp coding sequence (CDS) encoding a protein of 397 amino acids. Analysis of the protein sequence of CcCP1, performed using SignalP 3.0 server (<http://www.cbs.dtu.dk/services/SignalP/>) and the \"Conserved Domain Database\" (<http://www.ncbi.nlm.nih.gov/Structure/cdd/cdd.shtml>), shows it is a member of the peptidase_C1 superfamily/peptidase_C1A subfamily (papain family, clan CA) and it appears to have at least three distinct domains, a hydrophobic N-terminal signal peptide, followed by a predicted 56 amino acid long I29 inhibitor propeptide domain (with an ERFN(V/I)N like domain sequence) and a peptidase domain containing the catalytic triad of Cys, His, and Asn, as well as an active site Gln residue. The second coffee CP cDNA to be studied in detail has a 1365 bp long insert encoding a protein of 359 amino acids which we have called CcCP4 (pcccs46w7n5). The protein sequence indicates that CcCP4 is also a member of the C1 peptidase superfamily (C1A subfamily) and has the three distinct domains already described for CP1, plus an obvious ERFNIN like sequence which is located within the 55 AA long predicted I29 inhibitor domain. Figure [1A](#F1){ref-type="fig"} shows an alignment of the CcCP1 protein sequence with two very closely related cysteine proteinase sequences: VsCPR4 (CPR4 from *Vicia sativa*(CAB16316)) that is thought to be involved in storage protein mobilization \[[@B29]\] and AtCP (a putative CP protein from *Arabidopsis thaliana*(AAL49820)). Figure [1B](#F1){ref-type="fig"} shows the CcCP4 protein aligned with two highly related proteins; NtCP56-KDEL (*Nicotiana tabacum;*ACB70409) and SlCysEP-KDEL (*Solanum lycopersicum*; ABV22590). Zhang et al. \[[@B4]\] cloned a cDNA encoding NtCP56-KDEL from tobacco anthers and showed the recombinant protein produced in *E.coli*had cysteine proteinase activity after Ni-NTA purification under denaturing conditions, followed by refolding and acid activation. These authors proposed that NtCP56-KDEL was involved in pollen grain development because engineering transgenic plants to reduce expression of this gene led to plant sterility. The related SlCysEP was studied by Senatore et al. \[[@B5]\] and was shown to be expressed in anthers and was proposed to be associated with PCD (programmed cell death) during anther development in tomato. Figures [1A](#F1){ref-type="fig"} and [1B](#F1){ref-type="fig"} also indicate that both CcCP1 and CcCP4 possess key sequence elements associated with the C1A peptidase family; for example, the four catalytic residues Gln, Cys, His and Asn cited above plus the cathepsin propeptide inhibitor domain I29. ![**Alignment of *C. canephora*CP1 and CP4 sequences with highly homologous plant sequences**. The alignments were done using CLUSTAL W. The key conserved amino acid characteristics and motifs are noted. Amino acids shaded in grey indicate most conserved sequences. **Panel A:**Database accession numbers are *Coffea canephora*CP1 (GeneBank sequence \#AEQ54770), *Arabidopsis thaliana*AtCP (AAL49820) and V*icia sativa*VsCPR4 (CAB16316). The catalytic triad Cys, His, and Asn and also the Gln active site residues are indicated by an asterisk. The GCXGG motif is double underlined. The cathepsin propeptide inhibitor domain (I29) is shown in a rectangular box. Note: an ERFNIN-like sequence, shown in italics, exists in the propeptide region of CP1 (ie. ERFNAQ). **Panel B:**Database accession numbers: *Coffea canephora*CP4-KDDL (GeneBank sequence \#AEQ54771), *Nicotiana tabacum*NtCP56-KDEL (ACB70409) and *Solanum lycopersicum*SlCysEP-KDEL (ABV22590). Symbols for key amino acids and motifs are as above, plus the ERFNIN motif is indicated in italics and the KDEL (K358-L361 for NtCP56) motif is shown in a double lined rectangular box. Arrows indicates the site of auto-hydrolysis of NtCP56 reported by Zhang et al. (2009) \[[@B4]\].](1471-2229-12-31-1){#F1} Quantitative gene expression analysis for CP1 and CP4 in different coffee tissues and during grain development/germination -------------------------------------------------------------------------------------------------------------------------- The tissue and development specific expression of the coffee CP1 and CP4 genes was measured using TaqMan QRT-PCR. The expression results obtained for Robusta variety BP409 (Figure [2](#F2){ref-type="fig"}) indicate that both genes are expressed at increasingly higher levels as grain development progresses from large green grain stage to the mature red stage, with CP1 \> CP4 (RQ = 6.73 versus RQ = 0.96 at the red grain stage). Few CP1 and CP4 transcripts were detected in Robusta pericarp, leaves, or branch samples, although low levels were clearly detected in the roots (CP4 \> CP1; RQ = 0.39 versus RQ = 0.24). We also examined the expression of CP1 and CP4 in the germinating and post-germination grain. The material used for this experiment is illustrated in Additional file [1](#S1){ref-type="supplementary-material"}. CP1 and CP4 transcripts are found in germinating/post germination grain (Figure [2](#F2){ref-type="fig"}), but, in contrast to the developing grain, these genes clearly have different temporal expression patterns during these periods. First, the level of both transcripts rise from T1 (starting material; CP1 RQ = 1.1 and CP4 RQ = 0.45) to T3 (5 DAI; CP1, RQ = 4.4 and CP4, RQ = 6.4). From T3 to T5 (radial protrusion, radicle growth, and early cotyledon development), the levels of CP1 transcripts in the grain fell dramatically while the level of CP4 transcripts remained high. At T6, which is primarily the first pair of leaves (cotyledons plus remnants of the grain), little or no CP1 and CP4 transcripts were detected. Overall, these expression levels are in agreement with the previous \"in-silico\" expression data (data not shown). {#F2} CP1 and CP4 gene expression was also measured in an independent set of Robusta samples (data not shown). The results from this second sample set were globally in agreement with the results presented in Figure [2](#F2){ref-type="fig"}, with a few minor differences. For example, for the second independent set of BP409 samples covering grain development, overall CP1 expression was lower (RQ = 1.54 versus RQ = 6.73 at mature red stage) and CP4 expression was found to be higher at an earlier stage in this second grain sample set. It is likely that some of these transcript level differences result from slight differences in the precise development stages of the various samples. Comparison with a second germination sample set (from Robusta variety FRT05) showed that CP1 and CP4 expression were broadly similar, with the exception that the BP409 sample set showed increased CP1 transcript levels from T1 to T 3, but the FRT05 samples set shows the levels of CP1 transcripts are highest at T1 and then fall. Nonetheless, for both sample sets, CP1 transcripts levels are very low at T5. In both germination sample sets, CP4 transcript levels are low at T1 and rose to a maximum at T5. CP1 and CP4 transcripts were barely detected at T6 for the second sample set. Interestingly, the CP4 transcript levels were significantly higher in all the Robusta FRT05 samples examined during pre-germination/germination/post germination (including T1 to T5 samples) versus the equivalent samples of Robusta BP409. Production of recombinant CP1 and CP4 enzymes in E. coli: purification and activity testing ------------------------------------------------------------------------------------------- In order to study the functional properties of the CcCP1 and CcCP4 proteins, we expressed these proteins, minus their respective signal peptide sequences, in *E.coli*as N-terminal HIS-SUMO fusions (see Methods for details). The results shown in Figure [3](#F3){ref-type="fig"} indicate that both fusion proteins were expressed at high levels upon induction with IPTG and that their molecular weights are close to those predicted. However, subsequent analysis indicated that the vast majority of both the recombinant proteins produced were insoluble (data not shown). Therefore, we used a method similar to that described previously by Zhang et al. \[[@B4]\] to denature the fusion proteins and purify them by Ni-NTA affinity chromatography. The purified proteins were then re-natured as described in the methods. An SDS-PAGE analysis of samples from the different purification steps for HIS-SUMO-CP4 indicated that a significant level of purification was achieved (see Additional file [2](#S2){ref-type="supplementary-material"}). Similar results were obtained for the HIS-SUMO-CP1 recombinant protein. {#F3} The purified, soluble fusion proteins were then tested for proteinase activity with a general proteinase assay based on BSA hydrolysis. In this assay, cleavage of the BSA, detected by loss of the BSA band using SDS-PAGE, indicates the presence of a proteolytic activity. In the first assays, using the full length purified HIS-SUMO-CP1 and HIS-SUMO-CP4 after refolding, no activity was detected even after long incubation times. Because both coffee proteins are in the same super family as papain, we then decided to examine the possibility that these proteinases could be activated by a short acid treatment. Figure [4](#F4){ref-type="fig"} shows that the HIS-SUMO-CP4 proteinase can be activated by a short acid treatment yielding a significant level of protease activity, with the first signs of BSA degradation detected after a 1 min reaction time (clear detection of degradation products at ≈ 45 KDa and 34 KDa), and with a very pronounced degradation at T = 15 min with nearly complete disappearance of the BSA band. No BSA degradation was observed if the HIS-SUMO-CP4 polypeptide was not subjected to a low pH treatment. It was assumed that the acid treatment activates CP4 via an intrinsic autohydrolysis capability in the CP4 pro-peptide, leading to the release of an N-terminal CP inhibitor peptide as has been seen for the tobacco NtCP56 \[[@B4]\]. To verify that HIS-SUMO-CP4 was processed to generate a shortened, active polypeptide, the activation process was followed over time by SDS-PAGE analysis. The results obtained showed that recombinant HIS-SUMO-CP4 was processed to generate an approximately 32.6 kDa polypeptide that is presumed to be the active CP4 proteinase (Additional file [3](#S3){ref-type="supplementary-material"}). We confirmed that this \"activated\" CcCP4 fell into the cysteine proteinase class of proteases by showing that this activity was inhibited by the cysteine proteinase specific inhibitor E-64C (Additional file [4](#S4){ref-type="supplementary-material"}). A similar acid activation and protease test was carried out with purified and renatured CcCP1, but was not successful (data not shown). This result suggests that HIS-SUMO-CP1 may not have refolded correctly during the renaturation step. Alternatively, it is possible that the polypeptide refolded correctly, but this cysteine proteinase has another mode of activation, such as perhaps requiring the intervention of a specific proteinase to release its putative N-terminal inhibitory domain. {#F4} Identification of cysteine proteinase inhibitors expressed during grain maturation ---------------------------------------------------------------------------------- To find cDNA encoding coffee cysteine proteinase inhibitors, we carried out BLAST searches using the biochemically characterized *Helianthus annuus*cysteine protease inhibitor HaCPI (accession number JE0308 \[[@B30]\]), the *Dianthus caryophyllus*cysteine protease inhibitor DcCPIn (accession number AAK30004 \[[@B31]\]) and a putative cystatin-like inhibitor sequence from *Citrus × paradisi*(accession number AAG38521), as the query sequences. Using these criteria, 6 distinct unigene sequences were found (data not shown). 4 of these sequences were chosen for further study in this work. Plasmids potentially containing a complete ORF representing each unigene were isolated from the available EST libraries and fully sequenced to confirm the \"in-silico\" unigenes sequences. The respective gene sequences have been named CcCPI-1, CcCPI-2, CcCPI-3, and CcCPI-4. The plasmid names and the size of their inserts are presented in the Additional file [5](#S5){ref-type="supplementary-material"}. Quantitative gene expression analysis for CPI-1, CPI-2, CPI-3 and CPI-4 in different Robusta tissues and during grain development/germination --------------------------------------------------------------------------------------------------------------------------------------------- To obtain precise information on the expression profiles of the 4 cysteine proteinase inhibitor genes in Robusta, QRT-PCR was used to measure their expression levels in a number of tissues and during grain and pericarp maturation, and during germination/post germination. The data obtained is presented in Figure [5](#F5){ref-type="fig"}. CPI-1 is expressed at increasingly high levels during Robusta grain development (low transcript levels in small green grain with RQ~GSG~= 0.06, but increasing transcript levels from stages large green to red with RQ~GLG~= 0.87, RQ~GY~= 3.16 and RQ~GR~= 6.59 respectively). The results obtained in an independent experiment were similar except that this second set of BP409 samples showed more constant levels of CPI-1 transcripts between the large green and red stages (after a large increase between the small green and large green stages; data not shown). In both sample sets, few CPI-1 transcripts were detected in the Robusta pericarp, leaves, and branches, although a low level was seen for one of the red pericarp samples (which appeared to be \"softening\") plus a low level was detected in both root sample sets. (Figure [5](#F5){ref-type="fig"}\--red pericarp RQ~PR~= 0.12 and roots RQ~roots~= 0.13). Figure [5](#F5){ref-type="fig"} also shows that CPI-1 is expressed during Robusta grain germination, starting low at T = 1 (RQ~T1~= 0.25), then showing roughly similar transcript levels from T = 2 to T = 5 (RQ~T2~= 1.12 to RQ~T5~= 0.79), followed by a drop in T6 (RQ~T6~= 0.13): in first pair of young leaves/cotyledons. Quite similar expression levels were seen in the second set of samples for Robusta grain germination (FRT05; data not shown). {#F5} Both the CPI-2 and CPI-3 genes were found to be expressed to a small extent in all the tissues examined (Figure [5](#F5){ref-type="fig"}). A second independent RNA set showed a similar expression profile (data not shown). Thus, these two CPI genes do not appear to have any clear tissue specificity. Detailed examination of the QRT-PCR results in Figure [5](#F5){ref-type="fig"} show however that some expression differences exist for these two genes, for example there are higher CPI-2 transcript levels found at the SG grain stage (high RQ~GSG~= 1.71 for CPI-2 in Figure [5](#F5){ref-type="fig"}) and are slightly higher in leaves (both Robusta sample sets). One can also observe that CPI-3 transcript levels are very low at the T5 germination stage (this was also seen in the second set of Robusta samples). Another difference is the relatively high CPI-3 expression associated with the mature red coffee pericarp tissue (RQ~PR~= 2.29). Interestingly, the higher expression of CPI-3 is most noticeable in the sample used for Figure [5](#F5){ref-type="fig"}. As this sample could be more mature (and thus softer) than the other \"mature\" pericarp sample tested, it is entirely possible that increased CPI-3 expression is coupled with pericarp age and fruit softening. It will be interesting in the future to explore this possibility further by carrying out more detailed expression studies on this gene at the end of pericarp maturation, and to explore whether CPI-1 could also be involved (which also appears to rise somewhat at this stage). The QRT-PCR results in Figure [5](#F5){ref-type="fig"} show that there is little or no expression of CPI-4 in the developing grain or in the pericarp at any maturation stage examined (0 \< RQs \< 0.02). This was also observed with the second Robusta sample set studied (data not shown). However, Figure [5](#F5){ref-type="fig"} shows that significant levels of CPI-4 transcripts occur in Robusta roots (RQ~roots~= 0.53), leaves (R~leaves~= 0.20) and branches (RQ~branches~= 0.08). In contrast however, few CPI-4 transcripts were detected in the roots and branches of the second Robusta sample (data not shown), suggesting some possible maturity or other tissue sampling related differences could be involved. In the case of leaves, relatively low CPI-4 expression was seen in the leaves of the BP409 Robusta sample set used for Figure [5](#F5){ref-type="fig"}, but very high levels were seen for this gene in the leaves of the BP409 Robusta sample used for the second sample set (data not shown). This expression difference, which was hypothesized to be due to leaf maturity, is explored in more detail below. Another surprising aspect of the CPI-4 expression data obtained is the extremely high level of CPI-4 expression detected during the last post-germination stages examined (Figure [5](#F5){ref-type="fig"}: T5 stage, RQ = 4.59 and T6 stage consisting primarily of the first two young leaves/cotyledons with RQ = 31.97). There was no significant expression of CPI-4 from germination stages T1 to T4 (14DAI, grain showing its first sign of radicle emergence). The same results were seen in the second Robusta sample set analysed (FRT05; data not shown). This suggests CPI-4 could play some important role during this period of plantlet growth/emergence. Quantitative gene expression analysis of the cysteine protease inhibitor genes CPI-1\--CPI-4 at different stages of leaf development ------------------------------------------------------------------------------------------------------------------------------------ The very different transcript levels seen for CPI-4 in two independent leaf samples from Robusta BP409 noted earlier suggested that the expression of this gene could be influenced by differences in the developmental stages of the two leaf samples. The idea that CPI-4 expression could be significantly induced in some tissues was further strengthened by the fact that very high CPI-4 expression was associated with the first cotyledons of the germinated grain. Thus, we decided to explore the effect of leaf maturity on the expression of this gene, and the other coffee CPI genes, in leaves of *Coffea arabica*T2308. Samples corresponding to four different developmental stages were used; very young leaves (VYL), young leaves (YL), mature leaves (ML) and old leaves (OL). The expression results presented in Figure [6](#F6){ref-type="fig"} show that CPI-4 was very strongly induced as leaves enter their mature phase (RQ~ML~= 23.79 in set 1 and 26.31 in set 2). CPI-4 transcript levels then fell as the leaves aged further. As previously observed in *Coffea canephora*BP409 (Figure [5](#F5){ref-type="fig"}), very low transcript levels are seen for the CPI-1 gene in mature leaves, and this is also seen for all the other stages examined in this new experiment (Figure [6](#F6){ref-type="fig"}). While few transcripts were detected for CPI-2 and CPI-3 in the two early leaf stages studied here, a slightly variable, but still quite low expression was seen for CPI-2 and CPI-3 genes in the remaining two leaf stages. This latter result may indicate some slight inducible capability for the CPI-2 and CPI-3 genes, and thus it will be interesting in the future to re-examine the expression of these two CPI genes in leaf tissues subjected to different stresses (environmental and insect/fungal/bacterial attack). {#F6} Discussion ========== Cysteine proteases and their inhibitors have been studied in detail for many plants, but, to date, little work has been done on these genes or proteins from coffee. Here we describe full length cDNA representing two cysteine proteinase genes, called CcCP1 and CcCP4, which show relatively exclusive expression during grain maturation and germination. We also present the characterization of full length cDNA representing four cysteine proteinase inhibitor genes (CPI-1\--CPI-4) and describe the quantitative expression of these genes in coffee. Sequence comparisons indicated that the two Robusta cysteine proteinases described represent different CP proteinases; Blast analysis against the protein database indicates that CP1 is very closely related to a papain type CP called VsCPR4 from *Vicia sativa*(GenBank Accession [CAB16316](CAB16316)), as well as a putative CP of Arabidopsis (AT3G54940, GenBank Accession [NP_567010](NP_567010)). Examination of Figure [1A](#F1){ref-type="fig"} shows the protein sequence of CcCP1 and its putative homologues contain a partial ERFNIN box (ERFNAQ) within an N-terminal cathepsin propeptide inhibitor domain, indicating that these polypeptides fall within the CP subfamily having an I29 domain that is found at the N-terminus of some C1 peptidases, like Cathepsin L, where it acts as a propeptide. (<http://merops.sanger.ac.uk>\[[@B2]\]; <http://www.ncbi.nlm.nih.gov/Structure/cdd/cdd.shtml>\[[@B32]\]). The I29 domain of CcCP1 and its putative homologues are located just upstream of a clear peptidase C1 superfamily domain which have strong homologies with peptidase C1A_cathepsins_B/C/X. Several other CP specific elements are also completely conserved in the three highly similar protein sequences presented in Figure [1A](#F1){ref-type="fig"}. The CPR4 polypeptide of *Vicia sativa*has been shown to be expressed during seed maturation and during the early part of seedling germination/growth in both the embryonic axis and the cotyledons \[[@B29]\]. Although the functional activity of this protein has not been proven in a recombinant form, these authors nevertheless implied the processed, and thus presumably activated polypeptide, was involved in seed storage protein mobilization. Microarray expression analysis of the potential Arabidopsis CP1 homologue (AT3G54940) indicates this gene has significant expression in developing endosperm and in the embryos of developing and germinating seedlings (<https://www.genevestigator.com>\[[@B33]\]), supporting the idea that these highly related polypeptides play a role in storage protein modification and/or mobilization. Little expression was seen for the Arabidopsis gene in other tissues under normal conditions. The expression profile of CcCP1 transcripts (Figure [2](#F2){ref-type="fig"}) mirrors the expression of the candidate homologues VsCPR4 and AT3G54940, ie, CcCP1 is expressed in the later stages of grain development and during germination, and implying that CcCP1 performs a similar function as the putative homologue of the other two plants. To date, it has apparently not been possible to express and/or correctly activate any of the recombinant CcCP1 homologues in order to confirm their function. The most likely explanation for this inability to verify the activity of these proteins is that the precise conditions needed to process/activate these proteins have not yet been identified. The alignment of CcCP4 with two of the closest well characterized plant sequences (NtCP56, and SlCysEP) indicates that CcCP4 and its proposed homologues have a clear ERFNI/VN box within an N-terminal I29 type propeptide domain, followed by a peptidase C1 superfamily domain containing conserved cysteine proteinase specific sequence elements (see Figure [1B](#F1){ref-type="fig"} for details). All three polypeptides contain N-terminal signal sequences, and the two homologues have a C-terminal endoplasmic reticulum retention sequence (KDEL). Interestingly, the CcCP4 cDNA sequence characterized contains the C-terminal sequence KDDL. In order to explore whether the KDDL sequence was unusual in coffee, we examined the sequences of other cDNA in the coffee unigene (SGN-U613447). This unigene, which is the only clear hit obtained when the *Coffea canephora*Unigene set at <http://solgenomics.net> is blasted with the CcCP4 protein sequence, has 22 ESTs. Analysis of these ESTs showed that 14 have sequence data for the C-terminal end, and, interestingly 2 of these end with the KDEL sequence. This observation suggests that Robusta may have both KDDL and KDEL alleles of CcCP4. A preliminary PCR analysis of genomic fragments from this region of the CcCP4 gene in *Coffea eugenoides*and *Coffea arabica*suggests that the KDEL allele is more prominent in these species (data not shown). The significance of CP4 alleles with a C-terminal KDDL sequence is currently unclear. However, the fact that several other plant sequences in the protein database related to CcCP4 also have C-terminal KDEL or RDEL sequences (data not shown) suggests this is the more prominent form found in plants. Also, several groups have proposed that unprocessed CP-KDEL proteins are retained in the endoplasmic reticulum (ER) after synthesis, and are only processed/transported further upon specific signaling \[[@B34]-[@B36]\], and another group suggested that C-terminal KDDL proteins could be poorly retained in the ER \[[@B37]\]. These observations raise the possibility that an expressed CcCP4-KDDL protein might be poorly retained in the ER and thus could exist in unintended compartments of developing coffee grain cells, with unknown consequences. Future experiments comparing physiological or other differences between seeds of Robusta trees homozygous for the CcCP4-KDDL or CcCP4-KDEL genes could be illuminating. Both the proposed tobacco and tomato CcCP4 homologues are known to be involved in pollen development. Zhang et al. \[[@B4]\] confirmed that the tobacco protein (NtCP56) encoded a functional, acid activated CP proteinase and then went on to show that anti-sense suppression of this gene can disrupt normal pollen development and cause male sterility. The tomato SlCysEP gene product was also shown to encode an acid activated CP proteinase and to be an important component of the tomato ricinosome, which is a subcellular structure believed to orchestrate the final processing/recycling of cellular proteins during plant programmed cell death \[[@B5]\]. In each case, the recombinant CP polypeptides produced in *E. coli*were insoluble, and, as shown here for the coffee CcCP4, needed to be refolded to demonstrate auto-cleavage and cysteine protease activity. Analysis of SlCysEP transcripts showed that they could be detected in flowers at a specific period, and that this expression was primarily limited to the stamens \[[@B5]\]. While the expression of NtCP56 or SlCysEP was not studied in seeds, our examination of ESTs encoding SlCysEP (<http://solgenomics.net>) confirmed that cDNA representing this gene can be found in *Solanum lycopersicum*EST banks from fruit, seeds, young leaves, as well as flowers (with seed libraries having the highest number of ESTs). Tomato database analysis indicates the SlCysEP gene has three introns, and that two other highly related KDEL containing \"unigene\" sequences can be found which potentially represent other members of this specific CP gene family. Three potential CcCP4 homologues were also identified in the Arabidopsis genome (AT5G50260, AT3G48350, and AT3G48340). Examination of the expression patterns for these genes using microarray data \[[@B33]\] showed that AT5G50260 expression was limited to seeds, silique and stamen/anther, although lower levels could also be found in roots, but not in stems, from plants subjected to osmotic stress. Interestingly, some induction of AT5G50260 also appeared in nematode infested roots. No significant expression was seen for this gene in other tissues; in contrast, low levels of expression were found for the Arabidopsis CcCP4 like AT3G48350 gene in many tissues, suggesting this gene may play a more general role in plant cells. The only two situations that appeared to increase AT3G48350 transcript levels were treatment with uv and dramatic changes in light conditions (dark/light shifts). No probe sets were identified for the gene sequence AT3G48340, so the expression of this gene is not known. Overall, the CcCP4 expression data presented are consistent with our proposal that CcCP4 may be involved in the PCD associated with coffee grain germination and post-germination stages. Although no significant CcCP4 expression was detected in the Robusta BP409 flower sample tested (data not shown), this may be due to the limited developmental time frame we analysed. New analysis, using several different stages of flower development is clearly needed to clarify the expected participation of CcCP4 in coffee pollen development. It is interesting to note the *Vicia sativa*Proteinase A (CcCP4 homologue) was not detected during vetch seed development, but was detected in the cotyledons in the later stages of germination and post-germination (was also not detected in the seedling axis) \[[@B29],[@B38]\]. These observations, together with the fact that purified *Vicia sativa*Proteinase A was capable of completely digesting the vetch storage proteins vicilin and legumin, led the authors to propose that Proteinase A was not involved in seed development or in the early part of storage protein mobilization, but was important for later stages of germination which involved much more extensive proteolysis \[[@B38]\]. This contrasts with coffee where there appears to be two periods of CcCP4 expression, one in the developing grain (which may continue into the first part of germination), and another new burst of transcription beginning around the T2 stage of germination up to T5 stage. We currently do not know the significance of finding low CcCP4 expression in the developing grain, although we do note that the \"absence\" of Protein A in *Vicia sativa*in developing seeds \[[@B38]\] could be due to the less sensitive detection method used earlier (northern blotting versus QRT-PCR here). The quantitative expression analysis of the four CPI genes (Figure [5](#F5){ref-type="fig"}) showed that CcCPI-2 and CcCPI-3 are expressed in most tissues and that their levels of expression do not vary broadly. In contrast, CPI-1 had increasingly higher expression as grain development progresses (\> 100 fold increase from immature to mature stages) and also showed relatively strong expression during the T2 to T5 stages of grain germination and post germination. Little CcCPI-1 expression was detected in the other tissues tested. For CcCPI-4, extremely high levels of transcripts were seen exclusively in the T5 and T6 stages of post-germination, corresponding to stages in which the cotyledons are forming. The significance of this observation is not known, but one interesting line of future investigation will be to determine whether CPI-4 expression contributes to insect tolerance/resistance at this delicate stage of plantlet development. By examining gene expression at different stages of leaf development, we also found that while CcCPI-4 is weakly expressed in young leaves, its expression increases dramatically in mature leaves (Figure [6](#F6){ref-type="fig"}). No significant expression of CcCPI-4 was found for the other tissues tested, except a low level in the roots from Robusta BP409 cDNA set used in Figure [5](#F5){ref-type="fig"} (RQ = 0.53), which raises the interesting possibility that the higher levels of one or more CcCPI proteins could reduce damage by root pests like nematodes. Overall, the coffee CPI gene expression data suggest that CPI-2 and CPI-3 could be CP inhibitors with mostly \"house-keeping\" functions, while CPI-1 may play an important role during grain development, and CPI-4 could contribute to reducing damage by insects during the early life of the plantlets (first cotyledons), and perhaps in mature/old leaves and roots. Finally, as the peptide/amino acid profile of a coffee has an important impact on flavour and aroma generation during coffee grain roasting \[[@B24],[@B39]\], further research is warranted to investigate possible links that may exist between the allelic variation in genes encoding coffee cysteine proteinases and cysteine proteinase inhibitors and the flavour/aroma quality associated with the grain of different coffee varieties. Conclusions =========== Several cysteine proteinase and cysteine proteinase inhibitor genes with strong, relatively specific expression during coffee grain maturation and germination are presented. The temporal expression of the CcCP1 gene suggests it is involved in modifying proteins during late grain maturation and germination. The expression pattern of CcCP4, and its close identity with KDEL containing CP proteins, implies this proteinase may play a role in protein and/or cell remodelling during late grain germination, and that it is likely to play a strong role in the programmed cell death associated with post-germination of the coffee grain. Expression analysis of the cysteine proteinase inhibitor genes suggests that CcCPI-1 could primarily be involved in modulating the activity of grain CP activity; while CcCPI-4 may play roles modulating grain CP activity and in the protection of the young coffee seedlings from insects and pathogens. CcCPI-2 and CcCPI-3, having lower and more widespread expression, could be more general \"house-keeping\" CPI genes. The data generated opens up new avenues to explore the potential contribution of proteinases to coffee quality and facilitates new research to investigate the possibility that coffee cysteine proteinase inhibitors may help reduce damage caused by some plant pests. Abbreviations ============= CP: Cysteine proteinase; CPI: Cysteine proteinase inhibitor; QRT-PCR: Quantitative reverse transcriptase-PCR; PCD: Programmed cell death; VCR: Vacuolar receptor; PSV: Protein storage vesicles; DAI: Days after imbibition; IBS: Inclusion bodies; ER: Endoplasmic reticulum Authors\' contributions ======================= JMcC, ML and MBA were involved in the design, supervision and interpretation of the experiments, and all of the authors contributed to the analysis of the data. JMcC conceived the work and drafted the manuscript. MBA, GC and VC carried out the initial characterization and expression analysis of the gene sequences. ML and NM produced and characterized the recombinant CP1 and CP4 proteins. The quantitative transcript analysis and additional sequence analysis of these genes was done by ML, NM and VC All the authors have read and approved the manuscript. GenBank accession numbers ========================= *Coffea canephora*protein sequence data associated with this work article has been deposited in GenBank under following accession numbers: CcCPI-1 ([AEQ54766](AEQ54766)), CcCPI-2 ([AEQ54767](AEQ54767)), CcCPI-3 ([AEQ54768](AEQ54768)), CcCPI-4 ([AEQ54769](AEQ54769)), CcCP1 ([AEQ54770](AEQ54770)), CcCP4 (KDDL-tailed) ([AEQ54771](AEQ54771)) and CcCP4 (partial\--KDEL-tailed) ([AEQ54772](AEQ54772)). Supplementary Material ====================== ###### Additional file 1 **Robusta BP409 germination samples used for QRT-PCR**. The times of sampling are given for each stage. The \"T1\" sample is the sterilized and washed material obtained just before placing on the \"germination\" media. The \"T4\" sample (14 Days) showed the \"First Evidence\" of germination, ie. the radical has just started to protrude from the grain. In T1-T4, all the sample shown was used for RNA extraction. For T5 and T6 only the grain and first cotyledons (and remaining grain material) respectively were used to make RNA. ###### Click here for file ###### Additional file 2 **HIS-SUMO-CcCP4 recombinant protein purification steps**. Samples from the different purification steps were run on a 8-6% SDS-polyacrylamide gel and stained with Coomassie. Starting material was an induced BL21 (DE3) + pET-SUMO/CP4 whole cell lysate. Arrow indicates HIS-SUMO-CcCP4. **Panel A: Lane M**, Molecular marker proteins with the sizes shown on the left in kDa. **Lane 1**, Pellet of induced whole cell lysate (material insoluble in the extraction Buffer); **Lane 2**, Washed Inclusion bodies; **Lane 3-Lane 5**, Successive flow-through fractions collected during His-Tag column washes; **Lane 6**, First eluate fraction of the His-Tag column; **Lane 7**, Pool of second and third His-Tag column eluates before dialysis step. **Panel B: Lane M**, Molecular marker proteins. **Lane 1**, Purified recombinant HIS-SUMO-CP4 protease of pooled second and third eluates after dialysis step. ###### Click here for file ###### Additional file 3 **Aurto-catalytic processing/activation of recombinant HIS-SUMO-CP4 proteinase**. 10 μL (3.2 μg) of His-Tag column purified and dialysed recombinant HIS-SUMO-CP4 was added to 20 μL acid buffer (sodium formate 50 mM pH3), then either A) immediately stopped by the addition of 14 μl 5x loading buffer (Lane 1, T = 0), or B) incubated in a water-bath at 37°C for 30 sec (Lane 2, T = 30 sec) or 1 h (Lane 3, T = 1 h) followed by adding 14 μl 5x loading buffer to stop the reactions. The three samples were then heated at 95°C for 7 min and run on an 8-16% SDS-PAGE gel followed by silver staining with the SilverSNAP Stain Kit II (ThermoScientific). Arrow indicates processed, activated CP4 proteinase. The calculated size of the full length HIS-SUMO-CP4 was 59.9 kDa (its predicted size is 50.7 kDa), while the size of the processed, active CP4 indicated by the arrow was calculated to be 32.6 kDa (which is close to the 25.2 kDa size predicted if HIS-SUMO-CP4 is cleaved in a similar position to the that seen for *Nicotiana tabacum*NtCP56 recombinant protein (Zhang et al. ref \[[@B4]\]). ###### Click here for file ###### Additional file 4 **Effect of cysteine protease inhibitor E-64C on CcCP4 activity**. The inhibitory effect of the cysteine protease inhibitor E-64C on the activity of CcCP4 was tested as follows: 10 μL (3.2 μg) His-tag purified and dialysed recombinant HIS-SUMO-CP4 protease was added to 20 μl sodium formate (pH3) and incubated 30 sec in a 37°C water-bath. Then, either 100 μM E-64C (Panel A), or 10 μM E-64C (Panel B), or 1 μM E-64C (Panel C), or no E-64C (Panel D, control), were added, immediately followed by the addition of 6.7 μL BSA reaction buffer (see methods section). For each reaction, 3 μl samples were taken at the start (T = 0), and at T = 5 min, T = 10 min, T = 3 h, T = 4 h30, and immediately added to 5 ul 5x SDS gel loading buffer. The samples were subsequently run on 8-16% SDS-PAGE gels and stained by coomassie. **Lane M**, Molecular marker with the sizes shown on the left in kDa (Biorad Precision Protein™ Standards, prestained). ###### Click here for file ###### Additional file 5 **Plasmid names are given for all the cDNA described in the manuscript**. The sizes of the plasmid inserts and length of encoded proteins are also given. ###### Click here for file Acknowledgements ================ We wish to thank C. Lin and S. Tanksley for generating the Cornell EST clones described here, L. Mueller for bioinformatics and managing the coffee database within SGN, and Jérôme Spiral for providing grain samples for the germination-post germination gene expression analysis. We also thank Thomas Vinos Poyo for generating some of the QPCR data and Vincent Denis for the early work on CcCP1. Finally, we thank Vincent Pétiard and Pierre Broun for supporting the work.

Introduction {#s1} ============ The establishment of invasive organisms is a threat facing aquatic ecosystems world wide [@pone.0051249-Vitousek1]--[@pone.0051249-Strayer1]. Among the most significant--but often most difficult to quantify--ecological impacts of species invasions are alterations to food web structure and energy flow patterns [@pone.0051249-Strayer1]. Stable isotope analysis (SIA) of carbon and nitrogen in constituents of the food web offers a way to untangle the complex effects of species invasions and other perturbations on the trophic structure of ecosystems. Carbon isotope ratios are generally conserved between food sources and consumers, so ^13^C/^12^C ratios can be used to determine the contribution of different energy sources to consumer diets [@pone.0051249-Peterson1]--[@pone.0051249-Post1]. Nitrogen isotopic composition changes predictably with movement up the food web, allowing the determination of consumer trophic level if the nitrogen isotopic composition of basal resources in known [@pone.0051249-Post1], [@pone.0051249-Minagawa1]. Together, C and N stable isotopes have been used to characterize the effects of invasive organisms on the flow of energy through ecosystems [@pone.0051249-VanderZanden1], [@pone.0051249-Nilsson1], food web length [@pone.0051249-Gorokhova1], and resource partitioning [@pone.0051249-Limn1]. Few aquatic invaders have received as much attention from ecologists as the zebra and quagga mussels (*Dreissena polymorpha* and *D. rostriformis bugensis*, respectively). Yet, despite decades of study, the effects of dreissenids on energy flow and food web structure are not fully understood. Dreissenids are increasingly seen as habitat-couplers, redirecting nutrients, energy, and production from the water column to the littoral benthic region of lakes [@pone.0051249-Mills1]--[@pone.0051249-Higgins1]. Dreissenids are filter-feeders that graze on seston and release undigested and unwanted particles as feces and pseudofeces (material collectively known as biodeposits) on the lake bottom, thereby increasing the flux of sestonic material to the benthos [@pone.0051249-Izvekova1], [@pone.0051249-Gergs1]. By clearing the water column and excreting dissolved nutrients at the substrate-water interface dreissenids also increase light penetration and nutrient availability in the nearshore, stimulating primary production of benthic algae [@pone.0051249-Hecky2], [@pone.0051249-Malkin1]. Thus, dreissenids can compete with other filter-feeders for seston while increasing resource availability for littoral detritivores, grazers, and predators. Studies of the effect of dreissenids on benthic communities indeed show declines in the abundance of native filter feeders, and increased abundance of detritivores, some grazers, and benthic predators following dreissenid establishment [@pone.0051249-Higgins1], [@pone.0051249-Ward1], [@pone.0051249-Ozersky1]. In addition to evidence from studies of community structure, SIA approaches have shed some light on the way dreissenid establishment affects food webs. Because seston is often depleted in ^13^C relative to benthic algae [@pone.0051249-Hecky1], [@pone.0051249-France1], carbon isotope ratios can be used to determine how dreissenids affect the importance and availability of these two energy sources to food webs. SIA of carbon has been used to show that dreissenids can compete with native filter feeders for seston [@pone.0051249-Mitchell1], [@pone.0051249-Garton1], and that redirected sestonic material in the form of dreissenid biodeposits can contribute significantly to the diet and production of detritivores such as amphipods and chironomids [@pone.0051249-Limn1], [@pone.0051249-Gergs2]. However, there have been no comprehensive examinations of how dreissenid establishment affects the absolute importance of benthic primary production to food webs, or the relative balance between sestonic and benthic energy sources. The littoral zone can dominate production processes and the transfer of energy to higher trophic levels in lakes [@pone.0051249-Hecky1], [@pone.0051249-Vadeboncoeur1], making it important we understand how perturbations affect the food web structure and energy dynamics of littoral ecosystems. To investigate the effect of dreissenids on the littoral food web of a large lake we compared the abundance, biomass, and production of hard substrate-inhabiting littoral benthos collected prior to, and more than a decade following dreissenid establishment in Lake Simcoe, Ontario. We performed SIA of all common members of the pre- and postdreissenid benthos and used isotope mixing models to estimate the importance of sestonic material and benthic primary production to the pre- and postdreissenid littoral benthic communities. Our objectives, by combining information about the biomass, production, and energy sources of different faunal groups, were to i) determine whether dreissenid establishment has had an impact on food web structure of the littoral benthos and, ii) quantify the impact of dreissenid establishment on the relative and absolute importance of sestonic and benthic energy sources to sustaining the littoral benthic food web. Methods {#s2} ======= Ethics Statement {#s2a} ---------------- No specific permits were required for the described field studies. All field studies were carried out on public property and did not involve protected or endangered species. Study Site {#s2b} ---------- Lake Simcoe is a large (722 km^2^), oligo-mesotrophic lake in southern Ontario, Canada ([Fig. 1](#pone-0051249-g001){ref-type="fig"}). The extensive littoral zone of Lake Simcoe is dominated by rocky substrates to depths of about 8 m, while soft substrates prevail at greater depths. Dreissenid mussels were first observed in the benthos of Lake Simcoe in the fall of 1994, but did not become widespread and abundant until 1996 [@pone.0051249-Evans1]. {#pone-0051249-g001} Sampling the Benthos, Estimating Biomass and Production {#s2c} ------------------------------------------------------- Quantitative benthic samples were collected in 1993 (predreissenid period) and in 2007--2008 (postdreissenid period), using an airlift sampler [@pone.0051249-Barton1] operated by SCUBA divers. Fours sites ([Fig. 1](#pone-0051249-g001){ref-type="fig"}), selected to include mainland and island littoral habitats in different parts of the lake, were sampled in 1993 using a 0.25 m^2^ quadrat. The same locations were sampled again in 2007 for crayfish using a 0.25 m^2^ quadrat and in 2008 for other macroinvertebrates using a 0.0625 m^2^ quadrat. Three to four replicate samples were collected from 2-m depths at each site in the pre- and postdreissenid periods. Sampling in all years was carried out in late August and early to mid September. Crayfish were stored frozen until identification, measurement, and tissue removal for SIA (see below). Quantitative samples of smaller macroinvertebrates were fixed in 10% buffered formalin and then transferred to 90% ethanol prior to enumeration and identification. Preserved airlift samples collected in 1993 from three of the four sites were counted to the order level shortly after collection and then returned to the original sample jars. Samples collected in 2008 were counted to the taxonomic level of family or lower for most groups in order to provide finer resolution for food web reconstruction. To maintain consistency of taxonomic resolution, we resorted the samples collected in 1993, then enumerated and identified invertebrates to the same taxonomic level as samples collected in 2008. Fewer animals were found in most samples during this re-sorting, which we attribute to losses caused by the original handling and processing. To estimate 1993 abundances at a taxonomic level that would allow intercomparison with the postdreissenid benthic community, we assumed that the original counts at the order level were correct and proportionally adjusted the counts for lower taxonomic levels. For example, there were 88 amphipods in the original count of replicate A from Blackbird Point. The recount found 44 *Gammarus* spp. and 12 *Hyalella azteca*, or a total 56 amphipods. We assumed that both *Gammarus* spp. and *H. azteca* suffered similar loss rates, adjusting the numbers to 69 *Gammarus* spp. and 19 *H. azteca*. We believe the assumption of equal loss across lower level taxonomic groups is reasonable given that loss rates were not significantly different across orders (Kruskal-Wallis one-way analysis of variance, *p*\>0.05). Dry biomass of most organisms was estimated from site-specific, length-dry weight relationships ([Text S1](#pone.0051249.s001){ref-type="supplementary-material"}) and observed, period-specific size-frequency distributions. In cases where we did not have sufficient material to construct site-specific length-dry weight relationships, published relationships were used (references in [Text S1](#pone.0051249.s001){ref-type="supplementary-material"}). Annual production of benthic invertebrates was estimated using the empirical model of Morin and Bourassa [@pone.0051249-Morin1], which uses mean areal biomass, mean individual biomass, and mean annual water temperature to estimate taxon-specific secondary production. We report the abundance, biomass and production only for taxa that were relatively abundant (\>5% of total abundance) in either the pre- or postdreissenid period, or that were deemed to contribute greatly to biomass and production because of their large individual size (e.g., crayfish). For a complete listing of invertebrate taxa at these sites see [@pone.0051249-Ozersky1]. Collection and Processing of Samples for SIA {#s2d} -------------------------------------------- Seston for SIA was collected along the south-eastern shoreline of the lake from 8 sites of ∼1 m depth in July and August of 2008. Two 500 ml PET jars were submerged to a depth of ∼0.5 m, filled with lake water and placed on ice. In the lab seston samples were filtered onto precombusted quartz fibre filters, dried at 60°C for 24 hours and stored in a dessicator until preparation for isotope analysis. Periphyton scrapes were collected by a snorkeller from rocks at ∼1--1.5 m depth at McRae Point in September of 2008 ([Fig. 1](#pone-0051249-g001){ref-type="fig"}). A 60 ml syringe with an 18-gauge needle was used by a snorkeller to collect biodeposits and detritus from beneath mussel colonies. We observed mussels expelling fresh biodeposits during our sampling and gently collected this material using a syringe fitted with a 10 cm length of 2-mm Ø Tygon® tubing. In the lab biodeposit samples were filtered onto precombusted quartz fibre filters, dried at 60°C for 24 hours and stored in a glass dessicator until SIA. Samples collected from the surface and from beneath mussel colonies had similar stable isotope values, so these data were combined and are reported together. Most small macroinvertebrates used for SIA were collected in September of 1993 and 2008 from shallow depths (∼1 m) at McRae Point using D-nets with 500-µm nitex mesh and kick and sweep sampling. Macroinvertebrates were separated into general taxonomic groups, allowed to empty their guts for 24 hours in refrigerated containers with lake water, and stored frozen in de-ionized water until analysis. We did not have sufficient frozen material from McRae Point to cover all taxonomic groups, so frozen samples from the predreissenid period were supplemented with formalin-preserved invertebrates collected at McRae Point and adjacent Black Point at 2 m depth in the fall of 1993. SIA results of formalin-preserved samples were corrected for the effect of long-term formalin preservation, which was shown to result in a 2‰ depletion to δ^13^C values and no significant change to δ^15^N values of aquatic invertebrate tissues [@pone.0051249-Rennie1]. Our samples from 2008 were supplemented with specimens of common but less abundant invertebrates collected by kick and sweep sampling at McRae Point and Black Point in September of 2009. Where sufficient numbers of samples of the same taxa were available for a statistical comparison of isotope values of frozen material from 2008 versus 2009, and frozen versus formalin-preserved (and preservation effect-corrected) material from 1993 no significant differences in ^13^C values were found, and only small (\<1‰) differences in ^15^N values were seen in the case of postdreissenid *Gammarus* sp. and chironomids (two-sample independent t-tests at *α* = 0.05); we therefore combined the data. Crayfish for SIA were collected from depths of 2, 4, and 6 m at two sites adjacent to McRae Point (Strawberry Island and Grape Island, [Fig. 1](#pone-0051249-g001){ref-type="fig"}). A comparison of the results showed no statistically significant difference in crayfish isotopic composition with depth or between sites (one-way ANOVAs, *p*\>0.05), so crayfish from all depths were pooled for analysis. Samples of periphyton, crayfish dorsal abdominal muscle, and other invertebrates were prepared for SIA by drying for 24--48 hours at 60°C. Animals other than crayfish were either ground whole or after removal of shells in the case of gastropods and bivalves. When necessary to achieve a minimum sample weight of 0.25 mg, individuals within taxa were combined into a single isotope sample. Biodeposit samples and seston samples were analysed on quartz filters. Seston and biodeposit samples were split into two portions, one of which was acidified by fumigation in a glass dessicator with concentrated hydrochloric acid to remove carbonates [@pone.0051249-Lorrain1]. Periphyton samples were acidified in glass vials by slowly adding 10% HCl until visible bubbling stopped, and were then redried, ground and weighed. Samples were analyzed on a Delta Plus continuous flow mass spectrometer (Thermo Finnigan, Germany) coupled to Carlo Erba elemental analyzer (CHNS-O EA1108, Italy) at the University of Waterloo Environmental Isotope Laboratory. Replicate samples had a standard deviation of 0.21‰ for ^13^C and 0.28‰ for ^15^N. End-member Selection, Mixing Models {#s2e} ----------------------------------- A two-source, single-isotope (^13^C) linear mixing model (IsoError [@pone.0051249-Phillips1]) was used to estimate the relative importance of sestonic material and benthic primary production to taxa in the pre- and postdreissenid periods. No primary producers were available from the 1993 collections. Consequently snails, which are considered to be periphyton grazers [@pone.0051249-Post1], were used as the benthic end-member and filter-feeding hydropsychid caddisflies [@pone.0051249-Merritt1] as the sestonic end-member for the predreissenid mixing model. For the postdreissenid period we examined the effect of using two combinations of end-members on food web reconstruction. In our first approach we used seston and periphyton as the sestonic and benthic end-members (seston-periphyton model). In our second approach we used primary consumers as end-members, reasoning that since primary consumers were used as end-members for the predreissenid period, a postdreissenid mixing model based on primary consumers would enable a more direct comparison between periods. We used dreissenids as the sestonic end-member and psephenid beetle larvae (water pennies) as the benthic end-member because dreissenids appear to rely almost exclusively on seston, while psephenids are known to be grazers of periphyton [@pone.0051249-Merritt1], [@pone.0051249-Murvosh1] and were the closest to periphyton in ^13^C values among postdreissenid consumers (dreissenid-psephenid model).We decided not to use snails as the benthic end member in our postdreissenid reconstruction because snails were no longer the most enriched member of the fauna ([Fig. 2](#pone-0051249-g002){ref-type="fig"}), suggesting that they incorporated more sestonic material than snails in the predreissenid period. The percent contribution of sestonic and benthic material for each taxon obtained from the mixing models together with taxon-specific biomass and production estimates were used to determine the relative and absolute contribution of benthic and sestonic resources to the nearshore food web in the pre- and postdreissenid periods. {#pone-0051249-g002} Assumptions {#s2f} ----------- We did not perform SIA on sphaeriid clams, assuming 100% reliance on pelagic carbon. While the literature partially supports this assumption (e.g., [@pone.0051249-Hershey1], [@pone.0051249-Konkov1] and references therein), overestimation of the contribution of sestonic material to sphaeriids should not greatly affect overall conclusions because sphaeriids comprised only a small fraction of pre- (0.09 and 1.11%) and postdreissenid (0.02 and 0.13%) biomass and production, respectively. We also did not perform SIA on the crayfish *Orconectes virilis*, but results from other locations in Lake Simcoe show that *O. virilis* has similar δ^13^C and δ^15^N values to *O. rusticus* and *O. propinquus* (DO Evans, unpublished data). We assigned the isotopic value of *O. virilis* based on the average isotopic values of 5 randomly selected *O. rusticus* and 5 *O. propinquus* from our dataset. Only one sample of the amphipod *Crangonyx* sp. from 1993 was analysed; the average standard deviation of δ^13^C values of two other amphipods (*Gammarus* spp. and *H. azteca*) was used to approximate the standard deviation of δ^13^C values of *Crangonyx* sp. for use in our mixing model. Results {#s3} ======= Abundance, Biomass, Estimated Production {#s3a} ---------------------------------------- Abundance, biomass, and estimated production (EP) of the nearshore benthos were much higher in 2007--2008 than in 1993, the year immediately preceding dreissenid establishment in Lake Simcoe. The abundances of most individual taxa as well as total invertebrate abundance were significantly greater in the postdreissenid period ([Table 1](#pone-0051249-t001){ref-type="table"}; Wilcoxon rank-sum tests at *α* = 0.05). Total invertebrate abundance increased almost 33-fold from an average of 795.2 (±126.2 SE) to 25,807.0 (±1885.3 SE) individuals m^−2^, total biomass increased approximately 6-fold from 10.7 to 60.3 g dry wt m^−2^, and total EP increased 14-fold from 6.4 to 90.9 g dry wt m^−2^ year^−1^ ([Table 1](#pone-0051249-t001){ref-type="table"}). Dreissenids contributed to much of the increases in biomass and EP of the benthos, constituting 37.7% of the total biomass and 55.8% of total EP in the postdreissenid period ([Table 1](#pone-0051249-t001){ref-type="table"}). Two recent invaders, the crayfish *O. rusticus* and the amphipod *Echinogammarus ischnus*, together comprised 33% of non-dreissenid biomass and 19.7% of non-dreissenid EP in 2007--2008 ([Table 1](#pone-0051249-t001){ref-type="table"}). 10.1371/journal.pone.0051249.t001 ###### Mean abundance (standard error in parentheses), biomass, and annual production of benthic organisms in pre- and postdreissenid periods from 4 shallow (2 m) sites in Lake Simcoe. {#pone-0051249-t001-1} Period: Predreissenid Postdreissenid --------------------------- ------------------- ---------------- ------------- ------------------------ ----------- ----------- Amphipoda * Hyalella azteca* 117.1 (30.9) 11 92 4414.9 (882.1)\* 313 3280 * Gammarus* spp. 11.7 (4.1) 3.7 18 3820.6 (749.2)\* 1020 6850 * Echinogammarus ischnus* not present not present not present 2652.6 (691.3) 446 3500 * Crangonyx* sp. 6.2 (2.3) 0.70 4.6 1146.3 (408.6)\* 143 1230 Isopoda * Caecitodea racovitzai* 0.0 0 0 1516.6 (366.2)\* 385 2620 Decapoda * Orconectes propinquus* 17.6 (2.8) 9970 4870 19.0 (4.4) 17930 7710 * Orconectes rusticus* not present not present not present 10.8 (4.8) 11950 4420 * Orconectes virilis* 0.0 0 0 1.5 (0.7)\* 4040 1160 Gastropoda Physidae 15.9 (6.0) 76 154 51.4 (18.9) 142 427 Hydrobiidae 19.7 (6.3) 3.7 22 278.9 (88.6)\* 89 553 Pleuroceridae 26.2 (8.5)\* 520 699 0.0 0 0 Bivalvia * Dreissena* spp. not present not present not present 3366.3 (688.3) 22720 50760 Sphaeriidae 74.3 (12.7) 14 84 146.3 (45.6) 12 114 Insecta Chironomidae 296.8 (112.0) 12 124 3709.7 (698.4)\* 248 2640 Polycentropodidae 17.6 (4.4) 13 45 171.4 (46.8)\* 104 519 Heptageniidae 46.8 (8.3) 22 104 115.4 (48.1) 42 247 Elmidae 45.2 (11.0) 9.2 60 227.7 (135.0) 49 372 Oligochaeta 98.1 (43.1) 16 93 3798.9 (653.6)\* 565 372 Platyhelminthes 2.0 (1.3) 0.60 3.0 358.9 (120.3)\* 107 4170 **All taxa combined** **795.2 (126.2)** **10670** **6370** **25807.0 (1885.3)\*** **60300** **90900** Asterisks appear beside the significantly larger number as determined by Wilcoxon rank-sum test at α = 0.05. The composition of the benthic community shifted considerably after dreissenid establishment. The predreissenid community was numerically dominated by chironomids and amphipods, while crayfish, snails, mayflies, oligochaetes and amphipods accounted for the bulk of biomass and EP ([Table 1](#pone-0051249-t001){ref-type="table"}). Much of the enhanced postdreissenid abundance was due to large increases in the abundances of amphipods, isopods, chironomids, oligochaetes, and the introduction of dreissenids. Crayfish (33.9 g m^−2^) and dreissenids (22.7 g m^−2^) dominated the postdreissenid biomass, with amphipods, oligochaetes, and isopods also contributing considerably to total biomass. Dreissenids accounted for more than half of EP in the postdreissenid period, followed by small crustaceans, i.e. amphipods and isopods (19.2%), crayfish (14.6%), flatworms (4.6%), and chironomids and other Insecta (4.2%) ([Table 1](#pone-0051249-t001){ref-type="table"}). SIA Results {#s3b} ----------- Snails were the most enriched in ^13^C among the predreissenid benthic taxa (−21.4±0.5 SD ‰), and filter-feeding hydropsychid caddisflies the most depleted (−27.9±0.7 SD ‰). Chironomids were the most depleted group in ^15^N (3.2±1.2 SD ‰), and large lumbriculid oligochaetes the most enriched (14.0±0.85 SD ‰) ([Fig. 2A](#pone-0051249-g002){ref-type="fig"}, [Table 2](#pone-0051249-t002){ref-type="table"}). 10.1371/journal.pone.0051249.t002 ###### Summary of results of ^13^C and ^15^N isotope analysis (standard deviation in brackets) for taxa collected in the pre- (1993) and postdreissenid periods (2007--2009). {#pone-0051249-t002-2} Group Year of collection Preservation method *n* Average δ^13^C Average δ^15^N -------------------------- -------------------- -------------------------- ----- ---------------- ---------------- Chironomidae 1993 formalin 7 −21.76 (1.16) 3.17 (1.60) *Crangonyx* sp. 1993 formalin 1 −25.55 10.35 Elmidae 1993 frozen (1), formalin (6) 7 −26.19 (1.30) 5.19 (0.92) *Gammarus* spp. 1993 frozen (5), formalin (1) 6 −23.96 (1.01) 3.95 (1.01) Heptageniidae 1993 frozen (4), formalin (3) 7 −25.30 (0.40) 5.83 (1.17) *Hyalella azteca* 1993 formalin 5 −23.36 (0.52) 3.83 (0.41) Hydropsychidae 1993 frozen 5 −27.94 (0.67) 7.92 (1.71) *Orconectes propinquus* 1993 frozen 18 −23.54 (0.52) 10.54 (0.70) Oligochaeta 1993 frozen 2 −26.63 (0.22) 14.00 (0.85) Snails 1993 frozen 8 −21.37 (0.50) 5.35 (0.38) biodeposits 2008 frozen 6 −22.56 (0.79) 4.15 (0.25) periphyton 2008 frozen 5 −11.49 (0.72) 3.73 (0.29) seston 2008 frozen 8 −26.90 (1.45) 4.90 (0.56) Chironomidae 2008 (4), 2009 (7) frozen 11 −19.36 (0.48) 8.03 (0.30) *Dreissena* spp. 2008 frozen 7 −25.84 (0.40) 7.29 (0.25) *Echinogammarus ischnus* 2008 frozen 6 −18.60 (0.45) 6.05 (0.48) Elmidae 2008 frozen 4 −19.09 (0.61) 6.40 (0.22) *Gammarus* spp. 2008 (7), 2009 (2) frozen 9 −19.66 (0.55) 7.09 (0.78) Heptageniidae 2008 frozen 2 −22.17 (0.17) 7.13 (0.13) *Hyalella azteca* 2009 frozen 5 −19.99 (0.27) 6.59 (0.71) Hydropsychidae 2008 frozen 2 −24.27 (0.25) 10.29 (0.11) Isopoda 2008 (6), 2009 (1) frozen 7 −19.23 (0.27) 6.54 (0.52) *Orconectes propinquus* 2007 frozen 17 −18.49 (0.58) 10.29 (0.41) *Orconectes rusticus* 2007 frozen 18 −18.94 (1.63) 9.68 (0.67) Oligochaeta 2009 frozen 4 −18.39 (1.26) 8.19 (0.63) Platyhelminthes 2008 frozen 4 −20.54 (0.39) 11.55 (0.54) Polycentropodidae 2008 frozen 4 −21.08 (0.72) 9.77 (0.23) Psephenidae 2008 frozen 5 −15.82 (1.57) 5.65 (0.28) Snails 2008 frozen 14 −18.44 (1.04) 7.82 (0.60) Number in brackets beside year of collection and preservation method indicates how many of the total number of samples (*n*) were collected in that year or preserved using that method for taxa for which samples from more than one year or preservation methods were used. Postdreissenid δ^13^C values of all benthic taxa were bracketed by seston (−26.9±1.45 SD ‰) and periphyton (−11.5±0.7 SD ‰). Dreissenid biodeposits were found to be more enriched in δ^13^C than seston (−22.6±0.8 SD ‰), suggesting the possibility of inclusion of periphyton or macrophyte detritus in our biodeposit samples, preferential rejection by dreissenids of this ^13^C enriched fraction from their filtered intake, or enrichment during metabolic processing by mussels or bacteria. The most ^13^C enriched members of the post-dreissenid fauna were psephenid beetle larvae (−15.8±1.6 SD ‰ δ^13^C), and dreissenid mussels were the most depleted (−25.8±0.4 SD ‰ δ^13^C). Psephenid beetles (5.7±0.3 SD ‰ δ^15^N) and flatworms (11.6±0.5 SD ‰ δ^15^N) were the most depleted and enriched faunal groups in ^15^N, respectively ([Table 2](#pone-0051249-t002){ref-type="table"}). The nine groups that were sampled in both the pre- and postdreissenid periods, and had sufficient replicates for a statistical comparison (*Gammarus* spp., *Hyalella azteca*, *O. propinquus, Physa* sp., Chironomidae, Heptageniidae, Hydropsychidae, Elmidae and Oligochaeta) were consistently and significantly more enriched in ^13^C in the postdreissenid period (two-sample independent t-tests at *α* = 0.05) by an average of 4.47 (±0.66 SE) ‰ δ^13^C ([Table 2](#pone-0051249-t002){ref-type="table"}). Most of those groups also showed a significant, albeit smaller, enrichment in ^15^N (average 1.34 ‰ ±1.01 SE). Mixing Models {#s3c} ------------- Members of the predreissenid littoral food web displayed a range of reliance on benthic and sestonic resources ([Table 3](#pone-0051249-t003){ref-type="table"}). Filter-feeding hydropsychids and oligochaete worms were the most sestonic-reliant groups, and snails and chironomids the most benthic-reliant. Crayfish, which dominated the biomass and production of the predreissenid benthos were estimated to utilize 67% (±3.1 SE) benthic resources. The average, taxon-specific contribution of sestonic carbon across all taxa was 48.6%. The choice of end-members had an effect on estimates of the importance of benthic and sestonic energy sources to different taxa in the postdreissenid period. The seston-periphyton model showed a greater importance of sestonic material for most groups relative to the dreissenid-psephenid model ([Table 3](#pone-0051249-t003){ref-type="table"}). The average, taxon-specific contribution of sestonic carbon across all taxa was 55.3% in the seston-periphyton model and 41.9% in the dreissenid-psephenid model ([Table 3](#pone-0051249-t003){ref-type="table"}). 10.1371/journal.pone.0051249.t003 ###### Contribution of sestonic carbon (% sestonic) to members of the benthos in the pre- and postdreissenid periods in the shallow littoral zone of Lake Simcoe, as determined by stable isotope analysis and the IsoError isotope mixing model [@pone.0051249-Phillips1]. {#pone-0051249-t003-3} Period: Predreissenid Postdreissenid --------------------------- --------------- ---------------- ------------ Taxon Amphipods * Hyalella azteca* 30.3 (4.3) 55.2 (2.2) 41.6 (4.3) * Gammarus* spp. 39.4 (6.8) 53.0 (2.3) 38.3 (4.7) * Echinogammarus ischnus* -- 46.1 (2.3) 27.7 (5.6) * Crangonyx* sp. 63.6 (12.1) -- -- Isopods * Caecidotea racovitzai* -- 50.2 (2.1) 34.0 (4.8) Crayfish * Orconectes propinquus* 33.0 (3.1) 45.4 (2.1) 26.6 (5.3) * Orconectes rusticus* -- 48.3 (3.2) 31.1 (6.2) Snails 0 (3.8) 47.4 (3.0) 29.7 (6.0) Bivalves * Dreissena* spp. -- 93.1 (3.3) 100 (2.1) Insects Chironomids 5.9 (7.1) 51.1 (2.2) 35.3 (4.8) Polycentropodidae -- 62.2 (3.2) 52.5 (5.0) Heptageniidae 59.8 (4.0) 69.3 (2.5) 63.4 (3.0) Elmidae 73.4 (8.4) 49.3 (2.8) 32.6 (5.6) Psephenidae -- 28.1 (4.9) 0 (9.9) Hydropsychidae 100 (7.2) 82.9 (3.0) 84.3 (2.4) Oligochaeta 80.1 (4.7) 45.1 (2.6) 26.2 (5.9) Platyhelminthes -- 58.7 (2.5) 47.1 (4.2) Standard error is shown in brackets. The biomass (69.7%) and secondary production (70.6%) of the predreissenid littoral food web were supported mainly by benthic primary production ([Table 4](#pone-0051249-t004){ref-type="table"}, [Fig. 3](#pone-0051249-g003){ref-type="fig"}), due to the domination of biomass and production by mostly benthic-reliant crayfish and snails. Most (98.5%) of the sestonic carbon in the predreissenid benthos was also contained in crayfish. Amphipods, sphaeriid clams, heptageniid mayflies, elmid beetle larvae, and worms also contributed to storage and turnover of sestonic carbon prior to dreissenid establishment ([Table 3](#pone-0051249-t003){ref-type="table"}, [Table 4](#pone-0051249-t004){ref-type="table"}). 10.1371/journal.pone.0051249.t004 ###### Amount of sestonic- and benthic-derived biomass and production in different groups of littoral benthos of Lake Simcoe, in pre- and postdreissenid periods, using hydropsychids and snails as the sestonic and benthic end-members for the predreissenid period and two combinations of end-members for the postdreissenid period. {#pone-0051249-t004-4} Group sestonic-derived biomass mg m^−2^ benthic-derived biomass mg m^−2^ sestonic -derived production mg m^−2^ yr^−1^ benthic-derived production mg m^−2^ yr^−1^ ----------------------------------------------------------------------------- ----------------------------------- ---------------------------------- ---------------------------------------------- -------------------------------------------- *predreissenid; end-members: hydropsychids (sestonic)- snails(benthic)* Small crustaceans 5 10 37 78 Crayfish 3180 6790 1550 3320 Snails 0 600 0 880 Bivalves 14 0 84 0 Insects 20 23 110 180 Worms 12 4 72 21 **All combined** **3230** **7430** **1860** **4470** *postdreissenid; end-members: seston (sestonic)- periphyton (benthic)* Small crustaceans 1080 1070 8420 8270 Crayfish 16240 17600 6360 6980 Snails 100 130 440 540 Bivalves 20690 2040 46310 4570 Insects 240 200 1980 1800 Worms 310 360 2560 1980 **All combined** **38670** **21410** **66070** **24130** *postdreissenid; end-members: dreissenids (sestonic)- psephenids (benthic)* Small crustaceans 770 1390 6030 10660 Crayfish 9750 24090 3810 9530 Snails 60 170 260 720 Bivalves 22730 0 50870 0 Insects 180 260 1490 2290 Worms 200 480 2060 2480 **All combined** **33700** **26380** **64520** **25680** {#pone-0051249-g003} Contributions of sestonic material and benthic primary production to the food web shifted following dreissenid establishment. Estimates based on the seston-periphyton and dreissenid-psephenid mixing models were in agreement and revealed that, overall, sestonic material contributed more strongly than benthic primary production to the postdreissenid nearshore food web ([Table 4](#pone-0051249-t004){ref-type="table"}, [Fig. 3](#pone-0051249-g003){ref-type="fig"}). However, estimates based on seston-periphyton and dreissenid-psephenid models provided somewhat different estimates for the importance of sestonic material versus benthic primary production to the non-dreissenid benthos. The seston-periphyton model showed that benthic and sestonic energy sources were about equally important to the non-dreissenid benthos, whereas the dreissenid-psephenid model suggested benthic primary production is about twice as important in supporting non-dreissenid biomass and secondary production ([Table 4](#pone-0051249-t004){ref-type="table"}, [Fig. 3](#pone-0051249-g003){ref-type="fig"}). Discussion {#s4} ========== The establishment of *Dreissena* spp. has resulted in dramatic changes to the structure and energy base of the littoral food web in Lake Simcoe. Long-term dreissenid presence led to large increases in the abundance, biomass, and production of littoral macroinvertebrates. We observed changes in both the relative and absolute importance of sestonic and benthic energy sources to the littoral food web, with the relative importance of sestonic material increasing following dreissenid establishment. The results of isotope mixing models and the large increases in biomass and production of the benthos imply that the absolute contribution of both sestonic and benthic material to the food web increased dramatically following dreissenid establishment ([Fig. 3](#pone-0051249-g003){ref-type="fig"}). These results provide evidence that dreissenids transfer energy and nutrients from the water column to the littoral benthos of lakes though deposition of sestonic material, as well as through stimulation of benthic primary production. We found the predreissenid littoral fauna of Lake Simcoe to be typical of exposed rocky substrata in the Great Lakes region [@pone.0051249-Barton1] with crayfish and snails dominating biomass and secondary production, and benthic primary production supporting about 70% of biomass and secondary production. The establishment of dreissenids has frequently been associated with increases in the abundance and biomass of littoral benthos, and while the increase in abundance seen in our study is among the largest reported, the qualitative changes to the benthic community generally paralleled those observed elsewhere [@pone.0051249-Higgins1], [@pone.0051249-Ward1], with amphipods, isopods, chironomids, and oligochaetes undergoing large increases in absolute and relative abundance. It is thought that dreissenid enhancement of littoral benthos, especially on rocky substrates, results from a combination of habitat modification and increased food supply [@pone.0051249-Stewart1], [@pone.0051249-SilverBotts1]. Shells of living and dead dreissenids increase the availability of habitat [@pone.0051249-SilverBotts1] and provide refuge to invertebrates from predation by fish [@pone.0051249-Beekey1], while the increased flux of sestonic material [@pone.0051249-Gergs1], and enhanced benthic primary production rates [@pone.0051249-Hecky2], [@pone.0051249-Mayer1] provide greater food resources to benthic consumers. Until now, however, the relative and absolute importance of redirected sestonic material and benthic primary production to dreissenid-invaded food webs have not been well resolved. Previous studies have shown that dreissenid biodeposits are used as an energy source by some littoral benthic taxa [@pone.0051249-Izvekova1], [@pone.0051249-Gergs2], and our results demonstrate that redirected sestonic material forms a large portion of the energy budget of the postdreissenid littoral food web in Lake Simcoe. While sestonic carbon contributed most to the biomass and production of dreissenid mussels, it also supported between one third and half of the biomass and production of non-dreissenid benthos. The absolute importance of sestonic material to littoral biomass was about an order of magnitude greater in postdreissenid times, while the contribution of sestonic material to littoral secondary production increased by about 35-fold. Excluding dreissenids, the absolute importance of sestonic material to the biomass and production of the littoral benthos increased by between 3--5, and 7--11 fold, respectively, suggesting that dreissenid biodeposits are an important energy source to the nearshore. The notion that dreissenid biodeposits form the main source of sestonic carbon to the littoral food web is supported by the fact that water clarity increased and planktonic algal biovolumes in Lake Simcoe either decreased or remained unchanged since dreissenid establishment [@pone.0051249-Eimers1], [@pone.0051249-Winter1], making it highly unlikely that natural sedimentation rates of sestonic material increased appreciably in postdreissenid times. The absolute importance of benthic primary production to the benthos also increased considerably since predreissenid times. While increases in crayfish biomass accounted for much of the increase in the absolute importance of benthic carbon, all other benthic taxa (with the exception of snails) contributed to the increased utilization of benthic primary production in the postdreissenid food web. The absolute importance of benthic primary production to supporting littoral biomass approximately tripled, while its importance to supporting secondary production increased 4--6 fold. This result is consistent with observed increases in benthic primary production following dreissenid establishment [@pone.0051249-Higgins1], [@pone.0051249-Mayer1] and demonstrates that dreissenid-mediated increases in benthic primary production are translated into increased secondary production. Thus, while we saw moderate changes to the relative importance of benthic and sestonic primary production to non-dreissenid benthos, our results clearly demonstrate that the absolute importance of both increased substantially ([Fig. 3](#pone-0051249-g003){ref-type="fig"}). Our choice of mixing model end-members affected estimates of the relative and absolute importance of littoral and sestonic production to the postdreissenid food web. The relative merits of the two postdreissenid mixing models we used are open to debate. On the one hand mixing models based on primary consumers are recommended because consumers average the signatures of primary producers which can vary considerably over time [@pone.0051249-Post1], and offer a "time-integrated" picture of the food web. On the other hand, the use of primary consumers requires assumptions about their diets. Snails are recommended as the benthic-littoral end-member in mixing models because they are assumed to feed almost exclusively on periphyton. Feeding experiments, however, show that snails often feed opportunistically (e.g., [@pone.0051249-Lombardo1]), and our stable isotope results suggest that sestonic material may have comprised a considerable portion (30 to 48%) of snail diets in the postdreissenid period. The relatively high reliance of snails on sestonic material in the postdreissenid period raises the question whether snails are an appropriate benthic end-member for our reconstruction of the predreissenid food web. We believe that because sestonic material was less readily available to snails in the predreissenid period their reliance on benthic algae would have been higher than in postdreissenid times. There is still a possibility that by using snails we are overestimating the importance of benthic material to the predreissenid food web, but this is unlikely to affect the main conclusion of our study that the absolute importance of both sestonic material and benthic primary production to the littoral food web increased following dreissenid establishment. An unexpected finding of our study was the consistent enrichment in ^13^C and ^15^N of benthic organisms in the postdreissenid period, regardless of whether the organisms were filter-feeders, detritivores, or grazers. Our mixing models show that the relative importance of isotopically-heavy benthic primary production to most taxa did not increase significantly following dreissenid establishment ([Table 2](#pone-0051249-t002){ref-type="table"}), suggesting that benthic and sestonic primary producers became more enriched in ^15^N and ^13^C following dreissenid invasion. While we cannot confirm that such a shift in isotopic values of primary producers occurred, it would be consistent with increased rates of sestonic and benthic primary production often seen in dreissenid invaded systems [@pone.0051249-Higgins1]. As primary production rates increase fractionation against heavier isotopes can decrease drastically, leading to more enriched δ^13^C and δ^15^N values of benthic and sestonic primary producers [@pone.0051249-Laws1]--[@pone.0051249-Hill1], which is consistent with the isotopic enrichment seen in our study. The possibility that changes in primary production rates associated with dreissenid establishment and other ecological perturbations can lead to shifts in isotopic values of primary producers deserves further investigation, and may need to be considered in studies using isotopic approaches and archival samples to examine the effects of perturbations on food webs. Long-term background enrichment in ^15^N values [@pone.0051249-HiriartBaer1] associated with increasing human development of the watershed [@pone.0051249-Evans2] offers another possible explanation for the enrichment of ^15^N values in postdreissenid benthos, although the levels of ^15^N enrichment seen in sediment cores since dreissenid establishment in the lake [@pone.0051249-HiriartBaer1] are unlikely to account for the full magnitude of enrichment seen in our benthos samples. Several caveats should be mentioned. Firstly, empirical models of invertebrate production offer only general approximations of actual production rates [@pone.0051249-Benke1], so our secondary production results should be recognized as estimates rather than direct measurements. Secondly, we assumed that dreissenid establishment was the only major change that impacted the ecology of Lake Simcoe between 1993 and 2008. There is evidence that phosphorus loading into the lake has declined by 20--30% from the mid 1990's to the late 2000's [@pone.0051249-Evans2], [@pone.0051249-Young1], but it is unlikely that this relatively modest reduction could account for the observed changes to the food web, and we believe that dreissenid invasion offers the most parsimonious explanation for our results. Finally, our observations apply to the rocky nearshore of a relatively clear lake with an extensive littoral zone. We predict that in deeper portions of lakes, more turbid lakes, or lakes with a steeper morphometry, dreissenid enhancement of the benthic food web might be driven to a greater extent by redirection of sestonic material, and to a lesser extent by stimulation of littoral production than in our study system. This is the first study to combine SIA with biomass and production estimates to describe how dreissenids affect food web structure and energy sources in lakes. Our findings are consistent with the hypothesis that dreissenids redirect energy and material from the water column to the littoral areas of lakes [@pone.0051249-Hecky2], [@pone.0051249-Higgins1]. By providing increased habitat availability dreissenids create a physical "matrix" for enhancements in the abundance, biomass, and production of the littoral benthos. By changing the rates and sources of energy supply to the benthos, dreissenids modify the energy base and structure of the littoral food web. Biodeposition of sestonic material by dreissenid mussels creates a direct energy link from the water column to the littoral benthos and increases the amount of energy available to detritivores and other members of the benthic community. Increased water clarity and dreissenid remineralization of sestonic-derived nutrients in the nearshore creates an indirect energy link to the nearshore by stimulating benthic littoral primary production. The redirection of production from the sestonic realm to the nearshore by dreissenids has major implications for aquatic ecosystems and may explain some of the changes in the distribution, energy sources, and production of nearshore and offshore fish communities [@pone.0051249-Rennie2], [@pone.0051249-Campbell1], nutrient cycling patterns [@pone.0051249-Staczykowska1], [@pone.0051249-Ozersky2], and transfer of contaminants through food webs [@pone.0051249-Bruner1], [@pone.0051249-Morrison1] seen following dreissenid establishment. Supporting Information {#s5} ====================== ###### **Length-dry weight relationships used to estimate invertebrate biomass, methods used to construct length-dry weight relationships, and references for those relationships obtained from the literature.** (DOC) ###### Click here for additional data file. We wish to acknowledge the help rendered in the lab and in the field by Yvonne Allen, Victoria Kopf, Audie Skinner, Katrina Wisniewski, Stacey Zwiers (Ontario Ministry of Natural Resources), David Depew, Eric Dunford, Joel Harrison, Jennifer Hood, Adam Houben, Simona Lenard, Sairah Malkin, Bill Mark, Jessica Pang, and Lee Pinnel (University of Waterloo). Thoughtful comments from Robert Hecky and an anonymous reviewer helped in improving this manuscript. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TO DOE. Performed the experiments: TO DOE DRB. Analyzed the data: TO DOE DRB. Contributed reagents/materials/analysis tools: DOE DRB. Wrote the paper: TO DOE DRB. [^3]: Current address: Department of Biological Sciences, Wellesley College, Wellesley, Massachusetts, United States of America

Introduction {#s1} ============ Over the last 15 years, a large body of research has focused on the need for integrated, multidisciplinary, team-based and people-centred care. In May 2016, the *Framework on integrated people-centred health services* (IPCHS), adopted with overwhelming support by the Member States of the World Health Assembly (WHA), provided renewed direction and political commitment (WHO, [@ref50]) towards the role of primary care (PC) as integral to implementing integrated care. As part of the 40th anniversary of the World Health Organization's (WHO) Alma-Ata Declaration and the recent Astana Declaration, the WHO has called for a return to its basic tenets, reaffirming the key role of primary health care (PHC) (WHO, [@ref51]a). Revisiting and amending the PHC Declaration offers a unique opportunity to reflect on key achievements and to re-examine how integrated, people-centred care can catalyse PHC-driven universal health coverage (UHC). However, a deeper understanding of these fundamental concepts, and how to deploy them at the community, local, national and cross-border levels is still lacking in many countries. Inter-professional training and team building is also largely lagging behind. Furthermore, necessary supporting frameworks in terms of organisational structures, governance and legislation are either not in place or not comprehensive enough to support transition in many Member States. Available evidence from research efforts developed with significant investment from different European funding mechanisms is more often than not overlooked or underutilised by policymakers when making decisions. This is particularly relevant in terms of PHC and in the context of promoting tenets of the Alma-Ata and the current Astana Declaration (WHO, [@ref51]a). These issues are particularly important when discussing evidence-informed policy-making, especially in countries where accessibility disparities and service inequities are not systematically and adequately addressed. In this paper, we aim to explore why, despite all efforts undertaken and existing evidence, implementation of the Alma-Ata Declaration did not materialise in Greece, and what should be done differently after the Astana Global Conference. We believe Greece exemplifies the socio-economic, political and education drivers of many countries adversely affected by the financial crisis and austerity measures as well as of any country with underdeveloped PHC or lacking synergies in the context of integrated care provision. Moving beyond lessons learned for Greece, we aim to meaningfully discuss experiences to formulate recommendations to move forward so as to successfully meet the Sustainable Development Goal (SDG) 3, and more specifically, target 3.8 (SDG 3.8) \[*Achieve universal health coverage (UHC), including financial risk protection, access to quality essential health care services, and access to safe, effective, quality, and affordable essential medicines and vaccines for all*\]. More specifically, we report on the status of integrated people-centred PHC in Greece and describe actions taken to date, highlighting lessons learned from evidence-based research. In this context, we critically examine what appears to be barriers to the effective implementation of a comprehensive PHC system in Greece. We conclude with a set of evidence-informed policy recommendations on the basis of evidence from Greece and the international literature, especially from WONCA consensus papers and collaborative studies. Lessons learned and recommendations on how a country with limited capacity and resources could effectively implement and sustain integrated, people-centred care could be of value to other countries facing similar constraints and challenges. To serve the aim of this paper, we are using the terms PHC and PC. PHC is an approach to health policy and service provision that includes both services delivered to individuals (including patient pathways) and the general population (White, [@ref49]). PC refers to 'family doctor-type' services delivered to individuals, whereas some frameworks (White, [@ref49]; Kringos *et al*., [@ref16]a) use PC to assess PHC components. We have also adopted an operational integration model based on the Donabedian approach (Donabedian, [@ref9]), combining the basic PHC principles presented by Starfield ([@ref42]) and the Chronic Care Model (Boult *et al*., [@ref3]). The dimensions of PHC as reported in the work of Kringos *et al*. ([@ref17]b) was selected as the most fitting for the Greek health care system. The current situation in Greece {#s2} =============================== Greece first introduced a PC reform jointly with Spain and Portugal, as part of a WHO pilot project to assess different approaches through the operationalisation of key concepts of the Alma-Ata Declaration (WHO, [@ref52]b). In the early 2000s, the reform was widely discussed in the context of drafting and enacting legislation, and the necessity for integrated PC was highlighted for the first time (Souliotis and Lionis, [@ref41]). In 2009, a systematic review exploring the role of integrated PHC within the Greek health policy agenda concluded that it was still in its infancy, recommending a major restructuring of the National Health System (NHS) (Lionis *et al*., [@ref27]). In 2015, Tsiachristas *et al*. ([@ref45]) revisited the subject to provide an action plan that would bridge local policy and population needs considering experiences gained from other European settings. At that time, a critical debate, involving both politicians and professionals across specialties, was taking place as a result of major changes that were occurring in the country, and following a number of years with multiple measures of austerity implemented in the context of rationing versus rationalising care. Since 2010, Greece has been severely affected by the economic crisis and austerity with significant direct impact on the state-funded NHS. As a result, the country refocused its reform efforts on PC with two legislative bills. The first one, Law N° 4238/2014, had a clear fiscal direction on the National Primary Health Care Network (PEDY) and the scope of the Greek National Health Service (EOPYY). This reform introduced the integration of PHC structures and outposts with services of the Social Insurance System. Although conceptually on the right path, the reform faced several difficulties during implementation. It is still debated to what extent this is the direct result of serious lack of manpower, pushing citizens to higher utilisation of private services, including those of specialists, in a country with a long-standing tradition of mixed public and private services, and no gate-keeping system or comprehensive PHC system to start with. The second legislative bill, Law N° 4486/2017, was enacted within months of being posted for public is currently in its early phases of implementation. Partly because of the very low number of general practitioners (GPs), the direction is to strengthen the role of the family physician by assigning it to GPs, internists and paediatricians. As gatekeepers to costlier and specialised health care services and interventions, family physicians are the first point of contact, providing comprehensive and continuous care with a focus on disease prevention and health promotion. The law mandated the opening of more than 200 Local Primary Care Units (TOMY) in urban centres, jointly funded by the Greek State budget and the European Social Fund (ESF) under the National Strategic Reference Framework for a period of three years. These PC Units are team-based and comprise family physicians, nurses, health visitors, administrative staff and in some cases a midwife. A media campaign was launched in 2018 to raise public awareness. As of January 2019, just over 100 TOMY units were fully operational. These recent reform efforts have sparked an intense national debate over the lack of sufficient numbers of family physicians, financial incentives to ensure adequate recruitment and the sustainability of the units beyond the three-year period of funding support by the ESF. The latter necessitates further structural changes across levels of care and reallocation of resources to ensure sustainability. Despite the above challenges, establishing a fully operational and sustainable integrated PHC system in Greece is more relevant than ever before. A recent survey by the Observatory on Health Reforms, a joint initiative of the University of Peloponnese and the Health Policy Institute, revealed that there are serious and persistent barriers, primarily access to physicians and financial concerns, which have changed the public's attitudes towards GP-led care pathways (Kanavos and Souliotis, [@ref12]). There is evidence that GP-led pathway to care can act not only as a means to improve access to care for all, but also as a vehicle to guarantee fiscal sustainability for the health system as a whole (Kanavos and Souliotis, [@ref12]). Lessons learned from Greek observational research studies relevant to integrated multidisciplinary health care services {#s3} ======================================================================================================================= From 2012 to 2017, the Clinic of Social and Family Medicine at the Faculty of Medicine of the University of Crete (UoC) carried out several regional and national studies on PC services relevant to the implementation of an integrated PHC system in Greece. Specifically, a study exploring the *scope of practice and training needs* of PHC nurses working in rural Health Centres on the island of Crete found that nursing staff operated within a restricted and task-oriented framework (Markaki *et al*., [@ref30]; [@ref29]; [@ref28]). Job satisfaction was low in terms of daily interactions with colleagues and support from work environment. The need for timely recruitment of young, qualified nursing staff choosing a career in PHC explicitly emerged (Markaki *et al*., [@ref30]). Significant training needs were reported by all nursing personnel, mainly in clinical and research/audit skills. Most respondents expressed a strong desire for continuing education, despite a self-reported adequate knowledge level. Investigators recommended a systematic overview of skill deficits, in relation to skill requirements, implemented regularly by NHS authorities to enhance targeted on-the-job training, based on individual- and group-specific needs (Markaki *et al*., [@ref28]). *The need for multidisciplinary team-work* in managing health problems was also highlighted in research by Papadakaki *et al*. ([@ref31]; [@ref34]) who studied physician readiness to manage intimate partner violence (IPV). The PREMIS (Physician Readiness to Manage IPV) survey was translated and validated into Greek. A pre-post control group design was used to examine an intervention for addressing IPV which involved both GPs and residents in general practice in Crete (Papadakaki *et al*., 2011; [@ref31]). The training programme met a high acceptance by both groups with the intervention group of GPs performing better than the control group on 'perceived preparedness' and 'perceived knowledge' in both the post-intervention and the 12-month follow-up (Papadakaki *et al*., [@ref32]; [@ref31]). The study found no statistically significant improvements or between group differences in terms of the self-reported detection of IPV cases (Papadakaki *et al*., [@ref31]). This is a good example on how a training program could be designed based on a multidisciplinary approach and implemented effectively in low-resource settings. *Perceptions of PHC practitioners and patients* regarding the quality of services in rural Greece were also evaluated in a study involving 21 PHC units from two regions of Greece, namely, Epirus and Crete (Sbarouni *et al*., [@ref39]). This research generated recommendations for re-engineering the existing, at the time, system, including: (a) introduction of new technologies and emphasis on system interoperability, (b) GP empowerment, leadership reform and (c) mechanisms for evaluating the quality of services. Areas of concern regarding future development and utilisation of private PHC infrastructure and services were highlighted (Sbarouni *et al*., [@ref39]). A second large study explored patients' expectations and experiences in Greece using data collected as part of the international QUALICOPC study (Lionis *et al*., [@ref23]). Several barriers were identified in terms of waiting time for appointments, GP access to patient medical history, delivery of preventive services and patient involvement in decision-making (Lionis *et al*., [@ref23]). Patients with chronic disease reported a better experience than respondents without a chronic condition and highlighted the disease-focused orientation of health service delivery. Data gathered may be used to address the quality of services through an increased focus on patient-centred interventions, and people-centred approaches as well as on preventive service delivery. A nationally funded research project carried out by the UoC focused on *operational integration* and had a two-fold aim (1) to assess the level of operational integration within PHC units by utilising standardised mapping and evaluation of both unit-level and patient-level integration and (2) to develop an optimum model of operational integration tailored to the Greek PHC system (Sifaki-Pistolla *et al*., [@ref40]). Existing patient care pathways were mapped and analysed, and recommendations for an optimum model were made. The developed web platform, based on a strong theoretical framework, can serve as a robust integration evaluation tool and contribute towards optimising patient flow across levels of cares, beyond unit level. This work could serve as a first step towards restructuring and improving PHC services within a financially restrained environment (Sifaki-Pistolla *et al*., [@ref40]). A recently published paper by the UoC research team focused on how health care delivery data, collected from capacity building and research initiatives, can facilitate effective planning and integration of public health into PHC (Lionis *et al*., [@ref26]b). Research activities summarised here focus on the PHC working environment through the lens of physicians, nurses, other health care providers, and most importantly, patients and people across communities. All studies contribute towards enhancing knowledge, skills, cooperation and team work in the direction of building an integrated PHC model with the patient and family as an empowered and integral part of the care team. Thus, ensuring a dynamic relationship and establishing sound ground for shared decision-making are a key issue highlighted in the WHO Astana Declaration. Key challenges for establishing integrated people-centred PC in Greece {#s4} ====================================================================== Based on the study by Kringos *et al*., Greece was classified among the countries with weak PC in terms of access, continuity and comprehensiveness, while it was rated as strong on coordination (Kringos *et al*., [@ref18]). Although, the current NHS legislative framework promotes some of the key principles in the Alma-Ata Declaration, there are several barriers impeding progress. We review here the key barriers and challenges affecting PC services in Greece today. Availability of family physicians {#s4-1} --------------------------------- Human resources is considered a key factor to improving health care and the following demonstrates its relevance to integrated PC. While Greece has the highest per capita rate of licensed physicians among EU Member States (632 per 100 000 inhabitants in 2015), these are primarily specialists (Economou, [@ref6]). This ratio is considerably higher than any of the other EU Member States; Austria (510) and Portugal (461) had the next highest ratios of physicians licenced to practice and along with Lithuania (434) were the only other Member States to record over 430 physicians per 100 000 inhabitants. In contrast, Greece reports the lowest per capita rate of GPs (0.28 per family practitioner per capita) in Europe (Economou, [@ref6]; Eurostat, [@ref7]). The per capita ratio of registered nurses is also among the lowest in Europe (Economou, [@ref6]). The high density of specialists and disproportionate ratio of specialist to GPs create challenges and barriers to PC reform. Training programs with a standardised curriculum for general practice have only recently been developed, while the academic capacity of this specialty needs more support. Furthermore, the departure of large numbers of physicians during the austerity period (brain drain) has compounded the paucity of GPs. It has been reported that more than 18 000 Greek physicians have left Greece to seek employment and/or better work conditions (Economou, [@ref6]). In October 2013, the Athens Medical Association reported that approximately 7000 doctors of all specialties left between 2008 and 2013 after the onset of the financial crisis (Economou, [@ref6]). This leads to imbalances in service provisions and highlights the direct effect of 'brain drain' on efficient health care service delivery (Economou, [@ref6]). Information and Communication Technology (ICT) {#s4-2} ---------------------------------------------- Greece is not the only European country suffering from lack of interoperable systems -- indeed this is a well-recognised issue affecting cross-border health care across countries. Nevertheless, the lack of a uniform electronic health record (EHR) across regions and settings, linked effectively to the centralised prescribing and referral services, has been a significant impediment to progress. Other barriers to achieving integrated PC include (a) lack of a national electronic health monitoring platform to support communication and patient registries, (b) episodic health screening, (c) fragmented coordination of referrals for chronic disease management, in particular cancer care, (d) weak social services support, and (e) increased demand for services due to migration and refugee health problems. Collaborative projects on integrated care services have shown that delays in the transfer of ITC linkage and the socio-economic environment may contribute to negative results when advanced training initiatives are launched in different settings (Barberan-Garcia *et al*., [@ref2]; Hernandez *et al*., [@ref10]). Although Greece has adopted and applied medical informatics education, including multidisciplinary postgraduate courses and large participation in EU-programs informatics research, with palpable scientific contributions, these remain fragmented efforts. Although local experiences are reported in the literature (Kounalakis *et al*., [@ref15]; Samoutis *et al*., [@ref38]), the country still struggles to implement a uniform and universal EHR and interoperable e-health applications. According to research findings, major implementation barriers are physicians' beliefs that EHR adoption affects workflow, physicians' ethical concerns, lack of incentives, system breakdowns, software design deficiencies, transition difficulties and lack of familiarity with electronic devices (Samoutis *et al*., [@ref38]). Across Europe, countries that have invested in workforce and ICT infrastructure, understanding how cultural differences and capacity affect teamwork, have succeeded in reaching with health system improvement goals (Auffray *et al*., [@ref1]). Prevention and management of chronic disease {#s4-3} -------------------------------------------- Within the EU of 28 member countries, Greece reports among the highest rates of chronic diseases (lung cancer, lung diseases and CVD) and very high rates of CVD risk factors, including the highest rate of tobacco use and growing rates of obesity, metabolic syndrome and cancer. Preventing exposure to NCD-causing risks, such as physical inactivity, tobacco smoking, air pollution and other similar risks can be tackled through a toolkit of tested and affordable policy interventions. The 'natural' setting for utilising such a toolkit and implementing interventions is found at the PC level through PHC provision. PHC must work synergistically with strong public health policy and strategies ranging from strong enforcement of policies on smoke-free public places to increased taxation of unhealthy products. The primarily acute care-focused Greek NHS accounts for the very low rates of preventative health screening and vaccination relative to other EU countries. This results in delayed diagnosis and treatment of risk factors and disease. Strong working relationships between public health and PHC to support health screenings and pre-symptomatic care delivery are critical for addressing the burden of NCD nationally. Strengthening ICT system supports also plays a critical role in NCD management. For example, monitoring practice performance, phone and email contacts, development of communication and cultural interaction skills have all been found to improve rates of colorectal cancer screening delivery, offering a good example for implementing an integration care model (Triantafillidis *et al*., [@ref44]). Considering the low levels of systematic screening program deployment in Greece, we can presume that curative services have been disproportionally affected within NHS priorities, in comparison to preventive services. Service fragmentation with functional disruption between primary and secondary care may be attributed to the lack of any systematic effort of health integration for a long period of time. Examining the example of cardio-metabolic disease (CMD), a recently published European systematic review revealed that younger age, smoking, low education and attitudes, such as being or not being worried about the outcome, low perceived severity or susceptibility and negative attitude towards health checks or prevention were barriers for participation in cardiovascular health checks (deWaard *et al*., [@ref5]a). In a subsequent European study, despite the fact that most GPs considered selective CMD prevention as useful, it was not universally implemented. The biggest challenge was inviting individuals for risk assessment, while the involvement of stakeholders was identified as important (deWaard *et al*., [@ref4]b). Greece is still far from integrating self-management and self-care in chronic disease management. Investing in the development of expanded information and emotional support networks, and attendance of community organisations in decision-making processes have been shown to result in better self-management capabilities (Koetsenruijter *et al*., [@ref14]). Strengthening health system understanding and investment in known self-management strategies and interventions will be required to compliment actions in the broader NHS. Migrant and refugee health {#s4-4} -------------------------- Another challenge for the current PHC system in Greece has been introduced as a direct result of the migrant and refugee crisis in Europe. In many European countries, providing newly arrived migrants and refugees a more systematic health-reception, based on a holistic approach by a multidisciplinary team, will not only benefit migrants and refugees but also will protect the public health of host countries (Leder *et al*., [@ref19]). As part of EUR-HUMAN project, UoC explored the barriers and facilitators for an accessible, acceptable and appropriate -- culturally and linguistically relevant PHC (Lionis *et al*., [@ref25]a; vanLoenen *et al*., [@ref46]). This approach was informed and tested in five European countries with diverse health care systems. The inclusion of care recipients in early dialogue for needs, assessments and to elicit their wishes and preferences, results in higher engagement of both professionals and migrants/refugees and empowerment to take more proactive roles. Lessons learned included revisiting the feasibility assessment, and the toolkit generated could readily be used across settings to establish positive initial interaction and smoother conditions for the start of implementation journeys. Many shortcomings became evident, such as the absence of clear policies on entitlement to care, while legal restrictions on health care access for marginalised migrants resulted in numerous cases of ambiguous paths, decisions and delays in the provision of appropriate care. In particular, for irregular migrants that may remain undocumented until they reach the care system with an acute episode or for those having been refused asylum, the structural barriers were seriously impeding access and appropriate care delivery (Lionis *et al*., [@ref24]; Teunissen *et al*., [@ref43]; Papadakaki *et al*., [@ref33]). It is important to remember that UHC cannot be achieved in its true essence without providing for refugee and migrant groups. It is also important to consider an appropriate PHC versus other acute forms of care for these groups. Support at the primary level can facilitate not only smoother integration including entry into the health care system, but also appropriate surveillance and monitoring in the interest of public health, without compromising the rights of people. While the current economic conditions determine key aspects of Greek PC delivery, several practical and training limitations (e.g., number of interpreters, access to location, medical interpreting requires lengthy and specialised training) have surfaced. Interestingly, there is no single European coordinating mechanism to provide guidelines, high quality training and accreditation, ensuring the health of individual people and public health interests. Much of the implementation work to date has been the result of the early mobilisation of academic and not-for-profit organisations, without a centralised coordinating mechanism to facilitate inter-sectoral collaboration, public--private partnerships and rapid capacity increase. Barriers identified in the context of rapid capacity building have included duplication of efforts by groups and capacity limitations. The need for informational continuity of care and the importance of a centralised and well-managed mechanism that allows structures, institutions and care providers to cross-talk has been highlighted (Lionis *et al*., [@ref25]a; vanLoenen *et al*., [@ref46]). Discussion {#s5} ========== Although several actions have been undertaken by the Greek Ministry of Health and WHO over the past few years, PC in Greece is still far from embracing and utilising to the full extent the key concepts in consensus documents and WHO declarations (IOM, 2001; WHO, [@ref50]; WHO, [@ref51]a). Fragmentation of services and lack of consistent direction in the management of chronic diseases and major public health risks are on-going handicaps (Lionis *et al*., [@ref23]). Poverty, demographic issues, including population ageing, health care expenditures and limited resources and capacities are pressures that NHS will continue to face. Key questions, also, remain regarding the core attitude, knowledge and skill competencies for PC physicians required to serve the Greek population while being confronted with complex biopsychosocial conditions. In this regard, the core aspects of training needed to develop such competences ought to be systematically revised aiming towards an interdisciplinary team approach (Papadakaki *et al*., [@ref32]). The specialty training for GPs has recently expanded from a 4-year cycle to a 5-year cycle to expand baseline skills. However, this is only one of the steps required in terms of how both professionals and the public conceptualise care, PHC, and self-care. For GPs, more comprehensive training is needed to prepare for their role as family physicians in a manner that adequately addresses complex needs of individuals, families and communities. In addition, the Greek NHS still remains conventional, medically focused with a vertical hierarchical structure that promotes health care-seeking behaviours outside the PC level. The long-standing dominance of medically oriented health policy in Greece has impeded efforts to promote a more interdisciplinary integration of care. At the same time, potential for a strong contribution from the nursing profession to PC, through recommended optimum pathways in NCD management, preventive screening, health promotion and home care is still not achieved (Sifaki-Pistolla *et al*., [@ref40]). The recommendations below are designed to guide current health policy towards an effective integrated PHC model.(1)Effective human resource planning to increase the number of PC professionals and address existing skill mix imbalances between specialists and GPs and the lack of adequate nursing and allied health professions personnel.(2)Implementation of a fully operational e-communication, interoperable, system that is sensitive to the pragmatic conditions and accommodates the needs of multidisciplinary teams. Effective use of ICT with a comprehensive medical records system, as reported by Kounalakis *et al*. ([@ref15]), should become a high NHS priority.(3)Orientation of the new PHC units to address major public health issues (i.e., NCD spectrum, including cancer, CVD, diabetes, frailty, traffic accidents) and risk factors (i.e., smoking, obesity, driving behaviour, high consumption of sugar, alcohol, etc.), with interventions encompassing health promotion to change health care-seeking behaviours, prevention, screening and early diagnosis and management of health risks and disease.(4)Coordinated actions for integrated chronic disease care. The high percentage of people with two or more chronic conditions (i.e., multi-morbid people), necessitates optimal use of available resources and more sophisticated mechanisms to coordinate care. Also, mobilising resources beyond care structures; for example, through participatory initiatives at community level. This, in terms of policy, means substantial investment in ICT and training, with a strong coupling of what such a new approach offers via public awareness campaigns and professional education. A robust action plan could be operationally integrated using the model and tools proposed by Sifaki-Pistolla *et al*. ([@ref40]).(5)Emphasis on integrating public health and PHC, and information flow and exchange in order to inform priorities, opportunities and best practices. Such an effort can be supported by the IPCHS framework on IPCHS (Lionis and Petelos, [@ref22]).(6)Development of core competencies and implementation of a coordinated continuing education program for PHC professionals. This recommendation addresses the clear need to retrain the PHC practitioners in Greece with a focus on developing the PC team and a culture of interdisciplinary collaboration.(7)Interprofessional education through a national plan for restructuring PHC training programs, focusing on general practice (changing the structure, curriculum content, teaching and evaluation methods) and other health science disciplines.(8)Coordination of care by the regional and local health authorities in order to link health care services with other domains and sectors that impact both health promotion and disease prevention. All services should be well tailored with population health care needs and policy planning should consider public perception of PHC. It must be emphasised that multiple transitions in the context of reform may be difficult to implement successfully, particularly when this takes place rapidly with severely limited resources. Adopting integrated people-centred PHC as the norm (gold standard) throughout settings requires a major change in organisational culture at all levels of the health care system. Several policy papers published by WONCA ([@ref53]), including one on improving health globally (vanWeel and Rosser, [@ref47]), could support Greek policy makers towards this action. In addition, PHC policy should showcase contributions to UHC, as commented by vanWeel *et al*., ([@ref48]). This investment to PHC should be reflected in training of PHC professionals at community setting, and it is a true challenge for current reform in Greece. The four emerging priorities namely, community-based advocacy to policymakers, collaboration with universities to include PHC as a core component in medical curricula, collaboration with communities to improve public awareness of PHC and engagement with the private sector to focus on PHC and UHC are entirely relevant to the case of Greece. Lessons learned and recommendations made in this development paper are likely to be relevant to countries where PHC integration is emerging as a key issue in health policy reform. In particular, some of the PHC domains of cross-border relevance are service orientation towards chronic diseases and major public health challenges, integration of public health and advanced training of PHC professionals in multidisciplinary models of care. All of these issues deserve further attention by researchers who will conduct similar analyses in their own countries or settings (Lionis *et al*., [@ref26]b). Conclusions {#s6} =========== Unravelling Ariadne's thread shows a clear pathway towards sustainable PHC reform and UHC. Greece is emerging from a prolonged period of financial and refugee crisis during which the austerity measures and the rationing across state provisions have affected the socio-economic status of a large part of the population. At the same time, certain systematic efforts continue to reform PHC, conceptually having a sound base and meeting the tenets of Alma-Ata. However, these efforts necessitate strong political will, immediate and concrete actions, as well as substantial investments, to ensure that rapid transition to the proposed system is successful in 'leaving no one behind'. Serving as a testament to the resilience of overburdened health care professionals, systems, and, indeed people, this development paper intends to accelerate action for UHC. The authors would like to acknowledge the significant contribution of the academic and research staff and external scientific partners of the Clinic of Social and Family Medicine, School of Medicine, University of Crete, for their constant support and efforts in the aforementioned projects. Special thanks to the 7th Health Region of Crete for establishing an effective collaboration during aforementioned project's implementation as well as on their support in sustainability activities in the PHC setting in Crete. Author ORCIDs {#s7} ============= Christos Lionis [0000-0002-9324-2839](0000-0002-9324-2839), Adelais Markaki [0000-0002-2038-3139](0000-0002-2038-3139) Funding {#s8} ======= No funding support was received. Competing interests {#s9} =================== The authors have nothing to declare.

Background {#Sec1} ========== Globally, delivering in a facility has increased across Africa and Asia, among rural and urban, and rich and poor populations \[[@CR1]\]. In India, facility deliveries have impressively increased from about 20 % to over 70 % in a span of 10 years; however this has not resulted in the maternal and neonatal health improvements expected by health policy analysts \[[@CR2]\]. In 2013, the maternal mortality ratio in India was approximately 190 deaths per 100,000 live births and the infant mortality rate was 41 deaths per 1000 live births, one of the highest in the region \[[@CR3]\]. Moreover, there is significant heterogeneity across states in India, with Uttar Pradesh, located in Northern India \[[@CR4]\], reporting among the worst health outcomes in the country \[[@CR5]\] and low levels of women's autonomy \[[@CR6]\]. National programs in India, such as the Janani Suraksha Yojana (JSY), have been the major drivers of increases in facility deliveries, particularly among the poorest populations. JSY, launched in 2005, is a conditional cash transfer program that financially incentivizes women to deliver in a facility. Women are offered approximately \$15--\$30, depending on the state in India. While facility deliveries have shown dramatic improvements, quality of care and challenges with the scheme continue, including ensuring skilled attendant at birth, experiences of disrespectful care in facilities and discrimination due to social status of women, transportation barriers to facilities, and low levels of trust in public health facilities \[[@CR2]\]. Historically, much of the maternal health efforts have focused in rural areas; however, in 2013, in response to growing urban health disparities, India launched the National Urban Health Mission (NUHM), which focuses on meeting the healthcare needs of the urban poor, particularly slum populations. Nearly 100 million people in India live in slums: 17.4 % of all urban households, reaching as high as 41.3 % of the population of Mumbai. Barriers to healthcare utilization for maternal health in slum communities are well documented in the literature. Studies find that while slum dwellers prefer formal over informal maternal health services, there are significant barriers to accessing care, including geographical access, ineffective family household decision-making, safety concerns, high cost of health services, and perceived low quality of providers \[[@CR7]\]. Existing literature in Africa and parts of Asia suggests that slum-dwellers experience a number of adverse maternal and child health outcomes, including less access to antenatal care (ANC) services and facility deliveries. Urban populations have grown rapidly over the past two decades; it is estimated that by 2025 64 % of people in low-income countries will also live in urban areas \[[@CR8]\]. This process of urbanization presents a number of new health opportunities and challenges: although urban areas can provide residents with greater access to health and economic resources, urban residents are also exposed to new economic and environmental risks. This is particularly true in slum areas. Given the renewed focus on urban slum populations, it is important to better understand the healthcare challenges that this population faces, including decisions around where to deliver, support structures, financial challenges, and perceived quality of care. Few studies have gathered voices of women, husbands, and mothers-in-law in order to triangulate information \[[@CR9]\]. Past qualitative studies have found that financial burdens restrict place of delivery to public hospitals, and that cultural and social factors, such as expectations around respectful and safe care, were significant drivers for location of delivery. More recent studies also find that mothers-in-law, particularly paternal mothers-in-law, play a significant role in household decision-making, including health services \[[@CR10]\]. The role of husbands and men in particular has been under researched, but studies find that interventions targeting men during the time of delivery have positive health benefits to the mother and child \[[@CR11]\]. There is little information on decision-making among women, and how families, including husbands and paternal and maternal grandparents, navigate the continuum of maternal health services from antenatal care (ANC) to a facility delivery in slum areas. A recent qualitative study in rural India attempted to include information from recent mothers, their husbands and their mothers-in-laws and found that family members play an important role in where women seek delivery care \[[@CR12]\]. This study, however, was not able to link family members and therefore limited in understanding how household dynamics may influence shared decision-making. Triangulating these different voices and perspectives may shed light on points of interventions to improve maternal and child health in India, particularly in urban settings given rapid changing gender and social norms. Through qualitative interviews, this study seeks to examine the ways in which household decision-making influences the health utilization of maternal health services in urban slums and offers insights for points of interventions for the health sector. This study aims to understand the principal economic, environmental, and social factors that influence women's decisions to seek maternal health services and the quality of care received. This study first addresses decision-making around where to deliver, including home vs. facility or public vs. private sector; it then delves into recent mothers', husbands', and mothers-in-laws' experiences with maternal health services. Methods {#Sec2} ======= Study participants and procedures {#Sec3} --------------------------------- In total, we conducted 80 in-depth interviews, including 40 with recent mothers, 20 with their husbands, and 20 with their mothers-in-law. Purposeful sampling was conducted in order to obtain differences across delivery experiences (facility vs. home), followed by their family members. We recruited 30 women who delivered in a facility and ten women who delivered at home. Two recruitment lists were formed to provide the sampling frame for interviews. First, researchers worked with Anganwadi centers in the two study sites to develop lists of women who delivered in the past year. Second, researchers went door-to-door in the communities to recruit more women who had delivered in the past year in order to ensure that a variety of women were recruited from the community. Study participants were recruited from two slums in Uttar Pradesh, Lucknow. Inclusion criteria for recent mothers included women who delivered a child in the past year. Once women were recruited, they were asked whether their husband or mother-in-law lived in the same house and available to speak, and only if the woman gave permission did we recruit their partners and mothers-in-law. Interviews lasted approximately one hour, were tape-recorded, transcribed into Hindi, and translated into English. Four trained research assistants, including two men and two women, facilitated the interviews. Interviewers and participants were matched according to gender. All interviews were conducted between April and July 2014. Participants were asked a number of questions including about their delivery experiences, quality of care received, choice of provider and delivery location, support received before, during, and after the time of delivery, cost of care, and decision-making about maternal and child health services. Women, husbands, and mothers-in-law were asked similar questions in order to triangulate information where possible. Data analysis {#Sec4} ------------- Data was analyzed using Atlas-ti software by two trained researchers. Content analysis was used to analyze the textual information. The analysis was completed in three phases. First, two researchers read each transcript several times to examine the text as a whole, and to identify initial impressions. Second, two researchers then coded a subsample of the text in an iterative process. Two researchers first developed codes separately using ten transcripts, revised and refined codes together, and developed a common coding scheme. The tentative codes were discussed and revised by the researchers. The codes were then applied to all of the transcripts. Finally, codes were grouped into families, or broader themes. Researchers had discussions regarding broader themes, including where participant voices converged and diverged. Matrices were developed for specific themes in which a variety of participant types were questioned. Ethical approval {#Sec5} ---------------- The study and all accompanying study materials were reviewed and approved by the Institutional Review Boards at the University of California, San Francisco (UCSF) and the Centers for Operations Research and Training (CORT) in Gujarat, India. Verbal informed consent was obtained from all study participants. Results {#Sec6} ======= Participant demographic characteristics {#Sec7} --------------------------------------- The women interviewed for this study were, on average, 26 years old (Table [1](#Tab1){ref-type="table"}). In terms of religion, the large majority (*N* = 37, 93 %) of respondents were Hindu, and the rest were Muslim. Fifty percent (*N* = 20) belonged to a Scheduled Caste, 20 % (*N* = 8) to Other Backwards Caste (OBC), and 30 % (*N* = 12) to none of these. All women were married and currently living with their husbands, and 65 % also had their mothers-in-law living with them (*N* = 26). Twenty-eight percent had no education (*N* = 11), 23 % had less than 8 years (*N* = 9), 38 % had 8 to 11 years (*N* = 15) and 10 % had 12 or more years of education (*N* = 4). Women had, on average, 1.9 births in their lives, ranging from 1 to 7. Women spent an average of 7,721 rupees on services, supplies and medicines during their most recent delivery (interquartile range from 0 to 12500), and a mean of 334 rupees on transportation to the facility (IQR 0-200). The majority (*N* = 35, 88 %) were not currently working outside the home. None of the women had been living in Lucknow their entire lives, 8 % had lived there less than 1 year (*N* = 3), 20 % for 1 year (*N* = 8), 40 % for 2--5 years (*N* = 16), 18 % for 5 to 10 years (*N* = 7) and 15 % for more than 10 years (*N* = 6). The majority (*N* = 32, 80 %) had lived in a city before moving to Lucknow, 5 % in a town (*N* = 2) and 15 % in the countryside (*N* = 6). The majority of women had their most recent birth at a large government hospital (48 %), followed by 25 % at home (*N* = 10), 25 % at a private hospital (*N* = 10), and 2 % at a lower level facility (*N* = 1).Table 1Demographic Characteristics of Study ParticipantsWomen (*N* = 40)Husbands (*N* = 20)Mothers-in-law (*N* = 20)Age, mean (IQR)26 (23--29)30 (25 to 32)54 (45--61)Hindu (N (%))37 (93 %)19 (95 %)18 (90 %)Caste Scheduled Caste20 (50 %)6 (30 %) Scheduled Tribe00 Other backwards8 (20 %)7 (35 %) None12 (30 %)7 (35 %)Education No education11 (28 %)2 (10 %)16 (80 %) Less than 8 years9 (23 %)9 (45 %)3 (15 %) 8 to 11 years15 (38 %)9 (45 %)1 (5 %) 12 or more years of education4 (10 %)00Occupation Not working outside the home35 (88 %)1 (5 %)n/a Unskilled/skilled2 (5 %)9 (45 %) Services2 (5 %)5 (25 %) Sales05 (25 %) Professional, technical, or managerial1 (3 %)Migration Status Entire lives013 (65 %)3 (15 %)  \< 1 year3 (8 %)00 1 year8 (20 %)2 (10 %)0 2--5 years16 (40 %)01 (5 %) 5 to 10 years7 (18 %)1 (5 %)2 (10 %) More than 10 years6 (15 %)4 (20 %)14 (70 %)Place of residence before migration City32 (80 %)7 (100 %)15 (88 %) Town2 (5 %)00 Countryside6 (15 %)02 (12 %)Number of living children, mean (range)1.9 (1--7)n/an/a Place of most recent delivery, N (%)n/an/a Government Hospital19 (48 %) Private Hospital10 (25 %) Lower level Facility1 (2 %) Home10 (25 %)Mean cost spent on delivery7,721 (0--12500)n/an/aMean cost spent on transportation to delivery334 (0--200)n/an/aWidowed (Mother-in-laws only)n/an/a6 (30 %) Husbands were, on average, 30 years old. The majority of husbands were Hindu (*N* = 19, 95 %). In terms of educational attainment, 10 % had no education (*N* = 2), 45 % had 1--7 years (*N* = 9), 45 % had 8 to 11 years (*N* = 9) and none had 12 or more years of education. Almost half (45 %) worked as unskilled or skilled laborers (*N* = 9), 25 % worked in services (*N* = 5), 25 % worked in sales (*N* = 5) and 5 % were not working (*N* = 1). Sixty-five percent of the husbands had been living in Lucknow their entire lives (*N* = 13), 10 % for 1 year (*N* = 2), 5 % for 5-10 years (*N* = 1) and 20 % for more than 10 years (*N* = 4). Mothers-in-law were, on average, 54 years old. All but two (10 %) were Hindu while the others were Muslim. The majority (*N* = 16, 80 %) had no schooling, 15 % had 1--7 years (*N* = 3) and 1 (5 %) had 8--11 years. The majority (*N* = 14, 70 %) had been living in Lucknow for more than 10 years, 15 % had been living there their whole lives (*N* = 3), 1 (5 %) for 2--5 years and 10 % for 5--9 years (*N* = 2). Most of the mothers-in-law who had migrated came from a city (*N* = 15, 75 %) and 12 % from the countryside (*N* = 2). ### Decision-making about delivery location {#Sec8} Throughout the course of pregnancy, women and their families made a number of decisions regarding where to seek health services for antenatal and delivery care. In making decisions about where to deliver, respondents made choices about delivering at home or in a facility, and for those respondents who delivered at a facility, made choices about delivering at a public or private facility. These decisions were influenced by multiple factors, including prior experience of the woman or her family in receiving maternal health care, perceived accessibility of the facility -- based on both cost and proximity---and perceived quality of the facility. Intra-household dynamics and relationships between the woman, her husband and mother-in-law also influenced how these decisions were made. Household dynamics: joint decision-making common {#Sec9} ------------------------------------------------ Few of the women respondents made decisions about where to seek maternal health services on their own; rather these decisions were most commonly made together by the woman with her husband, mother-in-law, or both. These dynamics varied across households, with some families reporting that the mother-in-law was the primary decision-maker regarding maternal health care seeking, while in a smaller subset of families the woman and husband made these decisions independently of the mother-in-law. Among women who delivered at home, the majority reported that their mother-in-law was the primary decision-maker about where to deliver, while women who delivered in facilities were more likely to report that their husbands or they themselves were the primary decision-maker about where to deliver. In most cases, the previous experiences of other family members, including sisters, sisters-in-law, mothers, and mothers-in-laws, at different health facilities were important factors in care-seeking decisions. For example, as one mother-in-law explained regarding her decision to send her daughter-in-law to a particular hospital:"*My sister-in-law's children were also born there. My daughter's children as well as my elder sister-in-law's children were also born at \[that\] Hospital. My eldest grandson was also born there. (Mother-in-law, age 46)*" Similarly, a woman also explained her decision to attend a particular hospital based on her sister's prior positive experience receiving maternal care there."*My sister also went there for her delivery. From her pregnancy to delivery, she went there for all types of treatment. So on her suggestion, I went there. (Woman, age 34, 1 child)*" Choice of place of delivery was influenced by other people's experiences as well as their own perceptions of convenience, cost and quality. Home or facility delivery: complications, costs, and perceived quality {#Sec10} ---------------------------------------------------------------------- The majority of respondents (75 %) had their most recent delivery at a hospital or facility. The most common reason mentioned for selecting a facility for delivery, rather than home, was that facilities were considered a safer location for care in case of an emergency. As one respondent explained, her mother-in-law decided that she should go to a facility for the following reasons:"*Yes, she \[mother-in-law\] had also said that it is better to be in hospital \--what if you start feeling unwell suddenly, or something else, require some machine, something, then all these facilities are there, at home you wouldn't find all this (Woman, age 27, 2 children).*" However, despite the perception that facilities offered a safer location for delivery, a significant number of respondents (25 %) delivered at home. Among families who chose to have a home delivery, women, husbands, and mothers-in-law discussed two main reasons for this decision. The first, and most common, reason for choosing home delivery were the financial barriers to facility delivery, including both the cost of delivery services as well as the indirect costs associated with seeking treatment outside of the home such as for transportation."*If \[delivery\] is done somewhere else like hospital, it costs a lot of money. It can be done at home for less money. And it is not possible to give proper care at the hospital, which can be done at home... It was right to do it at home, because we do not have so much money to take her to the hospital. They require money everywhere. Who's going to give? It costs over twenty thousand rupees, how can we do it, tell us? That is why we do it at home. If it costs twenty thousand, from where do we get that kind of money? (Mother-in-law, age 60)There was a rumor in the village that I might have to pay after delivery---they \[facility\] might charge something. I am not able to pay; we don't have that much money. So, I didn't go there ever again....I am afraid that's why I didn't visit hospital; it's good to bear pain and expenses in home; at least I'll be happy at home. (Woman, age 35, 5 children)*" In addition to concerns about the cost of delivery services in facilities, many families reported that the indirect costs associated with seeking care in a facility were prohibitive to having a delivery outside of the home. The main concerns were regarding the difficulties and cost of arranging transportation to the facilities, and in arranging childcare for other children at home while the mother was at the facility."*We are poor and I have small children. If I stayed at home I would be able to take care of children and other thing, but if I stayed at hospital, how would my mother-in-law manage all these things? (Woman, age 35, 5 children)*" The second main reason respondents reported choosing to deliver at home was a concern about the quality of services provided at facilities. For many women delivering at home, the main quality concerns were about the way they would be treated at the facility, specifically fear of being disrespected, ignored, or treated poorly. As mentioned above, the prior experiences of family members and neighbors played a significant role in shaping these perceptions and influencing these decisions."*Someone in the neighborhood told her \[my mother-in-law\] that if we cry and howl too much in the hospitals, they start hitting us. She got scared that 'if my daughter-in-law gets troubled and if it is painful she will cry, and at that time if any nurse hits her.'...So she said it will be good if I deliver at home. (Woman, age 27, 2 children)*" Many families who selected home delivery due to their concerns about how they would be treated at a facility indicated that these concerns were related to issues of class and social status."*Rich people always get preference over poor. Poor people always stand in line while rich people get the direct entry. Poor people always suffer. That's why many poor people hesitate to go to a facility...(Woman, age 35, 5 children)*" For families preferring home delivery, it was this association between poor treatment and their social status that led some families to avoid seeking care at hospitals, in combination with the concerns about the cost of delivery and associated costs of seeking treatment outside the home. Public or private facility: perceived higher quality in private sector {#Sec11} ---------------------------------------------------------------------- Similar to decisions about home or facility delivery, the perceived quality and cost of services were main determinants in selecting the type of provider---either public or private. In general, women, mothers-in-law, and husbands perceived private facilities to offer higher quality services than public facilities. This was based on the perception that public facilities had longer wait times, poor interpersonal treatment, and lack of essential infrastructure."*Private is best\....In government there are no services. We have to stand in the queue for two hours, then your turn comes\.....they do a simple examination and then inform you to come again day after tomorrow.... If you go to private, only money will be spent. Facilities will be availed quickly. (Husband, age 28, 2 children)*" One husband described trying to go to a public facility when his wife was in labor, but after waiting for hours with no attention, taking her to a private facility, where she got attention quickly."*When we went there \[public hospital\], they admitted my wife inside. We sat there for almost 2-3 hours, my wife was suffering...I said it is no point tell me a place where I can take her..no one paid any attention to us, then I picked up my wife and got her to \[private hospital\]..then at \[private hospital\]..the doctor said for operation. (Husband, age 23, 1 child)*" However, private facilities were generally considered more expensive than public facilities. For this reason, even among respondents who reported that private facilities offered higher quality care, many chose to deliver in a public facility because of the high cost of private care. For some, delivering in a public facility enabled them to save money that they could use to attend a private facility if there was an emergency that required higher-level care."*Yes, they said "come on daughter let's go there in government hospital, if we go to private, it will be more expensive." And there was not much problem. If problem is there, then to save oneself a person can go to private. We go to private if there is any problem. (Woman, age 26, 2 children)*" When families could afford the cost of private care, or in cases when a woman was thought to require emergency or higher-level treatment, families made the decision to seek care in private facilities. However, in general, the overwhelming sentiment that private facilities offered higher quality services was not matched by a strong preference to seek care in these facilities given the cost barriers that were not present at public facilities. ### Experiences in facilities {#Sec12} In addition to the challenges women and their families experienced in accessing care, those who delivered in a facility also expressed a number of concerns about the quality of care they received. These concerns, similarly to the factors that influenced women's decisions about where to seek maternal health care, centered on quality and cost. The primary concern about quality was the poor treatment many women and family members reported; while the concerns about cost centered on the common practice of facilities requiring bribes or informal payments to secure treatment. Disrespectful care common {#Sec13} ------------------------- Women at both public and private facilities expressed dissatisfaction with the quality of interpersonal care at the facilities, in particular being treated disrespectfully by nurses and other health providers. Many women reported being yelled and shouted at; for example, one woman said there was "a lot of shouting and beating" (Woman, age 28, 1 child). The family members also reported disrespectful care from providers. Doctors told this same woman's husband that he and his family could not be near the woman while she was in labor."*The person who does delivery was speaking very rudely...they asked us to "get out from here, why are you gathering here, move, get out." We were not allowed to stand outside even. (Husband, age 40, 1 child)*" These concerns about disrespectful treatment from providers were reported equally across both public and private facilities. Although some women did report that they felt they were treated better in private facilities, many respondents indicated that there was no difference between type of facility and that disrespectful care was common across all facilities."*Wherever it is, in government hospital, all government hospitals and nowadays in private hospitals also, they say bad words\....Doctor, no its nurse, she says inappropriate words and bad words. (Woman, age 30, 3 children)*" However, as indicated above in the discussion about home delivery, there was a broad feeling that disrespectful treatment at facilities could be largely attributed to issues of social and economic status. As one husband explained:"*Only if you have resources, or influence in the government, only then will you be listened to \[at the facility\]. Wherever you go\-- private or government hospital\-- you will be asked who are you, what do you do. If you say you work as a laborer, then no one will give you any attention. (Husband, age 24, 1 child)*" It is important to note that not all respondents had negative experiences during their facility stays, with some reporting their treatment and experience to be positive. Although in some cases, this was acknowledged with a level of surprise, or seen as the exception to standard treatment practices."*Their services were very good. I mean whether it was staff related, doctor related. From conversation to everything was very good. My experience was very good. (Woman, age 34, 1 child)*" Financial barriers: bribes and tokens paid to facility staff {#Sec14} ------------------------------------------------------------ Respondents reported a number of financial barriers faced while seeking maternal health services. The primary challenge reported was the common practice of providers and other people at the facility seeking bribes or payments above the formal fees prior to providing essential care and services. Many respondents perceived the poor treatment by providers as linked to this system of informal payments; according to one woman, *"You have to give them \[facility providers\] bribes in order for them to attend to you" (Woman, age 30, 3 children).* Several respondents indicated that bribes were common throughout the hospital experience, and others reported that they were denied care if they did not provide financial "tokens" to facility staff."*The nurses demanded money\.... I had a baby girl and after the delivery they kept her on the scale and demanded Rs.300-400. If I will not pay money they will not return my baby to me. We are poor and how will we have this much money to give? (Woman, age 26, 3 children)The gate peon was asking for money, and when baby boy was born I was sitting outside the operation theater when I went to get him they were asking that we give money so we give Rs.600. We took the baby boy and went upstairs for cleaning. There, everyone was asking for money -- nurse, maid, everyone. Then they shifted her to another room, there were two people on the same bed...they asked for money again to shift her to an empty bed, then again for taking her to the room. So we gave money at 6-7 places...everywhere they asked for money. (Mother-in-law, age 50)*" These payments, and the informal format in which they were requested without families being able to anticipate the cost of services, was a source of both financial and emotional stress for many respondents. Financial barriers: payment challenges and difficulties with JSY payment {#Sec15} ------------------------------------------------------------------------ Respondents highlighted two additional sources of financial stress associated with a facility delivery. The first was the need to borrow money to cover the cost of services. Many respondents, both women and husbands, reported selling gold from their dowry or other household goods, or borrowing money from family or community members, in order to pay for their hospital stays."*How did we bear the expense? As my husband is the only earning member and there are 8 people dependent on him, so we had to sell some of our household goods. (Woman, age 35, 5 children)That arrangement- I had taken advance from my employer and also borrowed some money from my relatives...My people are really nice, my relatives, my boss, they did not show any reluctance to give the money..\[they\] gave money with ease. (Husband, age 23, 1 child)*" While for some families, borrowing money was a feasible option, for others the need to borrow money was seen as a burden. This was particularly true for those families who borrowed with the expectation of repayment, when this repayment would mean a significant financial burden for the family."*We had to borrow money...We did not feel nice. We had thought that in government hospital everything is free, so we won't have to pay anything. But they referred us to private hospital, so we had to borrow money, which made us sad. (Woman, age 23, 1 child)*" The second challenge women reported was in receiving the JSY conditional cash transfers. All women participants in this study were eligible to receive JSY payments; however, only 12 of the respondents reported receiving JSY, and of these, 4 of them reported not having cashed it yet. The primary reason respondents gave for their non-participation was the requirement to open a bank account in order to enroll in the JSY program. The cost of opening a bank account was reported to be equivalent to the amount of the JSY payment, reducing respondents' motivation and interest to participate."*I will be able get it \[bank account\] opened with one thousand rupees. And we will get Rs. 1000 Rupees. So what is the benefit of getting it opened? (Woman, age 20, 1 child)*" Among women who did receive JSY payments, many reported challenges in receiving the funds, such as having to wait many hours to receive the check or having to pay large bribes to hospital staff in order to receive their payment."*The service that we got, the cheque, we gave it all away. Everyone was asking for money, the nurse or the helper, everyone was asking for money. We gave them more than what we got. (Woman, age 20, 1 child)*" Many respondents also reported challenges when trying to cash the JSY payments, specifically due to not having the required identification for processing the payments."*I received a check but it is still not cleared. For clearance they are asking for \[identification card\]. They had given 3 months time, but to date I have not received the pehchan card, and the check is still lying with me uncleared. (Woman, age 26, 3 children)We had to get this scheme but we did not have any proof. They said if you have any residence proof or ID proof or anything, we did not have anything so we would have got the facility if we would have tried we would have got the money but we did not avail it, so we would say it is our mistake that we did not go again and try to get that money...When my first child was born I was given a voucher. We have to take it to the bank, I went there twice, \[my husband\] went twice to the bank and we spent 250 -- 300 rupees as well but we did not get the money. Now I still have the cheque with me. (Woman, age 22, 2 children)*" The association respondents made between these payments and the quality of care they received, as well as their social and economic status, was an additional source of concern and stress. Discussion {#Sec16} ========== This study is important because it triangulates voices from recent mothers, their husbands, and mothers-in-laws and sheds light on potential interventions for improving maternal health outcomes in slums in India. The purpose of this study was to identify women's motivations for delivering in a facility compared to their homes and to identify their experiences with care. Findings from this study suggest that the negative experiences women and their social networks face at the time of delivery were mostly related to financial challenges and quality of care. In particular, disrespectful care deterred some women from delivering in facilities and influenced their decisions to deliver at home vs. the facility. Our findings support other recent qualitative work in Karnataka, India, that found that perceptions of quality related to disrespect and cleanliness, as well as the costs of delivery, were main factors determining place of delivery and keeping women out of facilities \[[@CR9]\]. A recent systematic review explored multiple domains of mistreatment that women experience during childbirth and found that mistreatment is prevalent in many countries across the globe \[[@CR13]\]. Women in our study reported being shouted at, verbally abused, and being discriminated against because of their status. Our findings support this body of literature on mistreatment and disrespectful care, and provide additional evidence that these experiences, both those personally experienced and the experiences of others, impact decision-making and keep women from seeking care in facilities with trained providers. This study also highlighted the large burden on families from the costs of delivery, including bribes for providers and facility staff. Past research has highlighted that high out-of-pocket costs are one of the main barriers to women delivering in a facility or with a trained provider in India \[[@CR14]\]. Although the JSY program should have been available to all of the women in our study, many did not receive the benefit, or if they did, were not able to actually cash it or ended up giving most of it away to pay for additional bribes or costs of the delivery. Other studies have highlighted how the JSY program is not able to cover all costs associated with delivery, and in fact might have led to more out-of-pocket expenditures \[[@CR15]\]. Fears of costs were one of the primary deterrents from going to a facility, and also impacted decisions about private versus public facilities. Families resorted to taking loans or selling goods, leading to long-term impacts from the delivery on the family. The JSY program is structured to be an incentive to delivering in a facility, eliminating cost barriers associated with delivery services. It is clear that the JSY program must find ways to ensure that women and families actually are able to receive the benefit. Additionally, finding ways of eliminating the extra fees that families are being asked to pay is critical, including holding facilities accountable and punishing facilities that require extra payment. Our findings also highlight the importance of other family members of the decision-making process about where a woman delivers. Most research exploring choice of delivery focus on women's individual characteristics, or sometimes those of her husband; however, it is clear from our findings that these decisions are most often made at a family or couple level \[[@CR16]\]. Past research has shown that women's autonomy is an important factor on her health care utilization, and that women who live in nuclear households were more likely to deliver in a facility than women living in co-resident households (with in-laws) \[[@CR17], [@CR18]\]. Our qualitative data help explain these findings, since we found that women were more likely to deliver in a facility in households where women, or women and their husbands, were the main decision-makers. There are a number of limitations to this study. First, as is common in qualitative studies, the sample size is small and therefore not generalizable to other areas outside of this study site. It is important to note that there is significant heterogeneity across slums in India including differences in poverty levels, size, and population, race, and migration mix. This study describes the specific experiences of individuals living in particular slums in Lucknow, Uttar Pradesh. The sampling frame for the study attempted to reach both women who went to anganwadi centers as well as women in communities in general, with the goal of highlighting different types of women and families. Second, while we attempted to sample a mix of women who delivered in a facility and at home, other experiences, including levels of complication may influence results and were not captured systematically in this data. For example, women who experienced complications at the time of delivery may report more negative experiences compared to women without complications. Finally, eligibility for the study includes women who had delivered in the past year. This may lead to recall bias, as 1 year may be a long time for women, their husbands and family members to recall their decision regarding delivery and experiences of care. Despite these limitations, this study points to a number of programmatic and policy implications for urban slum populations. From a policy perspective, while women were aware of the JSY program, few women accessed this program. Therefore, targeted education in these communities regarding different public programs is warranted, including how to sign up for the program, how to receive incentives, and who is eligible. Follow up regarding how to access funds is particularly important. Future programs should include husbands and mothers-in-law in the education process. Many times, decision-making on where to deliver occurred in the context of families and social networks; therefore, programs should engage these broader contexts in health education and promotion. This may include partner communication counseling or educating mothers-in-law on the benefits of delivering in a facility. Finally, women in the study suggested that disrespectful care occurred in all types of facilities in India, both public and private facilities. This represents a significant challenge in India. Because experiences of disrespectful care were common in this population, future programs should focus on patient-provider interactions among urban, slum women in particular who may face discrimination due to their lower status. The World Health Organization (WHO) has outlined strategies to address disrespectful care which includes support through a companion of choice, access to foods and fluids, ensuring confidentiality and informed choice, and assuring high quality information for women to make informed decisions about care \[[@CR19]\]. Other strategies to work with low-status women may include working with facilities to improve women-centered quality of care, training providers on culturally competent care, and involving women and their families in care processes. Conclusions {#Sec17} =========== Financial barriers and disrespectful care were the main challenges experienced in urban, slum populations in regards to maternal health services. A number of programmatic and policy recommendations are highlighted from this study. Future endeavors should include a greater focus on health education and the JSY program, including educating women on how to access the program, who is eligible, and how to obtain funds. Families need to be educated on their rights and expectations in facilities. Future programs should consider the role of husbands and mother-in-law in reproductive decision-making as well as improving respectful care in facilities Abbreviations {#Sec18} ============= ANC, antenatal care; JSY, Janani Suraksha Yojana; UCSF, University of California, San Francisco We would like to thank Emily Treleaven for contributing to the analysis of data. Funding {#FPar1} ======= This study was funded by the Bill and Melinda Gates Foundation. Availability of data and materials {#FPar2} ================================== Data available upon request. Authors' contributions {#FPar3} ====================== MS conceived of the study, designed the study, analyzed data, and led the writing of the manuscript. NB contributed to the conception and design of the study and contributed to the writing of the manuscript. SB participated in the collection of data and contributed to the writing of the manuscript. NDS analyzed results and contributed to the writing of the manuscript. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== The study and all accompanying study materials were reviewed and approved by the Institutional Review Boards at the University of California, San Francisco (UCSF) and the Centers for Operations Research and Training (CORT) in Gujarat, India. Verbal informed consent was obtained from all study participants.

1. Introduction {#sec1} =============== Since 1992, when nitric oxide (NO) was nominated "molecule of the year" \[[@B1], [@B2]\], it continues to attract the interest of the scientific community. NO is the smallest gasotransmitter, recognized as an ubiquitous intercellular messenger; it is produced by three isoforms of NO synthases (NOS): endothelial NOS (eNOS) \[[@B3]\], neuronal NOS (nNOS) \[[@B4]\], and inducible NOS (iNOS) \[[@B5]\] and mitochondrial NOS (mtNOS) \[[@B6]\]. All NOS isozymes utilize L-arginine and oxygen and the reduced form of nicotinamide-adenine-dinucleotide phosphate (NADPH) as substrates and 6*R*-5,6,7,8-tetrahydro-L-biopterin (BH~4~) as essential cofactor to generate NO and L-citrulline \[[@B7], [@B8]\]. Then, the main downstream signaling pathway carried out by the NO is the activation of soluble guanylyl cyclase (sGC), which in turn generates cyclic guanosine monophosphate (cGMP) \[[@B9]\] ([Figure 1](#fig1){ref-type="fig"}). In the vascular system NO modulates blood flow \[[@B10]\], vascular tone \[[@B11]\], and platelet aggregation \[[@B12]\] exerting antihypertensive, antithrombotic, and atherosclerotic effects. It is also involved in the stimulation of the endothelial progenitor cells (EPCs) and proliferation of the smooth muscle cells (SMCs) \[[@B13]\]. Therefore, an impairment in the NO signaling is associated with the onset and perpetuation of the main clinical condition associated to cardiovascular diseases (CVDs) including endothelial dysfunction \[[@B14]\]. Given this premise, it is reasonable to consider NO as a therapeutic target for CVDs. Indeed, several approaches have been proposed to modulate NO pathways while preserving its physiological role \[[@B15]\]. From one side, the strategy consists in enhancing NO bioavailability, principally acting on NOS cofactors or avoiding NO breakdown; from the other side, different drugs act on the NO downstream signaling targets \[[@B16]\]. Data from epidemiological studies have suggested the existence of a relationship between physical exercise and/or specific diets with a reduction of CVDs prevalence and incidence \[[@B17]--[@B20]\]. In addition, clinical trials and experiments in animal models have indicated NO as the main mediator of the beneficial effects of certain natural derived compounds, such as the polyphenols \[[@B17], [@B21]\]. In the present review, we discuss the biochemistry and pathophysiology of signaling pathways of NO focusing our attention on the experimental data showing that some natural derived compounds could be effective in the prevention and possibly treatment of CVDs. 2. Molecular Pathways of NO {#sec2} =========================== Among the isoforms of NOS, eNOS represents the main source for the NO production in the vasculature. It is predominantly expressed in the endothelium but it has been also detected in kidney, human placenta, cardiomyocytes, platelets, and some neurons \[[@B22]\]. Several endogenous agonists, such as acetylcholine, bradykinin, and vascular endothelial growth factor (VEGF), as well as the shear stress induced by the blood flow, have been reported to activate eNOS \[[@B23]\]. Several studies have demonstrated that the phosphoinositide 3-kinase- (PI3K-) AKT pathway is mainly responsible for eNOS phosphorylation at Ser1177 especially in response to shear stress and VEGF \[[@B24]--[@B26]\]. Moreover, caveolin-1, the main component of the caveolae plasma membranes, has been reported as a negative regulator of eNOS \[[@B27], [@B28]\]. Another mechanism involved in the production of eNOS-derived NO is the activation of the *β*-adrenoreceptors \[[@B29]\] in response to the increase of catecholamines that are expressed at high levels in condition of oxidative stress associated with endothelial dysfunction \[[@B30], [@B31]\]. Neuronal NOS (nNOS) is expressed in specific neurons of the central nervous systems (CNS), as well as in the peripheral nervous systems (PNS) and in perivascular nerve fibers \[[@B32]\]. As in the case of eNOS, nNOS is responsible for the constitutively production of NO \[[@B33]\]. The inducible NOS (iNOS) is normally inactive in the vasculature \[[@B34]\], but its expression and activity can be induced in many cell types under oxidative and inflammatory stimuli; as a matter of fact several cytokines have been detected in the endothelium, in the media, and in the adventitia of blood vessels, as well as in neuronal cells and hepatocytes. Moreover, it is well known that NO produced by iNOS participates to the response of the immune system in killing bacteria and other exogenous compounds \[[@B35]\]. Several studies show the presence of a new isoform of eNOS enzyme in mitochondria (mtNOS) \[[@B36], [@B37]\]. This fourth isoform, the mtNOS, is responsible of the NO production in the mitochondria. It has been demonstrated that the NO-synthesizing capacity of mtNOS is higher than that derived from the combined activity of the all other NOS isoforms \[[@B38]\]. Moreover, recent findings suggest that an excessive stimulation of mtNOS leads to mitochondrial dysfunctions, which contribute to metabolic syndromes \[[@B39]\]. All NOS proteins are homodimers that transfer electrons from NADPH to the haem in the oxygenase domain where there are also binding sites for BH~4~, oxygen, and L-arginine; at the haem site, the electrons are used to reduced O~2~ and to oxidize L-arginine to L-citrulline and NO. Importantly, when oxidative stress increases, eNOS can lose its physiological properties in a process termed "eNOS uncoupling" \[[@B22], [@B40]\] ([Figure 2](#fig2){ref-type="fig"}). In such condition, NO reacts with superoxide O~2~ ^−^, leading to formation of peroxynitrite (ONOO^−^), potent inducers of cell death, and eNOS produces reactive oxygen species (ROS), mainly O~2~ ^−^, rather than NO \[[@B41]\]. Therefore, eNOS uncoupling not only leads to decreasing NO bioavailability, but contributes to enhancing the preexisting oxidative stress \[[@B42]\]. Different mechanisms have been suggested to explain eNOS uncoupling; among these, the oxidation of BH~4~ to the inactive form BH~3~ ^−^ by O~2~ ^−^ and ONOO^−^ together with depletion of L-arginine plays a prominent role \[[@B4]\]. In particular, the decrease of L-arginine is caused by the upregulation of arginase isoforms (Arg I and Arg II) expression and activity. As we will discuss in the next sections, oxidative stress associated to eNOS uncoupling and the changing of the eNOS phosphorylation status (summarized in [Figure 3](#fig3){ref-type="fig"}) are characteristics of clinical conditions commonly associated to CVDs, such as diabetes mellitus, hypertension, atherosclerosis, and cerebral ischemia \[[@B22], [@B43]\]. 2.1. Posttranslational Modifications of NOS {#sec2.1} ------------------------------------------- NOS enzymes are regulated by multiple interdependent mechanisms and signaling pathways, which can be calcium-dependent and/or calcium-independent. In particular, it has been demonstrated that the activity of eNOS is regulated by the increase of the cytosolic Ca^2+^ in endothelial cells, which leads to the activation of calmodulin that in turn binds eNOS, thus facilitating its function \[[@B23], [@B44]\]. Besides the increase of intracellular calcium, eNOS activity depends also on its phosphorylation status. In particular, It has been suggested that the phosphorylation of NOS isoforms at Tyr81 and Tyr657 represents a mechanism necessary to modulate the NO production above all during shear stress \[[@B45], [@B46]\]. Indeed, the phosphorylation is the major and most studied posttranslational modifications influencing the eNOS activity. Noteworthy, while the phosphorylation of serine at positions 617, 635, and 1179/1177 results in the activation of the eNOS, the same change at Ser116 and Thr497 reduces its function. Also acetylation of the eNOS influences its activity and, in general, acetylation/deacetylation balance represents a crucial homeostatic mechanism mediating the response to metabolic changes in the cell \[[@B47]\]. Other important posttranslational changes are acylation, nitrosylation, glycosylation, and glutathionylation. All of them are necessary and often interconnected in controlling the subcellular localization and/or activity of the eNOS and thus the NO bioavailability in response to a variety of physiologic and pathophysiologic signals \[[@B48]\]. 3. Physiopathological Role of NO in the Vascular System {#sec3} ======================================================= The role of NO in the maintenance of vascular homeostasis is well defined and it depends on both eNOS distribution pattern and NO production rate. Perturbation of NO signaling pathways represents one of the major determinants of endothelial dysfunction, which is characterized by the reduction of the NO bioavailability and oxidative stress increase with the resulting impairment of the endothelium-dependent vasodilation \[[@B49], [@B50]\]. The NO synthesized by eNOS diffuses from endothelial cells into the underlying SMCs in which it stimulates sGC, thus generating cGMP, which in turn activates downstream protein kinases. Protein kinases predominantly act on myosin light chain phosphatase, the enzyme that dephosphorylates myosin light chains and leads to smooth muscle relaxation and vasodilatation. Moreover, NO may diffuse also in the blood flow where it inhibits several processes normally impaired during thrombotic and atherosclerotic events including platelet aggregation and leukocyte adhesion and migration into vascular wall \[[@B51]\]. Interestingly, over the well-known involvement of NO in the main cardio- and cerebrovascular diseases, other minor vascular forms of vascular diseases had been associated with impairment of the NO signaling. In this regard, it has been widely recognized that NO plays a key role in the physiology of penile erection eliciting its effect on guanylate cyclase leading to the production of cGMP. About this mechanism, the impairment of NO activity is similar to that observed in other forms of vascular diseases or in patients with cardiovascular risk factors (e.g., dyslipidemia, diabetes, and hypertension) \[[@B52]\]. Another form of vascular alteration in which changes in NO production and bioavailability have been reported is represented by varicose vein disease \[[@B53]\]. Furthermore, recent studies have found a link between endothelial dysfunction and NO alterations in venous valve dysfunction \[[@B53]\]. In particular, processes associated with varicose vein disease are increased destruction of collagen and matrix proteins triggered by endothelial dysfunction, which in turn is characterized by loss of NO bioavailability and increase of inflammation and ROS build-up \[[@B54]\]. Based on these findings, it is imperative to identify a new therapeutic strategy aiming at stimulating NO production and preventing the reduction of its bioavailability. 3.1. NO in Ischemia and Heart Failure {#sec3.1} ------------------------------------- Many studies in animal models have documented the existence of a link between NO pathway impairment and CVDs. Kuhlencordt et al. showed that atherosclerosis, aortic aneurysm formation, and ischemic heart diseases can be accelerated as result of a chronic deficiency of eNOS \[[@B55]\]. The authors compared the atherosclerotic lesions occurring in two different knockout (KO) animal models, apolipoprotein E (apoE)/eNOS-double knockout (DKO) and apoE-KO, demonstrating that a genetic deficiency of eNOS significantly increased atherosclerosis in the apoE-KO mouse model. Of note, the location of the lesions, occurring mainly in the areas with disturbed flow, was similar in both KO models; therefore, the authors\' conclusion was that the absence of eNOS did not determine the site of lesion formation in the aorta but appeared to accelerate its development. In addition, the ApoE/eNOS-DKO animals showed a more marked increase in blood pressure, comparable to that of eNOS-KO mice, indicating that eNOS deficiency could reflect different degrees of endothelial dysfunction. These findings are very important because they suggested eNOS deficiency/endothelial dysfunction as a possible molecular mechanism linking hypertension to atherosclerosis \[[@B55]\]. Actually, Huang et al. have already demonstrated that in mice lacking the gene encoding eNOS the acetylcholine-induced relaxation was absent and the eNOS mutant mice had elevated blood pressure and developed hypertension \[[@B56]\]. Anti-ischemic actions of NO were also demonstrated by using of a transgenic (TG) mice model with cardiac specific overexpression of iNOS. After ischemia induced by coronary occlusion followed by 24 hours of reperfusion, the TG mice had a smaller infarct size compared to wild type. In addition, iNOS overexpression was able to attenuate the ROSs generation associated with reperfusion injury, in fact, the quantity of the ROSs trapped from reperfused hearts was lower in iNOS-TG than in wild type mice \[[@B57]\]. In another study performed in a model of eNOS-TG mice, it was demonstrated that a cardiomyocyte-specific overexpression of eNOS improved left ventricular performance and reduced compensatory hypertrophy after myocardial infarction (MI). Importantly, eNOS cardiac overexpression attenuated also a post-MI remodeling by reducing fibrosis in the noninfarcted area of the myocardium \[[@B58]\]. The beneficial role of the eNOS-derived NO has been demonstrated also in congestive heart failure (HF) in the study by Jones et al. in which the authors, by using a mouse model of infarct-induced HF, showed that eNOS overexpression enhanced animal survival, inhibited pulmonary edema, and improved cardiac function but did not attenuate the cardiac hypertrophy or improve cardiac contractility \[[@B59]\]. Moreover, it has been reported that the mitochondrial production of NO by mtNOS is reduced during ischemia because there is a lack of the O~2~, necessary to generate the NO \[[@B36]\]. These findings demonstrated that strategies aimed at increasing NO bioavailability in the heart might be useful to counteract the structural and functional damage induced by myocardial ischemia. 3.2. NO in Diabetes and Atherosclerosis {#sec3.2} --------------------------------------- NO production is reduced in diabetes mellitus and atherosclerosis, well-known risk factors for CVDs. In obese mouse model, eNOS activity was reported to be reduced by an enhanced phosphorylation at threonine 495 via PKC \[[@B60]\]. Similarly, Kashiwagi et al. showed that the lack of phosphorylation at serine 1176 residue was correlated with the development of obesity and insulin resistance in a mouse model \[[@B61]\]. Uncoupling eNOS also concurs to develop diabetes mellitus and, as mentioned above, both oxidation of BH~4~ and depletion of L-arginine are the cause of such phenomenon. In addition, BH~4~ was shown to be oxidized in diabetic mouse models, by a mechanism involving the activation of NADPH oxidases through PKC \[[@B62]\]. Similarly, in diabetic hypertensive rats, Alp et al. showed low levels of BH~4~ and decreasing in NO production \[[@B63]\]. Heitzer and colleagues demonstrated that a supplementation of BH~4~ improved endothelium-dependent vasodilation in patients with type II diabetes but not in control subjects. Of note, such beneficial effect was completely blocked by N(G)-monomethyl-L-arginine, a well-known inhibitor of NOS, suggesting that it was dependent on the NO production increase \[[@B64]\]. Also L-arginine deficiency has been reported in diabetic rats with a concomitant increase of the expression and activity of arginases, particularly, arginase, which has been recognized as the isoform responsible for eNOS uncoupling in diabetes \[[@B65]\]. In this regard, diabetic mice deficient of arginase I exhibited less endothelial dysfunction compared to wild type mice \[[@B66]\]. Notably, in the same way, in coronary arterioles of diabetic patients, arginase I was shown to contribute to the reduction of vasodilatation \[[@B67]\], and in plasma of patients with type II diabetes, arginase activity was reported to be elevated \[[@B68]\]. Interesting data in both humans and animal model have remarked the involvement of NO metabolism in the atherosclerosis. For example, depletion of BH~4~ has been demonstrated in hypercholesterolemic patients \[[@B69]\] and high level of superoxide anions produced by uncoupled eNOS and increased formation of aortic atherosclerotic plaque with the concomitant deficiency of BH~4~ were found in ApoE-KO mice where there were also observed an increased arginase II expression and activity \[[@B70]\]. Similarly, in human endothelial cells exposed to thrombin, Yang et al. found an enhancement of the arginase enzymatic activity \[[@B71]\]. In ApoE-KO mice, Ming et al. demonstrated that the small G protein RhoA and its effector ROCK play a role in the regulation of arginases activity involved in atherosclerotic process \[[@B72]\]. Moreover, posttranslational modifications of the eNOS have been shown to play a crucial role during atherosclerosis and diabetes \[[@B73]\]. Importantly, recent investigations have highlighted that phosphorylation and acetylation of the eNOS might concur to mediate the beneficial effects of some drugs. In this regard, Romero et al. have investigated the effects of BM-573, a compound that combines thromboxane synthase inhibition and thromboxane receptor antagonism, on endothelial dependent relaxation during early stage of atherosclerosis in apoE-KO mouse model. The authors demonstrated that BM-573 was able to ameliorate endothelial dysfunction by reducing oxidative stress and improving the NO bioavailability by increasing the eNOS phosphorylation \[[@B74]\]. Moreover, it has been demonstrated that lysine acetylation of the eNOS mainly contributes to the well-known atherothrombotic effects of low-dose acetylsalicylic acid \[[@B75]\]. The eNOS posttranslational modifications are necessary also in mediating the antidiabetic effects of several therapeutic interventions. For example, a diet supplementation with l-arginine and sepiapterin along with salsalate has been proved to increase the eNOS phosphorylation and improved vasorelaxation of thoracic/abdominal aorta in type-1 diabetic mice \[[@B76]\]. Furthermore, Ding et al. showed that cardiac overexpression of SIRT1, a NAD+-dependent deacetylases, reduced diabetes-exacerbated myocardial ischemia reperfusion injury and oxidative stress in diabetic rats via eNOS activation and that such effect was mediated by increase of the eNOS phosphorylation and reduction of the eNOS acetylation \[[@B77]\]. It has been also showed that the eNOS phosphorylation might be also important in mediating the beneficial effects of metformin and thiazolidinediones into microvasculature. In this regard, Ghosh et al. demonstrated that a brief 3 h exposure to metformin induced changes in eNOS signaling in mouse microvascular endothelial cells by reducing the ratio of phosphorylated (p-eNOS)/eNOS, but not the expression of total eNOS \[[@B78]\]. Xu et al. investigated the effects of ciglitazone in rat microvascular endothelial cells, finding that such antidiabetic drug was able to reverse the decrease of eNOS levels in the cells stressed with oxidized LDL thus improving the NO bioavailability \[[@B79]\]. 3.3. NO and Hypertension {#sec3.3} ------------------------ A decreased NO bioavailability is one of the mechanisms involved in the pathogenesis of hypertension. Indeed, the phosphorylation of eNOS at threonine 495 residue was shown to be enhanced in angiotensin II- (Ang II-) induced hypertensive rats \[[@B80]\]. Landmesser et al. showed an increase of BH~4~ oxidation caused by the activation of the p47phox subunit of NADPH oxidase in a model of salt-induced hypertension rat \[[@B42]\]. A similar enhanced expression of NAPDH oxidase has been also shown in spontaneously \[[@B22]\] and in angiotensin II-induced hypertensive rats \[[@B81]\]. In addition, an oral administration of BH~4~ was shown to suppress the hypertension in spontaneously hypertensive rats thanks to the reduction of ONOO^−^ and O~2~ ^−^ accumulation \[[@B82]\]. Similarly, a supplementation of BH~4~ increased acetylcholine-dependent endothelium vasodilatation in hypertensive patients to the level of normal control subjects \[[@B83]\]. Besides the oxidation of BH~4~, the depletion of L-arginine could contribute to hypertension causing eNOS/NO impairment. Indeed, in spontaneously hypertensive rats, as well as in the aorta of mineralocorticoid and salt-induced hypertensive rats, the expression/activity of arginases was found to be enhanced \[[@B84]--[@B86]\]. Moreover, angiotensin II, via stimulation of AT~1~, receptor is reported to be a molecular pathway responsible for the increased expression/activity of arginases in hypertension. In particular, in arginases knockout mice Shatanawi et al. showed that the p38 MAPK is the downstream effectors of AT~1~, leading to endothelial dysfunction \[[@B87]\]. Intravenous administration of L-arginine produces a vasodilatory effect by increasing the NO production in hypertensive individuals \[[@B88]\], as well as the arginase inhibitor*N-*(omega)-hydroxy-nor-l-arginine prevents the hypertension, lowering the blood pressure in a hypertensive rat model \[[@B89], [@B90]\]. 3.4. NO and Cerebrovascular Diseases {#sec3.4} ------------------------------------ Several experimental evidences have underlined the protective role of eNOS/NO pathways in neuronal injury after cerebral ischemia as well as in the prevention of stroke and severe subarachnoid hemorrhage (SAH) \[[@B91], [@B92]\]. In physiological conditions, eNOS-derived NO is the main molecule responsible for the control of the cerebral blood flow (CBF). In this regard, it has been shown that ischemic injury increases eNOS activity and NO availability, which in turn leads to the improvements of the CBF and to decreasing neuronal injury \[[@B93]\]. Osuka et al. \[[@B94]\], in rat cerebral models of ischemia, found increased level of phosphorylation at eNOS Ser1177 residue in microvessels, with temporary expression of VEGF. Similarly, in eNOS knockout mice, after middle cerebral artery (MCA) occlusion, Huang et al. demonstrated an enlargement of infarct size and showed that systemic administration of nitro-L-arginine prevented brain damage \[[@B43]\]. Moreover, thrombotic cerebral infarctions have been found in eNOS^+/−^ mice after three--six months of age \[[@B95]\]. Other authors underlined the importance of NO in the angiogenesis and neurogenesis occurring after cerebral stroke; for example, neovascularization after stroke was found to be impaired in eNOS deficient mice, indicating that endothelial NO mediates this effect \[[@B96]\]. 4. Main Modulators of NO Pathways {#sec4} ================================= Several therapeutic strategies have been proposed to ameliorate the NO homeostasis. Currently, the best strategy is based on the drugs administration in order to activate downstream effectors of eNOS/NO from one side and to reduce eNOS uncoupling \[[@B16]\], improving BH~4~ and L-arginine bioavailability and regulating post-translational modifications of eNOS, from the other side. Nevertheless, it is important to remark that a helpful strategy for the prevention and attenuation of CVDs is to make a good lifestyle and, in this context, physical exercise and specific diets such as diet rich in polyphenols have been suggested to improve the NO pathways. The inhibition of the renin-angiotensin-aldosterone system is widely recognized as an effective therapy in CVDs \[[@B97]\]. In animal models, angiotensin-converting enzyme inhibitors (ACE-I) and AT1 receptor blockers (ARBs) are able to reduce eNOS uncoupling, while restoring BH~4~ bioavailability \[[@B98]\], and to protect against cerebral ischemia via upregulation of the eNOS in middle cerebral artery \[[@B99]\] and cerebral infarct size via eNOS activation \[[@B100]\]. The renin-angiotensin system blockers exert also NO-dependent antithrombotic effects. In this regard, Kucharewicz et al. demonstrated that angiotensin 1--7, a component of the renin-angiotensin system, caused an increased production of NO, which contributes to reduction of thrombosis in rats \[[@B101]\]. Also the cholesterol-lowering drugs, the statins, improve endothelial functions by enhancing the NO bioavailability thanks to their antioxidant, anti-inflammatory, and antiatherosclerotic properties \[[@B102], [@B103]\]. For example, in hypercholesterolemic patients treated with fluvastatin, John et al. demonstrated an improvement of endothelial vasodilatation through increase of the NO production \[[@B104]\]. Moreover, in a rat experimental model of MI, statins were found to enhance NO bioavailability by restoring mobilization of EPCs, myocardial neovascularization, and, ultimately, increasing survival \[[@B42]\] and statins were also showed to decrease eNOS uncoupling through a reduction of vascular O~2~ ^∙−^ and BH~4~ oxidation \[[@B105]\]. Another way to ameliorate endothelial homeostasis is the activation of the *β*-adrenoreceptor subtype 3 (*β* ~3~), which leads to eNOS activation and thus to the NO generation by increasing the levels of cAMP and Ca^2+^ \[[@B30], [@B31], [@B106]\]. Nebivolol, a third-generation *β*-adrenoreceptor blocker, is a promising drug able to improve NO pathways thanks to its ability to antagonize *β* ~1~ and to activate *β* ~3~ receptors. Maffei et al. \[[@B107]\] showed that nebivolol induced endothelial NO production in both conductance and resistance rats arteries in a calcium-dependent manner. In another study, the same authors measured in mice the heart production of the NO consequent to the stimulation of *β* ~3~ receptor and iNOS increased activity, thus indicating nebivolol as therapeutic strategy for hypertension and heart failure \[[@B108]\]. Another aspect that deserves attention is the link between adrenergic pathway, NO bioavailability, and oxidative stress and, in this context, the beneficial effects of the nebivolol are attributable to its well-recognized antioxidant properties, which are considered an additional factor for increasing the NO bioavailability. For example, nebivolol and atenolol (a second-generation *β*-blockers) similarly reduced blood pressure values in hypertensive patients, but oxidative stress markers, such as LDL hydroperoxides, 8-isoprostanes, and ox-LDL were significantly improved only in patients treated with nebivolol \[[@B109], [@B110]\]. Moreover, in hypertensive patients, Okamoto et al. demonstrated that nebivolol lowered blood pressure \[[@B111]\], while in elderly patients with heart failure it was shown to reduce mortality and morbidity \[[@B112]\]. Interestingly, Falciani et al. highlighted also the role of nebivolol in inhibiting platelet aggregation by increasing L-arginine/NO, remarking also an antithrombotic effect of this *β*-blocker \[[@B113]\]. Nowadays, researchers are paying particular attention to the nonpharmacological strategies, including adoption of specific diet habits and exercise programs for the management of several chronic diseases. In this context, several experimental and epidemiological findings have underlined the role of physical exercise (PE) in decreasing the oxidative stress associated with aging and in the prevention and attenuation of CVDs-associated risk factors \[[@B114]--[@B117]\]. It was suggested that the reduction of oxidative stress triggered by PE could be associated with the improvement of the NO function \[[@B118]\]. In this regard, in patients with chronic heart failure and coronary artery disease, Laurent et al. showed that water-based exercises increased NO metabolism by improving cardiorespiratory capacity and endothelial function \[[@B119]\]. Recently, a regular exercise was demonstrated to activate eNOS and nitrite production and to reduce oxidative stress in spontaneously hypertensive rats \[[@B120]\]. PE was also suggested to have a cardioprotective effects; in ischemic rats, high-intensity interval training increased NO metabolites levels and reduced myocardial infarction \[[@B121]\]. Different molecular mechanisms, such as phosphorylation status and transcription rate of eNOS, have been proposed to explain the effects of PE on the NO production. For example, in rats subjected to acute and chronic aerobic training eNOS mRNA levels were found to be upregulated \[[@B122]\]. Other authors underlined the role of *β* ~3~ adrenoreceptor in mediating the effects of PE on the NO production; in particular, Calvert et al. demonstrated that exercise could improve the cardiac function in ischemic rats via *β* ~3~ adrenoreceptor by increasing the eNOS phosphorylation \[[@B123]\]. Another molecular mechanism is represented by the NO-dependent changes in the vascular redox state and oxidative stress even if the beneficial role of the NO in this context could be complex to elucidate. In this regard, Farah et al. suggested that certain level of eNOS uncoupling could be required for exercise-induced myocardial cardioprotection during ischemia reperfusion. In particular, in such study, it was showed that eNOS uncoupling was associated with the improved myocardial antioxidant capacity that prevented excessive NO synthesis limiting the reaction between NO and O~2~ ^∙−^ to form peroxynitrites \[[@B124]\]. 5. Crosstalk between NO and the Other Gasotransmitters {#sec5} ====================================================== Besides the NO, other two gaseous molecules, hydrogen sulfide (H~2~S) and carbon monoxide (CO), have been recognized as "gasotransmitters" \[[@B125]\]. Much like their predecessor NO, H~2~S and CO have been historically considered as highly toxic and harmful agents; afterward, many investigations have showed that they not only play various physiological roles but could be effective against a number of diseases, including CVDs \[[@B126]\]. Indeed, also CO and H~2~S mediate muscle relaxation and vasodilatation, the first, as well as the NO, through activation of GMP and consequent elevation of cGMP levels and the second through a cGMP-independent mechanism \[[@B127], [@B128]\]. Compelling evidence has demonstrated that each member of this triad of gasotransmitter can influence each other. For example, the inhibition of the NO synthesis might increase the CO production \[[@B129]\], while low-dose CO has been showed to decrease the eNOS mRNA expression \[[@B130]\]. Recently, particular attention is paid to the role of crosstalk existing among the gasotransmitters in determining cytoprotective effects in the heart and vessels. It was demonstrated that NO, CO, and H~2~S act in concert to preserve the cardiovascular homeostasis thanks for instance to their antioxidant and anti-inflammatory properties \[[@B131]\]. Noteworthy, the gaseous nature of these compounds makes them attractive candidates for the treatment of several pathological conditions, especially ischemia reperfusion injury. In this regard, H~2~S have been shown to stimulate vascular remodeling after ischemia in mice by enhancing the NO production \[[@B132]\]. Donnarumma et al. have recently investigated in murine and swine models of ischemia reperfusion the effects of an oral administration of zofenopril, an ACE-I containing a sulfhydrylic group. The authors found that zofenopril reduced myocardial infarct size in both animal models and preserved blood flow in swines and such effects were associated with an elevation of the H~2~S and NO plasmatic levels \[[@B133]\]. Importantly, there are conflicting evidences on the anti-ischemic effects of the ACE-I and some studies have revealed such effects only for sulfhydryl-containing agents \[[@B134]\]. Moreover, it was demonstrated that an early treatment of the acute myocardial infarction with zofenopril is able to reduce morbidity and mortality any more than ramipril, dicarboxylate-containing ACE-I \[[@B135]\]. Therefore, the understanding of the mechanisms involved in the cytoprotective effects of all gasotransmitters either individually or together is necessary to fully exploit their therapeutic potential. 6. Natural Derived Compounds and NO {#sec6} =================================== Growing evidence leads to considering a healthy dieting regimen as an helpful strategy to reduce CVDs-associated risk factors acting, as well as the aforementioned drugs, via modulation of the NO pathways. The Mediterranean diet, rich in fruits and vegetables and based on high consumption of red wine, was associated with a good prognosis in patients with CVDs \[[@B136]--[@B139]\]. In particular, in subjects who usually consume large amounts of fruits, vegetables, red wine, tea, chocolate, and nuts, a significant improvement of endothelial function has been reported, which in turn contributes to reduction of blood pressure, atherosclerosis, and cardiovascular mortality \[[@B140], [@B141]\]. Interestingly, the beneficial properties of the red wine were recognized as the solution of the "French paradox," a term used to describe the observation that the French population had a low incidence of CVDs, despite a diet predominantly characterized by a high consumption of wine and saturated fat food \[[@B142]\]. The protective effects against CVDs have been attributed, at least in part, to the high content in these specific foods of polyphenols, a class of chemicals characterized by the presence of phenol units in their chemical structure \[[@B143]\]. 6.1. Classification and Source of Polyphenols {#sec6.1} --------------------------------------------- ### 6.1.1. Flavonoids {#sec6.1.1} Flavonoids represent a large group of polyphenols, characterized by two benzene rings linked via a heterocyclic pyran ring. The latter gives reason of the differences between the various classes of flavonoids. According to their chemical structure, the flavonoids can be subdivided in (i) flavones, such as apigenin (bilberry, raspberry, strawberry, plum, cherry, blackberry, red pepper, and tomato skin) \[[@B144]\], (ii) flavonols that include quercitin (red onions, tea, wine, apples, cranberries, buckwheat, and beans) \[[@B145]\], (iii) isoflavones, including genistein (soy, legumes) and coumestrol (soy, red clover), also known as phytoestrogens \[[@B146]\], and (iv) flavanols that include catechins and epicatechin (tea, apple juice, wine, and cocoa) \[[@B147]\]. Interestingly, these compounds have been found also in medical plants, such as*Aloe vera* \[[@B148]\] and*Cannabis sativa* \[[@B149]\]. Flavonoids have been demonstrated to exert a plethora of beneficial effects both*in vivo* and*in vitro* and to regulate specific molecular pathways and target several genes \[[@B150], [@B151]\]. In particular, the best characterized biological property for all flavonoids, as well as for polyphenols in general, is their ability to act as antioxidants, inhibiting ROS accumulation, acting either by scavenging ROS or inhibiting enzymes involved in the ROS production or by enhancing the natural antioxidant defenses \[[@B152]\]. Moreover, several studies have shown and are underlining anticancer activities of the flavonoids. For example, quercitin has been shown to inhibit cell proliferation in several human cells, such as lymphoid, colon, ovarian, and gastric cells, through modulation of several genes involved in cancer progression \[[@B153]\]. Moreover, genistein has been recently proposed as a chemopreventive agent especially against prostate cancer, thanks to several interesting results deriving from*in vitro* and epidemiological studies \[[@B154], [@B155]\]. ### 6.1.2. Stilbenoids {#sec6.1.2} Stilbenoids are a class of phenolic compounds synthesized as defense agents from the plants expressing stilbene synthase. Resveratrol is the most studied stilbenoid, but more than 400 compounds have been identified; most of them are currently used in Chinese traditional medicine \[[@B156]\]. Generally, stilbenoids are classified on the basis of the number of the C6-C2-C6 units (monomer, dimers, trimers, tetramers, and examers). They are present abundantly in berries (grape, blueberry, bilberry, cowberry, cranberry, and strawberries) and peanuts but are also detectable in cocoa powder, dark chocolate, and white tea \[[@B157], [@B158]\]. As mentioned for the flavonoids, several researches have suggested a role of stilbenoids as anticancer, antioxidant, and antiaging agents or as positive modulators of several human degenerative diseases \[[@B159], [@B160]\]. In this regard, resveratrol has been shown to induce apoptosis in breast cancer and prostate cells, by induction of caspases, Bax proteins, and p53 \[[@B161], [@B162]\]. Of note, a natural analog of resveratrol has been documented to inhibit growth of several cancers, such as pancreatic \[[@B163]\] and colon \[[@B164]\] cancer. ### 6.1.3. Curcuminoids {#sec6.1.3} Cucrcuminoids are chemical compounds extracted from the rhizome of*Curcuma Longa*Linn. They are characterized by a linear structure (diarylheptanoid) with two phenolic groups (C6-C7-C6) and are widely used as colorants for vegetables. Curcumin and its derivates have been demonstrated to possess numerous pharmacological activities, including anti-inflammatory, antioxidant, and antitumorigenic effects. In particular, it has been reported that curcumin is able*in vitro* to downregulate the expression of cyclin D and E and to upregulate p53 and p21, which in turn contribute to arresting cell proliferation/migration and promoting apoptosis \[[@B165], [@B166]\]. Concerning its antioxidant properties, curcumin acts prevalently as superoxide radical scavenger \[[@B167]\]. ### 6.1.4. Phenylethanoids {#sec6.1.4} Phenylethanoids are polyphenols characterized by a phenethyl alcohol structure. Typical examples of phenylethanoids are tyrosol and its derivate oleuropein, present prevalently in olive oil and olive leaf. Oleuropein is the most abundant polyphenol in olives and thus it is receiving particular attention by the scientific community because extra virgin olive oil is an essential component of Mediterranean diet. Several studies have demonstrated that the oleuropein possesses a wide range of pharmacological properties such as antiatherogenic \[[@B168]\], hypotensive \[[@B169]\], and antidiabetic \[[@B170]\], as well as anticancer activity and antioxidant effects \[[@B171], [@B172]\]. Moreover, also hydroxytyrosol, a metabolite of oleuropein, has been shown to possess antioxidant properties \[[@B173]\] as well as anti-inflammatory, antiplatelet aggregation, antiatherogenic and cardioprotective, antimicrobial, antiviral, and anticancer activities \[[@B174]\]. 7. Polyphenols and the NO Signaling {#sec7} =================================== Concerning the effects on vascular physiology, several data suggest that polyphenols act on the NO signaling and metabolism, improving eNOS expression and activity, as well as reducing eNOS uncoupling. Nowadays, one of the limits during the characterization of the molecular pathways activated by polyphenols is that most of the experiments have been conducted with the total extracts of food, such as wine, cocoa powder, or olive leafs extracts; therefore often it is very difficult to identify the specific compound exerting protective effects. Nevertheless, some studies measured the effects of single compounds, such as resveratrol, quercitin, or curcumin \[[@B175]\]. Irrespective of their source, one of the main effects exerted by the polyphenols is the NO-dependent vasodilatation ([Figure 4](#fig4){ref-type="fig"}). For example, in isolated arteries of rabbits, Karim et al. demonstrated that cocoa extracts increased levels of intracellular Ca^2+^, leading to L-arginine conversion in citrulline and to the eNOS activation \[[@B176]\]. Similarly, plant-derived polyphenols have been reported to induce vasodilatation of porcine coronary arteries through NO generation \[[@B177]\]. Moreover, in bovine endothelial cells, catechins of green tea activated eNOS by phosphorylation at Ser1179 and dephosphorylation at Thr495 in a PKA-Akt dependent manner \[[@B178], [@B179]\]. In addition, such compounds were also shown to exert protective effects in diabetic rats thanks to the reduction of oxidative stress obtained by downregulation of NADPH oxidase \[[@B180]\]. Interestingly, catechins were found to reduce platelet aggregation and to reverse endothelial dysfunction in patients with coronary artery disease, thus exerting antiatherosclerotic properties \[[@B181], [@B182]\]. Moreover, polyphenols of the black tea were found to enhance the activity of eNOS via p38 MAPK-dependent phosphorylation in porcine aortic endothelial cells. In fact, both pharmacological and genetic inhibition of p38 MAPK attenuated both eNOS activation and phosphorylation changes in response to these polyphenols \[[@B183]\]. Among plant-derived polyphenols, fruit extracts of*Camelia japonica*(CJF), a plant widely distributed in Asia and well known for its antioxidant properties \[[@B184]\], have been demonstrated to induce the NO production via Akt pathways in endothelial cells and to activate eNOS via phosphorylation at Ser1179. In the same study, CJF inhibited VSMCs proliferation and migration, suggesting its beneficial role in the prevention of atherosclerosis \[[@B185]\]. Similarly, polyphenols of the tropical plant*Terminalia* have been reported to induce a calcium-dependent activation of eNOS \[[@B186]\]. Interestingly, Appeldoorn et al. by using an*in vitro* screening to discover the potential effects of different polyphenols have found that quercitin, abundant in many vegetables and fruits, is one of the major stimulator of the NO production \[[@B187]\]. Indeed, the effects of quercitin have been extensively investigated in animal models of CVDs, especially with regard to its antihypertensive effects. For example, a reduction of blood pressure after administration of quercitin in spontaneously hypertensive rats has been showed \[[@B188]\], as well as in salt-hypertensive \[[@B189], [@B190]\] and NO deficient rats \[[@B191]\]. Recently, it has been reported that quercitin is able to ameliorate arterial erectile dysfunctions in rats via NOS regulation restoring, almost in part, the function of NO-cGMP pathway in the process of penis erection \[[@B192]\]. The molecular mechanism involved in the antihypertensive effect of the flavonoid quercitin was attributed to the inhibition/downregulation of NADPH oxidase. Concerning this, Perez-Vizcaino et al. demonstrated that quercitin was able to induce the lowering of blood pressure by diminishing superoxide-driven NO inactivation via downregulation of aortic p47phox, a regulatory subunit of NADPH oxidase, which is the main source of vascular superoxide \[[@B193]\]. These results are in accordance with others showing that quercitin decreased NADPH oxidase-mediated superoxide anion generation, as a consequence of inhibition of p47 protein subunit expression in \[[@B194]\]. In isolated rat aortic ring, Jin et al. found that apigenin, a polyphenol abundant in many plants, enhanced the NO bioavailability via reduction of oxidative stress. Apigenin evoked a concentration-dependent relaxation in aortas, which was specifically inhibited by L-NAME, a direct inhibitor of NOS. Of note, vasodilation occurred concomitantly with inhibition of superoxide anion and increasing of the NO levels \[[@B195]\]. In a similar way, curcumin has been reported to increase relaxation in porcine coronary arteries, probably thanks to mechanism involving NO, cGMP, and adrenergic *β*-receptor and, also in this case, such relaxant effect was specifically inhibited by L-NAME \[[@B196]\]. The involvement of caveolin-1 in polyphenols-mediated effects on the NO pathways has also been reported. Li et al. demonstrated in endothelial cells that green tea extracts downregulated the caveolin expression via activation of ERK and deactivation of p38 MAPK kinases \[[@B197]\]. Similarly, Vera et al. found in hypertensive rats that genistein, a soy isoflavone, was able to enhance eNOS activity via inhibition of caveolin-1 and NADPH oxidase and favoring O~2~ ^−^ reduction, thereby leading to decrease in blood pressure \[[@B198]\]. Moreover, soy isoflavones has also been demonstrated to improve the NO metabolism in carotid and cerebral rat arteries \[[@B199]\] as well as to enhance eNOS mRNA expression \[[@B200]\]. NO-mediated antihypertensive effects were also reported in rats after administration of other soy isoflavones, such as glucosyl hesperidin \[[@B201]\]. Yamamoto et al. found that the hypotensive effects of this natural compound were associated with reduction of oxidative stress and improvement of the NO metabolism \[[@B202]\]. In this regard, hesperidin was found to significantly prevent endothelial damage and leucocytes adhesion in animal models of ischemia reperfusion. Concomitantly, an increase of NO bioavailability and a reduction of inflammatory molecules which contribute to ameliorate edema and other symptoms of venous diseases have been reported \[[@B203]\]. Polyphenol-rich cocoa extracts have been demonstrated to reduce blood pressure in spontaneously hypertensive rats \[[@B204]\] and, similarly, in hypertensive patients, as well in healthy subjects, the intake of black cocoa extracts has been reported to reduce blood pressure and improve endothelial function through increase of the NO bioavailability \[[@B205]--[@B208]\]. Moreover, in patients with high cardiovascular risk it was showed that the administration of two different diets, one rich in polyphenols deriving from extra virgin olive oil and another rich in nuts, was shown to reduce systolic and diastolic pressure concomitantly with an increase of the NO plasma levels \[[@B209]\]. 7.1. Red Wine Polyphenols and NO Pathways {#sec7.1} ----------------------------------------- Red wine is one of the main sources of the natural polyphenols. As mentioned above, epidemiological studies have suggested that the high consumption of red wine correlates with a reduction of the CVDs risk factors. The evidence corroborating vascular effects of red wine polyphenols (RWPs), as well as grape seed extracts (GSEs) and grape juice polyphenols (GJPs), is the induction of NO-dependent relaxation in isolate arteries and the activation of NO signaling pathways in endothelial cells \[[@B210]--[@B212]\]. Leikert et al. found that RWPs enhanced eNOS expression and release of NO in human endothelial cells \[[@B213]\]. In the same way, NO production and intracellular Ca^2+^ release have been shown in bovine endothelial cells treated with RWPs \[[@B214]\] and an increase of eNOS and Akt phosphorylation were also reported in endothelial cells exposed to GSEs \[[@B215]\]. Similar eNOS activation was also demonstrated in isolated arteries. For example, in porcine coronary arteries Madeira et al. showed endothelium relaxation induced by GSEs via Akt/eNOS phosphorylation \[[@B216]\], and also in isolated porcine coronary arteries, RWPs were found to enhance phosphorylation of eNOS at Ser1177, resulting in the increase of the NO production \[[@B217]\]. Interestingly, in rat femoral arteries, RWPs were shown to induce vasodilatation and to increase the NO levels in a concentration-dependent manner \[[@B218]\]. Moreover, RWPs were demonstrated in rat aorta to enhance NO bioavailability and to increase intracellular Ca^2+^ and cGMP concentrations \[[@B218], [@B219]\]. Several molecular mechanisms have been proposed to explain in both animal models and humans the beneficial effects of the RWPs in vascular physiology. In this regard, Bernátová et al. in hypertensive NO deficient rats showed that RWPs restored endothelial functions thanks to a reduction of blood pressure induced by increased eNOS activity in the left ventricle and aorta \[[@B220]\]. Similarly, in salt-induced hypertensive rats, RWPs were shown to improve vascular physiology by inhibiting NADPH oxidase \[[@B221]\]. The inhibition of NADPH oxidase was also reported in Ang II hypertensive rats treated with RWPs in which a reduction of superoxide anions level occurred concomitantly with restoration of the NO bioavailability \[[@B222]\]. RWPs have been demonstrated to exert protective effects also in animal models of ischemia and atherosclerosis. For example, in ischemic rats, RWPs were shown to reduce the angiogenic process \[[@B223]\], and, in hypercholesterolemic mice, Napoli et al. showed that low doses of RWPs reduced atherosclerosis by eNOS activation \[[@B224]\]. Interestingly, with an*in vitro* model of human atherosclerosis, Magrone et al. have reported enhanced production of the NO, after administration of red wine. The authors tested some red wines for their ability to trigger NO production from human healthy peripheral blood mononuclear cells, finding that flavonoids and resveratrol, abundant in the red wine, once absorbed at intestinal level and entered into circulation, induced monocytes to produce the NO \[[@B225]\]. Few clinical trials have planned with the aim to investigate the effects of a dietary regimen based on moderate consumption of wine about NO related improvement in vascular physiology in both healthy patients and patients with high risk of CVDs. For example, in healthy subjects, an oral supplementation of grape juice was found to inhibit platelet aggregation with decreased production of superoxide and enhanced NO levels \[[@B226], [@B227]\]. Moreover, besides its antithrombotic activity, red wine has also been suggested to exert cardiovascular protective effects by enhancing circulating endothelial progenitor cells thanks to a mechanism involving an increase of the NO bioavailability, as reported in studies performed in healthy individuals by Huang et al. \[[@B228]\]. In addition, red wine consumption has been shown to significantly decrease blood pressure and enhance plasma NO levels in hypertensive patients \[[@B229]\]. Interestingly, Karatzi et al. demonstrated that in smokers a consumption of red wine counterbalanced the endothelial dysfunction caused by oxidative stress induced by cigarettes smoke, in a pathway probably mediated by NO \[[@B230]\]. 7.2. Resveratrol and NO Pathways {#sec7.2} -------------------------------- Among the RWPs, resveratrol (RSV) is one of the best characterized members. It has been used in the Indian medical herb named "Darakchasava" from about 4500 years ago and the clinical effects described in the past for "Darakchasava" are the same attributed to RSV today \[[@B231]\]. RSV was firstly described for its antitumorigenic properties \[[@B232]\]; it is present especially in grape skin and red wine, but also in peanuts, pistachios, and pine trees \[[@B233]\]. The interest of the scientific community for RSV derives from the observation that its administration mimics the effects of calorie restriction, a tool widely recognized to prevent the endothelial dysfunction, thereby attenuating atherosclerosis, hypertension, diabetes, and CVDs risk factors and aging-associated diseases in general \[[@B234]--[@B236]\]. Thanks to some experiments conducted*in vitro* in endothelial cells, RSV has been shown to regulate several target molecules, such as the NAD^+^-dependent deacetylases named sirtuins, acting at transcriptional and posttranscriptional levels \[[@B237]--[@B239]\]. Although the studies underlining the vascular protective effects exerted by RSV did not study the involvement of the NO signaling \[[@B157], [@B234]\], several findings, obtained in animal models of CVDs, have proposed the NO as the main downstream target mediating such effects. For example, Xia et al. demonstrated in ApoE deficient mice that RSV was able to modulate the oxidative stress responsible for atherosclerosis. From one side, NADPH oxidases were downregulated; from the other side superoxide dismutases (SOD) were upregulated. Moreover, oxidation of BH~4~ was found to be reduced, attenuating the increase of eNOS uncoupling levels \[[@B240]\]. Other beneficial effects were shown in many different clinical settings reinforcing the idea that RSV could be considered an optimal therapeutic strategy against CVDs. For example, in hypercholesterolemic rabbits, RSV improved endothelial function in parallel with an increase of NO plasma levels \[[@B241]\]. In addition, RSV has been suggested to contrast the endothelial dysfunction correlated with metabolic syndromes. In this regard, in endothelial cells RSV was demonstrated to suppress superoxide generation and to activate eNOS through phosphorylation at Ser1177 thereby increasing the NO generation \[[@B242]\]. In aortas of diabetic mice, RSV restored vasodilatation by enhancing eNOS activity and inhibiting the tumor necrosis factor *α*- (TNF*α*-) induced activation of NADPH oxidase \[[@B243]\]. In the same way, a treatment in rats with RSV has been showed to increase muscle microvascular recruitment via an NO-dependent mechanism blocked by TNF*α* \[[@B244]\]. Also, RSV was shown to reduce blood pressure in obese rats and to enhance the expression of eNOS via AMPK and reduction of TNF*α* in adipose tissue \[[@B245]\]. Similarly, in rats fed with high fructose diet, RSV decreased blood pressure via AMPK-Akt-NOS pathway \[[@B246]\]. Interestingly, in the myocardium of diabetic mice, RSV reduced Cav-1 expression, which in turn contributes to enhance eNOS activity \[[@B247]\], and the same effects on Cav-1 expression were found in hypercholesterolemic rats \[[@B248]\]. Furthermore, RSV was shown to protect heart from ischemic reperfusion injury. Hattori et al. demonstrated that RSV reduced infarct size in rat hearts by enhancing iNOS expression \[[@B249]\]. The cardioprotective effects of the RSV has also been showed in spontaneously and angiotensin Ang II-induced hypertensive rats, in which RSV contributes to the upregulation of the eNOS activity and reduction of pressure and cardiac hypertrophy \[[@B250]\]. Moreover, the antihypertensive effect of the RSV was also shown to be mediated by the attenuation of eNOS uncoupling via reduction of L-arginine levels and oxidative stress \[[@B251]\]. The antithrombotic activity of the RSV has been also reported in human platelets. Gresele et al. showed that RSV stimulated platelet NO production through inhibition of p38 MAPK, NADPH oxidases, and superoxide formation, thus decreasing peroxynitrite accumulation \[[@B252]\]. RSV was also shown to mobilize endothelial progenitor cells in a NO-dependent manner, thus contributing to repairing the damage occurring in vessels after ischemic injuries \[[@B253]\]. In the arteries of patients with hypertension and dyslipidemia, Carrizzo et al. characterized many of the downstream effectors of the RSV-dependent NO generation. The authors found an enhanced vasodilatation of arteries due to the activation of AMPK and reduction of eNOS uncoupling via increasing levels of BH~4~ and, in the same study, RSV was found to reduce vascular oxidative stress trough upregulation of manganese superoxide dismutase in a pathway mediate by nuclear factor erythroid-derived 2-like 2 \[[@B254]\]. Some authors have also suggested the potential therapeutic use of RSV for the prevention of stroke; for example, in rat models of stroke, RSV reduced brain damage in a NO-dependent manner \[[@B255]\]. Similarly, in rats subjected to focal cerebral ischemia Tsai et al. provided the evidence that RSV might enhance plasma levels of the NO and upregulate eNOS expression while it might downregulate iNOS expression and that these effects were abolished by the coadministration of selective NOS inhibitors \[[@B256]\]. 8. Bioavailability of Polyphenols {#sec8} ================================= Although the use of the polyphenols represents a promising tool for increasing the NO production and activity against CVDs, one of the biggest challenges for their employ in the clinical practice is to enhance their low bioavailability. In this regard, it has been shown that when orally administered, polyphenols concentration appears not to be sufficient to ensure therapeutic effects \[[@B257]\]. For example, the plasmatic levels of the resveratrol from dietary intake are often undetectable or very low when compared with the concentrations employed during*in vivo*and*in vitro* experiments \[[@B258]\]. Similarly, the pharmacological properties of curcumin are drastically restricted mainly because of its low water solubility and absorption from the gut, short half-life, and extremely poor bioavailability. To overcome such problems, one of the best approach could be developing new pharmaceutical formulations, for example, polyphenols conjugated with cyclodextrins, or encapsulated in nanoparticles (NP), such as poly(lactic-co-glycolic acid) (PLGA) based NP or liposomes. In this regard, many of these formulations have been demonstrated to improve solubility, systemic half-life, resistance to metabolic degradation, and ultimately the bioavailability of the polyphenolic compounds in order to potentiate their biological activities \[[@B259], [@B260]\]. However, while the differences between polyphenols monoadministered or administered in encapsulated formulations have been extensively studied for what concerns the polyphenols antioxidant and anticancer properties, no experiments have been carried out on the effects of these formulations on the NO metabolism. 9. Conclusion {#sec9} ============= Targeting the gasotransmitter NO is becoming a new challenge in cardiovascular medicine. We here reviewed some of the experimental evidences that have indicated several natural compounds as suitable activators of the NO signaling pathways. It is necessary to remark that for most of them the molecular mechanism, as well as the precise concentration to obtain beneficial effects, especially because of their low bioavailability remains to be determined. Nevertheless, these agents, mainly the polyphenols, doubtless possess a great therapeutic potential above all when you consider that the available drugs, although effective, did not act exclusively on the NO pathways often causing deleterious side effects. Moreover, most of the investigations on the natural compounds have involved*in vitro* studies; thus it is difficult to draw definite conclusions about their therapeutic usefulness. Although accumulating evidence suggests that the polyphenols exert beneficial effects against vascular diseases by restoring the impairment of the NO production and/or bioavailability, much remains to be clarified. Doubtless, many gaps must be filled in understanding the complex chemistry, biochemistry, and molecular biology of such natural agents in order to introduce such NO signaling modulators in the clinical practice. Competing Interests =================== The authors declare that they have no competing interests. {#fig1} {#fig2} {#fig3} {#fig4} [^1]: Academic Editor: Yong Ji

In 2008, there were 12.7 million people diagnosed worldwide with cancer and 7.6 million deaths associated with cancer \[[@CR1]\]. Of all cancers, lung cancer is not only the most commonly diagnosed in the world with 1.61 million new cases per year which accounts for 12.7 % of all new cancers but also the most deadly with 1.38 million deaths (representing 18.2 % of all cancer-related deaths) \[[@CR1]\]. The most common treatments for lung cancer include chemotherapy, radiation and surgical resection. For patients undergoing any surgery, including lung surgery, patient education is seen as critical to their recovery. Education is known to be an essential tool to provide patients with information concerning their health condition, their treatment and their recovery \[[@CR2]\]. Patients who receive structured patient education, such as patient education classes, are more likely to follow through with their therapeutic regiment compared with patients who receive unstructured education \[[@CR3]\]. Certainly, health care professionals (HCPs) are aware of the importance of providing effective information and advice for patients undergoing surgery \[[@CR4]\]. Education material can be presented prior to surgery, during the hospital stay and/or after surgery, and post-discharge. Studies show that preoperative teaching has a beneficial effect on postoperative outcomes for patients undergoing thoracic surgery \[[@CR5], [@CR6]\]. Specifically, preoperative anaesthesia teaching has been shown to increase patients' understanding of options for anaesthesia care, decrease anxiety and fear and decrease analgesic requirements, postoperatively \[[@CR7], [@CR8]\]. Preoperative physiotherapy teaching has been shown to decrease complications and even shorten hospital stays \[[@CR9], [@CR10]\]. Preoperative patient education can be provided by HCPs in a group setting or in individual teaching sessions. Both settings have their advantages as in groups; patients can decrease their anxiety and gain reassurance while discussing with others having similar conditions, but individual teaching allows for patients to obtain specific information more pertinent to them \[[@CR11], [@CR12]\]. Whyte and Grant \[[@CR13]\] advanced that providing preoperative education in a coordinated format allows for patients to receive the same information prior to surgery with the overall goal to have the patient become more knowledgeable about their condition and ultimately become a member of the decision making team. We know from adult learning principles that the amount of information retained from various sources differs across patients \[[@CR14], [@CR15]\]. Therefore, patient education needs to be provided using a variety of formats \[[@CR16]--[@CR17]\]. As well, due to other factors that affect patients' learning, including the stress of having cancer, information needs to be reinforced at different points along the clinical journey (preoperative, postoperative stay), multiple times, and by different HCPs \[[@CR10], [@CR13]\]. Yet, despite this knowledge, there is often a disconnection between what HCPs think patients need to know and what the patients want to know \[[@CR18]\]. Van Weert \[[@CR19]\] demonstrated this discrepancy in patient education for those undergoing cardiac surgery. A key area of discord was around the issue of emotional support. HCPs provided mainly medical information and patients were seldom asked about their fears, anxiety level, understanding of the information or expectations. Furthermore, little reassurance or emotional support was provided. To further understand and uncover the issues surrounding patient education for patients having lung surgery due to lung cancer, this study was undertaken. The specific purpose of this study was to understand what information patients who were about to undergo lung surgery wanted to learn before and after their surgery and to establish what information is currently provided to these patients by HCPs. Methods {#Sec1} ======= Research Site {#Sec2} ------------- The study was conducted at a cancer assessment clinic (CAC) that uses structured preoperative group and individual education sessions to prepare patients for their upcoming surgery and cancer treatments. The interprofessional bilingual (French and English) CAC provides care for a diverse population within a large geographic area. When a person is diagnosed with a lung cancer that can potentially be surgically resected, they are referred to the CAC where they receive the required evaluation and tests for surgery. They also receive a group teaching session that includes a tour of the hospital by a volunteer and a formal presentation by nurse educators. This presentation also includes two short sections presented by the social worker and the physiotherapist. Afterwards, each patient meets one-on-one with the social worker, the physiotherapist and finally the anaesthetist to receive more individualized surgical preparation information. Methodological Approach {#Sec3} ----------------------- The study was approved by the Ottawa Hospital Research Ethics Board and the University of Ottawa Research Ethics Board. This qualitative study used a generalized qualitative approach and employed semi-structured interviews to gather data \[[@CR20]\]. Ten patients were interviewed preoperatively and postoperatively to obtain insight into their patient education needs, and 11 HCPs were interviewed to uncover their education practices and their self-perceived role in patient education. Qualitative analysis of the interviews was completed in order to extrapolate common themes. Participants {#Sec4} ------------ Patient participants were recruited in person at the CAC by a bilingual research assistant. Inclusion criteria included being over 18 years of age, either English or French speaking, scheduled to have lung surgery due to lung cancer at the hospital and willing to discuss their patient education needs. Exclusion criteria included plans to have a surgery for a reason other than lung cancer. HCP participants' inclusion criteria included being over 18 years of age, either English or French speaking, providing patient education to patients awaiting lung surgery and willing to discuss the patient education they provided. HCPs were excluded if they did not provide patient education. Data Collection {#Sec5} --------------- To reduce patient participants' time at the hospital, they had the option of completing the preoperative interview in person after their CAC session or by telephone at a later time the same day. All postoperative interviews were conducted over the telephone. Both interviews were guided by the same questionnaire that included semi-structured and close-ended questions. During the postoperative interview, patients were reminded of their preoperative comments. HCPs' interviews were conducted in person and also included both close-ended and semi-structured questions. Data Analysis {#Sec6} ------------- Interviews were analysed using a qualitative naturalistic inquiry approach \[[@CR21]\]. Inductive analysis of the data allowed for the identification of themes. Each of the patient's and HCP's interviews was read individually to determine a narrative response to each of the major questions asked. For the patient participants, there was a focus on the information they had found most helpful as well as information they had found unclear or missing. As well, the interviews of each of the participants were read to determine themes emerging across the interviews. For the HCPs, a descriptive narrative was developed of all the information and resources that they used in teaching patients. Findings {#Sec7} -------- ### Health Care Professionals Results {#Sec8} #### Demographics {#d30e360} Of the 11 HCPs recruited for the study, there were two registered nurses (RN), one physiotherapist (PT), two social workers (SW), two physicians and four thoracic surgeons. Five of the HCPs interviewed had been with the CAC since its inception in 2006, and ten of the 11 HCPs had more than 3 years of experience working with patients diagnosed with lung cancer. #### Topics Discussed in Education Sessions {#d30e365} A variety of topics were discussed with patients during education sessions. The topics covered by each HCP depended on their profession but were also specific to the patient's phase of recovery (i.e. before or after surgery, during hospital stay or after discharge). Preoperatively, the HCPs repeated similar educational information addressing the topics of, namely, medications, pain control, exercises, preoperative procedures, common complications, role of the HCPs, role of the patient, role of family members, discharge planning and follow-up. Other issues covered preoperatively included diet, postoperative procedures and home care services. Postoperatively during the hospital stay, fewer HCPs provided information for patients; here, the topics included pain control and pain medication, the role of the specific HCP and patient follow-up after discharge. There were other issues discussed which were similar to those presented preoperatively. Following the patient's discharge from the hospital, the patients would attend a follow-up visit at the CAC. At this follow-up visit, they would receive information about further treatment they may require. #### Format and Resources {#d30e370} While all HCPs used verbal communication to transmit information, they also used written materials to supplement their verbal communication; the aim of using both methods was to support patients' understanding of the material. Written information included the Canadian Cancer Society's resources, a specific exercise booklet (*Post-Op Thoracotomy Exercise*), and a printout of the Power Point presentation used in the group education session. Some of these resources had pictures and drawings while others did not. The most commonly used booklet was *Lung Cancer: Information Guide and Personal Record*, a compilation of lung cancer information that covers such topics as diagnosis, personal thoughts and coping strategies. Additional materials included the admission booklet from the hospital, S*ternotomy Restrictions* (a hospital produced activity guide), a chemotherapy and radiation booklet and the pre-admission unit booklet for medication information. The social workers also provided patients with logistical details regarding options for convalescent care postoperatively as well as information about parking and finances. #### HCP's Perception of their Role {#d30e384} Each of the HCPs discussed the primary goals of their education sessions. Three common themes surfaced as follows: HCPs all wanted to adequately prepare patients for their surgery and recovery, make sure patients understood the HCPs unique role and responsibilities, and support and reassure patients about the surgery. These findings could be categorized as HCPs wanting to prepare, inform and empower their patients. ### Patient Participants Results {#Sec9} #### Demographics {#d30e392} Ten patient participants, five men and five women, were recruited from four preoperative group education sessions at the CAC. Their ages varied between 50 and 79 years old. Seven patient participants primarily spoke English, three spoke French and either Vietnamese or Italian at home. Eight patient participants lived with a spouse and two lived alone. Six patient participants required more than 40 min to drive to the CAC, while the rest lived between 10 to 40 min from the hospital. Nine patient participants were retired, and only one patient continued to work fulltime. Hospital length of stay varied from 1 to more than 7 days. Postoperatively, the interviews were conducted after the patient had been home for at least 5 days. This time was chosen to allow enough time for the patient to get home and into some sort of routine, but not so much time that they might forget details of their experience. Three patient participants were receiving home care services at the time of the postoperative interview. #### Topics Discussed {#d30e397} When patient participants were asked to discuss the most helpful and most surprising information, as well as what was unclear in the education presentation preoperatively, there were no common themes that emerged. Although all patient participants found the information to be very helpful, they reported that the information about the surgery itself and the session with the physiotherapist where they learned about breathing exercises were most helpful. When asked about the most surprising information, each participant's answer was unique and ranged from the importance of physiotherapy interventions including exercises, to details about the surgery, information about the functioning of the lungs and how a cancer in the lung can originate elsewhere. In general, participants felt there was no unclear information and one participant noted that: "if it was unclear, you got to ask on the spot and they answered your questions and explained it to you". When asked about what information was the most helpful, each participant responded with the same type of information they found most helpful preoperatively. All the participants mentioned that they still were keeping up with the exercises prescribed by the physiotherapist. All participants confirmed that the information was clear. Two participants wished that they had received their preoperative test results and the surgery plan prior to surgery. The most notable difference between preoperative and postoperative comments was postoperative comments regarding a wish to know more about the level and timing of pain associated with surgery prior to the surgery. This appears to be a considerable need not met. One participant expressed the need to know "how long things are going on after surgery, like pain. How long will I have pain?" #### Format and Resources {#d30e402} In terms of teaching format, the majority of patient participants stated they liked receiving information directly from a HCP and reading on their own both preoperatively and postoperatively. Patient participants commented that they found the *Pulmonary Resection Booklet* and *Post-Thoracotomy Exercise Booklet* the most helpful of all the print material given to them preoperatively. Looking back at their journey, patients reflected that the booklet *Post-Thoracotomy Exercise Booklet* and the *Lung Cancer: Information Guide and Personal Record* were helpful in answering their questions. #### Overall Experience {#d30e419} The CAC education session was an overall positive experience for the participants in the face of being diagnosed with cancer. When asked to give feedback on their confidence and satisfaction, all patients participants felt very satisfied with the information provided and very prepared for their surgery. From their experiences, participants provided a number of suggestions to improve the experience for future patients. These suggestions included allowing for a break in the long day, having a doctor present for the education session and simplifying some of the language while adding pictures to the presentation to improve their understanding of the material. Discussion {#Sec10} ========== An interprofessional team approach to preoperative patient education allows for various topics to be presented to patients by different members of the team. These topics are profession specific. Studies have found that to enhance information retention, topics need to be discussed on several occasions, by more than one HCP \[[@CR10], [@CR12], [@CR13]\]. This may have been supported in this study by the fact that two of the topics presented both preoperatively and postoperatively, regarding surgery and importance of physiotherapy, were perceived as the most helpful information presented at both time periods. It is important to note that although repetition is likely to enhance retention, there is a limit to the amount of new information patients will remember \[[@CR22]\]. Repeated information should be carefully selected. As well, all HCPs who present a topic to the patient should be aware of what other members of the team are saying about it, to ensure patients receive consistent messages. Pain and pain management were topics the majority of HCPs discussed preoperatively, yet preoperatively, patients did not perceive this as one of the most helpful pieces of information. Following their preoperative education sessions, patients did not mention that this information was unclear. However, after returning home postoperatively, patients wished they had been better informed regarding pain before their surgery. Although the topic was addressed repeatedly in the preoperative and again postoperative phase by HCPs, the fact that patients perceived they lacked information regarding pain may be as a result of how HCPs described the pain to patients. Perhaps, pain simply needs to be addressed further or differently preoperatively and postoperatively. As well, it may be the case that patients are more concerned with surviving the surgery preoperatively than concerned with their postoperative recovery. This is noted from the HCPs who commented that the questions that patients ask them are around cure and if they can "get the cancer out", with the surgery. When meeting with patients and families, HCPs provided or referred to various resources in order to promote a better understanding of what was taught. The amount of information can be overwhelming and retention may be suboptimal as patients become saturated with information. One patient participant noted that""I think there is too much information in the green leaf book (*Lung Cancer: Information Guide and Personal Record*). It's a difficult operation and I don't want to remember some things"." Multiple patient education methods are needed to transfer knowledge, including information provided in a written format, the use of drawings and the repetition of information to facilitate retention \[[@CR13]\]. Therefore, as most of the HCPs did prior to surgery, it is imperative to provide written material and visual aids for patients to consult after the education sessions. But, this information needs to be designed using plain language and clear design so that it is accessible to people who might be living with limited literacy \[[@CR23]\]. The gap between the literacy demands of the system and the literacy abilities of patients must be narrowed to improve both the effectiveness of the health care system and the quality of life for the patients \[[@CR24]\]. In terms of learning formats, preoperatively, the most popular method to receive information was directly from a HCP, while after their surgery, patients preferred reading independently. This supports the idea that adult learners need different methods to meet their needs. A few of the patients both preoperatively as well as postoperatively noted that they or their family members seek information from the internet. Peterson and Fretz \[[@CR25]\] have reported an increased use of the internet by patients to obtain information pertinent to their cancers. Certainly, patients, as part of their self-directed learning, are using the internet as a source of health information about their condition \[[@CR26]\]. Perhaps the age range of the patients could explain the limited use of the internet in this study; these patients may not have known how to search for additional information. Alternatively, our patient participants may not have wanted to learn more. At least preoperatively, they may have been finding their cancer diagnosis and planned surgery overwhelming; the thought of additional information was too stressful. In fact, one participant noted, "Friends looked for me and found things on the internet but they thought it was too scary so they didn\`t give it to me". Lilja \[[@CR27]\] found that patients who had received detailed information preoperatively had higher levels of stress postoperatively than the group who had received less information. The lack of attempts by the participants to find external resources of information may also reflect that patients at that stage are primarily concerned with survival versus recovery. Conclusions {#Sec11} =========== For a patient education session to be efficient and achieve the goals of decreasing anxiety, empowering through knowledge and improving postoperative outcomes, information provided has to be perceived by the patient as relevant. Good communication is required between the HCPs to cover all required topics and only repeat those of utmost importance. The information needs to be shared with patients at various phases or the continuum of care using different formats to maximize learning. In the case of this study, patients reported satisfaction with their preoperative education when they reflected on this both preoperatively and postoperatively. Furthermore, all patients felt very prepared before their surgery and when looking back once the surgery had been done, five out of six still felt very prepared while one now felt he had been moderately prepared. It is hoped that the results of this study will help inform HCPs of the education needs of patients who are undergoing lung surgery so that they may better tailor education sessions to meet patient needs and expectations. Although a limitation of this study was a small sample size and there was not a uniformity in the patient's length of hospital stay and post-op management. These varied experiences did not seem to impact the perceptions of the participants. Based on the education needs expressed by the participants in this study, pain management is definitely a topic that needs to be addressed more fully both preoperatively and postoperatively. Our next step in this research program will be to further examine the topic of pain including how the issue is discussed by HCPs with patients including the words used to describe the pain experience by both HCPs and patients. Furthermore, we look forward to comparing the current data gathered from an Ottawa clinic to data from across Canada. The investigators express their profound gratitude to all patient and health care participants involved in this project. The authors are also indebted to Dr. Mary Egan, for her thoughtful comments and constructive feedback during preparation of the manuscript.

Introduction ============ Gestational diabetes mellitus (GDM) is defined as a glucose intolerance of varying severity with an onset or first diagnosis during pregnancy ([@b1-etm-07-01-0236]). GDM occurs in 4--10% of pregnancies and is associated with maternal and fetal complications, as well as long-term consequences, including metabolic syndrome, type 2 diabetes mellitus (T2DM) and cardiovascular disease ([@b2-etm-07-01-0236],[@b3-etm-07-01-0236]). Although the pathogenesis of GDM has not been completely elucidated, the innate immune system has been reported to be involved ([@b4-etm-07-01-0236]). In addition, several types of cytokines have been shown to have the ability to interfere with insulin signaling, as well as the development of insulin resistance (IR), in patients with GDM ([@b5-etm-07-01-0236]). Toll-like receptors (TLRs) recognize preserved patterns and are important in the regulation of innate immune responses and inflammation ([@b6-etm-07-01-0236]). TLRs are expressed in numerous types of cells, including adipose cells and monocytes. These cells are the predominant cells of the innate immune system and are pivotal in diabetes ([@b7-etm-07-01-0236]). Among the TLRs, TLR4 is a particularly important receptor. TLR4 is the receptor for lipopolysaccharide (LPS) from Gram-negative bacteria, and affects the innate immune response, the prevalence of T2DM and the metabolic system ([@b8-etm-07-01-0236]). TLR4 expression has been identified in numerous cells and tissues, primarily in monocytes ([@b9-etm-07-01-0236]). TLR4 is important for the regulation of the immune response and inflammatory reaction ([@b10-etm-07-01-0236]). In addition, TLR4 induces the production of proinflammatory cytokines, leading to an impairment of tissue-specific effects ([@b11-etm-07-01-0236],[@b12-etm-07-01-0236]). However, whether TLR4 is expressed in maternal monocytes of patients with GDM has not been evaluated. The current study was conducted to investigate whether TLR4 is expressed in maternal monocytes of patients with GDM and to elucidate the roles of TLR4 in the pathogenesis of GDM. Novel insights into the involvement of TLR4 in the pathogenesis of GDM may provide an opportunity to trace the underlying pathogenesis of GDM and, if proven, may be conducive to improving the treatment of the disease. Materials and methods ===================== Subjects -------- A total of 62 females (21--39 years old) at ≥37 weeks of gestation were involved in this study, including 32 females with GDM and 30 healthy pregnant females. Females with GDM were selected by a screening and diagnostic program according to the criteria of the Fourth Workshop Conference of Gestational Diabetes ([@b13-etm-07-01-0236]). Females with multiple pregnancies, fetal anomalies, preexisting hypertension or DM, or chronic disease were excluded. All females with GDM were treated with insulin and IR was estimated using the homoeostasis model assessment (HOMA)-IR. The protocol was approved by the local Ethics Committee of College of life and Technology, Jinan University (Guangzhou, China) and written informed consent was obtained from all females. Isolation of monocytes ---------------------- Maternal peripheral blood was collected using tubes treated with EDTA and the peripheral blood monocytes were isolated by density gradient centrifugation using Ficoll-Paque PLUS (Amersham, Piscataway, NJ, USA). The negative isolation of human monocytes was performed using a Monocyte Isolation kit II (Miltenyi Biotech, Bergisch Gladbach, Germany), in accordance with the manufacturer's instructions. Following centrifugation at 400 × g for 30 min, the cells were washed with phosphate-buffered saline at 4°C. The viabilities of monocytes at \>90% confluence were calculated using 0.4% trypan blue. The monocytes were stored at −80°C until use. RNA isolation and quantitative polymerase chain reaction (qPCR) --------------------------------------------------------------- Total RNA from the monocytes was isolated using a commercial kit (Omega Bio-Tek, Norcross, GA, USA), in accordance with the manufacturer's instructions. The RNA concentration and purity were determined using 1.0% agarose gel electrophoresis with an optical density 260/280 absorption ratio of \>1.8. cDNA was synthesized in a 20-μl reaction mixture containing 2 μg total RNA, using the Omniscript reverse transcription kit (Takara Bio, Inc., Otsu, Japan) and oligo(dT) primers, following the manufacturer's instructions. TLR4 mRNA expression was measured with the ABI PRISM 7300 Sequence Detection System using the SYBR^®^ Green PCR Master mix (Applied Biosystems, Foster City, CA, USA). The following primers were used for analysis: TLR4 mRNA forward, 5′-AGTGTGTGTGTCCGCATGAT-3′ and reverse, 5′-CCACTTGGGGTCTAAGAACG-3′; 18S rRNA forward, 5′-TTCGGAACTGAGGCCATGAT-3′ and reverse, 5′-CGAACCTCCGACTTTCGTTT-3′. qPCR was performed with a 20-μl reaction mixture, using the SYBR Premix *Ex Taq*™ II kit (Takara Bio, Inc.), according to the manufacturer's instructions. The PCR profile was obtained as follows: Initial activation step at 95°C for 5 min, followed by 40 cycles of denaturation, annealing and amplification (95°C for 15 sec, 60°C for 30 sec and 72°C for 15 sec, respectively). With regard to the internal control, the expression of the house-keeping gene, 18S rRNA, was examined under the same reaction conditions. The experiment was repeated in triplicate. Following amplification, melting curve analysis was conducted for the product formed. Western blotting ---------------- Western blotting was performed using the same monocytes as qPCR. Total protein was extracted from the monocytes using a cell lysis buffer and protease inhibitor cocktail. Following centrifugation at 10,000 × g and 4°C for 15 min, the protein concentration was assessed using the Bradford protein assay kit (Bio-Rad Laboratories, Hercules, CA, USA). The protein samples (20 μg) were loaded onto a 12% SDS-PAGE gel and transferred onto polyvinylidene difluoride (PVDF) membranes (Amersham Biosciences, Piscataway, NJ, USA). The membranes were blocked using 5% non-fat dry milk in a Tris-buffered sodium chloride-Tween-20 (TBST) solution (20 mmol/l Tris, pH 7.6, 137 mmol/l sodium chloride and 0.1% Tween-20) at room temperature for 1 h. The PVDF membranes were subsequently incubated with monoclonal antibodies against human TLR4 (1:1,000; Abcam, Cambridge, MA, USA) overnight at 4°C. Following this, the membranes were incubated with secondary antibody goat anti-rabbit horseradish peroxidase conjugate (1:2,000; Abcam) at room temperature for 2 h. Following three 10-min washes in TBST, the immunoreactive bands were detected using western blotting chemiluminescence luminol reagents (Santa Cruz Biotechnology, Inc., Santa Cruz CA, USA). The band intensities were quantified using scanning densitometry (Bio-Rad Quantity One software; Bio-Rad). Serum tumor necrosis factor (TNF)-α level analysis -------------------------------------------------- The maternal peripheral blood samples were collected and transferred to centrifuge tubes. The blood samples were centrifuged at 3,000 × g for 15 min to separate the plasma and maintained at −80°C until analysis. Serum TNF-α levels were measured using a commercial TNF-α ELISA kit (BioSource International, Inc., Camarillo, CA, USA), in accordance with the manufacturer's instructions. Statistical analysis -------------------- All experimental data are expressed as the mean ± standard deviation. Statistical analyses were performed with SPSS 17.0 statistical software (SPSS, Inc., Chicago, IL, USA). Inter-group differences were compared using the Student's t-test and the correlation analyses were conducted using Pearson's linear correlation analysis. P\<0.05 was considered to indicate a statistically significant difference. Results ======= Patient characteristics ----------------------- The characteristics of the studied patients are listed in [Table I](#tI-etm-07-01-0236){ref-type="table"}. There were no significant differences in age, body mass index, blood pressure, gestational age or birth weight between the normal control and GDM groups (P\>0.05). However, the blood glucose levels and HOMA-IR score were significantly higher in the GDM group than in the normal control group (P\<0.05). qPCR and western blotting ------------------------- The TLR4 mRNA expression results are shown in [Fig. 1](#f1-etm-07-01-0236){ref-type="fig"}. TLR4 mRNA expression levels were significantly higher in the GDM group than in the normal control group (P\<0.05). TLR4 protein expression was observed in all monocyte samples ([Fig. 2A](#f2-etm-07-01-0236){ref-type="fig"}). TLR4 protein expression levels were significantly increased in the GDM group compared with the normal control group (P\<0.05; [Fig. 2B](#f2-etm-07-01-0236){ref-type="fig"}). Serum TNF-α levels ------------------ Significantly elevated serum TNF-α levels were observed in the GDM group (9.50±1.73 pg/ml) compared with the control group (7.27±0.45 pg/ml) (P\<0.05, [Fig. 3](#f3-etm-07-01-0236){ref-type="fig"}). Correlation between TLR4 expression and serum TNF-α levels ---------------------------------------------------------- TLR4 mRNA levels correlated with serum TNF-α levels in all participants. There was a positive correlation between the serum TNF-α levels and TLR4 mRNA expression in monocytes ([Fig. 4](#f4-etm-07-01-0236){ref-type="fig"}). Discussion ========== To the best of our knowledge, the present is the first to compare the levels of TLR4 gene expression between females with GDM and healthy pregnant females. The results showed that TLR4 gene expression levels were significantly increased in patients with GDM compared with healthy pregnant females. Serum TNF-α levels were higher in the GDM group than in the normal control group. In addition, a positive correlation was also observed between TLR4 mRNA expression and serum TNF-α levels in this study. Inflammation is characterized by increased levels of circulating biomarkers of inflammation and monocyte activity. The relationships between inflammation, hyperglycemia and IR have been widely studied in patients with diabetes and have been hypothesized to be mediated via the activation of the innate immune system. TLR4 is an important mediator in the innate and cytokine-activated immune systems. In addition, TLR4 is required for the adaptive immune response, and its activation mediates a signaling pathway involved in the transcriptional expression of proinflammatory cytokines and chemokines ([@b14-etm-07-01-0236]). TLR4 has also been identified as the primary receptor for LPS and is expressed in numerous types of cells. TLR4 gene expression and activation have been demonstrated to be increased in the monocytes of patients with hyperglycemia ([@b15-etm-07-01-0236]). In addition, upregulated TLR4 expression is likely to lead to disease-resistant effects, and TLR4 has a proinflammatory role in the occurrence of diabetes and its complications ([@b11-etm-07-01-0236],[@b16-etm-07-01-0236]). It has been suggested that GDM is a state of chronic inflammation ([@b17-etm-07-01-0236]). In DM, the aggravated inflammation may be mediated by TLRs via the activation of the innate immune pathway ([@b18-etm-07-01-0236]). TLR4 activation and downstream cytokine production may lead to the development of diabetes ([@b19-etm-07-01-0236]). However, the underlying mechanisms remain unknown. Increased TLR4 expression has been observed to correlate with enhanced nuclear factor (NF)-κB activation in response to the TLR4 ligand, resulting in elevated levels of proinflammatory cytokines ([@b19-etm-07-01-0236]). NF-κB is a well-recognized transcription factor that regulates the production of proinflammatory cytokines, including TNF-α and interleukin 1 (IL-1) ([@b20-etm-07-01-0236]). TLR4 also activates the mitogen-activated protein kinase (MAPK) signaling pathway, leading to the increased transcription of genes involved in inflammation ([@b21-etm-07-01-0236]). In addition, TLR4 activation affects chemokine (C-X-C motif) ligand 10, inducing apoptosis of islet β cells ([@b22-etm-07-01-0236]). These observations suggest that TLR4 may lead to the activation of various downstream pathways and cause diverse pathophysiological events. These events may induce the development of GDM. GDM is characterized by decreased maternal insulin sensitivity or increased IR. In a previous study, the majority of females with GDM appeared to exhibit β-cell dysfunction under a background of chronic IR ([@b23-etm-07-01-0236]). IR was revealed to be associated with the innate immune response ([@b24-etm-07-01-0236]). A recent study suggested that TLR4 and downstream pathways (MAPK and NF-κB) are important in the pathogenesis of IR ([@b25-etm-07-01-0236]). In addition, abnormal TLR4 expression may induce an inflammatory response in insulin-resistant subjects ([@b26-etm-07-01-0236]). These results indicate that TLR4 expression in females with GDM may result in IR. In patients with GDM, IR may lead to hyperglycemia, fetal macrosomia, a higher likelihood of developing obstetric complications and a higher risk of stillbirth ([@b27-etm-07-01-0236]). The function of the TLR4 gene has been widely studied in immune system cells, including monocytes and macrophages. Monocytes effectively regulate specific and non-specific immunological responses, which avoid damage to the organisms by pathogens. Thus, phagocytosis by monocytes is important in the innate immune response. Monocytes phagocytose a portion of debris left from the digestion of a pathogen and present it as an antigen to the adaptive immune system ([@b28-etm-07-01-0236]). The upregulation of TLR4 expression has been demonstrated to promote the phagocytic capacity of monocytes ([@b29-etm-07-01-0236]). Hence, TLR4 is expressed in monocytes and TLR4 signaling is necessary for phagocytosis by monocytes. TNF-α is a proinflammatory cytokine secreted predominantly by monocytes and macrophages. The results of the present study are consistent with a previous observation that serum TNF-α levels were increased in females with GDM compared with healthy pregnant females ([@b30-etm-07-01-0236]). Therefore, hyperglycemia-induced TNF-α release in patients with GDM may contribute to the underlying pathogenesis of GDM. It has been suggested that the production of TNF-α is induced by the activation of TLR4. The activation of TLR4 has been shown to lead to the activation of NF-kB, which results in the production of TNF-α ([@b31-etm-07-01-0236]). In addition, the activation of monocytes, induced by the TLR4-mediated c-Jun N-terminal kinase signaling, may also cause the secretion of TNF-α ([@b32-etm-07-01-0236]). These results suggest that the activation of TLR4 is associated with the upregulation of TNF-α. Therefore, the elevated expression of TLR4 is able increase serum TNF-α level. An increase in serum TNF-α levels from early to late pregnancy was correlated with a decrease in insulin sensitivity ([@b33-etm-07-01-0236]), which suggested that TNF-α is associated with the development of IR. TNF-α is a significant predictor of IR in patients with GDM through its ability to decrease the tyrosine kinase activity of the insulin receptor ([@b27-etm-07-01-0236]). In conclusion, the novel observations of the present study indicate that TLR4 expression increases in monocytes in GDM and a positive correlation exists between TLR4 mRNA expression in monocytes and serum TNF-α level in females with GDM. These results suggest that a selective interference with TLR4 may present an opportunity for the treatment of IR and GDM. {#f1-etm-07-01-0236} {#f2-etm-07-01-0236} {#f3-etm-07-01-0236} {#f4-etm-07-01-0236} ###### Demographic data in the GDM and normal control groups. Characteristics Normal control (n=32) GDM (n=30) P-value -------------------------------- ----------------------- --------------------------------------------------------- --------- Age, years 29.6±3.3 30.9±4.7 NS Gestational age, days 272.6±4.5 273.9±5.6 NS Fetal weight, g 3,230.0±304.7 3,373.0±418.4 NS Systolic blood pressure, mmHg 114.5±14.7 108.6±8.3 NS Diastolic blood pressure, mmHg 72.9±8.7 72.5±6.8 NS BMI, kg/m^2^ 22.1±1.9 23.3±2.8 NS Fasting glucose, mmol/l 4.1±0.5 5.8±0.8[a](#tfn2-etm-07-01-0236){ref-type="table-fn"} \<0.05 Serum fasting insulin, μU/ml 8.6±4.2 14.5±10.2[a](#tfn2-etm-07-01-0236){ref-type="table-fn"} \<0.05 HOMA-IR 1.58±0.7 4.6.0±2.2[a](#tfn2-etm-07-01-0236){ref-type="table-fn"} \<0.05 Results are presented as the mean ± standard deviation. P\<0.05 was set as statistically significant. GDM, gestational diabetes mellitus; NS, not significant; BMI, body mass index; HOMA-IR, homoeostasis model assessment-insulin resistance.